Patch Analysis v2.1

简介

该模块计算蛋白质表面静电和疏水作用相对富集的区域，用于显示出在蛋白质-蛋白质相互作用中其重要作用的区域，这对于预测基于非共价弱相互作用的可逆的聚集现象尤其有用。尤其是，疏水相互作用长期被认为是大分子间吸引相互作用的主要组成部分。对于抗体，静电相互作用牵涉到了自聚集，而偶极-偶极相互作用被认为是导致β-折叠的纤维化的原因。同时，也可以通过WeView界面对蛋白结构进行Patch分析。

v2.1 更新内容

• 支持设定PH值
• 支持CDR编号，高亮CDR残基，输出CDR patch面积。

参数说明

Structure PDB File

蛋白的结构文件，PDB格式

pH

pH值，用于蛋白质子化判断

Antibody Numbering

抗体编号方法，其中 no_use 不使用编号

Hydrophobic Cutoff

Hydrophobic cutoff是一个以疏水性氨基酸（通常包括Leu，Ile，Val，Phe，Trp和Met）为基础定义的截断值，用于将表面上疏水性氨基酸的数量与表面面积相比较，从而筛选出可能具有重要生物学功能的区域。一般来说，Patch区域中获得的化学性质信息会根据其表面密度和具有高疏水性氨基酸的数量而有所变化。

Positive Cutoff

Positive Cutoff是一个以阳离子氨基酸为基础定义的截断值，用于将表面上阳离子氨基酸的数量与表面面积相比较，从而筛选出可能具有重要生物学功能的区域。一般来说，positive cutoff方式用于筛选出可能参与离子相互作用的蛋白质表面区域。

Negative Cutoff

Negative Cutoff是一个以阴离子氨基酸为基础定义的截断值，用于将表面上阴离子氨基酸的数量与表面面积相比较，从而筛选出可能具有重要生物学功能的区域。一般来说，negative cutoff方式用于筛选出可能参与离子相互作用的蛋白质表面区域。

SASA Cutoff

SASA Cutoff是一个以溶剂可及表面积为基础定义的截断值，低于截断值的patch会被过滤掉。

Distance Cutoff

Distance Cutoff是原子距离截断值，低于截断值的才会认为属于同一聚集块。值越小，聚集块patch越小。

Min Distance Cutoff

Min Distance Cutoff是patch之间的距离截断值，距离小于截断值的归为同一个patch。

Result Type

输出文件格式，csv或者json
通俗地讲，cutoff代表静电势能或疏水势能的强度阈值，单位是kcal/mol，超过阈值才会被计入面积。阈值越小，则patch越多。

Keep Original

不添加缺失原子（包括氢原子）和结构优化。

Neutral N-terminus

使得N-氮端的蛋白残基中性化。

Neutral C-terminus

使得C-氮端的蛋白残基中性化。

结果说明

输出结果包括：

其中patch_list.csv，包含信息如下：

其中patch_list_sum.csv，包含信息如下：

参考文献

• Karshikoff, A.; Non-Covalent Interactions In Proteins; Imperial College Press (2006) pp. 348. DOI: 10.1142/p477
• Heiden, W., Moeckel, G., Brickmann, J.; A New Approach to Analysis and Display of Local Lipophilicity/Hydrophilicity Mapped on Molecular Surfaces; J. Comput. Aided Mol. Des. 7 (1993) 503–514. DOI: 10.1007/BF00124359
• Wildman, S.A., Crippen, G.M.; Prediction of Physiochemical Parameters by Atomic Contributions; J. Chem. Inf. Comput. Sci. 39 (1999) 868–873. DOI: 10.1021/ci990307l

Patch Analysis v2.1

Introduction

Protein Patches calculates both electrostatic (excess charge) and hydrophobic surface patches to show regions of significance with respect to protein-protein interactions. This can be particularly useful in the prediction of reversible aggregation, which typically arises from relatively weak non-covalent interactions. In particular, hydrophobic interactions have long been recognized as major contributors in high affinity interactions between macromolecules. In antibodies, electrostatic interactions have been implicated in forming self-associated aggregates [Karshikoff 2006], while dipole-dipole interactions are believed to be the cause of fibrillogenic association of β-sheets. At the same time, protein structures can also be analyzed for patches through the WeView interface.
Electrostatic patches.
The surface electrostatic field is estimated using an exponentially decaying Debye-Hückel field with a screening length of λD=3.5Å.
The map thus obtained is one mostly of excess charge close to the molecular surface.
Significant patches are established by cutting the surface along isocontour lines of absolute field value equal to 40 kcal/mol/C, keeping regions above. Finally, a default minimal patch area of 40Å2 filters out smaller, presumably less relevant, patches.
Hydrophobicity map.
The hydrophobic potential is calculated from the Wildman and Crippen octanol-water partition coefficients f=log P [Wildman 1999]:

where fi is the coefficient of atom i and g(ri) is a Fermi-type distance-dependent weighting function proposed by Heiden et al. [Heiden 1993]:

with rcut=5Å and α=1.5.
Similarly to electrostatic (excess charge) patches, significant hydrophobic patches are established by cutting the surface along isocontour lines, retaining only the portion above a potential threshold value of 0.09 kcal/mol, and filtering for a minimal patch area of 50Å2.

v2.1 updates

• Supports setting the pH value
• Supports CDR numbering, highlights CDR residues, and outputs the CDR patch area.

Parameters

Structure PDB File

Protein structure file in PDB format.

pH

pH value for protein protonation

Antibody Numbering

Antibody Numbering type, no_use indicates no antibody numbering applied.

Hydrophobic Cutoff

Hydrophobic Cutoff defined based on hydrophobic amino acids (usually including Leu, Ile, Val, Phe, Trp, and Met) is used to compare the number of hydrophobic amino acids on a surface with the surface area, so as to screen out areas that may have important biological functions. Generally speaking, the chemical property information obtained in the Patch region will vary according to its surface density and the number of highly hydrophobic amino acids.

Positive Cutoff

Positive Cutoff is a cut-off value defined based on cationic amino acids, which is used to compare the number of cationic amino acids on the surface with the surface area to screen out areas that may have important biological functions. Generally speaking, the positive cutoff method is used to screen protein surface regions that may be involved in ion interactions.

Negative Cutoff

Negative Cutoff is a value defined based on anionic amino acids, which is used to compare the number of anionic amino acids on a surface with the surface area, to screen out areas that may have important biological functions. Generally speaking, the negative cutoff method is used to screen the surface regions of proteins that may participate in ion interactions.

SASA Cutoff

SASA Cutoff is a cutoff value defined on the basis of polar surface area, which is used to screen the surface regions of proteins that have enough polar surface areas.

Distance Cutoff

Distance Cutoff is a cutoff value defined on the basis of neighbor atoms, which is used to adjust the size of patches. Lower values result in smaller patches.

Min Distance Cutoff

Min Distance Cutoff is the cutoff value for neighbor patch point distance (Å). Patches with distances lower than the cutoff value would be merged.

Result Type

output file format, json or csv

Keep Original

Do no atom addition and optimization.

Results

The output includes:

Details of patch_list.csv:
The file contains the following information:

Details of patch_list_sum.csv:
The file contains the following information:

References

• Karshikoff, A.; Non-Covalent Interactions In Proteins; Imperial College Press (2006) pp. 348. DOI: 10.1142/p477
• Heiden, W., Moeckel, G., Brickmann, J.; A New Approach to Analysis and Display of Local Lipophilicity/Hydrophilicity Mapped on Molecular Surfaces; J. Comput. Aided Mol. Des. 7 (1993) 503–514. DOI: 10.1007/BF00124359
• Wildman, S.A., Crippen, G.M.; Prediction of Physiochemical Parameters by Atomic Contributions; J. Chem. Inf. Comput. Sci. 39 (1999) 868–873. DOI: 10.1021/ci990307l

Immunogenicity Prediction (WeADApt v4.2)

简介

WeADApt (Wecomput ADA prediction) 是唯信开发的基于多模融合深度学习架构的免疫原性预测系统（也被熟知为AlphaMHC）。

该方法采用全新的多模融合深度神经网络架构，整合了近10亿条与免疫原性相关的湿实验数据（包括亲和力数据、呈递数据、NGS数据、质谱数据等）进行训练，有机地将多个与免疫原性相关的模型融合，构成一个高效的免疫反应模拟系统，可准确地模拟蛋白、抗体、多肽、疫苗等生物药的免疫原性，并能鉴别潜在的免疫原性的T细胞表位（引起临床人体免疫应答的肽段），实现了从序列到临床免疫原性风险的端到端的预测，并通过了数百条来自FDA、EMA的临床真实免疫原性数据（包括单/多特异性抗体和重组蛋白等）的验证测试。

在同样的42个抗体分子的临床ADA数据集上，WeADApt(v4)预测的相关性超过了知名的商业软件EpiMatrix（R2=0.49 vs R2=0.42)。

v4.2版本

该版本是截止到2025/04/30的最新主力版本。
相比v4.1进一步提升了预测的特异性，且对不同风险水平的表位的区分度更高，结果对于去免疫原性改造更有指导性。

V4.2版本相对于上个版本v4.1主要有以下改进：

• 算法架构优化
• 测试集规模扩大1倍
• 分类能力F1提升：18%
• 特异性提升：26%
• 敏感性提升：4%

性能

测试数据：

从FDA和EMA的临床试验中收集了200余个已知免疫原性的分子及其ADA的分布，计算模型预测值与真实ADA发生率的相关性，以测试其预测性能。

单抗 mAb

使用唯信收集整理的200多个临床及上市单抗的ADA数据的测试结果如下图所示，预测分数与ADA发生率的Pearson相关性达到R=0.76。

0.2分适合作为单抗的高/低风险的阈值（>20% ADA定义为高风险）。

双抗 BsAB

WeADApt v4被设计为兼容各类的分子形式，不论是对称还是非对称、是否有重复结构域的任意蛋白分子，仅需输入不重复的链即可（重复链全部输入也会自动处理）。

使用唯信收集整理的双抗ADA数据集的测试表现如下图所示，预测分数与ADA发生率的Pearson相关性达到R=0.60。

注意区别于v4.1的一点是，由于分布的变化，该版本以0.4的分数作为分界线时，可以较好的区分高、低风险的双抗分子。

本系统仅从序列水平预测产生的影响，因此尤其适合同类靶点分子的相对比较和筛选。

PPI Binding Energy & Contacts

简介

该模块结合界面接触特征与非相互作用表面（NIS）特征，用于预测蛋白-蛋白结合亲和力，并可输出接触界面的残基信息。模块基于PRODIGY模型，该模型通过线性回归利用界面接触点和NIS的物理化学性质来估算结合亲和力，这些性质已被验证对亲和力具有显著影响。

以下为亲和力的计算公式：

公式中的 ICsxxx/yyy 表示在相互作用的两个蛋白之间检测到的界面接触点数，xxx/yyy表示接触残基的类型（带电/极性/非极性等），例如 ICscharged/apolar 表示带电残基与非极性残基之间的接触点数量。若两个残基之间任意重原子的距离小于5.5 Å，则视为发生了接触。

该模型在81个复合物的数据集上进行了验证，预测亲和力与实验值之间的皮尔逊相关系数为0.73（p < 0.0001），均方根误差（RMSE）为1.89 kcal/mol。

参数说明

Structure

蛋白复合物的结构文件，格式支持 .pdb 或 .cif。支持多个复合物结构打包进行批量预测，压缩格式支持 .tar、.tar.gz 或 .zip。

Group

用于将结构中的多个链组合为组，组内链作为整体，仅计算组与组之间的结合亲和力。组合格式为：组内链名用逗号分隔，组与组之间用空格分隔。
示例：H,L A 表示将链 H 和 L 作为一组，链 A 作为另一组，计算这两组之间的亲和力。

注意：

1. 若不设置该参数，则默认对结构中所有发生接触的链对进行亲和力计算。
2. 在进行抗体-抗原亲和力计算时，应将抗体的重链与轻链合并为一个整体（即为一组），并与抗原链之间计算亲和力。

Contacts

输出链间接触界面的残基对信息。

Output

预测结果文件名，默认值为 prodigy_output.csv。

Output_CRP

接触界面残基对的结果文件名，默认值为 contacts.txt。

结果说明

预测结果文件 prodigy_output.csv 包含以下信息：

可选接触界面结果文件 Contacts.txt，每行记录一个接触残基对，包含残基名称、编号及所在链名。

若启用批量模式，输出将为以下两个打包文件：

• prodigy_output.tar.gz：亲和力预测结果
• Contacts.tar.gz：接触残基对结果

参考文献

• Xue LC, Rodrigues JP, Kastritis PL, Bonvin AM, Vangone A. PRODIGY: a web server for predicting the binding affinity of protein-protein complexes. Bioinformatics. 2016 Dec 1;32(23):3676–3678. DOI: 10.1093/bioinformatics/btw514

PPI Binding Energy & Contacts

Introduction

This module predicts protein-protein binding affinity by combining interfacial contact features with non-interacting surface characteristics. It also provides residue-level information for the contact interface. The module is based on the PRODIGY model, which applies linear regression using properties of interfacial contacts and non-interacting surfaces (NIS), both of which have been shown to influence binding affinity.

The binding affinity is calculated using the following formula:

ICsxxx/yyy represent the number of interfacial contact points found between interacting protein 1 and interacting protein 2, categorized by the polarity/charge of the interacting residues (e.g., ICscharged/apolar indicates the number of interfacial contact points between charged and apolar residues). Two residues are considered to be in contact if any of their heavy atoms are within 5.5 Å of each other.

The model’s prediction accuracy was evaluated using a dataset of 81 complexes. The Pearson correlation coefficient between predicted and experimental binding affinities is 0.73 (p < 0.0001), with a root-mean-square error (RMSE) of 1.89 kcal/mol⁻¹.

Parameters

Structure

The protein complex structure in PDB or CIF format. Multiple complex structures can be packaged together for batch prediction. Supported package formats: .tar, .tar.gz, or .zip.

Group

Allows grouping of multiple chains in the structure. Chains in the same group are treated as a single unit, and binding affinity is only calculated between groups. Use chain IDs to define groups: separate chains in the same group with commas, and separate groups with spaces.
Example: H,L A means chains H and L are treated as one group, and chain A as another group. The binding affinity is then calculated between these two groups.

Note:

1. If this parameter is not specified, binding affinity will be calculated for all contacting chain pairs in the complex.
2. For antibody-antigen binding affinity calculations, the heavy and light chains of the antibody should be grouped together using this parameter to compute affinity with the antigen chain.

Contacts

Outputs residue pairs at the inter-chain contact interface.

Output

Filename for the binding affinity prediction result. Default: prodigy_output.csv

Output_CRP

Filename for the contact interface residue pairs. Default: contacts.txt

Results

The binding affinity prediction result is saved in prodigy_output.csv, which includes the following columns:

The optional contact interface file Contacts.txt lists one contacting residue pair per line, including residue names, numbers, and chain IDs.

In batch mode:

• Binding affinity results are packaged in prodigy_output.tar.gz
• Contact interface results are packaged in Contacts.tar.gz

References

Xue LC, Rodrigues JP, Kastritis PL, Bonvin AM, Vangone A. PRODIGY: a web server for predicting the binding affinity of protein-protein complexes. Bioinformatics. 2016 Dec 1;32(23):3676–3678. DOI: 10.1093/bioinformatics/btw514

MolSeeker: 大模型多智能体驱动 ADMET 预测

简介

基于大模型多智能体的 ADMET 预测，通过构建分布式智能协作系统，精准预测药物在体内的吸收（Absorption）、分布（Distribution）、代谢（Metabolism）、排泄（Excretion）及毒性（Toxicity）性质。深度融合大语言模型（LLM）的深度语义理解能力与多智能体系统（MAS）的分工协作。将 ADMET 预测任务拆解为数据清洗、分子表征、性质建模、决策推理等多个子模块，从分子结构输入到成药性预测的全流程智能化闭环。

参数

1. 物理化学性质（Physicochem）: LogD, LogP, pKa, pKb, Solubility_ Kinetic, Solubility_ FASSIF
2. 吸收（Absorption）: Caco2_A2B (Cls), Caco2_A2B (Reg), PAMPA (Reg)
3. 分布（Distribution）: BBB (Cls), MDCK_Efflux (Reg)
4. 代谢（Metabolism）: HLM (Cls), HLM (Reg), hHep (Cls), hHep (Reg)
5. 毒性（Toxicity）: hERG (Cls), hERG (Reg), AMES (Cls), Hepatotoxicity (Cls)

结果说明

Back Mutation Grouping v2.5

简介

该模块是抗体人源化设计流程中分组模块，根据Mutation Score模块输出的回复突变评分表对回复突变进行分组，并返回突变后的序列。

更新内容：

• 新增参数Combination Max Cutoff，高于改截断值的突变自动进行回复突变，
• 新增参数Combination Site Cutoff，每条链回复突变打分在Combination Min Cutoff与Combination Max Cutoff之间的选择打分前n个位置进行组合突变

参数说明

Grafted Chain

抗体CDR区嫁接后序列文件，FASTA格式，由Grafting模块生成

Raw Chain

抗体序列文件，FASTA格式

Mutation Score

人源化突变评分文件，CSV格式，由Mutation Score模块生成

Output File

指定输出的突变序列文件名称，FASTA格式

Cutoff

打分分组的截断值，逗号分割，例如：2,5,10表示将氨基酸突变评分大于10的为一组，5~10的氨基酸为一组，小于2的氨基酸分为一组。

Output Policy

指定输出的回复突变的文件

Type

普通抗体Antibody或者纳米抗体Nanobody

Combination Min Cutoff

突变组合的截断值，Mutation Score模块中输出的氨基酸回复突变打分大于截断值的氨基酸参与生成突变组合

Combination Max Cutoff

高于截断值的突变自动进行回复突变

Combination Site Cutoff

每条链回复突变打分在Combination Min Cutoff与Combination Max Cutoff之间的选择打分前n个位置进行组合突变

结果说明

根据不同截断值得到突变分组结果文件mutate_policy.json。

根据组合突变截断值得到的突变分组结果文件combination_mutate_policy.json，高通量人源化设计流程。

Back Mutation Grouping v2.5

Introduction

Back mutation grouping in the antibody humanization, which groups the back mutations based on the mutation scoring table generated by the Mutation Score module.

Parameters

Grafted Chain

Sequence file of the antibody CDR region after grafting, in FASTA format, generated by the Grafting module.

Raw Chain

Sequence file of the antibody, in FASTA format.

Mutation Score

Humanization mutation score file, in CSV format, generated by the Mutation Score module.

Output File

Specify the name of the output mutation sequence file, in FASTA format.

Cutoff

Cutoff values for scoring grouping, separated by commas. For example, 2,5,10 indicates grouping amino acid mutations with scores greater than 10 in one group, amino acids with scores between 5 and 10 in another group, and amino acids with scores less than 2 in a separate group.

Output Policy

Specify the file for the output of back mutations.

Type

Antibody or Nanobody

Combination Min Cutoff

Cutoff value for mutation combinations. Amino acids with scores (generated from Mutation Score module) greater than the cutoff value are involved in the mutation combinations.

Combination Max Cutoff

Mutations above the cutoff value automatically undergo reversion mutations.

Combination Site Cutoff

For each chain, select the top n positions with back mutation scores between the Combination Min Cutoff and the Combination Max Cutoff for combination mutations.

Results

The mutation grouping results file mutate_policy.json is generated based on different cutoff values.
The mutation grouping results file combination_mutate_policy.json is generated based on combination cutoff values.

Protease (MMP) Cleavage Prediction

简介

该模块具有两方面的功能：
1，用于预测肽段（长度不超过10个氨基酸）被18种基质金属蛋白酶（MMPs）切割的效率。
2，基于指定的目标切割谱（如：仅被MMP13切割），生成相应的多肽底物。

模块基于CleaveNet模型实现，CleaveNet是一种基于深度学习的蛋白酶底物设计工具，通过整合预测与生成技术，实现了从“虚拟筛选”到“智能设计”的转变。

CleaveNet包含两个核心模块：
预测模块

• 基于Transformer架构，训练于大规模mRNA展示肽段库数据。
• 针对18种基质金属蛋白酶（MMPs），能够预测肽段被特定蛋白酶切割的效率，测试集Pearson相关系数达0.80，优于传统二分类模型。
• 模型不仅复现了已知的酶切基序，还发现了新的底物偏好，例如甲硫氨酸在P4位的作用，拓展了对蛋白酶特异性的理解。

生成模块

• 采用条件化生成技术，用户可通过条件标签指定目标切割谱（如“对MMP13高活性、对其他MMPs低活性”）。
• 通过注意力机制调整生成方向，生成的6-mer肽段新颖度达89%，突破了训练数据的局限性。
• 与传统虚拟筛选相比，生成效率提升约5.5倍，支持复杂设计需求，如“双蛋白酶逻辑门”底物。

这一端到端的设计流程显著提高了底物设计的效率和精准性，为蛋白酶研究提供了一种全新的计算驱动方法。

实验验证
为评估CleaveNet的实际应用能力，研究团队以MMP13（一种与癌症转移、伤口愈合和骨关节炎相关的胶原酶）为目标，设计并合成了95条肽段底物，并通过荧光共振能量转移（FRET）技术验证其切割效率。实验结果表明：

• 切割效率：所有CleaveNet设计的MMP13底物均能被有效切割，其中一条底物（DL73）的切割效率比训练集中最优底物高出39%（p<0.01）。
• 特异性：3条底物（如DL41）实现了对MMP13的绝对特异性，不被其他MMPs切割；5条底物（如DL48）同时表现出高活性和高选择性，填补了传统方法的空白。
• 机制洞察：分析生成序列后，发现了P2位亮氨酸偏好和P3’位天冬氨酸的作用，为MMP13的特异性机制提供了新的研究方向。

这些结果验证了CleaveNet在设计高效且特异性底物方面的能力，同时也展示了其揭示未知底物偏好的潜力。

参数说明

Prediction

Peptide Sequence

必填参数，多肽序列，txt或fasta格式，支持多条（txt格式时，每行放置一条多肽）。注意：多肽长度不能超过10个残基，超过长度的多肽序列会自动被过滤掉。
txt格式实例如下：

LRVFL
FMPLNFTASG
LGPYAMTSRG
AARFKKFATE

Output

可选参数，预测得到的MMPs酶切概率结果文件名称，默认为“pred_cleavage.csv”。

Generation

Number of Peptides

可选参数，指定需要生成的多肽数量，默认为50。

Z-score of MMPs

可选参数，指定多肽生成的酶切条件，CSV文件格式。包含每种MMP酶的酶切概率Z-score值，值越大表示酶切的可能性越高，值可为负，一般阈值为2.5，大于该阈值时，表示极大可能被酶切。模型会根据设置的各种MMPs酶的酶切概率Z-score值进行多肽生成。注意：18种MMPs的Z-score数值都必须设定，不能缺少任意一种。
文件内容实例如下：

MMP1,MMP10,MMP11,MMP12,MMP13,MMP14,MMP15,MMP16,MMP17,MMP19,MMP2,MMP20,MMP24,MMP25,MMP3,MMP7,MMP8,MMP9
2.0,1.2,2.2,2.9,3.3,4.1,2.2,3.1,2.2,1.91,3.6,2.83,2.7,0.2,1.8,2.1,0.61,4.2

以上内容为一组条件，也支持多组条件同时输入，每行一组条件即可。每组条件都会生成指定数量的多肽。多组条件示例如下：

MMP1,MMP10,MMP11,MMP12,MMP13,MMP14,MMP15,MMP16,MMP17,MMP19,MMP2,MMP20,MMP24,MMP25,MMP3,MMP7,MMP8,MMP9
2.0,1.2,2.2,2.9,3.3,4.1,2.2,3.1,2.2,1.91,3.6,2.83,2.7,0.2,1.8,2.1,0.61,4.2
3.33,2.5,3.6,2.7,2.9,5.2,3.4,4.2,2.7,2.0,3.5,2.6,4.0,3.1,3.5,0.61,2.1,2.9

Temperature

可选参数，指定生成的温度条件，用于控制生成多肽序列的多样性，默认为1.0，越大表示多样性越高。如果希望多样性低一些，推荐0.7，如果希望多样性再高一些，推荐1.2~1.5。

Output

可选参数，指定序列输出文件名称，fasta或txt格式，默认为“gen_seqs.fasta”。

结果说明

Prediction

预测得到的MMPs酶切概率结果文件，默认为pred_cleavage.csv。包含如下内容：

Generation

生成的序列文件，默认为“gen_seqs.fasta”。

参考文献

• Deep learning guided design of protease substrates. Carmen Martin-Alonso, Sarah Alamdari, Tahoura S. Samad, Kevin K. Yang, Sangeeta N. Bhatia, Ava P. Amini bioRxiv 2025.02.27.640681; DOI: 10.1101/2025.02.27.640681

Protease (MMP) Cleavage Prediction

Introduction

This module has two functions:
Predicting the cleavage efficiency of peptides (≤10 amino acids) by 18 matrix metalloproteinases (MMPs).
Generating corresponding peptide substrates based on a specified cleavage profile (e.g., only cleaved by MMP13).
Built on the CleaveNet model, a deep - learning - based protease substrate design tool, it integrates prediction and generation, shifting from “virtual screening” to “intelligent design”.
CleaveNet has two core modules:
Prediction Module
Trained on a large - scale mRNA - displayed peptide library using a Transformer architecture.
Predicts peptide cleavage efficiency by 18 MMPs, with a test - set Pearson correlation of 0.80, outperforming traditional binary - classification models.
Reproduces known cleavage motifs and reveals new substrate preferences (e.g., methionine at P4), enhancing understanding of protease specificity.
Generation Module
Uses conditional generation. Users can set target cleavage profiles (e.g., “high MMP13 activity, low other MMP activities”) via conditional tags.
Adjusts generation direction with attention mechanisms. Generated 6 - mer peptides have 89% novelty, surpassing training data limits.
Is about 5.5 times more efficient than traditional virtual screening, supporting complex designs like “dual - protease logic gate” substrates.
This end - to - end design process improves substrate design efficiency and accuracy, offering a new computation - driven method for protease research.
Experimental Validation
To assess CleaveNet’s practicality, the team targeted MMP13 (a collagenase linked to cancer metastasis, wound healing, and osteoarthritis). They designed and synthesized 95 peptide substrates, validating cleavage efficiency via fluorescence resonance energy transfer (FRET). Results showed:
All CleaveNet - designed MMP13 substrates were efficiently cleaved. One (DL73) had 39% higher efficiency than the best training - set substrate (p<0.01).
Three substrates (e.g., DL41) were absolutely specific to MMP13, and five (e.g., DL48) had both high activity and selectivity, addressing traditional method gaps.
Analysis of generated sequences revealed leucine preference at P2 and aspartic acid’s role at P3’, offering new insights into MMP13’s specificity mechanism.
These results confirm CleaveNet’s ability to design efficient, specific substrates and its potential to uncover unknown substrate preferences.

Parameters

Prediction

Peptide Sequence

Required parameter, peptide sequence, in txt or fasta format, supporting multiple sequences (when in txt format, place each peptide on a separate line). Note: The length of the peptide cannot exceed 10 residues.
An example in txt format is as follows：

LRVFL
FMPLNFTASG
LGPYAMTSRG
AARFKKFATE

Output

Optional parameter, the file name of the predicted MMPs cleavage probability results, default is “pred_cleavage.csv”。

Generation

Number of Peptides

Optional parameter, specify the number of peptides to be generated, default is 50.

Z-score of MMPs

Optional parameter, specify the cleavage conditions for peptide generation in CSV file format. It includes the Z-score values of cleavage probabilities for each type of MMP enzyme. A higher value indicates a higher likelihood of cleavage. The value can be negative. The general threshold is 2.5. When the value is above this threshold, it indicates a very high probability of being cleaved. The model will generate peptides based on the set Z-score values of cleavage probabilities for various MMPs enzymes. Note: The Z-score values for all 18 types of MMPs must be set, and none can be missing.

An example of the file content is as follows：

MMP1,MMP10,MMP11,MMP12,MMP13,MMP14,MMP15,MMP16,MMP17,MMP19,MMP2,MMP20,MMP24,MMP25,MMP3,MMP7,MMP8,MMP9
2.0,1.2,2.2,2.9,3.3,4.1,2.2,3.1,2.2,1.91,3.6,2.83,2.7,0.2,1.8,2.1,0.61,4.2

The above content is a set of conditions, and multiple sets of conditions can also be input simultaneously. Just place each set of conditions on a separate line. Peptides of the specified quantity will be generated for each set of conditions. An example of multiple sets of conditions is as follows：

MMP1,MMP10,MMP11,MMP12,MMP13,MMP14,MMP15,MMP16,MMP17,MMP19,MMP2,MMP20,MMP24,MMP25,MMP3,MMP7,MMP8,MMP9
2.0,1.2,2.2,2.9,3.3,4.1,2.2,3.1,2.2,1.91,3.6,2.83,2.7,0.2,1.8,2.1,0.61,4.2
3.33,2.5,3.6,2.7,2.9,5.2,3.4,4.2,2.7,2.0,3.5,2.6,4.0,3.1,3.5,0.61,2.1,2.9

Temperature

Optional parameter, specify the temperature condition for controlling the diversity of the generated peptide sequences. The default value is 1.0. A higher value indicates higher diversity. If you want lower diversity, it is recommended to use 0.7. If you want higher diversity, it is recommended to use a value between 1.2 and 1.5.

Output

Optional parameter, specify the output file name for the sequences in fasta or txt format. The default is “gen_seqs.fasta”.

Results

Prediction

The predicted MMPs cleavage probability results file, default is pred_cleavage.csv. It contains the following content:

Generation

The generated sequence file, default is “gen_seqs.fasta”.

References

• Deep learning guided design of protease substrates. Carmen Martin-Alonso, Sarah Alamdari, Tahoura S. Samad, Kevin K. Yang, Sangeeta N. Bhatia, Ava P. Amini bioRxiv 2025.02.27.640681; DOI: 10.1101/2025.02.27.640681

Computing Electrostatic Surfaces

简介

该模块用于分析和可视化蛋白质表面的静电特性，这对分子识别、蛋白质溶解性、粘度和抗体的可开发性等过程至关重要。它主要通过定义“Patch”来识别和量化蛋白质表面的静电势，这些Patch是具有统一正或负电势值的连接区域。
主要功能和特点：

• 静电势计算：
  该工具使用APBS（自适应泊松-玻尔兹曼求解器）来计算静电势。此外，它还可以接受用户提供的势图或基于疏水性尺度的映射。
• 分子表面生成：
  工具生成分子表面，并将计算的静电势映射到该表面。然后，可以通过颜色编码来可视化该表面，以指示正负区域。
• Patch识别：
  识别和量化蛋白质表面上不同的正电和负电静电Patch，这对于理解蛋白质-蛋白质相互作用和抗体开发非常重要。

参数说明

Structure PDB

蛋白结构文件，PDB格式。

Surface Type

分子表面的类型：sas或者ses。以下是两个选项的解释：

• 溶剂可及表面（SAS，Solvent-Accessible Surface）：SAS 是溶剂探针（通常是水分子）在分子表面滚动时，其中心轨迹形成的表面。
• 溶剂排除表面（SES，Solvent-Excluded Surface）：SES 是溶剂探针围绕分子滚动时，其最靠近分子的外部轮廓所形成的表面。

Probe Radius

探针半径，单位为纳米（默认：0.14）。

Size Cutoff

Patch面积（area ）阈值，单位为Ų。如果 Size Cutoff = 0，则不过滤任何 patch，即所有 patch 都会被保留。

pH Value

pH 值。

Output Patch

输出Patch文件名称

结果说明

输出结果包括：

其中patches.csv包括信息如下：

参考文献

• Hoerschinger VJ, Waibl F, Pomarici ND, Loeffler JR, Deane CM, Georges G, Kettenberger H, Fernández-Quintero ML, Liedl KR. PEP-Patch: Electrostatics in Protein-Protein Recognition, Specificity, and Antibody Developability. J Chem Inf Model. 2023 Nov 27;63(22):6964-6971. DOI: 10.1021/acs.jcim.3c01490

Computing Electrostatic Surfaces

Introduction

This module is designed for analyzing and visualizing the electrostatic properties of protein surfaces, which are critical for processes such as molecular recognition, protein solubility, viscosity, and antibody developability. It primarily identifies and quantifies the electrostatic potential on protein surfaces by defining “patches,” which are connected regions with uniform positive or negative potential values.
Key Features:

• Electrostatic Potential Calculation:
  This tool uses APBS (Adaptive Poisson-Boltzmann Solver) to compute electrostatic potentials. Additionally, it can accept user-provided potential maps or mappings based on hydrophobicity scales.

• Molecular Surface Generation:
  The tool generates molecular surfaces and maps the calculated electrostatic potentials onto these surfaces. The surface can then be visualized using color coding to indicate positive and negative regions.

• Patch Identification:
  It identifies and quantifies different positive and negative electrostatic patches on the protein surface, which are crucial for understanding protein-protein interactions and antibody development.

Parameter

Structure PDB

The protein structure file in PDB format.

Surface Type

The type of molecular surface: SAS or SES. Below are explanations for the two options:

• Solvent-Accessible Surface (SAS): SAS represents the surface formed by the center trajectory of a solvent probe (usually a water molecule) rolling over the molecular surface.
• Solvent-Excluded Surface (SES): SES represents the outer contour closest to the molecule formed when the solvent probe rolls around the molecule.

Probe Radius

The radius of the probe, measured in nanometers (default: 0.14).

Size Cutoff

Patch area threshold (area), measured in Ų. If Size Cutoff = 0, no patch will be filtered, meaning all patches will be retained.

pH Value

The pH value.

Output Patch

The name of the output file for identified patches.

Result

The output includes the following files:

The patches.csv file includes the following information:

References

• Hoerschinger VJ, Waibl F, Pomarici ND, Loeffler JR, Deane CM, Georges G, Kettenberger H, Fernández-Quintero ML, Liedl KR. PEP-Patch: Electrostatics in Protein-Protein Recognition, Specificity, and Antibody Developability. J Chem Inf Model. 2023 Nov 27;63(22):6964-6971. DOI: 10.1021/acs.jcim.3c01490

Patch Analysis v2

简介

该模块计算蛋白质表面静电和疏水作用相对富集的区域，用于显示出在蛋白质-蛋白质相互作用中其重要作用的区域，这对于预测基于非共价弱相互作用的可逆的聚集现象尤其有用。尤其是，疏水相互作用长期被认为是大分子间吸引相互作用的主要组成部分。对于抗体，静电相互作用牵涉到了自聚集，而偶极-偶极相互作用被认为是导致β-折叠的纤维化的原因。同时，也可以通过WeView界面对蛋白结构进行Patch分析。

V2 更新内容

• 优化原子参数，提高计算准确性。

参数说明

Structure PDB File

蛋白的结构文件，PDB格式

Hydrophobic Cutoff

Hydrophobic cutoff是一个以疏水性氨基酸（通常包括Leu，Ile，Val，Phe，Trp和Met）为基础定义的截断值，用于将表面上疏水性氨基酸的数量与表面面积相比较，从而筛选出可能具有重要生物学功能的区域。一般来说，Patch区域中获得的化学性质信息会根据其表面密度和具有高疏水性氨基酸的数量而有所变化。

Positive Cutoff

Positive Cutoff是一个以阳离子氨基酸为基础定义的截断值，用于将表面上阳离子氨基酸的数量与表面面积相比较，从而筛选出可能具有重要生物学功能的区域。一般来说，positive cutoff方式用于筛选出可能参与离子相互作用的蛋白质表面区域。

Negative Cutoff

Negative Cutoff是一个以阴离子氨基酸为基础定义的截断值，用于将表面上阴离子氨基酸的数量与表面面积相比较，从而筛选出可能具有重要生物学功能的区域。一般来说，negative cutoff方式用于筛选出可能参与离子相互作用的蛋白质表面区域。

SASA Cutoff

SASA Cutoff是一个以溶剂可及表面积为基础定义的截断值，低于截断值的patch会被过滤掉。

Distance Cutoff

Distance Cutoff是原子距离截断值，低于截断值的才会认为属于同一聚集块。值越小，聚集块patch越小。

Min Distance Cutoff

Min Distance Cutoff是patch之间的距离截断值，距离小于截断值的归为同一个patch。

Result Type

输出文件格式，csv或者json
通俗地讲，cutoff代表静电势能或疏水势能的强度阈值，单位是kcal/mol，超过阈值才会被计入面积。阈值越小，则patch越多。

Keep Original

不添加缺失原子（包括氢原子）和结构优化。

Neutral N-terminus

使得N-氮端的蛋白残基中性化。

Neutral C-terminus

使得C-氮端的蛋白残基中性化。

结果说明

输出结果包括：

其中patch_list.csv，包含信息如下：

其中patch_list_sum.csv，包含信息如下：

参考文献

• Karshikoff, A.; Non-Covalent Interactions In Proteins; Imperial College Press (2006) pp. 348. DOI: 10.1142/p477
• Heiden, W., Moeckel, G., Brickmann, J.; A New Approach to Analysis and Display of Local Lipophilicity/Hydrophilicity Mapped on Molecular Surfaces; J. Comput. Aided Mol. Des. 7 (1993) 503–514. DOI: 10.1007/BF00124359
• Wildman, S.A., Crippen, G.M.; Prediction of Physiochemical Parameters by Atomic Contributions; J. Chem. Inf. Comput. Sci. 39 (1999) 868–873. DOI: 10.1021/ci990307l

Patch Analysis v2

Introduction

Protein Patches calculates both electrostatic (excess charge) and hydrophobic surface patches to show regions of significance with respect to protein-protein interactions. This can be particularly useful in the prediction of reversible aggregation, which typically arises from relatively weak non-covalent interactions. In particular, hydrophobic interactions have long been recognized as major contributors in high affinity interactions between macromolecules. In antibodies, electrostatic interactions have been implicated in forming self-associated aggregates [Karshikoff 2006], while dipole-dipole interactions are believed to be the cause of fibrillogenic association of β-sheets. At the same time, protein structures can also be analyzed for patches through the WeView interface.
Electrostatic patches.
The surface electrostatic field is estimated using an exponentially decaying Debye-Hückel field with a screening length of λD=3.5Å.
The map thus obtained is one mostly of excess charge close to the molecular surface.
Significant patches are established by cutting the surface along isocontour lines of absolute field value equal to 40 kcal/mol/C, keeping regions above. Finally, a default minimal patch area of 40Å2 filters out smaller, presumably less relevant, patches.
Hydrophobicity map.
The hydrophobic potential is calculated from the Wildman and Crippen octanol-water partition coefficients f=log P [Wildman 1999]:

where fi is the coefficient of atom i and g(ri) is a Fermi-type distance-dependent weighting function proposed by Heiden et al. [Heiden 1993]:

with rcut=5Å and α=1.5.
Similarly to electrostatic (excess charge) patches, significant hydrophobic patches are established by cutting the surface along isocontour lines, retaining only the portion above a potential threshold value of 0.09 kcal/mol, and filtering for a minimal patch area of 50Å2.

V2 updates

• Optimized atoms parameters and improved the accuracy.

Parameters

Structure PDB File

Protein structure file in PDB format.

Hydrophobic Cutoff

Hydrophobic Cutoff defined based on hydrophobic amino acids (usually including Leu, Ile, Val, Phe, Trp, and Met) is used to compare the number of hydrophobic amino acids on a surface with the surface area, so as to screen out areas that may have important biological functions. Generally speaking, the chemical property information obtained in the Patch region will vary according to its surface density and the number of highly hydrophobic amino acids.

Positive Cutoff

Positive Cutoff is a cut-off value defined based on cationic amino acids, which is used to compare the number of cationic amino acids on the surface with the surface area to screen out areas that may have important biological functions. Generally speaking, the positive cutoff method is used to screen protein surface regions that may be involved in ion interactions.

Negative Cutoff

Negative Cutoff is a value defined based on anionic amino acids, which is used to compare the number of anionic amino acids on a surface with the surface area, to screen out areas that may have important biological functions. Generally speaking, the negative cutoff method is used to screen the surface regions of proteins that may participate in ion interactions.

SASA Cutoff

SASA Cutoff is a cutoff value defined on the basis of polar surface area, which is used to screen the surface regions of proteins that have enough polar surface areas.

Distance Cutoff

Distance Cutoff is a cutoff value defined on the basis of neighbor atoms, which is used to adjust the size of patches. Lower values result in smaller patches.

Min Distance Cutoff

Min Distance Cutoff is the cutoff value for neighbor patch point distance (Å). Patches with distances lower than the cutoff value would be merged.

Result Type

output file format, json or csv

Keep Original

Do no atom addition and optimization.

Results

The output includes:

Details of patch_list.csv:
The file contains the following information:

Details of patch_list_sum.csv:
The file contains the following information:

References

• Karshikoff, A.; Non-Covalent Interactions In Proteins; Imperial College Press (2006) pp. 348. DOI: 10.1142/p477
• Heiden, W., Moeckel, G., Brickmann, J.; A New Approach to Analysis and Display of Local Lipophilicity/Hydrophilicity Mapped on Molecular Surfaces; J. Comput. Aided Mol. Des. 7 (1993) 503–514. DOI: 10.1007/BF00124359
• Wildman, S.A., Crippen, G.M.; Prediction of Physiochemical Parameters by Atomic Contributions; J. Chem. Inf. Comput. Sci. 39 (1999) 868–873. DOI: 10.1021/ci990307l

Molecular Docking (AutoDock-GPU v2)

简介

该模块是一种用于分子对接模拟工具，主要用于预测分子之间的结合模式和相互作用，得到分子对接的能量和结合亲和力等信息。它还可以计算和比较多个分子之间的结合能力，用于药物分子的筛选、设计和优化。AutoDock-GPU是AutoDock4.2.6的OpenCL和Cuda加速版本，其利用可并行的LGA，从而通过在多个计算单元上并行处理配体-受体结合构象。

参数说明

支持自行上传小分子文件（Private Ligand Library）或者选择公共分子虚筛库（Public Ligand Library）。

Private Ligand Library (Comp＜100)

Binding Mode

对接模式为刚性配体对接（rigid）或者柔性配体对接（flex），
刚性配体对接：配体自身保持刚性，经平移、旋转，在口袋内寻找合适的结合取向。
柔性配体对接：配体在固定某些非关键部位的键长、键角的前提下允许其构象发生一定程度的变化。

Receptor

受体结构文件，PDB格式

Private Ligand

配体结构文件，支持SDF、PDB、MOL格式。只会计算前100的分子。

Box Center

对接口袋中心的三维坐标（XYZ），空格分割。例如：0 0 0。

Box Size

对接口袋长方体盒子的大小，必须是整数，空格分割，例如 24 22 32。

Number of Poses

每个分子保留的最大结合模式数量

TopN

虚拟筛选中保留打分排名前n个分子。

Unbound Model

未结合状态模型选择：

• bound：适用于已知结合模式的精确优化，假设配体初始构象接近结合状态。
• extended：适用于探索结合模式的中等灵活配体，从自由分子状态开始搜索。
• compact：适用于高度灵活或折叠配体，提供最大范围的结合模式探索，但计算成本最高所需时间最长。

Keep Heterogens

保留非标准氨基酸，格式为[链名]:[残基名称]-[残基编号]，如A:UNL-311。不能包含特殊离子的小分子结构。

Private Ligand Library (Comp＜10,000)

Private Ligand

配体结构文件，支持SDF、PDB、MOL格式。只会计算前10,000的分子。
其余参数与**Private Ligand Library (Comp＜100)**模式一致。

Public Ligand Library模式

Public Ligand

提供17个公共分子虚筛库用于分子对接，包括：

1. Alinda：~77万库存分子，源自中国香港的Alinda Chemical公司，致力于分子砌块和新颖筛选化合物的研发供应。
2. Analyticon：~4万库存分子，源自德国的天然产物品牌，专注天然产物提取及类似物合成工作，产品质量稳定。
3. Asinex：~57万库存分子，源自美国的品牌，多年来致力于类先导化合物及分子砌块的研发供应，价格较贵。
4. Bionet：~30万库存分子，源自英国的品牌，拥有多年的有机合成经验。
5. Chembridge：~137万库存分子，源自美国的化合物品牌，总部位于圣地亚哥，拥有多样性库、大环库等多种热门化合物库。
6. Chemdiv：~156万库存分子，全球最大的化合物品牌之一，拥有5000多种化合物骨架结构和100多种化合物库，性价比高。
7. Enamine：~407万库存分子，源自乌克兰的化合物品牌，具有较强的化合物研发能力，有高性价比化合物和高价值化合物两类产品。
8. Eximed：~6万库存分子，源自乌克兰的化合物品牌，近20年来致力于提供高通量筛选化合物及相关服务。
9. HTS：~6万库存分子，源自德国的HTS Biochemie Innovationen化合物品牌，致力于为制药、农业和生物技术公司开发独特的化合物。
10. IBS：~55万库存分子，源自俄罗斯的InterBioScreen化合物品牌，拥有多种天然产物及衍生物。
11. Life_Chemicals：~54万库存分子，源自加拿大的化合物品牌，拥有2900多种化合物骨架结构，化合物规格较齐全且有对应价格。
12. Maybridge：~5万库存分子，源自英国的化合物品牌，Thermofisher旗下，产品数量少而专，每种产品均具有较大库存。
13. Otava：~29万库存分子，源自加拿大的化合物品牌，专门从事特色化合物，生物化学药品和生物分析试剂的开发和生成。
14. Princeton：~153万库存分子，源自美国的化合物品牌，20多年来设计独特的小分子化合物用于药物开发。
15. Specs：~20万库存分子，源自荷兰的化合物品牌，价格优势明显。
16. UORSY：~68万库存分子，源自乌克兰的化合物品牌，产品主要用于高通量筛选和药物发现，价格与Enamine接近。
17. Vitas-m：~140万库存分子，源自美国的化合物品牌，在香港拥有发货中心，到货速度快，价格适中。

其他参数与Private Ligand Library模式相同，公共库只允许刚性对接。

结果说明

输出结果包括：

其中TopNScores.csv包括信息如下：

其中TopNScores_Molecule_Info.csv包含TopNScores.csv的信息和SDF格式小分子原有信息。

参考文献

Santos-Martins D, Solis-Vasquez L, Tillack AF, et al., Accelerating AutoDock4 with GPUs and Gradient-Based Local Search. J Chem Theory Comput. 2021 Feb 9;17(2):1060-1073.

Molecular Docking (AutoDock-GPU v2)

Introduction

This module is a molecular docking simulation tool primarily used for predicting molecular binding modes and interactions. It provides information on docking energy and binding affinity. Additionally, it allows for the calculation and comparison of binding abilities among multiple molecules, facilitating the screening, design, and optimization of drug molecules.

AutoDock-GPU is the OpenCL and CUDA-accelerated version of AutoDock 4.2.6, utilizing parallelizable LGA (Lamarckian Genetic Algorithm) to process ligand-receptor binding conformations in parallel across multiple computing units.

Parameter

It supports private ligand file uploads (Private Ligand Library) or the selection of public virtual screening libraries (Public Ligand Library).

Private Ligand Library (Comp <100)

Binding Mode

Docking mode can be either rigid docking or flexible docking:

• Rigid docking: The ligand remains rigid, undergoing translation and rotation within the binding pocket to find an optimal binding orientation.
• Flexible docking: The ligand is allowed to undergo conformational changes while keeping certain non-critical bond lengths and angles fixed.

Receptor

• Format: PDB

Private Ligand

• Formats Supported: SDF, PDB, MOL
• Limitation: Only the top 100 molecules will be processed.

Box Center

• The XYZ coordinates of the docking pocket center, separated by spaces.
  • Example: 0 0 0

Box Size

• The size of the docking pocket, represented as a rectangular box with integer values separated by spaces.
  • Example: 24 22 32

Number of Poses

• The maximum number of binding modes retained for each molecule.

TopN

• The number of top-scoring molecules retained from the virtual screening.

Unbound Model

Defines the unbound state model:

• bound: Assumes the initial ligand conformation is close to the bound state, suitable for precise optimization with known binding modes.
• extended: Begins from a free molecular state, suitable for moderately flexible ligands to explore binding modes.
• compact: Best for highly flexible or folded ligands, allowing the broadest exploration of binding modes but with higher computational costs and longer runtime.

Keep Heterogens

• Retains non-standard amino acids.
• Format: [Chain Name]:[Residue Name]-[Residue Number], e.g., A:UNL-311.
• Restriction: Cannot include small molecular structures containing special ions.

Private Ligand Library (Comp <10,000)

Private Ligand

• Formats Supported: SDF, PDB, MOL
• Limitation: Only the top 10,000 molecules will be processed.

🔹 Other parameters are identical to those in Private Ligand Library (Comp <100) mode.

Public Ligand Library

Public Ligand

Provides 17 public virtual screening libraries for molecular docking, including:

1. Alinda (~770,000 molecules) - Hong Kong-based company specializing in molecular building blocks and novel screening compounds.
2. Analyticon (~40,000 molecules) - German brand specializing in natural product extraction and analog synthesis.
3. Asinex (~570,000 molecules) - US-based company focused on lead-like compounds and molecular building blocks, but relatively expensive.
4. Bionet (~300,000 molecules) - UK-based company with extensive organic synthesis expertise.
5. Chembridge (~1.37 million molecules) - US-based company with a diverse compound collection, including macrocycles.
6. Chemdiv (~1.56 million molecules) - One of the largest compound brands globally, offering over 5,000 scaffolds and 100+ libraries.
7. Enamine (~4.07 million molecules) - Ukraine-based company known for cost-effective and high-value compounds.
8. Eximed (~60,000 molecules) - Ukraine-based company providing high-throughput screening compounds.
9. HTS (~60,000 molecules) - German company developing unique compounds for pharmaceutical, agricultural, and biotech applications.
10. IBS (~550,000 molecules) - Russian company specializing in natural products and derivatives.
11. Life Chemicals (~540,000 molecules) - Canadian company with diverse scaffolds and transparent pricing.
12. Maybridge (~50,000 molecules) - UK-based ThermoFisher subsidiary focusing on high-quality compounds.
13. Otava (~290,000 molecules) - Canadian company specializing in biochemical drugs and reagents.
14. Princeton (~1.53 million molecules) - US-based company with 20+ years of expertise in small molecule drug discovery.
15. Specs (~200,000 molecules) - Dutch company known for its cost-effective compounds.
16. UORSY (~680,000 molecules) - Ukraine-based company with a price range similar to Enamine.
17. Vitas-m (~1.4 million molecules) - US-based company with a Hong Kong shipping center, offering fast delivery and moderate pricing.

🔹 Other parameters are identical to Private Ligand Library, but only rigid docking is allowed.

Result

The docking results include:

📌 TopNScores.csv Fields:

The TopNScores_Molecule_Info.csv file contains the information from TopNScores.csv along with the original data of small molecules in SDF format.

References

Santos-Martins D, Solis-Vasquez L, Tillack AF, et al., Accelerating AutoDock4 with GPUs and Gradient-Based Local Search. J Chem Theory Comput. 2021; 17(2): 1060-1073.

Antibody Design (RFAntibody)

简介

RFantibody 是目前最先进的抗体从头生成方法，通过人工智能（AI）技术实现抗体的从头（de novo）设计，包括单域抗体（VHH）和单链抗体片段（scFv），能够精准结合用户指定的目标表位，并已通过湿实验验证其功能。

RFantibody基于蛋白质结构预测模型RoseTTAFold2（RF2）和蛋白质生成模型RFdiffusion，通过对原始RFdiffusion进行微调，开发出专用于抗体设计的RFdiffusion版本。其核心原理如下：

• 抗体结构特性利用：RFdiffusion在蛋白质数据库（PDB）中的抗体结构数据（约8100个抗体结构）上进行微调，重点训练抗体特有的互补决定区（CDR）loop 区域，同时保持框架结构接近用户指定的优化框架。训练过程中，通过逐步添加噪声（Cα 坐标加入三维高斯噪声，残基方向加入 SO(3) 布朗运动），网络学习预测去噪后的结构。

• 表位靶向设计：通过引入"热点"（Hotspot）特征，用户可指定目标蛋白上的表位，网络通过CDR loop与表位的相互作用进行设计。训练时，抗体框架以全局坐标无关的方式提供（通过二维距离和二面角矩阵表示），允许网络自由设计CDR Loop构象及抗体与目标的刚体定位。

• 序列设计与验证：结构设计后，使用ProteinMPNN生成CDR loop区序列，优化与目标表位的相互作用。设计的抗体通过微调后的RF2进行结构预测和自一致性验证，筛选高潜力候选分子。

• 支持 VHH 和 scFv 设计：RFdiffusion 不仅支持单域抗体（VHH）的设计，还可应用于单链抗体片段（scFv）的设计。scFv 设计涉及重链和轻链的所有六个 CDR 的设计。

通过上述方法，RFantibody能够生成多样化的抗体结构，显著区别于训练数据集，同时实现与目标表位的高度形状互补性和功能性结合。

RFantibody项目针对多个疾病相关表位进行了VHH和scFv设计，并通过表面等离子共振（SPR）、冷冻电镜（cryo-EM）、中和实验等手段验证了设计的有效性。以下是具体实验结果及分析：

1, 单域抗体（VHH）设计与实验验证

实验选择了多个疾病相关靶点，包括流感血凝素（HA）、呼吸道合胞病毒（RSV）位点I和III、SARS-CoV-2受体结合域（RBD）、艰难梭菌毒素B（TcdB）和IL-7Rα。以下为关键结果：

• 结合亲和力（KD）：

  • 流感HA：针对HA茎部表位的VHH设计中，最高亲和力结合体（VHH_flu_01）KD值为78 nM，其他结合体KD值分别为546 nM、698 nM和790 nM。实验使用昆虫细胞表达的单体HA（模拟去糖基化状态）以匹配计算设计条件。
  • SARS-CoV-2 RBD：最佳VHH结合体KD值为5.5 μM，通过竞争实验（与已知结合体AHB2竞争）确认结合至目标表位。
  • TcdB：针对Frizzled-7表位的VHH最佳结合体KD值为260 nM，结合特异性高，未观察到与同源性70%的Clostridium sordellii毒素L（TcsL）的交叉反应。

• 中和活性（EC50）：

  • TcdB：针对TcdB的VHH在中和实验中表现出功能性，在CSPG4敲除细胞中中和TcdB毒性，EC50值为460 nM，表明其潜在的治疗应用价值。

• 结构准确性（cryo-EM）：

  • 流感HA：通过cryo-EM解析了VHH_flu_01与原生糖基化HA三聚体的复合物结构（分辨率3.0 Å）。66%的HA颗粒结合了至多两个VHH，部分未结合可能由于N296糖基的遮挡。实验结构与设计模型高度一致，整体RMSD为1.45 Å，CDR3 RMSD为0.8 Å，关键CDR3残基（V100、V101、S103、F108）与HA茎部表位的相互作用如设计预期。
  • TcdB：针对TcdB的原始设计（VHH_TcdB_H2）和亲和力成熟后版本（VHH_TcdB_H2_ortho）进行了cryo-EM分析。原始设计确认结合至Frizzled-7表位，成熟后版本（分辨率5.7 Å）显示更高的结合比例，结构符合设计预期。
  • SARS-CoV-2 RBD：亲和力成熟后的VHH（VHH_RBD_D4_ortho19）结合至RBD"上"构象表位（分辨率3.9 Å）。

• 亲和力成熟（OrthoRep）：

  • 使用OrthoRep系统对TcdB、流感HA和SARS-CoV-2 RBD的VHH进行亲和力成熟，结合亲和力提升约两个数量级，同时保留了原始表位特异性。

2, 单链抗体片段（scFv）设计与实验验证

进一步扩展至scFv设计，涉及重链和轻链六个CDR的设计，采用结构导向的组合库策略以提高成功率。实验靶点包括TcdB的Frizzled-7表位和Phox2b/HLA-C*07:02复合物。

• 结合亲和力（KD）：

  • TcdB：通过组合库筛选出针对Frizzled-7表位的scFv，最高亲和力结合体（scFv6）KD值为72 nM，其他结合体的KD值未详细列出。竞争实验（与Frizzled-7竞争）确认结合至目标表位，未与无关受体CSPG4竞争。
  • Phox2b/HLA-C*07:02：针对神经母细胞瘤相关表位的scFv结合体KD值为400 nM（SPR）和1 μM（ITC），特异性结合至Phox2b肽，未结合R6A突变肽。尝试将其转化为CAR-T细胞未显示细胞毒性，可能因亲和力不足或抗原密度低。

• 结构准确性（cryo-EM）：

  • TcdB：两个scFv（scFv5和scFv6）结合至Frizzled-7表位的cryo-EM结构验证了设计准确性。scFv6的分辨率为3.6 Å，整体RMSD为0.9 Å，六个CDR的骨架RMSD分别为CDRH1=0.4 Å、CDRH2=0.3 Å、CDRH3=0.7 Å、CDRL1=0.2 Å、CDRL2=1.1 Å、CDRL3=0.2 Å，侧链构象及相互作用符合设计。scFv5（分辨率6.1 Å）以不同接近角度结合，实验结构与设计模型一致。

3, 实验结果分析

• 结构多样性：设计的VHH和scFv的CDR区与自然抗体显著不同，且针对TcdB的Frizzled-7表位无已知抗体，表明RFdiffusion实现了真正的从头设计。
• 功能性与应用潜力：TcdB VHH的中和活性（EC50=460 nM）和scFv的高亲和力（KD=72 nM）显示出治疗潜力，但Phox2b scFv的CAR-T应用失败表明需进一步优化亲和力或抗原表达。

4, 总结

RFantibody通过微调RFdiffusion网络，实现了从头设计VHH和scFv的目标，能够靶向多种疾病相关表位。实验结果显示设计的抗体具有较高的结构准确性（RMSD低至0.9 Å）和功能性（KD低至72 nM，EC50为460 nM）。cryo-EM验证了设计的原子级精度，而亲和力成熟和组合库策略进一步提升了成功率。

参数说明

Complex

用于抗体设计的抗体-抗原复合物结构，PDB格式。如果指定了该参数，后续的Antigen，Antibody参数不用再指定。如果不指定该参数，则需要分别输入Antigen与Antibody的结构。
注意：
1，当前只支持单链抗原，如存在多链时会提示错误，可以使用蛋白编辑工具去掉抗原多余的链，保留单链抗原即可。
2，抗体编号方式只支持Chothia，会自动转成Chothia编号。

Antigen

指定抗原的结构文件，PDB格式。
说明：抗原结构通常需要截短以减少计算开销，建议保留表位周围约 10Å 的区域即可。

Antibody

指定抗体的结构文件，PDB格式。

Number of designs

指定设计的抗体数量，默认为20。

H-CDR1, H-CDR2, H-CDR3, L-CDR1, L-CDR2, L-CDR3

分别指定需要设计的抗体重、轻链CDR区的长度范围。格式为：起始长度-终止长度（如:5-13），或单一长度（如:7）。
说明：这些参数定义了每个CDR区的允许长度范围，如果设置的是起始长度-终止长度（如:5-13）,模型将从中均匀采样长度。如果设置的是单一长度（如：7），则该CDR将以指定长度进行设计。如果不指定某个CDR的长度范围（如：不设置H-CDR1的长度），则该CDR将保持原始结构和序列不被设计。
对于VHH设计，仅需指定H-CDR1, H-CDR2, H-CDR3；对于scFv设计，可指定所有六个CDR。长度选择可参考自然抗体的CDR 长度分布，推荐较短的H-CDR3（如:5-13），以降低设计难度。

Hotspot

指定抗原上的结合位点残基，用于定义抗体结合的表位。格式为：逗号分隔的残基列表，格式为 305,456

• 说明：结合位点残基帮助模型聚焦于特定表位。选择时建议挑选表位中3个以上疏水性残基，避免过多极性或糖基化区域。

结果说明

经过抗体设计后，得到的抗体-抗原复合物结构，并根据质量评估指标进行排序。包括：

结构文件：按结构质量排序的PDB格式抗体-抗原复合物结构的打包文件 de_novo_antibody.tar及最优的设计结果rank_1.pdb
结构评分：CSV格式的评估指标表格 cdr_sequences.csv，包含如下信息：

输出示例

Design_ID,CDR_H3,ipAE,pLDDT
rank_1,IAYTPGAPLF,8.91,0.92
rank_2,VAPSKTDALF,9.29,0.92

参考文献

Bennett NR, Watson JL, Ragotte RJ, Borst AJ, See DL, Weidle C, Biswas R, Shrock EL, Leung PJY, Huang B, Goreshnik I, Ault R, Carr KD, Singer B, Criswell C, Vafeados D, Sanchez MG, Kim HM, Torres SV, Chan S, Baker D. Atomically accurate de novo design of antibodies with RFdiffusion. bioRxiv. 2024.03.14.585103.

Antibody Design (RFAntibody)

Introduction

RFantibody is the most advanced de novo antibody generation method currently available. Through artificial intelligence (AI) technology, it achieves de novo design of antibodies, including single-domain antibodies (VHH) and single-chain antibody fragments (scFv), capable of precisely binding to user-specified target epitopes, with functionality validated through wet lab experiments.

RFantibody is based on the protein structure prediction model RoseTTAFold2 (RF2) and the protein generation model RFdiffusion. By fine-tuning the original RFdiffusion, a specialized version for antibody design has been developed. Its core principles are as follows:

• Utilization of Antibody Structural Features: RFdiffusion is fine-tuned on antibody structural data (approximately 8,100 antibody structures) from the Protein Data Bank (PDB), focusing on training the antibody-specific complementarity-determining region (CDR) loops while maintaining framework structures close to user-specified optimized frameworks. During training, noise is gradually added (3D Gaussian noise to Cα coordinates, SO(3) Brownian motion to residue orientations), and the network learns to predict the denoised structure.

• Epitope-Targeted Design: By introducing “Hotspot” features, users can specify epitopes on target proteins, and the network designs through interactions between CDR loops and the epitope. During training, the antibody framework is provided in a globally coordinate-independent manner (represented by 2D distance and dihedral angle matrices), allowing the network to freely design CDR loop conformations and rigid-body positioning of the antibody relative to the target.

• Sequence Design and Validation: After structural design, ProteinMPNN is used to generate sequences for CDR loop regions, optimizing interactions with the target epitope. The designed antibodies are validated through structure prediction and self-consistency verification using the fine-tuned RF2, screening for high-potential candidates.

• Support for VHH and scFv Design: RFdiffusion supports not only the design of single-domain antibodies (VHH) but also single-chain antibody fragments (scFv). scFv design involves designing all six CDRs of the heavy and light chains.

Through these methods, RFantibody can generate diverse antibody structures that significantly differ from the training dataset while achieving high shape complementarity and functional binding to target epitopes.

Experimental Validation

The RFantibody project has conducted VHH and scFv designs targeting multiple disease-related epitopes and validated their effectiveness through surface plasmon resonance (SPR), cryo-electron microscopy (cryo-EM), neutralization assays, and other methods. The following are specific experimental results and analyses:

1, Single-Domain Antibody (VHH) Design and Experimental Validation

Experiments selected multiple disease-related targets, including influenza hemagglutinin (HA), respiratory syncytial virus (RSV) sites I and III, SARS-CoV-2 receptor-binding domain (RBD), Clostridioides difficile toxin B (TcdB), and IL-7Rα. Key results include:

• Binding Affinity (KD):

  • Influenza HA: Among VHH designs targeting the HA stem epitope, the highest affinity binder (VHH_flu_01) had a KD value of 78 nM, with other binders having KD values of 546 nM, 698 nM, and 790 nM. Experiments used insect cell-expressed monomeric HA (simulating deglycosylated state) to match computational design conditions.
  • SARS-CoV-2 RBD: The best VHH binder had a KD value of 5.5 μM, confirmed to bind to the target epitope through competition experiments (competing with known binder AHB2).
  • TcdB: The best VHH binder targeting the Frizzled-7 epitope had a KD value of 260 nM, with high binding specificity and no observed cross-reactivity with Clostridium sordellii toxin L (TcsL), which has 70% homology.

• Neutralization Activity (EC50):

  • TcdB: VHHs targeting TcdB demonstrated functionality in neutralization assays, neutralizing TcdB toxicity in CSPG4 knockout cells with an EC50 value of 460 nM, indicating potential therapeutic applications.

• Structural Accuracy (cryo-EM):

  • Influenza HA: Cryo-EM resolved the complex structure of VHH_flu_01 with native glycosylated HA trimer (resolution 3.0 Å). 66% of HA particles bound up to two VHHs, with partial non-binding possibly due to N296 glycan shielding. The experimental structure highly aligned with the design model, with an overall RMSD of 1.45 Å, CDR3 RMSD of 0.8 Å, and key CDR3 residues (V100, V101, S103, F108) interacting with the HA stem epitope as designed.
  • TcdB: Cryo-EM analysis was performed on the original design (VHH_TcdB_H2) and affinity-matured version (VHH_TcdB_H2_ortho) targeting TcdB. The original design confirmed binding to the Frizzled-7 epitope, while the matured version (resolution 5.7 Å) showed higher binding proportions, with structures conforming to design expectations.
  • SARS-CoV-2 RBD: The affinity-matured VHH (VHH_RBD_D4_ortho19) bound to the RBD “up” conformation epitope (resolution 3.9 Å).

• Affinity Maturation (OrthoRep):

  • The OrthoRep system was used for affinity maturation of VHHs targeting TcdB, influenza HA, and SARS-CoV-2 RBD, improving binding affinity by approximately two orders of magnitude while maintaining original epitope specificity.

2, Single-Chain Antibody Fragment (scFv) Design and Experimental Validation

Further expansion to scFv design involved designing six CDRs of heavy and light chains, adopting a structure-guided combinatorial library strategy to increase success rates. Experimental targets included the Frizzled-7 epitope of TcdB and the Phox2b/HLA-C*07:02 complex.

• Binding Affinity (KD):

  • TcdB: Through combinatorial library screening, scFvs targeting the Frizzled-7 epitope were identified, with the highest affinity binder (scFv6) having a KD value of 72 nM. KD values for other binders were not detailed. Competition experiments (competing with Frizzled-7) confirmed binding to the target epitope, with no competition with the unrelated receptor CSPG4.
  • Phox2b/HLA-C*07:02: scFvs targeting the neuroblastoma-related epitope had KD values of 400 nM (SPR) and 1 μM (ITC), specifically binding to the Phox2b peptide but not to the R6A mutant peptide. Attempts to convert it to CAR-T cells did not show cytotoxicity, possibly due to insufficient affinity or low antigen density.

• Structural Accuracy (cryo-EM):

  • TcdB: Cryo-EM structures of two scFvs (scFv5 and scFv6) binding to the Frizzled-7 epitope validated design accuracy. scFv6 had a resolution of 3.6 Å, overall RMSD of 0.9 Å, and backbone RMSDs for the six CDRs of CDRH1=0.4 Å, CDRH2=0.3 Å, CDRH3=0.7 Å, CDRL1=0.2 Å, CDRL2=1.1 Å, CDRL3=0.2 Å, with side chain conformations and interactions conforming to design. scFv5 (resolution 6.1 Å) bound with a different approach angle, with the experimental structure consistent with the design model.

3, Analysis of Experimental Results

• Structural Diversity: The designed VHHs and scFvs had CDR regions significantly different from natural antibodies, and there were no known antibodies for the Frizzled-7 epitope of TcdB, indicating that RFdiffusion achieved true de novo design.
• Functionality and Application Potential: The neutralization activity of TcdB VHH (EC50=460 nM) and high affinity of scFv (KD=72 nM) demonstrated therapeutic potential, but the failure of Phox2b scFv in CAR-T applications indicated the need for further optimization of affinity or antigen expression.

4, Summary

RFantibody, through fine-tuning the RFdiffusion network, has achieved the goal of de novo designing VHHs and scFvs capable of targeting various disease-related epitopes. Experimental results show that the designed antibodies have high structural accuracy (RMSD as low as 0.9 Å) and functionality (KD as low as 72 nM, EC50 of 460 nM). Cryo-EM validated the atomic-level precision of the designs, while affinity maturation and combinatorial library strategies further improved success rates.

Parameter

Complex

The structure of the antibody-antigen complex used for antibody design, in PDB format. If this parameter is specified, the subsequent Antigen and Antibody parameters do not need to be specified. If this parameter is not specified, the structures of Antigen and Antibody need to be input separately.

Antigen

The structure file of the antigen, in PDB format.
Note: The antigen structure usually needs to be truncated to reduce computational cost. It is recommended to retain only the region within approximately 10 Å around the epitope.

Antibody

The structure file of the antibody, in PDB format.

Number of designs

The number of antibodies to be designed, with a default value of 20.

H-CDR1, H-CDR2, H-CDR3, L-CDR1, L-CDR2, L-CDR3

Specify the length range of the CDR regions in the heavy and light chains to be designed. The format is: start length-end length (e.g., 5-13), or a single length (e.g., 7).
Note: These parameters define the allowed length range for each CDR region. If a range is specified (e.g., 5-13), the model will uniformly sample lengths within this range. If a single length is specified (e.g., 7), the CDR will be designed with the given length. If the length range of a CDR is not specified (e.g., H-CDR1 is not set), that CDR will retain its original structure and sequence without being designed.
For VHH design, only H-CDR1, H-CDR2, and H-CDR3 need to be specified; for scFv design, all six CDRs can be specified. The length selection can refer to the natural distribution of CDR lengths in antibodies. It is recommended to use a shorter H-CDR3 (e.g., 5-13) to reduce design complexity.

Hotspot

Specify the binding site residues on the antigen to define the epitope for antibody binding. The format is: a comma-separated list of residues, e.g., 305,456.
Note: Binding site residues help the model focus on specific epitopes. It is recommended to select more than three hydrophobic residues within the epitope and avoid areas with excessive polarity or glycosylation.

Result Description

After antibody design, the antibody-antigen complex structures are obtained and sorted based on quality assessment metrics. These include:
Structure Files: The packed file of antibody - antigen complex structures in PDB format sorted by structural quality is de_novo_antibody.tar, and the optimal design result rank_1.pdb.
Structure Scores: A CSV file cdr_sequences.csv containing the assessment metrics, with the following information:

Example

Design_ID,CDR_H3,ipAE,pLDDT
rank_1,IAYTPGAPLF,8.91,0.92
rank_2,VAPSKTDALF,9.29,0.92

References

Bennett NR, Watson JL, Ragotte RJ, Borst AJ, See DL, Weidle C, Biswas R, Shrock EL, Leung PJY, Huang B, Goreshnik I, Ault R, Carr KD, Singer B, Criswell C, Vafeados D, Sanchez MG, Kim HM, Torres SV, Chan S, Baker D. Atomically accurate de novo design of antibodies with RFdiffusion. bioRxiv. 2024.03.14.585103.

Nanobody Thermostability Prediction

简介

该模块用于预测纳米抗体的热稳定性（Tm值）。模型使用多种序列嵌入（如ESM-1b、one-hot、VHSE）来表示蛋白质序列，并通过不同的回归模型（如ridge、GPR、RF、SVR）进行处理，最后进行集成学习得到最终模型。通过文献整合和新测量，构建了一个包含640个独特纳米抗体序列的熔化温度数据集。具体来说，从NbThermo数据库中添加了511个独特序列点，并通过实验生成了129个新数据点。在测试集上表现出较高的预测准确性，Pearson相关系数为0.853，Spearman相关系数为0.832，MAE为4.1°C，SDR为0.86。
模型的整体架构如下图所示：

模型预测效果如下图所示：

参数说明

Nanobody Sequence

纳米抗体的序列文件，FASTA格式

Output

输出结果文件名，默认为Tm_pred.csv。

结果说明

输出结果文件为Tm_pred.csv，包含信息如下：

备注：部分纳米抗体无法预测其热稳定性。

参考文献

Ramon A, Ni M, Predeina O, Gaffey R, Kunz P, Onuoha S, Sormanni P. Prediction of protein biophysical traits from limited data: a case study on nanobody thermostability through NanoMelt. MAbs. 2025 Dec;17(1):2442750.

Nanobody Thermostability Prediction

Introduction

This module is designed to predict the thermostability (Tm value) of nanobodies. The model uses various sequence embeddings, such as ESM-1b, one-hot, and VHSE, to represent protein sequences. These are processed by different regression models, including ridge, GPR, RF, and SVR, and finally integrated to produce the final model. A dataset comprising the melting temperatures of 640 unique nanobody sequences was constructed through literature integration and new measurements. Specifically, 511 unique sequence points were added from the NbThermo database, and 129 new data points were generated experimentally. The model demonstrates high prediction accuracy on the test set, with a Pearson correlation coefficient of 0.853, a Spearman correlation coefficient of 0.832, an MAE of 4.1°C, and an SDR of 0.86.
The overall architecture of the model is shown in the figure below:

The model’s prediction performance is illustrated in the figure below:

Parameter

Nanobody Sequence

The sequence file of the nanobody, in FASTA format.

Output

The name of the output result file, default is Tm_pred.csv.

Result

The output result file is Tm_pred.csv, containing the following information:

** Note ** : Some nanoantibodies cannot predict their thermal stability.

References

Ramon A, Ni M, Predeina O, Gaffey R, Kunz P, Onuoha S, Sormanni P. Prediction of protein biophysical traits from limited data: a case study on nanobody thermostability through NanoMelt. MAbs. 2025 Dec;17(1):2442750.

MD Solvation v2

简介

对MD体系进行溶剂化操作，添加水盒子和离子。

参数说明

Receptor Topology

输入的受体拓扑文件，可由GMX Receptor Parameterization模块生成。

Receptor GRO

输入的受体结构文件，可由GMX Receptor Parameterization模块生成。

Receptor ITP

输入的受体参数(压缩)文件，可由GMX Receptor Parameterization模块生成。

Ligand GRO

输入的配体结构(压缩)文件，可由GMX Ligand Parameterization模块生成。

Ligand ITP

输入的配体参数(压缩)文件，可由GMX Ligand Parameterization模块生成。

Ions

需要添加的离子，支持钠离子NA，钾离子K，氯离子CL，钙离子CA，镁离子MG，锌离子ZN，同时添加多个使用英文冒号:分割，如NA:K:MG

Number of Ions

需要添加的离子数目，添加多种离子时，和Ions参数对应，使用英文冒号:分割，如15:20:30
说明：Number of Ions与Concentration of Ions，选择其中一种输入，不要同时输入

Concentration of Ions

需要添加的离子浓度，添加多种离子时，和Ions参数对应，使用英文冒号:分割，如0.15:0.3:0.1
说明：Number of Ions与Concentration of Ions，选择其中一种输入，不要同时输入

Output Topology

输出的体系总的拓扑文件

Output GRO

输出的体系总的结构文件

Output ITP

输出的体系参数的(压缩)文件

Distance Restraints

距离限制，仅当Disre不为no时生效，格式如下所示：

[AtomIndex1]  [AtomIndex2]  [Type]  [Index]  [Type]  [Low]  [Up1]  [Up2]  [Factor]

其中，AtomIndex1和AtomIndex2为在system.gro的原子编号；Type为施加约束类型，通常设置为1，Type类型见表1；Index是计算顺序；Low、Up1、Up2为原子间限制距离，Low到Up1区间的原子距离是不受限制的，但是不能超过Up2，单位为nm；Factor为因子，将Factor乘以“Disre Force Constant”即为限制力的大小，单位为kJ/mol/nm2。
例如：

10     16      1       0       1      0.0     0.3     0.4     1.0
10     46      1       1       1      0.0     0.3     0.4     1.0
16     22      1       2       1      0.0     0.3     0.4     2.5

表1：GROMACS中三种约束类型对原子对进行限制

Angle Restraints

角度限制是两对原子间角度的限制，仅当Disre不为no时生效，格式如下所示：

[AtomIndex1]  [AtomIndex2]  [AtomIndex3]  [AtomIndex4]  [Type]  [Theta0]  [Force Constant]  [Multiplicity]

其中，AtomIndex1-AtomIndex2是第一对原子编号；AtomIndex3-AtomIndex4为第二对原子编号；Type在这里无用，定义为1即可；Theta0为约束的角度，单位为deg；Force Constant为约束力常数，单位为kJ/mol；Multiplicity为多重度。
例如

2642     2643     2635     2652     1     67.0     1500     1

Dihedral Restraints

二面角限制，仅当Disre不为no时生效，格式如下所示：

[AtomIndex1]  [AtomIndex2]  [AtomIndex3]  [AtomIndex4]  [Type]  [Label]  [Phi]  [dPhi]  [KFactor]  [Power]

其中，AtomIndex1-AtomIndex4为组成二面角的原子编号；Type为约束类型函数，总是为1；Label无效；Phi为参考角，dPhi为超出参考角的角度值，单位为deg；KFactor为因子，将KFactor乘以“Disre Force Constant”即为限制力的大小，单位为 kJ/mol/rad2；Power无效。
例如：

2642      2643      2635      2652      1      67.0      1500      1

约束势函数如下所示：

其中，Φ’为参考角Phi，ΔΦ为超出参考角的值dPhi，K_dihr为限制力的大小KFactor。

结果说明

输出结果包括：

参考文献

GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, E. Lindahl, SoftwareX, 1, (2015), 19-25. DOI: 10.1016/j.softx.2015.06.001

Human Germline BLAST (v2.1)

简介

通过序列比对在人类生殖系数据库中搜索与目标抗体序列最接近的同源模板，输出对应的模板序列以及序列一致性信息。

参数说明

Sequence String模式

Input Sequence

抗体的序列（纯序列信息，非FASTA格式文件）。

Type

抗体编号类型：kabat、chothia、imgt。

TopHits

输出同源性最高的n条序列。

Fasta File模式

FASTA File

抗体的序列文件，FASTA格式。

Type

抗体编号类型：kabat、chothia、imgt。

TopHits

输出同源性最高的n条序列。

结果说明

相关内容

抗体常用的germline模板：

临床后期及已上市抗体的germline配对分布情况（统计自Adimab数据集）：

Human Germline BLAST (v2.1)

Introduction

This module performs sequence alignment to search for the closest homologous template in the human germline database for a given target antibody sequence. It outputs the corresponding template sequence along with sequence similarity information.

Parameter Description

Sequence String Mode

Input Sequence

The antibody sequence (pure sequence information, not in FASTA format).

Type

Type of antibody numbering: kabat, chothia, imgt.

TopHits

Number of top hits to output.

Fasta File Mode

FASTA File

Antibody sequence file in FASTA format.

Type

Type of antibody numbering: kabat, chothia, imgt.

TopHits

Number of top hits to output.

Result Description

Related Content

Commonly used germline templates for antibodies:

Distribution of germline pairing for late-stage and marketed antibodies (statistics from the Adimab dataset):

Grafting v2.4

简介

Grafting模块是移植抗体的CDR到特定的框架区模板上，通常用于人源化设计。版本：v2.4

参数说明

Antibody Sequence File

抗体序列文件，FASTA格式

Numbering Type

抗体编号规则：kabat，imgt，chothia

Output File

指定输出抗体graft后的序列文件名称，FASTA格式

Output Policy

指定输出graft策略文件，JSON格式

Germline Score

指定输出抗体FR区序列比对同源性打分文件

Germline

指定轻链或重链使用特定germline模板，也可都指定，写法如下：

seq_name1:germline_name1,seq_name2:germline_name2

其中链名来自于流程第一步输入的fasta文件。
例1：以下语句为链"Infliximab.H"指定了模板"IGHV3-7*01"：

Infliximab.H:IGHV3-7*01

例2：以下语句为两条链分别指定了模板：

Infliximab.H:IGHV3-7*01,Infliximab.L:IGKV1-39*01

Template Sequence

指定参考模板序列，FASTA格式

Germline Hits

指定输出FR区序列比对结果文件，FASTA格式

Number of Hits

指定输出命中序列的数目

结果说明

输出结果包括：

Grafting v2.4

Introduction

The Grafting module is used to graft the CDRs of an antibody onto a specific framework region template, typically used for humanization design. Version: v2.4

Parameter Description

Antibody Sequence File

Antibody sequence file in FASTA format.

Numbering Type

Antibody numbering rule: kabat, imgt, chothia.

Output File

Specify the output file name for the grafted antibody sequence in FASTA format.

Output Policy

Specify the output grafting strategy file in JSON format.

Germline Score

Specify the output file for the homology scores of the antibody FR region sequences.

Germline

Specify the specific germline template to be used for the light chain or heavy chain, or both, in the format:

seq_name1:germline_name1,seq_name2:germline_name2

Where the chain names come from the FASTA file input in the first step of the process.
Example 1: The following statement specifies the template “IGHV3-7*01” for the chain “Infliximab.H”:

Infliximab.H:IGHV3-7*01

Example 2: The following statement specifies templates for two chains separately:

Infliximab.H:IGHV3-7*01,Infliximab.L:IGKV1-39*01

Template Sequence

Specify the reference template sequence in FASTA format.

Germline Hits

Specify the output file for the FR region sequence alignment results in FASTA format.

Number of Hits

Specify the number of sequences to output.

Result Description

The output includes:

Mutation Energy of Stability (Pythia)

简介

该模块基于Pythia模型实现，该模型是一种针对零样本 ∆∆G 预测量身定制的自监督图神经网络。

蛋白质突变效应预测是解码分子进化机制、优化蛋白质工程改造的关键物理量。然而，传统预测方法面临两大挑战：一是基于物理力场的计算方法（如自由能微扰）计算复杂度高，难以满足大规模筛选需求；二是依赖于实验数据的监督学习方法易受训练集偏差影响，泛化能力受限。

为了应对这些问题，研究团队提出了Pythia框架，它结合了图神经网络与注意力机制，能够直接从蛋白质的三维结构中学习氨基酸之间的相互作用。通过这种“零监督”预训练策略，Pythia突破了传统方法对标记数据的依赖，成功捕捉了蛋白质折叠过程中隐藏的物理化学约束规律。

Pythia的模型架构采用了将蛋白质局部结构转化为k近邻图的方式，每个氨基酸作为节点，通过欧几里得距离连接其32个最近的氨基酸。节点的特征包括氨基酸类型以及主链的二面角，边的特征则涉及主链原子之间的距离、序列位置和链信息。通过消息传递神经网络（MPNN）架构，Pythia可以高效地更新每个氨基酸节点的信息，并对突变的稳定性变化进行准确预测。

与传统的基于物理力场的方法相比，Pythia能够在单核计算中实现每分钟预测约50,000个突变，速度提升了5个数量级。其在标准测试集S2648上的Spearman相关系数为0.616，Pearson相关系数为0.598，表现优于现有的所有对比模型。这一进展为大规模蛋白质序列空间扫描提供了强大的计算支持，能够处理多达2600万个高质量蛋白质结构数据，显著加深了我们对蛋白质序列空间的理解。

在实验验证中，Pythia表现出了比传统能量函数方法高出一倍的成功率，充分证明了其在实际应用中的可靠性。同时，Pythia的可解释性也为蛋白质工程提供了宝贵的生物学见解，使其更易于应用于复杂的蛋白质工程任务。

模型架构：Pythia将蛋白质局部结构转换为k近邻图，其中每个氨基酸作为一个节点，并通过欧几里得距离连接其32个最近的氨基酸。节点的特征包括氨基酸类型和主链的二面角（φ、ψ、ω），边的特征包括主链原子之间的距离、序列位置和链信息。

训练目标：Pythia的训练目标是预测中心节点的自然氨基酸类型，使用来自节点和边的信息。

消息传递神经网络（MPNN）：Pythia采用消息传递神经网络（MPNN）架构，具体为带有注意力机制的消息传递层（AMPL）。在每个AMPL层中，顶点表示通过注意力块更新，然后与边表示连接以派生消息表示，最终通过另一个注意力块进一步细化节点表示。

损失函数：通过估计特定位置处每个氨基酸的概率来实现ΔΔG的预测。

在与其他自监督预训练模型和基于力场的方法的比较基准中，Pythia以极高的相关性超越其他同类算法，同时以最少的参数运行，使得计算速度显着加快，高达105倍。Pythia的功效通过其在预测柠檬烯环氧水解酶 (LEH) 的热稳定突变中的应用得到证实，实验成功率显着提高。
S2648数据集上的性能：Pythia在S2648数据集上的Spearman相关系数为0.616，Pearson相关系数为0.598，优于所有测试的模型。
S669数据集上的性能：在S669数据集上，Pythia的Spearman相关系数为0.66，在所有评估的方法中表现最佳。

大规模数据集上的性能：在一个包含约100万个突变的百万级数据集上，Pythia的Spearman相关系数为0.602，Pearson相关系数为0.633，AUROC为0.83，AUPRC为0.88。
计算速度：Pythia的计算速度比传统的力场方法快105倍，能够在20秒内完成S2648数据集的计算，单核速度约为50,000个突变/分钟。

参数说明

Structure PDB

蛋白结构文件，PDB格式。不支持含有非标准氨基酸的蛋白。

Chain

指定要突变扫描的链名，可多链，用英文逗号分隔，如：A,B，默认为空，表示全部链都扫描。

Output

输出文件名称，默认mutation_energy.csv。

Output_fmt

特定格式化的输出文件名称，默认mutation_energy_fmt.csv。

备注：当前24GB的GPU显存支持计算的残基数量在2000个左右。

结果说明

输出mutation_energy.csv结果文件，包含以下信息：

输出mutation_energy_fmt.csv结果文件，包含如下信息：

输出结果对应的热图mutation_energy_[chain].png

参考文献

• Jinyuan Sun, Tong Zhu, Yinglu Cui, Bian Wu, Structure-based self-supervised learning enables ultrafast protein stability prediction upon mutation, The Innovation, Volume 6, Issue 1, 2025, 100750, ISSN 2666-6758, DOI: 10.1016/j.xinn.2024.100750

Mutation Energy of Stability (Pythia)

Introduction

This module is implemented based on the Pythia model, which is a self-supervised graph neural network specifically designed for zero-shot ∆∆G prediction.

Predicting the effects of protein mutations is a key factor in decoding molecular evolution mechanisms and optimizing protein engineering modifications. However, traditional prediction methods face two major challenges: first, computational methods based on physical force fields (such as free energy perturbation) have high computational complexity, making them unsuitable for large-scale screening; second, supervised learning methods that rely on experimental data are susceptible to training set biases, limiting their generalization ability.

To address these issues, the research team proposed the Pythia framework, which combines graph neural networks with attention mechanisms to learn interactions between amino acids directly from the three-dimensional structure of proteins. Through this “zero-supervision” pre-training strategy, Pythia overcomes the traditional methods’ dependence on labeled data and successfully captures the hidden physicochemical constraints in the protein folding process.

The architecture of Pythia converts the local structure of proteins into k-nearest neighbor graphs, where each amino acid acts as a node connected to its 32 nearest amino acids based on Euclidean distance. Node features include amino acid type and backbone dihedral angles, while edge features involve distances between backbone atoms, sequence positions, and chain information. Using a message-passing neural network (MPNN) architecture, Pythia efficiently updates information for each amino acid node and accurately predicts changes in mutation stability.

Compared to traditional physical force field-based methods, Pythia can predict approximately 50,000 mutations per minute on a single-core processor, achieving a speed increase of five orders of magnitude. On the standard test set S2648, it achieves a Spearman correlation coefficient of 0.616 and a Pearson correlation coefficient of 0.598, outperforming all existing comparative models. This advancement provides powerful computational support for large-scale scanning of protein sequence space, capable of handling up to 26 million high-quality protein structure data points, significantly deepening our understanding of protein sequence space.

In experimental validation, Pythia demonstrated a success rate twice as high as traditional energy function methods, fully proving its reliability in practical applications. Additionally, Pythia’s interpretability offers valuable biological insights for protein engineering, making it more applicable to complex protein engineering tasks.

Model Architecture: Pythia transforms the local structure of proteins into a k-nearest neighbor graph, where each amino acid is represented as a node, connected to its 32 nearest amino acids by Euclidean distance. The features of the nodes include the amino acid type and the backbone dihedrals (φ, ψ, ω), while the features of the edges include the distances between backbone atoms, sequence positions, and chain information.

Training Objective: The training objective of Pythia is to predict the natural amino acid type of the central node, using information from both nodes and edges.

Message Passing Neural Network (MPNN): Pythia employs a message passing neural network (MPNN) architecture, specifically an Attention-based Message Passing Layer (AMPL). In each AMPL layer, the vertices are updated through an attention block, and then connected to edge representations to derive message representations, which are further refined through another attention block.

Loss Function: The prediction of ΔΔG is achieved by estimating the probability of each amino acid at specific positions.

In benchmark comparisons with other self-supervised pre-training models and force-field-based methods, Pythia outperforms other similar algorithms with high correlation while operating with minimal parameters, significantly accelerating computational speed by up to 105 times. The effectiveness of Pythia is demonstrated through its application in predicting thermally stable mutations of limonene epoxide hydrolase (LEH), with a notable increase in experimental success rates.
Performance on the S2648 Dataset: Pythia achieves a Spearman correlation coefficient of 0.616 and a Pearson correlation coefficient of 0.598 on the S2648 dataset, outperforming all tested models.
Performance on the S669 Dataset: On the S669 dataset, Pythia achieves a Spearman correlation coefficient of 0.66, performing the best among all evaluated methods.

Performance on Large-scale Datasets: On a large dataset containing approximately 1 million mutations, Pythia achieves a Spearman correlation coefficient of 0.602, a Pearson correlation coefficient of 0.633, an AUROC of 0.83, and an AUPRC of 0.88.
Computational Speed: Pythia is 105 times faster than traditional force-field methods, capable of completing calculations on the S2648 dataset in 20 seconds, with a single-core speed of approximately 50,000 mutations per minute.

Parameters

Structure PDB

Protein structure file in PDB format. Proteins containing non-standard amino acids are not supported.

Chain

Specify the chain names to be scanned for mutations. Multiple chains can be listed, separated by commas, e.g., A,B. The default is empty, which means all chains will be scanned.

Output

Output file name, mutation_energy.csv is the default.

Results

Outputs a mutation_energy.csv file containing the following information:

References

• Jinyuan Sun, Tong Zhu, Yinglu Cui, Bian Wu, Structure-based self-supervised learning enables ultrafast protein stability prediction upon mutation, The Innovation, Volume 6, Issue 1, 2025, 100750, ISSN 2666-6758, DOI: 10.1016/j.xinn.2024.100750

Mutation Energy of Binding (Pythia-PPI)

简介

Mutation Energy of Binding (Pythia-PPI)模块基于Pythia-PPI模型实现，该模型基于深度学习，结合了多任务学习和自蒸馏策略，以克服实验数据稀缺的瓶颈，并提高预测准确性。Pythia-PPI由两个模块组成：预训练的结构图编码器模块和ΔΔG预测模块。该模型使用k-最近邻（k-NN）图将蛋白质或蛋白质-蛋白质复合物的局部结构转换为图表示，每个氨基酸作为一个节点，与其32个最近的氨基酸基于C-alpha原子的欧几里得距离建立连接。输入的结构图编码器结合了氨基酸类型的一热编码，以及使用正弦和余弦函数表示的主链二面角（φ、ψ和ω）作为节点特征。边特征则考虑了五个主链原子（C-alpha、C、N、O和C-beta）之间的距离，以及序列位置和链信息。通过结构图编码器，节点和边输入特征被转换为嵌入，这些嵌入与预训练模块中的氨基酸概率相结合，形成ΔΔG预测模块的输入向量。Pythia-PPI采用迁移学习和多任务学习相结合的方法，共享结构编码器层以预测突变对PPI结合亲和力和蛋白质稳定性的影响。

使用了SKEMPI数据集进行基准测试，并与其他方法进行了比较。结果显示，Pythia-PPI在SKEMPI数据集上的皮尔逊相关系数从0.6447提高到0.7850，在病毒-受体数据集上的皮尔逊相关系数从0.3654提高到0.6051。这些结果表明Pythia-PPI是一个分析蛋白质-蛋白质相互作用适应性景观的有力工具。

参数说明

Structure PDB

蛋白复合物结构文件，PDB格式。不支持含有非标准氨基酸的蛋白。

Chain

指定要突变扫描的链名，可多链，用英文逗号分隔，如：A,B，默认为空，表示全部链都扫描。

Output

输出文件名称，默认mutation_ddg.csv。

Output_fmt

特定格式化输出的结果文件名称，默认mutation_ddg_fmt.csv。

备注：当前24GB的GPU显存支持计算的残基数量在1500个左右。

结果说明

输出mutation_ddg.csv结果文件，包含以下信息：

输出mutation_ddg_fmt.csv结果文件，包含如下信息：

输出结果对应的热图mutation_ddg_[chain].png

参考文献

• Reliable prediction of protein-protein binding affinity changes upon mutations with Pythia-PPI, Fangting Tao, Jinyuan Sun, Bian Wu, George F Gao bioRxiv 2024.10.28.620752; DOI: 10.1101/2024.10.28.620752

Mutation Energy of Binding (Pythia-PPI)

Introduction

The Mutation Energy of Binding (Pythia-PPI) module is implemented based on the Pythia-PPI model, which utilizes deep learning and combines multi-task learning with a self-distillation strategy to overcome the bottleneck of scarce experimental data and improve prediction accuracy. Pythia-PPI consists of two modules: a pre-trained structural graph encoder module and a ΔΔG prediction module. The model uses a k-nearest neighbors (k-NN) graph to convert the local structure of proteins or protein-protein complexes into a graph representation, where each amino acid is represented as a node, connected to its 32 nearest amino acids based on the Euclidean distance of C-alpha atoms. The input structural graph encoder combines one-hot encoding of amino acid types with backbone dihedrals (φ, ψ, and ω) represented using sine and cosine functions as node features. Edge features take into account the distances between five backbone atoms (C-alpha, C, N, O, and C-beta), as well as sequence positions and chain information. Through the structural graph encoder, the input features for nodes and edges are transformed into embeddings, which are combined with amino acid probabilities from the pre-trained module to form the input vector for the ΔΔG prediction module. Pythia-PPI employs a combination of transfer learning and multi-task learning, sharing structural encoder layers to predict the effects of mutations on PPI binding affinity and protein stability.

Benchmarking was conducted using the SKEMPI dataset and compared with other methods. The results show that Pythia-PPI improved the Pearson correlation coefficient from 0.6447 to 0.7850 on the SKEMPI dataset, and from 0.3654 to 0.6051 on the virus-receptor dataset. These results indicate that Pythia-PPI is a powerful tool for analyzing the adaptive landscape of protein-protein interactions.

Parameters

Structure PDB

Protein complex structure file in PDB format. Proteins containing non-standard amino acids are not supported.

Chain

Specify the chain names to be scanned for mutations. Multiple chains can be listed, separated by commas, e.g., A,B. The default is empty, which means all chains will be scanned.

Output

Output file name, mutation_ddg.csv is the default.

Output_fmt

Formatted output file name, mutation_ddg_fmt.csv is the default.

Results

Outputs a mutation_ddg.csv file containing the following information:

Outputs a mutation_ddg_fmt.csv file containing the following information:

The heatmap output mutation_ddg_[chain].png

References

• Reliable prediction of protein-protein binding affinity changes upon mutations with Pythia-PPI, Fangting Tao, Jinyuan Sun, Bian Wu, George F Gao bioRxiv 2024.10.28.620752; DOI: 10.1101/2024.10.28.620752

Antibody (Off-) Target Prediction

简介

Antibody (Off-) Target Prediction模块对输入的抗体进行潜在靶点预测，基于丰富的抗体-抗原相互作用数据库，寻找与输入抗体在序列及结构上高度相似的一系列抗体。基于相似性原理（相似抗体可能具有相似靶点），这些高度相似的抗体对应的抗原靶点可能是输入抗体的潜在靶点。当前抗体-抗原相互作用数据库包含16万对抗原-抗体复合物，主要来源于文献、专利等开源数据。

参数说明

Antibody Structure

待预测靶点的抗体结构文件，PDB格式或CIF格式。推荐使用AF3-like相关的结构预测模块进行抗体结构预测，如：Protenix

Mode

搜索模式，支持4种模式（默认为模式3）：

• 模式1: 完整抗体模式，以完整的抗体重轻链结构进行数据库检索。
• 模式2: 抗体CDR模式，仅提取抗体的CDR区域结构进行数据库检索。
• 模式3: 抗体重链CDR模式，仅提取抗体的重链CDR区域结构进行数据库检索。
• 模式4: 抗体重链CDR3模式，仅提取抗体的重链CDR3区域结构进行数据库检索。

TopN

保留打分排名最高的前N个结果，默认为50。

Species

物种信息过滤：

• Human表示仅保留人源靶点。
• Any表示不做任何限制。

Output

输出结果的文件名，默认为“hits.csv”

结果说明

检索结果文件默认为hits.csv，包含信息如下：

参考文献

• van Kempen M, Kim S, Tumescheit C, Mirdita M, Lee J, Gilchrist CLM, Söding J, and Steinegger M. Fast and accurate protein structure search with Foldseek. Nature Biotechnology, doi:10.1038/s41587-023-01773-0 (2023)
• Schneider, C., Raybould, M.I.J., Deane, C.M. (2022) SAbDab in the Age of Biotherapeutics: Updates including SAbDab-Nano, the Nanobody Structure Tracker. Nucleic Acids Res. 50(D1):D1368-D1372
• Brennan Abanades et al. “The Patent and Literature Antibody Database (PLAbDab): an evolving reference set of functionally diverse, literature-annotated antibody sequences and structures”. In: Nucleic Acids Research, Volume 52, Issue D1, 5 January 2024, Pages D545-D551
• Steinegger M and Soeding J. MMseqs2 enables sensitive protein sequence searching for the analysis of massive data sets. Nature Biotechnology, doi: 10.1038/nbt.3988 (2017)

Antibody (Off-) Target Prediction

Introduction

The Antibody (Off-) Target Prediction module predicts potential targets for the input antibody. Based on a rich database of antibody-antigen interactions, it identifies a series of antibodies that are highly similar to the input antibody in both sequence and structure. Following the principle of similarity (similar antibodies may have similar targets), the antigen targets corresponding to these highly similar antibodies could be potential targets for the input antibody. The current antibody-antigen interaction database contains 160,000 antigen-antibody complexes, primarily sourced from open-source data such as literature and patents.

Parameter

Antibody Structure

Antibody structure file for the target to be predicted, in PDB or CIF format.

Mode

Search Modes, supporting 4 modes (default is Mode 3):

• Mode 1: Full Antibody Mode, where the complete heavy and light chain structure of the antibody is used for database search.
• Mode 2: Antibody CDR Mode, where only the CDR regions of the antibody are extracted for database search.
• Mode 3: Antibody Heavy Chain CDR Mode, where only the CDR regions of the heavy chain are extracted for database search.
• Mode 4: Antibody Heavy Chain CDR3 Mode, where only the CDR3 region of the heavy chain is extracted for database search.

TopN

Retain the top N results with the highest scores, with the default being 50.

Species

Species Information Filtering:

• Human: Retain only human-derived targets.
• Any: No restrictions.

Output

The name of output file, default is “hits.csv”.

Result

The search result file hits.csv contains the following information:

Reference

• van Kempen M, Kim S, Tumescheit C, Mirdita M, Lee J, Gilchrist CLM, Söding J, and Steinegger M. Fast and accurate protein structure search with Foldseek. Nature Biotechnology, doi:10.1038/s41587-023-01773-0 (2023)
• Schneider, C., Raybould, M.I.J., Deane, C.M. (2022) SAbDab in the Age of Biotherapeutics: Updates including SAbDab-Nano, the Nanobody Structure Tracker. Nucleic Acids Res. 50(D1):D1368-D1372
• Brennan Abanades et al. “The Patent and Literature Antibody Database (PLAbDab): an evolving reference set of functionally diverse, literature-annotated antibody sequences and structures”. In: Nucleic Acids Research, Volume 52, Issue D1, 5 January 2024, Pages D545-D551
• Steinegger M and Soeding J. MMseqs2 enables sensitive protein sequence searching for the analysis of massive data sets. Nature Biotechnology, doi: 10.1038/nbt.3988 (2017)

Protein Design (LigandMPNN)

简介

LigandMPNN是一种基于深度学习的蛋白质序列设计方法，专门用于模拟蛋白质与非蛋白质组分（如小分子、核苷酸和金属）之间的相互作用。它是 ProteinMPNN的升级版，能够在蛋白质设计中加入非蛋白质的组分，从而提升对非蛋白-蛋白相互作用的理解。
主要特点和优势：
全面建模：LigandMPNN结合了蛋白质图、配体图和蛋白质-配体图三种图结构，全面建模蛋白质与非蛋白质组分的相互作用。
高性能：在恢复与小分子、核苷酸和金属相互作用的原生背景序列方面，LigandMPNN的表现优于传统方法如Rosetta和ProteinMPNN。
侧链预测：除了生成蛋白质序列，LigandMPNN还能生成侧链构象，允许对结合相互作用进行更详尽的评估。
LigandMPNN 在酶设计、小分子结合剂开发以及生物传感器的设计中具有广阔的应用前景。其高效性和准确性使其成为蛋白质工程领域的重要工具。

参数说明

Structure PDB

蛋白的结构文件，PDB格式。

Chain

指定需要设计的链，多条链用空格分割，例如：A,B。

Number of Sequences

输出设计的序列数目。

Sampling Temp

氨基酸采样温度，T=0.0表示取argmax，T>>1.0表示随机采样。建议的取值为0.1、0.15、0.2、0.25、0.3。较高的值会导致更多的多样性。

Position Type

设计残基模式：

• 固定（Fix）指定下一步Position中的残基在设计时保持不变。
• 设计（Design）指定下一步Position中的残基可进行设计而其他未指定残基在设计时保持不变。

Position

可选参数，设置氨基酸序号，对设置的氨基酸根据Position Type选项进行固定或设计。当参数Chain设置为A,C时，此参数如果设置为1 2 3 4 5 6 7 8 23 25, 10 11 12 13 14 15 16 17 18 19 20 40意味着对链A中的残基1 2 3…25和链C中的残基10 11 12…40进行固定或设计。
注意：同一条链的氨基酸序号用空格分隔，不同链的氨基酸用逗号分隔。根据原始 PDB 编号设计的，并且支持插入代码。

Omit_AAS

指定在生成的结果序列中不许出现的氨基酸种类。

Other Parameter

可选参数，可指定设计时参考的模式。具体含义如下：
–homomer：基于同源多聚体进行序列设计；
–pack_side_chains：对设计的序列生成侧链结构。

结果说明

最终设计的序列文件result.fasta，里面包含最终设计的序列。
其中序列名称：

1. overall_confidence：设计序列的全序列的置信度评分，数值在0~1.0之间，数值越大表示序列置信度越高
2. ligand_confidence：设计序列的所有已设计残基的置信度评分，数值在0~1.0之间，数值越大表示已设计部分序列的置信度越高
3. seq_rec：序列恢复率（与原序列的相似程度），0-1之间，越高表示与原序列越相似

指定参数--pack_side_chains时，输出设计后的结构打包文件packed_side_chains.tar.gz，包含最终设计的序列对应的复合物结构PDB文件。

参考文献

Justas Dauparas, Gyu Rie Lee, Robert Pecoraro, Linna An, Ivan Anishchenko, Cameron Glasscock, D. Baker. Atomic context-conditioned protein sequence design using LigandMPNN. bioRxiv 2023.12.22.573103;

Protein Design (LigandMPNN)

Introduction

LigandMPNN is a deep learning-based protein sequence design method specifically designed to simulate interactions between proteins and non-protein components (such as small molecules, nucleotides, and metals). It is an upgraded version of ProteinMPNN and can incorporate non-protein components into protein design, thereby enhancing the understanding of non-protein-protein interactions.

Key Features and Advantages:

• Comprehensive Modeling: LigandMPNN integrates three types of graph structures: protein graphs, ligand graphs, and protein-ligand graphs, to comprehensively model interactions between proteins and non-protein components.
• High Performance: In terms of recovering native background sequences interacting with small molecules, nucleotides, and metals, LigandMPNN outperforms traditional methods such as Rosetta and ProteinMPNN.
• Side Chain Prediction: In addition to generating protein sequences, LigandMPNN can also generate side chain conformations, allowing for a more detailed assessment of binding interactions.

LigandMPNN has broad application prospects in enzyme design, small molecule binder development, and biosensor design. Its efficiency and accuracy make it an important tool in the field of protein engineering.

Parameter

PDB File

Protein structure file in PDB format.

Chain

Specify the chain to be designed, multiple chains are separated by spaces, for example: A,B.

Number of Sequences

Output the number of sequences designed.

Position Type

Residue Design Mode:

• Fix specifying that the residues in the next Position step remain unchanged during design.
• Design specifying that the residues in the next Position step can be designed while other unspecified residues remain unchanged during design.

Position

Optional parameter to set the amino acid sequence number for fixing or designing amino acids based on the Position Type option. When the parameter Chain is set to A C, if this parameter is set to 1 2 3 4 5 6 7 8 23 25, 10 11 12 13 14 15 16 17 18 19 20 40, it means that residues 1 2 3…25 in chain A and residues 10 11 12…40 in chain C are fixed or designed.

Note: Amino acid sequence numbers of the same chain are separated by spaces, while amino acids from different chains are separated by commas. The position is designed according to the original PDB numbering and also supports insertion codes.

Sampling Temp

Amino acid sampling temperature, T=0.0 means argmax, T>>1.0 means random sampling. The suggested values are 0.1, 0.15, 0.2, 0.25, 0.3. Higher values result in more diversity.

Omit_AAS

Optional parameter specifying the types of amino acids that are not allowed to appear in the generated sequence.

Other Parameter

Optional parameter specifying the reference mode for design. Specific meanings are as follows:
–homomer: Sequence design based on homologous oligomers;
–pack_side_chains：Generate side chain structures for the designed sequence。

Result

The output file is result.fasta and contains the final design sequence.
Where the sequence name:

1. overall_confidence: The confidence score for the entire designed sequence, with values ranging from 0 to 1.0. A higher value indicates a higher confidence in the sequence.
2. ligand_confidence: The confidence score for all designed residues in the sequence, with values ranging from 0 to 1.0. A higher value indicates a higher confidence in the designed portion of the sequence.
3. seq_rec: Sequence recovery rate (similarity to the original sequence), ranging from 0 to 1. A higher value indicates greater similarity to the original sequence.

When the parameter --pack_side_chains is specified, the output is a packed structure file named packed_side_chains.tar.gz, which includes the PDB file of the final designed sequence’s corresponding complex structure.

Reference

Justas Dauparas, Gyu Rie Lee, Robert Pecoraro, Linna An, Ivan Anishchenko, Cameron Glasscock, D. Baker. Atomic context-conditioned protein sequence design using LigandMPNN. bioRxiv 2023.12.22.573103;

Thermostability Prediction

简介

基于TemBERTure开发的Thermostability Prediction是一个用于预测蛋白质热稳定性的深度学习工具，专注于氨基酸序列分析。它包括两个模型：TemBERTureCLS和TemBERTureTm。TemBERTureCLS是一个分类模型，用于预测蛋白质序列的热类别，即判断其是嗜热的还是非嗜热的。TemBERTureTm是一个回归模型，用于根据蛋白质序列预测其熔点温度（Tm）。这两个模型都基于protBERT-BFD语言模型，该模型在大量蛋白质序列数据集上进行了预训练。通过基于适配器的方法进行高效微调，使得TemBERTure能够在不需要广泛重新训练的情况下，稳健地适应特定任务。

TemBERTureCLS与其他常用模型的预测结果比较

TemBERTureTm与其他常用模型的预测结果比较

参数说明

Protein Sequence

蛋白的序列文件，FASTA格式

结果说明

默认输出结果文件为predicted_Tm.csv，包含信息如下：

参考文献

• Chiara Rodella, Symela Lazaridi, Thomas Lemmin, TemBERTure: advancing protein thermostability prediction with deep learning and attention mechanisms, Bioinformatics Advances, Volume 4, Issue 1, 2024, vbae103 DOI:10.1093/bioadv/vbae103

Thermostability Prediction

Introduction

Thermostability Prediction, developed based on TemBERTure, is a deep learning tool designed to predict protein thermostability, focusing on amino acid sequence analysis. It includes two models: TemBERTureCLS and TemBERTureTm. TemBERTureCLS is a classification model used to predict the thermal category of a protein sequence, determining whether it is thermophilic or non-thermophilic. TemBERTureTm is a regression model used to predict the melting temperature ™ of a protein based on its sequence. Both models are based on the protBERT-BFD language model, which has been pre-trained on a large dataset of protein sequences. By using an adapter-based fine-tuning approach, TemBERTure can efficiently and robustly adapt to specific tasks without the need for extensive retraining.

Comparison of TemBERTureCLS with other common models’ prediction results

Comparison of TemBERTureTm with other common models’ prediction results

Parameter

Protein Sequence

The protein sequence file in FASTA format.

Result

The output result file is predicted_Tm.csv, containing the following information:

Reference

• Chiara Rodella, Symela Lazaridi, Thomas Lemmin, TemBERTure: advancing protein thermostability prediction with deep learning and attention mechanisms, Bioinformatics Advances, Volume 4, Issue 1, 2024, vbae103 DOI:10.1093/bioadv/vbae103

GMX Metadynamics Generation

简介

GMX Metadynamics Generation模块是生成可用于Metadynamics模拟的输入文件。

参数说明

GRO File

提交模拟体系的gro文件。该文件可以从MD Solvation模块获取。

PBC

Metadynamics模拟阶段是否考虑周期性边界条件，yes或者no。

CV Group1

组成集合变量CV的第一个组所包含的原子。

CV Group2

组成集合变量CV的第二个组所包含的原子。

CV Group3

组成集合变量CV的第三个组所包含的原子。

CV Group4

组成集合变量CV的第四个组所包含的原子。
备注：

• Group1和Group2组成DISTANCE集合变量，Group1，Group2和Group3组成ANGLE集合变量，Group1,Group2,Group3和Group4组成TORSION集合变量
• Group的书写规则：a5表示GRO文件中第5个原子，a5-10表示GRO文件中第5-10个原子，aCA表示GRO文件中名字为CA的原子，同理r5, r5-10, rASP分别表示GRO文件中第5位残基，第5-10位残基和名字位ASP的残基，一些特殊的字符如Protein，Protein-H, MainChain等亦可使用，也可以合并使用，但需用逗号隔开，如"a5-10,r5,r8-10,UNK"表示GRO文件中第5-10位原子、第5位残基、第8-10位残基以及名字叫UNK的分子
• 多个CV的处理方式：如果要定义多个集合变量，则在Group定义中用"//"将不同集合变量对应的原子组进行分割，如a5//r5-10表示a5是第一个集合变量对应的原子组，r5-r10是第二个集合变量对应的原子组，当集合变量在某个Group没有对应的原子组时，用none表示，比如第一个集合变量是DISTANCE，第一个集合变量是ANGLE,那么第一个DISTANCE集合变量在Group3中没有对应的原子组，此时在Group3可以写none//r5-10,表示第一个集合变量在Group3中没有对应的原子组，而第二个集合变量在Group3中对应的原子组为r5-10

Component

集合变量DISTANCE对应的成分，其成分有x，y，z和xyz，分别表示计算DISTANCE仅考虑x，y，z维度以及xyz三个维度都考虑，有多个集合变量时用"//"进行分割。

Metad Height

施加的沉积高斯函数的高度，默认1.0

Metad Width

施加的沉积高斯函数的宽度或者标准差，有多个集合变量时用"//"进行分割，默认0.05

Metad Frequency

施加的沉积高斯函数的频率，默认500，即每500个时间步长进行一次高斯函数沉积

CV Min

集合变量的边界最小值，有多个集合变量时用"//"进行分割。无默认值时即不考虑边界，此时计算量会增加，强烈建议设置边界。

CV Max

集合变量的边界最大值，有多个集合变量时用"//"进行分割，无默认值时即不考虑边界，此时计算量会增加，强烈建议设置边界。

CV Space

集合变量的窗口大小，有多个集合变量时用"//"进行分割，默认等于metad_width的1/5

CV Bin

集合变量的窗口数量，有多个集合变量时用"//"进行分割，默认等于150，CV Space和CV Bin的相乘等于CV Max和CV Min的差值，因此当CV Space和CV Bin同时设置时以对应窗口数最多的为准

Adaptive

是否考虑施加自适应沉积函数, geom或者diff，默认为不填，即不考虑自适应。

Sigma Min

施加的自适应高斯函数的宽度或者标准差的最小值，有多个集合变量时用"//"进行分割，默认等于0。

Sigma Max

施加的自适应高斯函数的宽度或者标准差的最大值，有多个集合变量时用"//"进行分割，默认等于0。

Reweight

是否考虑重加权以获得重加权因子，对获得归一化偏势，yes或者no，默认no，即不考虑重加权，一般在体系收敛后才考虑重加权。

Reweight Ngauss

计算重加权因子时施加的高斯函数的个数，默认等于50。

Reweight Bin

计算重加权因子时集合变量的窗口数量，其值不能小于CV Bin的值，有多个集合变量时用"//"进行分割，默认等于CV Bin。

Well Tempered

是否考虑回火metadynamics模拟，yes或者no。

Temperature

回火metadynamics模拟时对应的基础温度，默认等于300K

Bias Factor

回火Metadynamics模拟时对应的偏置因子，其值等于(T+deltaT)/T，默认等于1，此时未进行偏置模拟，若进行偏置模拟，偏置因子应大于1

TAU

回火Metadynamics模拟时对应的施加的沉积高斯函数的高度，Height=kbDeltaTFrequency*TimeStep/TAU，默认等于0，即直接使用设置的沉积函数的高度代替。

Step

Metadynamics模拟时指定的输出步长，默认100。

Gauss File

Metadynamics模拟时指定的沉积高斯函数的输出文件名。

CV File

Metadynamics模拟时指定的集合变量的输出文件名。

PLUMED Index File

Metadynamics模拟时指定的CV Group的输出文件名，该文件中包含所有的CV Group的原子组，用于下一步Metadynamics的输入文件。

PLUMED Data File

Metadynamics模拟时指定的参数的输出文件名，该文件中包含计算时所需的参数，用于下一步Metadynamics的输入文件。

结果说明

输出结果包括：

上述两个生成的文件将作为下一步metadynamics模拟的输入文件。

GMX Metadynamics Generation

Introduction

The GMX Metadynamics Generation module is used to generate input files for Metadynamics simulations.

Parameter

GRO File

Submit the gro file of the simulation system. This file can be obtained from the MD Solvation module.

PBC

Whether to consider periodic boundary conditions during the Metadynamics simulation phase, yes or no.

CV Group1

Atoms included in the first group that makes up the collective variable (CV).

CV Group2

Atoms included in the second group that makes up the collective variable (CV).

CV Group3

Atoms included in the third group that makes up the collective variable (CV).

CV Group4

Atoms included in the fourth group that makes up the collective variable (CV).
Note:

• Group1 and Group2 form the DISTANCE collective variable, Group1, Group2, and Group3 form the ANGLE collective variable, and Group1, Group2, Group3, and Group4 form the TORSION collective variable.
• The notation for Groups: a5 represents the 5th atom in the GRO file, a5-10 represents atoms 5 to 10 in the GRO file, aCA represents the atom named CA in the GRO file. Similarly, r5, r5-10, and rASP represent the 5th residue, residues 5 to 10, and the residue named ASP in the GRO file, respectively. Some special characters like Protein, Protein-H, MainChain, etc., can also be used and can be combined, separated by commas. For example, “a5-10,r5,r8-10,UNK” represents atoms 5 to 10, the 5th residue, residues 8 to 10, and a molecule named UNK in the GRO file.
• Handling multiple CVs: If you want to define multiple collective variables, separate the corresponding atom groups for different collective variables in the Group definition using “//”. For example, a5//r5-10 indicates that a5 corresponds to the atom group for the first collective variable, and r5-10 corresponds to the second collective variable. If there is no corresponding atom group for a collective variable in a Group, use “none” to indicate this. For instance, if the first collective variable is DISTANCE and the second is ANGLE, and the first DISTANCE collective variable has no corresponding atom group in Group3, you can write none//r5-10 in Group3 to indicate that the first collective variable has no corresponding atom group, while the second collective variable corresponds to r5-10 in Group3.

Component

The components corresponding to the DISTANCE collective variable, which can be x, y, z, and xyz, representing calculations of DISTANCE considering only the x, y, z dimensions or all three dimensions, respectively. Use “//” to separate multiple collective variable components.

Metad Height

The height of the deposited Gaussian function, default is 1.0.

Metad Width

The width or standard deviation of the deposited Gaussian function. Use “//” to separate multiple collective variable widths, default is 0.05.

Metad Frequency

The frequency of depositing the Gaussian function, default is 500, meaning a Gaussian function deposition occurs every 500 time steps.

CV Min

The minimum boundary value of the collective variable. Use “//” to separate multiple collective variable minimums. If there is no default value, the boundary will not be considered, which will increase the computational load. It is strongly recommended to set boundaries.

CV Max

The maximum boundary value of the collective variable. Use “//” to separate multiple collective variable maximums. If there is no default value, the boundary will not be considered, which will increase the computational load. It is strongly recommended to set boundaries.

CV Space

The window size of the collective variable. Use “//” to separate multiple collective variable window sizes, default is 1/5 of metad_width.

CV Bin

The number of windows for the collective variable. Use “//” to separate multiple collective variable bin counts, default is 150. The product of CV Space and CV Bin equals the difference between CV Max and CV Min. Therefore, when both CV Space and CV Bin are set, the one with the highest corresponding window count will prevail.

Adaptive

Whether to consider applying an adaptive deposition function, geom or diff, default is not filled, which means adaptive deposition is not considered.

Sigma Min

The minimum width or standard deviation of the applied adaptive Gaussian function. Use “//” to separate multiple collective variable minimums, default is 0.

Sigma Max

The maximum width or standard deviation of the applied adaptive Gaussian function. Use “//” to separate multiple collective variable maximums, default is 0.

Reweight

Whether to consider reweighting to obtain the reweighting factor for normalization of the bias potential, yes or no, default is no, which means reweighting is not considered. Reweighting is generally considered only after the system has converged.

Reweight Ngauss

The number of Gaussian functions applied when calculating the reweighting factor, default is 50.

Reweight Bin

The number of windows for the collective variable when calculating the reweighting factor, which cannot be less than the value of CV Bin. Use “//” to separate multiple collective variable bin counts, default is equal to CV Bin.

Well Tempered

Whether to consider simulated annealing in the Metadynamics simulation, yes or no.

Temperature

The base temperature corresponding to the simulated annealing Metadynamics simulation, default is 300K.

Bias Factor

The bias factor corresponding to the simulated annealing Metadynamics simulation, which equals (T + deltaT) / T, default is 1, meaning no bias simulation is performed. If a bias simulation is performed, the bias factor should be greater than 1.

TAU

The height of the deposited Gaussian function applied during the simulated annealing Metadynamics simulation, Height = kb * DeltaT * Frequency * TimeStep / TAU, default is 0, meaning the set deposition function height is used directly.

Step

The specified output step length during the Metadynamics simulation, default is 100.

Gauss File

The output file name for the deposited Gaussian function during the Metadynamics simulation.

CV File

The output file name for the collective variable during the Metadynamics simulation.

PLUMED Index File

The output file name for the CV Group during the Metadynamics simulation, which contains all the atom groups of the CV Group for the next step’s Metadynamics input file.

PLUMED Data File

The output file name for the parameters during the Metadynamics simulation, which contains the parameters required for calculations for the next step’s Metadynamics input file.

Result

The output results include:

The two generated files above will serve as input files for the next step of the Metadynamics simulation.

Free Energy Surface Analysis

简介

Free Energy Surface Analysis模块是对基于PLUMED元动力学模拟后得到的模拟结果进行自由能计算。

参数说明

Input File

基于PLUMED元动力学模拟后输出的沉积高斯函数文件，默认为HILLS.dat文件。

Histogram

对沉积高斯函数文件进行自由能计算时是否考虑直方图分布方法，yes或者no，默认no。

Sigma

当考虑直方图分布方法时高斯函数的宽度值，有多个集合变量(即CV)时用"//"进行分割，比如0.35//0.35。只有当Histogram值为no时Sigma参数才会生效，当有多个CV而只设置了一个宽度值时，则表示该宽度值适用于所有CV。默认0.05。

CV Name

CV名称，对沉积高斯函数文件进行自由能计算时只考虑该指定的CV。当不指定CV时则考虑沉积高斯函数文件中包含的所有CV，当指定CV时则不能考虑直方图分布方法。

CV Min

集合变量的边界最小值，有多个集合变量时用"//"进行分割，比如0.1//0.3，强烈建议设置边界。当有多个CV而只设置了一个边界最小值时，则表示该最小值适用于所有CV。

CV Max

集合变量的边界最大值，有多个集合变量时用"//"进行分割，比如0.1//0.3，强烈建议设置边界。当有多个CV而只设置了一个边界最大值时，则表示该最大值适用于所有CV。

Grid Size

集合变量的窗口大小，有多个集合变量时用"//"进行分割，比如0.1//0.3。仅当设置了CV Min和CV Max值时，Grid Size才会生效。当有多个CV而只设置了一个窗口大小值时，则表示该窗口大小值适用于所有CV。

Bin

集合变量的窗口数量，有多个集合变量时用"//"进行分割，比如150//300。仅当设置了CV Min和CV Max值时，Bin才会生效。当有多个CV而只设置了一个窗口数量值时，则表示该窗口数量值适用于所有CV。Grid Size和Bin相乘等于CV Max和CV Min的差值，因此当Grid Size和Bin同时设置时以对应窗口数最多的为准。

Temperature

温度，对沉积高斯函数文件进行自由能计算时使用的温度值，默认300K

Min to Zero

是否对输出的自由能数据进行归零处理，即将自由能数据进行相对移动以保证最小值移动到0的位置，yes或者no，默认no。

Stride

沉积高斯函数的数量，在对沉积高斯函数文件进行自由能计算时，每隔该指定的沉积高斯函数的数量进行一次自由能计算。当不设置该数量值时表示对所有的沉积高斯函数在整体上只进行一次自由能计算。

Output File

输出结果文件，文件中包含随CV变化的自由能数据，默认为FES.csv文件。当指定了Stride值时，默认文件为FES.dat.tar.gz。

结果说明

输出结果包括：

Free Energy Surface Analysis

Introduction

The Free Energy Surface Analysis module is used to to calculate the free energy based on the simulation results outputed from the metadynamics simulations.

Parameter

Input File

The deposited Gaussian function file outputed from the metadymamics simulations. Default “HILLS.dat”.

Histogram

Whether considers the Historgram method when calculates the free energy based on the deposited Gaussian function file. “yes” or “no”, default “no”.

Sigma

Width of Gaussian Function used by the Historgram method, if there are multiple CVs, you can separated them by “//”, such as 0.35//0.35. Only effective when Historgram method is used. When there are multiple CVs and only one width value is set, it means that the width value will be applied to all CVs. Default 0.05.

CV Name

The specified CV considered in the free energy calculation based on the deposited Gaussian function file. When CV is not specified, all CVs contained in the deposited Gaussian function file will be considered, and when CV is specified, histogram distribution methods cannot be considered.

CV Min

The minimum boundary value of the collective variable. Use “//” to separate multiple collective variable minimums, such as 0.1//0.3. It is strongly recommended to set boundaries. When there are multiple CVs and only one minimum value is set, it means that the minimum value will be applied to all CVs.

CV Max

The maximum boundary value of the collective variable. Use “//” to separate multiple collective variable maximums, such as 0.1//0.3. It is strongly recommended to set boundaries. When there are multiple CVs and only one maximum value is set, it means that the maximum value will be applied to all CVs.

Grid Size

The window size of the collective variable. Use “//” to separate multiple collective variable window sizes, such as 0.1//0.3. Only effective when CV Min and CV Max values are set. When there are multiple CVs and only one window size value is set, it means that the window size value will be applied to all CVs.

Bin

The window number of the collective variable. Use “//” to separate multiple collective variable bin counts, such as 150//300. Only effective when CV Min and CV Max values are set. When there are multiple CVs and only one window number value is set, it means that the window number value will be applied to all CVs.The product of Grid Size and Bin equals the difference between CV Max and CV Min. Therefore, when both CV Space and CV Bin are set, the one with the highest corresponding window count will prevail.

Temperature

The temperature value used in the free energy calculation based on the deposited Gaussian function file. Default 300K.

Min to Zero

Whether mintozeros the obatined free energy data calculated based on the deposited Gaussian function file. “yes” or “no”, default “no”.

Stride

Specified number of the deposition Gauss function. When calculates the free energy based on the deposition Gauss function file, the free energy will be calculated every specified number of the deposition Gauss function. When this stride value is not set, it means that only one free energy calculation is performed for all deposition Gaussian functions as a whole.

Output File

The specified output file. The output file contains free energy data that varies with CV. Default FES.csv file. When the Stride value is specified, default FES.dat.tar.gz file.

Result

The output results include:

MD Clustering

简介

MD Clustering是对动力学轨迹进行归簇分析。

参数说明

Path File

MD模拟后得到的路径文件，可以在GMX MD Run (GMX2023)模块或者AlphaAutoMD (GMX2023)模块中获取。

Cutoff

聚类时结构的RMSD截断值(nm)

Cluster Method

聚类算法：linkage, jarvis-patrick, monte-carlo, diagonalization, gromos， 默认使用gromos算法。

System Group

选择需要计算的结构组别：Backbone，Protein，DNA，RNA。
可以根据PDB中小分子的名称填写组别名称。

Custom Resid

自定义需要计算的残基编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10，15

Custom Atom

自定义需要计算的原子编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10，15

Skip Time (ns)

每一帧的间隔时间（单位ns）

结果说明

输出结果包括：

MD Clustering

Introduction

MD Clustering is a clustering analysis of molecular dynamics trajectories.

Parameter Description

Path File

Path file obtained after MD simulation, can be obtained from the GMX MD Run (GMX2023) module or the AlphaAutoMD (GMX2023) module.

Cutoff

RMSD cutoff value for clustering (in nm).

Cluster Method

Clustering algorithm: linkage, jarvis-patrick, monte-carlo, diagonalization, gromos. The default method is gromos.

System Group

Select the structural group for calculation: Backbone, Protein, DNA, RNA. You can specify group names based on the names of small molecules in the PDB file.

Custom Resid

Custom residue numbers for calculation. Use “-” for continuous residues and commas to separate non-continuous residues. For example: 1-10, 15.

Custom Atom

Custom atom numbers for calculation. Use “-” for continuous atoms and commas to separate non-continuous atoms. For example: 1-10, 15.

Skip Time (ns)

Time interval between frames (in ns).

Result Description

The output results include:

MD Hbond

简介

MD Hbond模板对于指定组别之间的氢键分析。

参数说明

Path File

MD模拟后得到的路径文件，可以在GMX MD Run (GMX2023)模块或者AlphaAutoMD (GMX2023)模块中获取。

System Group1

选择需要计算的氢键组别1：Protein，DNA，RNA。如果两个组相同则分析的是组内氢键。
可以根据PDB中小分子的名称填写组别名称。

System Group2

选择需要计算的氢键组别2：Protein，DNA，RNA。如果两个组相同则分析的是组内氢键。
可以根据PDB中小分子的名称填写组别名称。

Custom Resid1

自定义需要计算的组1残基编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10,15

Custom Atom1

自定义需要计算的组1原子编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10,15

Custom Resid2

自定义需要计算的组2残基编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10,15

Custom Atom2

自定义需要计算的组2原子编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10,15

Skip Time (ns)

每一帧的间隔时间（单位ns）

结果说明

输出结果包括：

其中hbnum.csv包括信息如下：

MD Hbond

Introduction

MD Hbond template is used for analyzing hydrogen bonds between specified groups.

Parameter Description

Path File

Path file obtained after MD simulation, can be obtained from the GMX MD Run (GMX2023) module or the AlphaAutoMD (GMX2023) module.

System Group1

Select the hydrogen bond group 1 for calculation: Protein, DNA, RNA. If both groups are the same, it analyzes intra-group hydrogen bonds. You can specify group names based on the names of small molecules in the PDB file.

System Group2

Select the hydrogen bond group 2 for calculation: Protein, DNA, RNA. If both groups are the same, it analyzes intra-group hydrogen bonds. You can specify group names based on the names of small molecules in the PDB file.

Custom Resid1

Custom residue numbers for group 1 calculation. Use “-” for continuous residues and commas to separate non-continuous residues. For example: 1-10, 15.

Custom Atom1

Custom atom numbers for group 1 calculation. Use “-” for continuous atoms and commas to separate non-continuous atoms. For example: 1-10, 15.

Custom Resid2

Custom residue numbers for group 2 calculation. Use “-” for continuous residues and commas to separate non-continuous residues. For example: 1-10, 15.

Custom Atom2

Custom atom numbers for group 2 calculation. Use “-” for continuous atoms and commas to separate non-continuous atoms. For example: 1-10, 15.

Skip Time (ns)

Time interval between frames (in ns).

Result Description

The output results include:

The hbnum.csv file includes the following information:

MD Trajectory

简介

可根据起始帧数、结束帧数以及间隔帧数对平衡模拟进行轨迹提取，从而将其转换为GRO或者PDB轨迹文件。

参数说明

Path File

MD模拟后得到的路径文件，可以在GMX MD Run模块或者AlphaAutoMD模块中获取。

Type

文件输出类型：GRO或者PDB。

Water

输出文件是否保留水盒子。

Start Time (ps)

起始位置（单位ps）。

End Time (ps)

结束位置（单位ps）。

Skip Time (ps)

间隔时间，单位ps。

Index File

索引文件，ndx格式。对于膜体系的轨迹提取是必填项。

Keep Heterogens

是否保留体系中的溶剂（Water以及Ion）：不保留（none），都保留（all），指定保留溶剂范围（specify）。

Specify Heterogens

指定需要保留的特殊组别如：水（Water），离子（Ion）；或者指定保留组别的范围，规定格式为：需要保留的溶剂组别（Water或者Ion）:限定距离（单位Å）:目标组别，中间使用冒号（:）进行分隔，例如Water:3:ligand。
注：组别名称可以通过MD Solvation模块的index文件查询；若目标组别是小分子，可以根据PDB中小分子的名称填写组别名称，多个小分子可填写ligand表示。

结果说明

输出结果包括：

参考文献

GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, E. Lindahl, SoftwareX, 1, (2015), 19-25.

MD Trajectory

Introduction

The MD Trajectory module allows for the extraction of trajectories from equilibrium simulations based on the starting frame number, ending frame number, and frame interval, converting them into GRO or PDB trajectory files.

Parameter

Path File

Path file obtained after MD simulation, can be obtained from the GMX MD Run module or the AlphaAutoMD module.

Type

File output type: GRO or PDB.

Water

Whether to retain the water box in the output files.

Start Time (ps)

Starting time (in ps).

End Time (ps)

Ending time (in ps).

Skip Time (ps)

Time interval, in ps.

Index File

Index file in ndx format. This is a required parameter for extracting trajectories in membrane systems.

Keep Heterogens

Whether to retain the solvents in the system (Water and Ion) : none (none), all (all), specify the solvent range (specify).

Specify Heterogens

Specify special groups to be retained: Water, Ion; Or specify the range of reserved groups in the format: solvent group to be retained (Water or Ion) : limit distance (unit Å) : target group, separated by a colon (:), e.g., Water:3:ligand.
Note: The group name can be queried through the index file of the MD Solvation module. If the target group is a small molecule, the group name can be filled in according to the name of small molecule in PDB, and the ligand representation can be filled in for multiple small molecules.

Result

The output results include:

Reference

GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, E. Lindahl, SoftwareX, 1, (2015), 19-25.

MD Gyration

简介

MD Gyration回旋半径分析，可用来衡量体系模拟时的质权平均半径。

参数说明

Path File

MD模拟后得到的路径文件，可以在GMX MD Run (GMX2023)模块或者AlphaAutoMD (GMX2023)模块中获取。

System Group

选择需要计算的结构组别：Backbone，Protein，DNA，RNA。
可以根据PDB中小分子的名称填写组别名称。

Custom Resid

自定义需要计算的残基编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10，15

Custom Atom

自定义需要计算的原子编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10，15。

Skip Time (ns)

每一帧的间隔时间（单位ns）

Index File

索引文件，格式为ndx

结果说明

输出结果包括：

其中gyrate.csv包括信息如下：

MD Gyration

Introduction

MD Gyration is a radius of gyration analysis used to measure the mass-weighted average radius of a system during simulation.

Parameter Description

Path File

Path file obtained after MD simulation, can be obtained from the GMX MD Run (GMX2023) module or the AlphaAutoMD (GMX2023) module.

System Group

Select the structural group for calculation: Backbone, Protein, DNA, RNA. You can specify group names based on the names of small molecules in the PDB file.

Custom Resid

Custom residue numbers for calculation. Use “-” for continuous residues and commas to separate non-continuous residues. For example: 1-10, 15.

Custom Atom

Custom atom numbers for calculation. Use “-” for continuous atoms and commas to separate non-continuous atoms. For example: 1-10, 15.

Skip Time (ns)

Time interval between frames (in ns).

Index File

Index file in ndx format.

Result Description

The output results include:

The gyrate.csv file includes the following information:

MD SASA

简介

MD SASA模块是计算指定组别的溶剂可及表面积（solvent accessible surface area，SASA）。

参数说明

Path File

MD模拟后得到的路径文件，可以在GMX MD Run (GMX2023)模块或者AlphaAutoMD (GMX2023)模块中获取。

System Group

选择需要计算的结构组别：Backbone，Protein，DNA，RNA。
可以根据PDB中小分子的名称填写组别名称。

Custom Resid

自定义需要计算的残基编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10，15

Custom Atom

自定义需要计算的原子编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10，15。

Skip Time (ns)

每一帧的间隔时间（单位ns）

Index File

索引文件，格式为ndx

结果说明

输出结果包括：

其中area.csv包括信息如下：

MD SASA

Introduction

The MD SASA module calculates the solvent accessible surface area (SASA) of specified groups.

Parameter Description

Path File

Path file obtained after MD simulation, can be obtained from the GMX MD Run (GMX2023) module or the AlphaAutoMD (GMX2023) module.

System Group

Select the structural group for calculation: Backbone, Protein, DNA, RNA. You can specify group names based on the names of small molecules in the PDB file.

Custom Resid

Custom residue numbers for calculation. Use “-” for continuous residues and commas to separate non-continuous residues. For example: 1-10, 15.

Custom Atom

Custom atom numbers for calculation. Use “-” for continuous atoms and commas to separate non-continuous atoms. For example: 1-10, 15.

Skip Time (ns)

Time interval between frames (in ns).

Index File

Index file in ndx format.

Result Description

The output results include:

The area.csv file includes the following information:

MD Distance

简介

MD Distance是针对分子动力学轨迹的距离分析模块，输出两个组之间距离 (质心距离或几何中心距离) 随时间的变化情况。自定义组别时需要注意，如果只需测量两个原子之间的距离则填写Custom Atom1和Custom Atom2即可；当同时填写Custom Resid1和Custom Atom1时，组别1的原子数是Custom Atom1与Custom Resid1交集的原子。

参数说明

Path File

MD模拟后得到的路径文件，可以在GMX MD Run (GMX2023)模块或者AlphaAutoMD (GMX2023)模块中获取。

System Group1

选择需要计算的组别1：Protein，DNA，RNA。
可以根据PDB中小分子的名称填写组别名称。

System Group2

选择需要计算的组别1：Protein，DNA，RNA。
可以根据PDB中小分子的名称填写组别名称。

Custom Resid1

自定义需要计算的组1残基编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10,15。自定义组别之间是并集。

Custom Atom1

自定义需要计算的组1原子编号，连续参数可用“-”表示，不连续原子用逗号隔开，例如：1-10,15。自定义组别之间是并集。

Custom Resid2

自定义需要计算的组2残基编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10,15。自定义组别之间是并集。

Custom Atom2

自定义需要计算的组2原子编号，连续参数可用“-”表示，不连续原子用逗号隔开，例如：1-10,15。自定义组别之间是并集。

Skip Time (ns)

每一帧的间隔时间（单位ns）。

结果说明

输出结果包括：

其中dist.csv包括信息如下：

MD Distance

Introduction

MD Distance is a distance analysis module for molecular dynamics trajectories, providing the variation of distance (center-of-mass distance or geometric center distance) between two groups over time. When defining custom groups, it is important to note that if you only need to measure the distance between two atoms, you can fill in Custom Atom1 and Custom Atom2. When both Custom Resid1 and Custom Atom1 are filled in, the number of atoms in group 1 is the intersection of Custom Atom1 and Custom Resid1.

Parameter Description

Path File

Path file obtained after MD simulation, available in the GMX MD Run (GMX2023) module or AlphaAutoMD (GMX2023) module.

System Group1

Select the group 1 for calculation: Protein, DNA, RNA.
You can enter the group name based on the name of the small molecule in the PDB.

System Group2

Select the group 2 for calculation: Protein, DNA, RNA.
You can enter the group name based on the name of the small molecule in the PDB.

Custom Resid1

Custom residue numbers for group 1 that need to be calculated. Use “-” for continuous parameters and separate non-continuous residues with commas, for example: 1-10,15. Custom groups are unions of specified residues.

Custom Atom1

Custom atom numbers for group 1 that need to be calculated. Use “-” for continuous parameters and separate non-continuous atoms with commas, for example: 1-10,15. Custom groups are unions of specified atoms.

Custom Resid2

Custom residue numbers for group 2 that need to be calculated. Use “-” for continuous parameters and separate non-continuous residues with commas, for example: 1-10,15. Custom groups are unions of specified residues.

Custom Atom2

Custom atom numbers for group 2 that need to be calculated. Use “-” for continuous parameters and separate non-continuous atoms with commas, for example: 1-10,15. Custom groups are unions of specified atoms.

Skip Time (ns)

Time interval for each frame (in ns).

Result Description

The output includes:

The dist.csv file includes the following information:

MMPBSA

简介

MMPBSA计算受体与配体之间的结合自由能，并且提供能量分解数据、结合常数（Ka）、抑制剂常数（Ki）。熵的计算采用的是张增辉教授的相互作用熵的方法，该方法直接从分子动力学模拟计算结合自由能的熵组分（相互作用熵或-TΔS），但是需要选取结构稳定部分进行计算。
本模块提供四种计算方法，其中 Trajectory 是针对进行分子动力学模拟轨迹后计算其受配体之间的结合自由能；One Structure 是针对一帧pdb结构或者对接得到的PDB结构计算受配体之间的结合自由能，MMPBSA of One Structure 计算流程中可以直接输入PDB进行计算。
Index Name 是输入受配体组别名称；Custom Name 则是输入受配体的在PDB中的残基编号。

参数说明

Trajectory方法

Path File

MD模拟后得到的路径文件，可以在GMX MD Run (GMX2023)模块或者AlphaAutoMD模块中获取。

Receptor Name

受体名称，可以为Protein、DNA、RNA。

Ligand Name

配体名称，可以为Protein、DNA、RNA。如果为小分子，填写其在PDB中的名称。如果体系中除了蛋白以外为配体（包括小分子）可用Other表示。

Start Time (ps)

起始帧时间，单位ps。最好选取RMSD稳定时区进行计算，以消除结构不稳定时导致的整体熵偏大。

End Time (ps)

结束帧时间，单位ps。最好选取RMSD稳定时区进行计算，以消除结构不稳定时导致的整体熵偏大。

Skip Time (ps)

间隔时间，单位ps。

Index File

索引文件，ndx格式。当体系中存在膜结构的时候可以通过Membrane Solvation模块获取index.ndx文件。膜模拟中的受体为receptor，配体为ligand，膜为membrane。

Custom Receptor

定义受体组别进行结合能计算。组别中填写的为蛋白氨基酸或者核酸碱基的序号。例如1-213或者1-211,212-213。蛋白氨基酸或者核酸碱基序号从1开始重新编号，与初始pdb氨基酸编号无关。

Custom Ligand

定义配体组别进行结合能计算。组别中填写的为蛋白氨基酸或者核酸碱基的序号。例如1-213或者1-211,212-213。蛋白氨基酸或者核酸碱基序号从1开始重新编号，与初始pdb氨基酸编号无关。

One Structure方法

System Topology

拓扑文件，由MD Solvation模块或者Membrane Solvation模块得到。

System GRO

结构文件，.gro格式，由MD Solvation模块或者Membrane Solvation模块得到。

System ITP

体系参数压缩文件，tar.gz格式。由MD Solvation模块或者Membrane Solvation模块得到。

结果说明

输出结果包括：

参考文献

GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, E. Lindahl, SoftwareX, 1, (2015), 19-25.
Duan L, Liu X, Zhang JZ. Interaction Entropy: A New Paradigm for Highly Efficient and Reliable Computation of Protein-Ligand Binding Free Energy. J Am Chem Soc. 2016 May 4;138(17):5722-8.
https://github.com/Electrostatics/apbs/blob/main/docs/index.rst

MMPBSA

Introduction

MMPBSA calculates the binding free energy between a receptor and a ligand, providing energy decomposition data, binding constants (Ka), and inhibitor constants (Ki). The entropy calculation uses the method of interaction entropy proposed by Professor Zhang Zenghui, which directly calculates the entropy component of the binding free energy (interaction entropy or -TΔS) from molecular dynamics simulations, but requires selecting stable structural regions for calculation. This module provides four calculation methods. Among them, Trajectory calculates the binding free energy between the receptor and ligand after molecular dynamics simulation trajectories; One Structure calculates the binding free energy between the receptor and ligand for a single frame PDB structure or a docked PDB structure; MMPBSA of One Structure allows direct input of a PDB for calculation. Index Name is the input receptor-ligand group name, while Custom Name is the input residue numbers of the receptor-ligand in the PDB.

Parameter Description

Trajectory Method

Path File

Path file obtained after MD simulation, available in the GMX MD Run (GMX2023) module or AlphaAutoMD module.

Receptor Name

Name of the receptor, can be Protein, DNA, or RNA.

Ligand Name

Name of the ligand, can be Protein, DNA, or RNA. If it is a small molecule, enter its name in the PDB. Use ‘Other’ if the system contains a ligand (including small molecules) other than proteins.

Start Time (ps)

Start frame time in ps. It is best to select a stable RMSD region for calculation to eliminate large overall entropy caused by structural instability.

End Time (ps)

End frame time in ps. It is best to select a stable RMSD region for calculation to eliminate large overall entropy caused by structural instability.

Skip Time (ps)

Time interval in ps.

Index File

Index file in ndx format. When there is a membrane structure in the system, you can obtain the index.ndx file through the Membrane Solvation module. In membrane simulations, the receptor is ‘receptor’, the ligand is ‘ligand’, and the membrane is ‘membrane’.

Custom Receptor

Define receptor groups for binding energy calculation. Enter the sequence numbers of protein amino acids or nucleic acid bases in the group. For example, 1-213 or 1-211,212-213. Protein amino acid or nucleic acid base numbers start from 1 and are independent of the initial pdb amino acid numbering.

Custom Ligand

Define ligand groups for binding energy calculation. Enter the sequence numbers of protein amino acids or nucleic acid bases in the group. For example, 1-213 or 1-211,212-213. Protein amino acid or nucleic acid base numbers start from 1 and are independent of the initial pdb amino acid numbering.

One Structure Method

System Topology

Topology file obtained from the MD Solvation module or Membrane Solvation module.

System GRO

Structure file in .gro format obtained from the MD Solvation module or Membrane Solvation module.

System ITP

System parameter compression file in tar.gz format. Obtained from the MD Solvation module or Membrane Solvation module.

Result Description

The output includes:

References

GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, E. Lindahl, SoftwareX, 1, (2015), 19-25.
Duan L, Liu X, Zhang JZ. Interaction Entropy: A New Paradigm for Highly Efficient and Reliable Computation of Protein-Ligand Binding Free Energy. J Am Chem Soc. 2016 May 4;138(17):5722-8.
https://github.com/Electrostatics/apbs/blob/main/docs/index.rst

RMS

简介

通过计算平衡模拟轨迹的均方根偏差（RMSD，Root Mean Square Deviation）和均方根波动（RMSF，Root Mean Square Fluctuation），从而分析结构的稳定性和结构变化情况。

参数说明

Path File

MD模拟后得到的路径文件，可以在**GMX MD Run (GMX2023)模块或者AlphaAutoMD (GMX2023)**模块中获取。

Analysis Type

选择分析类型：RMSD或者RMSF（可多选）。

System Group

选择需要计算的组别。

Custom Resid

自定义需要计算的残基编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10,15。

Custom Atom

自定义需要计算的原子编号，用逗号隔开，例如：CA,O,H。与Custom Resid是交集关系。

Skip Time (ps)

Index File

索引文件，可由Membrane Solvation模块得到。

结果说明

输出结果包括：

RMS

Introduction

By calculating the Root Mean Square Deviation (RMSD) and Root Mean Square Fluctuation (RMSF) of equilibrium simulation trajectories, the stability and structural changes of the system can be analyzed.

Parameter Description

Path File

The path file obtained after MD simulation, which can be obtained from the GMX MD Run (GMX2023) module or the AlphaAutoMD (GMX2023) module.

Analysis Type

Select the type of analysis: RMSD or RMSF (multiple selections possible).

System Group

Select the group to be calculated.

Custom Resid

Custom residue numbers to be calculated, continuous parameters can be represented by “-”, non-continuous residues are separated by commas, for example: 1-10,15.

Custom Atom

Custom atom numbers to be calculated, separated by commas, for example: CA, O, H. This intersects with Custom Resid.

Skip Time (ps)

Index File

Index file obtained from the Membrane Solvation module.

Result Description

The output results include:

MD PCA

简介

N个原子的柔性大体系如蛋白，其运动轨迹需要3N维笛卡尔坐标来描述，这样高维的数据很难理解和直观分析。MD PCA（Principal component analysis，PCA）模块可以从高维数据中分析出主要的影响因素 (本征向量) 。前几个本征向量（主成分，如前两个主成份则为 PC1，PC2) 一般可以描述分子运动的大部分信息。

参数说明

Path File

MD模拟后得到的路径文件，可以在GMX MD Run (GMX2023)模块或者AlphaAutoMD (GMX2023)模块中获取。

System Group

选择需要计算的结构组别：Backbone，Protein，DNA，RNA。
可以根据PDB中小分子的名称填写组别名称。

Custom Resid

自定义需要计算的残基编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10，15

Custom Atom

自定义需要计算的原子编号，连续参数可用“-”表示，不连续残基用逗号隔开，例如：1-10，15。

Skip Time (ns)

每一帧的间隔时间（单位ns）

Index File

索引文件，格式为ndx

结果说明

得到结果文件，每种类型的文件如果包含PNG、CSV以及XVG后缀，相同名称只是表现形式不同，数据一样

MD PCA

Introduction

For a large flexible system with N atoms such as a protein, its motion trajectory requires 3N-dimensional Cartesian coordinates to describe, making it difficult to understand and analyze high-dimensional data. The MD PCA (Principal Component Analysis, PCA) module can analyze the main influencing factors (eigenvectors) from high-dimensional data. The first few eigenvectors (principal components, such as the first two principal components PC1, PC2) can generally describe most of the information about molecular motion.

Parameter Description

Path File

Path file obtained after MD simulation, can be obtained from the GMX MD Run (GMX2023) module or the AlphaAutoMD (GMX2023) module.

System Group

Select the structural group for calculation: Backbone, Protein, DNA, RNA. You can specify group names based on the names of small molecules in the PDB file.

Custom Resid

Custom residue numbers for calculation. Use “-” for continuous residues and commas to separate non-continuous residues. For example: 1-10, 15.

Custom Atom

Custom atom numbers for calculation. Use “-” for continuous atoms and commas to separate non-continuous atoms. For example: 1-10, 15.

Skip Time (ns)

Time interval between frames (in ns).

Index File

Index file in ndx format.

Result Description

Obtain the following result files. If the files have PNG, CSV, and XVG suffixes, they contain the same data but in different formats.

GMX MD Run (GMX2024)

简介

提交GROMACS对应文件，从而进行分子动力学模拟，得到平衡模拟后得到的轨迹文件。

参数说明

GRO File

提交模拟体系的gro文件。该文件可以从MD Solvation或者Membrane Solvation模块获取。

Topology File

提交模拟体系的top文件。该文件可以从MD Solvation或者Membrane Solvation模块获取。

ITP File

提交模拟体系的itp文件。该文件可以从MD Solvation或者Membrane Solvation模块获取。

Minimize MDP File

提交进行最小化的参数化文件，文件格式为mdp。可以根据GMX MDP Generation模块的Minimization方法或者**GMX MDP Generation (Auto)**生成。

NPT MDP File

提交进行等压等温的参数化文件，文件格式为mdp。可以根据GMX MDP Generation模块的NPT方法或者**GMX MDP Generation (Auto)**生成。

MD MDP File

提交进行平衡模拟的参数化文件，文件格式为mdp。可以根据GMX MDP Generation模块的MD方法或者**GMX MDP Generation (Auto)**生成。

结果说明

输出结果包括：

参考文献

GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, E. Lindahl, SoftwareX, 1, (2015), 19-25.

GMX MD Run (GMX2024)

Introduction

Submit corresponding files to GROMACS to perform molecular dynamics simulations and obtain trajectory files after equilibrium simulations.

Parameter Description

GRO File

Submit the gro file of the simulated system. This file can be obtained from the MD Solvation or Membrane Solvation modules.

Topology File

Submit the top file of the simulated system. This file can be obtained from the MD Solvation or Membrane Solvation modules.

ITP File

Submit the itp file of the simulated system. This file can be obtained from the MD Solvation or Membrane Solvation modules.

Minimize MDP File

Submit the script file for minimization, in mdp format. This file can be generated using the GMX MDP Generation module with the Minimization method or GMX MDP Generation (Auto).

NPT MDP File

Submit the script file for NPT (isothermal-isobaric) simulation, in mdp format. This file can be generated using the GMX MDP Generation module with the NPT method or GMX MDP Generation (Auto).

MD MDP File

Submit the script file for equilibrium simulation, in mdp format. This file can be generated using the GMX MDP Generation module with the MD method or GMX MDP Generation (Auto).

Result Description

The output results include:

Reference Literature

GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, E. Lindahl, SoftwareX, 1, (2015), 19-25.

Structure Prediction (Protenix)

简介

Protenix是字节跳动公司AML AI4Science团队复现的pytorch版本的AlphaFold3模型。以下是ByteDance AML AI4Science团队的主要贡献概要：
- 模型性能：将Protenix与现有的模型进行了基准测试。Protenix在不同分子类型的结构预测中表现出强大的性能。作为一个完全开源的模型，它使研究人员能够生成新的预测并对模型进行微调，以满足特定的应用需求。
- 方法：在复现过程中，依据AF3的描述实现了Protenix，并优化了一些模糊步骤，纠正了排版错误，并根据模型行为进行了有针对性的调整。通过分享复现经验，希望支持社区在这些改进的基础上进一步推动该领域的发展。
- 可访问性：已将Protenix开源，提供了模型权重、推理代码和可训练代码供研究用途。

参数说明

Protein Sequence

蛋白序列文件，FASTA格式，支持多条序列。
注意：多蛋白复合物结构预测，其氨基酸序列输入格式如下：

>protein1
AASJ...
>protein2
AASJ...
>peptide
ASDF...

DNA Sequence

DNA序列文件，FASTA格式，支持多条序列。

RNA Sequence

RNA序列文件，FASTA格式，支持多条序列。

备注：当前支持计算的残基/碱基数量在1400个左右。

Ligand

文本文件包含小分子信息，TXT格式。支持SMILES或 CCD Code（化学组分词典编号）。如果使用SMILES格式，每行应包含一个小分子；如果使用CCD Code，每行可以包含一个或多个小分子，使用逗号分隔，并加上CCD前缀。示例如下：

CC(=O)OC1C[NH+]2CCC1CC2
CCD,ATP,HY3,P1L
CCD,MG

注意：不适用于配体蛋白或多肽的氨基酸序列格式输入。

Modification

包含翻译后修饰（PTM）信息的文本文件，TXT格式。每行放置一个PTM信息，每个PTM信息由三部分组成：

• 发生PTM序列的顺序编号
• PTM类型的CCD编号
• 发生PTM的残基位置编号
  三部分由逗号分隔，例如：1,HY3,1 表示第一条序列的第一个残基，发生了类型为HY3（CCD编号，为3-羟基脯氨酸，为脯氨酸的羟基化）的PTM
  备注：
• 序列的顺序编号，是依次按上述参数Protein、DNA、RNA中的序列顺序与数量，从1开始进行编号，例如：当有2条蛋白序列，1条DNA序列，1条RNA序列时，各序列对应的编号为：第一条蛋白序列编号为1，第二条蛋白序列编号为2，DNA序列编号为3，RNA序列编号为4
• CCD的介绍参考https://www.wwpdb.org/data/ccd 编号查询网址为https://www.ebi.ac.uk/pdbe-srv/pdbechem/
  包含多个PTM信息的文件内容示例如下：

1,HY3,1
1,P1L,5
2,HY3,3

Covalent Bond

共价键信息的文本文件，TXT格式。每行放置一个共价键信息，每个共价键信息包含两个原子信息，每个原子信息由三部分组成：

• 原子所在序列或小分子的顺序编号（编号规则在Modification中定义的序列编号规则基础上，在最后加入小分子的顺序即可）
• 原子所在残基的位置编号（如残基为小分子时，编号为1）
• 原子的标准名称：
  • 默认是CCD中定义的原子标准名称
  • 如果配体是SMILES，则是SMILES字符串中原子对应的从0开始位置序号。

三部分由逗号分隔，例如：3,1,CA表示第三个实体（序列或小分子）中的第一个残基（或小分子）的CA原子
一个共价键是由两个原子信息组成，原子间用分号分隔，如：1,1,CA;2,1,CA
表示一个共价键，该共价键由两个原子组成，第一个原子为1,1,CA，第二个原子为2,1,CA
包含多个共价键信息的文件内容示例如下：

1,1,CA;2,1,CA
1,1,CA;3,1,CHA

Ion

离子名称，可以包含一个或多个离子，需写在一行文本中，不同的离子使用英文逗号分隔，支持输入离子数量，使用英文冒号分隔。示例如下：

MG:2,ZN,CU:3

表示2个MG离子，1个ZN离子，3个CU离子

Format

输出结构的格式，支持PDB或CIF格式，默认为PDB格式。

Enhanced Mode

该模式下，会默认使用1000个随机种子，每个随机种子进行5个结构采样，共进行5000个结构的大批量采样，并从中选择评分靠前的多个预测结构，最终获得更高精度的预测结构。该模式特别适用于抗原-抗体复合物结构的高精度预测，有研究表明该模式下抗体-抗原复合物结构预测准确性提升60%。该模式的输入参数与Single Mode一致，一次运行时间约10~20小时。

备注:

序列总长度不可超过1300。

结果说明

输出结果文件为排名前5的复合物结构rank_1-5.cif和pred_scores_protenix.csv，csv中包含信息如下：

参考文献

https://github.com/bytedance/Protenix

Structure Prediction (Protenix)

Introduction

Protenix is the PyTorch version of the AlphaFold3 model reproduced by the AML AI4Science team at ByteDance. Here is a summary of the main contributions from the ByteDance AML AI4Science team:
- Model Performance: Protenix has been benchmarked against existing models, demonstrating strong performance in structure prediction across different types of molecules. As a fully open-source model, it enables researchers to generate new predictions and fine-tune the model to meet specific application needs.
- Methodology: During the reproduction process, Protenix was implemented based on the description of AF3, optimizing some ambiguous steps, correcting typographical errors, and making targeted adjustments based on model behavior. By sharing our reproduction experience, we hope to support the community in further advancing the field based on these improvements.
- Accessibility: Protenix has been open-sourced, providing model weights, inference code, and training code for research purposes.

Parameter

Protein Sequence

A sequence file for proteins in FASTA format, supporting multiple sequences.
Note: For multi-protein complex structure prediction, the amino acid sequence input format is as follows:

>protein1
AASJ...
>protein2
AASJ...
>peptide
ASDF...

DNA Sequence

A sequence file for DNA nucleic acids in FASTA format, supporting multiple sequences.

RNA Sequence

A sequence file for RNA nucleic acids in FASTA format, supporting multiple sequences.

Note:The currently supported number of residues/bases for calculation is around 1,400.

Ligand

A text file containing information about small molecules in TXT format. It supports SMILES or CCD Code (Chemical Component Dictionary number). If using the SMILES format, each line should contain one small molecule; if using CCD Code, each line can contain one or more small molecules, separated by commas, and prefixed with CCD. Examples are as follows:

CC(=O)OC1C[NH+]2CCC1CC2
CCD,ATP,HY3,P1L
CCD,MG

Note: Not applicable for amino acid sequence format input of ligand proteins or polypeptides.

Modification

An optional parameter that includes a text file with post-translational modification (PTM) information in TXT format. Each line contains one PTM entry, which consists of three parts:

• The sequential number of the sequence where the PTM occurs
• The CCD number for the PTM type
• The position number of the residue where the PTM occurs
  These three parts are separated by commas. For example, 1,HY3,1 indicates that a PTM of type HY3 (CCD number, which is 3-hydroxyproline, a hydroxylation of proline) occurs at the first residue of the first sequence.
  Notes:
• The sequential number of the sequence is assigned based on the order and quantity of sequences in the parameters Protein, DNA, and RNA, starting from 1. For example, if there are 2 protein sequences, 1 DNA sequence, and 1 RNA sequence, the corresponding numbers are: the first protein sequence is numbered 1, the second protein sequence is numbered 2, the DNA sequence is numbered 3, and the RNA sequence is numbered 4.
• For an introduction to CCD, refer to https://www.wwpdb.org/data/ccd, and for the number query website, visit https://www.ebi.ac.uk/pdbe-srv/pdbechem/.
  An example of a file containing multiple PTM entries is as follows:

1,HY3,1
1,P1L,5
2,HY3,3

Covalent Bond

A text file containing covalent bond information in TXT format. Each line contains one covalent bond entry, which consists of two atom entries. Each atom entry consists of three parts:

• The sequential number of the sequence or small molecule containing the atom (the numbering rule is based on the sequence numbering defined in Modification, with the order of small molecules added at the end)
• The position number of the residue containing the atom (if the residue is a small molecule, the number is 1)
• Standard Names of Atoms:
  • By default, the standard names of atoms are defined in CCD (Chemical Component Dictionary).
  • If the ligand is represented as a SMILES string, the standard names correspond to the position indices of atoms in the SMILES string, starting from 0.

These three parts are separated by commas. For example, 3,1,CA indicates the CA atom of the first residue (or small molecule) in the third entity (sequence or small molecule).
A covalent bond is represented by two atom entries separated by a semicolon, such as: 1,1,CA;2,1,CA, indicating a covalent bond composed of two atoms, with the first atom being 1,1,CA and the second atom being 2,1,CA.
An example of a file containing multiple covalent bond entries is as follows:

1,1,CA;2,1,CA
1,1,CA;3,1,CHA

Ion

Ion names can include one or more ions, which should be written in a single line of text, with different ions separated by commas. It is also possible to specify the quantity of ions, using a colon to separate the ion name and its quantity. Examples are as follows:

MG:2,ZN,CU:3

Format

The output structure format supports PDB or CIF, with PDB format as the default.

Enhanced Mode

In this mode, a default of 1000 random seeds will be used, with each seed conducting 5 structural samplings, totaling 5000 structures for large-scale sampling. From these, multiple predicted structures with high scores will be selected to ultimately obtain a more accurate predicted structure. This mode is particularly suitable for high-precision prediction of antigen-antibody complex structures, and studies have shown that the accuracy of antibody-antigen complex structure prediction can be increased by 60% in this mode. The input parameters for this mode are consistent with those in Single Mode, and the runtime for one session is approximately 10 to 20 hours.

Note:

The total length of the sequence cannot exceed 1300.

Result

The output result files are the structures of the top 5 complexes, rank_1-5.cif and pred_scores_protenix.csv. The CSV file contains the following information:

Reference

https://github.com/bytedance/Protenix

Generate Humanized Variants

简介

抗体人源化设计中基于Grafting以及Back Mutation Grouping的结果批量生成人源化后的序列。

参数说明

Graft Policy

Grafting模块生成的Graft Policy文件，JSON格式

Mutate Policy

Back Mutation Grouping模块生成的组合突变的Policy文件（combination_mutate_policy.json），JSON格式

结果说明

输出人源化后的序列文件humanized_variants_esmfold.fasta，将轻重链的序列通过冒号:拼接成一条链，便于直接用于ESMFold模块进行批量结构预测。示例：

>L1H1
EIVLTQSPATLSLSPGERATLSCRASQFVGSSIHWYQQKPGQAPRLLIYYASESMSGIPARFSGSGSGTDFTLTISSLEPEDFAVYYCQQSHSWPFTFGQGTKLEIK:EVQLVESGGGLVQPGGSLRLSCAASGFIFSNHYMSWVRQAPGKGLEWVSVIRSKSINSATYYADSVKGRFTISRDNSKNTLYLQMNSLRAEDTAVYYCARNYYGSTYDYWGQGTTVTVSS
>L1H2
EIVLTQSPATLSLSPGERATLSCRASQFVGSSIHWYQQKPGQAPRLLIYYASESMSGIPARFSGSGSGTDFTLTISSLEPEDFAVYYCQQSHSWPFTFGQGTKLEIK:EVQLVESGGGLVQPGGSLRLSCAASGFIFSNHYMSWVRQAPGKGLEWVSEIRSKSINSATYYADSVKGRFTISRDNSKNTLYLQMNSLRAEDTAVYYCARNYYGSTYDYWGQGTTVTVSS

Generate Humanized Variants

Introduction

Generate humanized variant sequences based on the Grafting and Back Mutation Grouping results.

Parameters

Graft Policy

Graft policy file in JSON format generated by the Grafting module.

Mutate Policy

Combination mutate policy file generated by Back Mutation Grouping module in JSON format.

Results

The output file humanized_variants_esmfold.fasta in which sequences of the light and heavy chains are concatenated into a single chain using a colon (:). This format facilitates direct use in the ESMFold module for batch structural prediction.

>L1H1
EIVLTQSPATLSLSPGERATLSCRASQFVGSSIHWYQQKPGQAPRLLIYYASESMSGIPARFSGSGSGTDFTLTISSLEPEDFAVYYCQQSHSWPFTFGQGTKLEIK:EVQLVESGGGLVQPGGSLRLSCAASGFIFSNHYMSWVRQAPGKGLEWVSVIRSKSINSATYYADSVKGRFTISRDNSKNTLYLQMNSLRAEDTAVYYCARNYYGSTYDYWGQGTTVTVSS
>L1H2
EIVLTQSPATLSLSPGERATLSCRASQFVGSSIHWYQQKPGQAPRLLIYYASESMSGIPARFSGSGSGTDFTLTISSLEPEDFAVYYCQQSHSWPFTFGQGTKLEIK:EVQLVESGGGLVQPGGSLRLSCAASGFIFSNHYMSWVRQAPGKGLEWVSEIRSKSINSATYYADSVKGRFTISRDNSKNTLYLQMNSLRAEDTAVYYCARNYYGSTYDYWGQGTTVTVSS

Humanization Report v2.4

简介

Humanization Report v2.4是抗体人源化设计报告生成模块，用于生成最终的抗体人源化设计报告以及相应的专利实施例段落。相比v2.3，新增RMSD和能量信息。

参数说明

Graft Policy

Grafting模块生成的Graft Policy文件。

Mutate Policy

Back Mutation Grouping模块生成的Policy文件。

Antibody Type

抗体类型，Antibody 标准双链抗体，Nanobody 纳米抗体。

Germline Score File

Grafting模块生成的score文件，JSON格式

Mutation Score File

Mutation模块生成的score文件，CSV格式

Antibody RMSD File

抗体结构RMSD文件，由Antibody RMSD模块生成，CSV格式

Antibody RMSD Top

从RMSD排序中取前N个RMSD值小的抗体

Folding Stability File

Absolute Folding Stability模块预测生成的蛋白稳定性文件，CSV格式

结果说明

输出结果包括：

其中batch_registration_template.xlsx包含如下信息：

其中hotspot_summary.xlsx包含如下信息：

Humanization Report v2.4

Introduction

The Humanization Report v2.4 is a module for generating reports on antibody humanization design, including the final antibody humanization design report and corresponding paragraphs for patent implementation examples. Compared with v2.3, RMSD and energy information are added.

Parameter Description

Graft Policy

The Graft Policy file generated by the Grafting module.

Mutate Policy

The Policy file generated by the Back Mutation Grouping module.

Antibody Type

Antibody type, Antibody or Nanobody

Germline Score File

Graft germline score file in JSON format generated by the Grafting module

Mutation Score File

Mutation score file in csv format generated by the Mutation module

Antibody RMSD File

Antibody structure RMSD file generated by Antibody RMSD module

Antibody RMSD Top

Select the top N antibodies with the smallest RMSD values from the RMSD ranking

Folding Stability File

Protein folding stability file generated by Absolute Folding Stability module in CSV format

Result Description

The output results include:

The batch_registration_template.xlsx file contains the following information:

The hotspot_summary.xlsx file contains the following information:

Structure Prediction (HelixFold3)

简介

百度螺旋桨PaddleHelix团队研发的HelixFold3，在常规的小分子配体、核酸分子（包括DNA和RNA）以及蛋白质的结构预测精度上已与AlphaFold3相媲美。为了评估其在蛋白质-配体结构预测中的效果，HelixFold3与其他主流方法在PoseBusters数据集上的表现进行了对比。HelixFold3即便在没有指定蛋白质结构的情况下，仍然展示出卓越的表现，成功率甚至超过了依赖已知蛋白质结构的方法，其预测精度与目前顶尖的AlphaFold3相当，这表明HelixFold3在蛋白质-配体相互作用预测领域的出色潜力。HelixFold3在蛋白质-蛋白质复合体结构预测方面已经略微超越了AlphaFold-Multimer的表现，展示出更强的预测能力。

参数说明

Protein Sequence

蛋白的序列文件，FASTA格式，支持多条序列。
注意：多蛋白复合物结构预测，其氨基酸序列输入格式如下：

>protein1
AASJ...
>protein2
AASJ...
>peptide
ASDF...

DNA Sequence

DNA核酸的序列文件，FASTA格式，支持多条序列。

RNA Sequence

RNA核酸分子的序列文件，FASTA格式，支持多条序列。

备注:

序列总长度不可超过2000。

Ligand

文本文件包含小分子信息，TXT格式。HF3支持绝大多数重核数量不超过50的配体，支持SMILES或 CCD Code（化学组分词典编号）。如果使用SMILES格式，每行应包含一个小分子；如果使用CCD Code，每行可以包含一个或多个小分子，使用逗号分隔，并加上CCD前缀，示例如下：

CC(=O)OC1C[NH+]2CCC1CC2
CCD,ATP,HY3,P1L
CCD,MG

注意：不适用于配体蛋白或多肽的氨基酸序列格式输入。

备注:

水、助剂和少量特殊的配体目前是模型所不支持的。模型会将这些配体从CCD列表中除去，如果您通过输入SMILES的方式进行了这些输入，可能会造成结果的表现下降。具体不支持的配体的CCD列表参见HF3 FAQ https://paddlehelix.baidu.com/app/tut/guide/all/helixfold3faq。每种配体数量不能超过50。

Modification

包含翻译后修饰（PTM）信息的文本文件，TXT格式。每行放置一个PTM信息，每个PTM信息由三部分组成：

• 发生PTM序列的顺序编号
• PTM类型的CCD编号
• 发生PTM的残基位置编号
  三部分由逗号分隔，例如：1,HY3,1 表示第一条序列的第一个残基，发生了类型为HY3（CCD编号，为3-羟基脯氨酸，为脯氨酸的羟基化）的PTM。不同的氨基酸支持的CCD不同，具体参考https://paddlehelix.baidu.com/app/tut/guide/all/helixfold3json里的说明。

备注：

• 序列的顺序编号，是依次按上述参数Protein、DNA、RNA中的序列顺序与数量，从1开始进行编号，例如：当有2条蛋白序列，1条DNA序列，1条RNA序列时，各序列对应的编号为：第一条蛋白序列编号为1，第二条蛋白序列编号为2，DNA序列编号为3，RNA序列编号为4
• CCD的介绍参考https://www.wwpdb.org/data/ccd 编号查询网址为https://www.ebi.ac.uk/pdbe-srv/pdbechem/
  包含多个PTM信息的文件内容示例如下：

1,HY3,1
1,P1L,5
2,HY3,3

ion

目前支持的离子为：MG, ZN, CL, CA, NA, MN, MN3, K, FE, FE2, CU, CU1, CU3, CO
不支持分行书写，需要写在一行，可以包含一个或多个离子，不同的离子使用逗号分隔，冒号后对应的是离子的数量，每种离子数量不能超过50。示例如下：

MG:2,ZN,CU:3

结果说明

输出结果文件为排名前5的复合物结构rank_1-5.cif和ranking_scores.csv，csv中包含信息如下：

参考文献

Liu, L., Zhang, S., Xue, Y., Ye, X., Ye, X., Zhu, K., Li, Y., Li, Y., Zhao, W., Yu, H.,Wu, Z., Zhang, X., & Fang, X. (2024). Technical Report of HelixFold3 for Biomolecular Structure Prediction. DOI: arxiv-2408.16975

Structure Prediction (HelixFold3)

Introduction

HelixFold3, developed by the Baidu PaddleHelix team, is comparable to AlphaFold3 in the accuracy of structure prediction of conventional small molecule ligands, nucleic acid molecules (including DNA and RNA), and proteins. To assess HelixFold3’s performance in protein-ligand structure prediction, we compared it against other leading methods using the PoseBusters dataset. HelixFold3 demonstrated exceptional performance even without specifying protein structures, surpassing methods that rely on known protein structures. Its prediction accuracy is comparable to the state-of-the-art AlphaFold3, indicating HelixFold3’s outstanding potential in protein-ligand interaction prediction. Currently, HelixFold3 has slightly surpassed AlphaFold-Multimer in protein-protein complex structure prediction, demonstrating stronger predictive power.

Parameter

Protein Sequence

The sequence file of proteins in FASTA format, supporting multiple sequences.
Note: For multi-protein complex structure prediction, the amino acid sequence input format is as follows:

>protein1
AASJ...
>protein2
AASJ...
>peptide
ASDF...

DNA Sequence

The sequence file of DNA nucleic acids in FASTA format, supporting multiple sequences.

RNA Sequence

The sequence file of RNA nucleic acid molecules in FASTA format, supporting multiple sequences.

Note:

The maximum length of this module is 2000.

Ligand

A text file containing small molecule information in TXT format. HF3 supports most ligands with fewer than 50 heavy atoms. It supports SMILES or CCD Code (Chemical Component Dictionary number). If using the SMILES format, each line should contain one small molecule; if using the CCD Code, each line can contain one or more small molecules, separated by commas and prefixed with CCD. An example is as follows:

CC(=O)OC1C[NH+]2CCC1CC2
CCD,ATP,HY3,P1L
CCD,MG

Note: Not applicable for amino acid sequence format input of ligand proteins or polypeptides.

Note:

Water, adjuvants, and certain special ligands are not supported by the model. These ligands are excluded from the CCD list. If you input them using SMILES, it may result in degraded performance. For a list of unsupported ligands, refer to the HelixFold3 FAQ https://paddlehelix.baidu.com/app/tut/guide/all/helixfold3faq. The number of each ligand is not to exceed 50.

Modification

A text file containing post-translational modification (PTM) information in TXT format. Each line contains one PTM information entry, consisting of three parts:

• Sequence order number where the PTM occurs
• CCD number of the PTM type
• Residue position number where the PTM occurs
  The three parts are separated by commas. For example, 1,HY3,1 indicates that the first residue of the first sequence undergoes a PTM of type HY3 (CCD number, which is 3-hydroxyproline, a hydroxylation of proline). Different amino acids support different CCDs, as shown in tutorial https://paddlehelix.baidu.com/app/tut/guide/all/helixfold3json.

Note:

• The sequence order number is numbered sequentially according to the order and number of sequences in the parameters Protein, DNA, and RNA, starting from 1. For example, if there are 2 protein sequences, 1 DNA sequence, and 1 RNA sequence, the sequence numbers are: the first protein sequence is numbered 1, the second protein sequence is numbered 2, the DNA sequence is numbered 3, and the RNA sequence is numbered 4.
• For CCD introduction, refer to https://www.wwpdb.org/data/ccd. The CCD number lookup website is https://www.ebi.ac.uk/pdbe-srv/pdbechem/.

An example of a file containing multiple PTM information entries is as follows:

1,HY3,1
1,P1L,5
2,HY3,3

ion

Currently supported ions include: MG, ZN, CL, CA, NA, MN, MN3, K, FE, FE2, CU, CU1, CU3, CO

It does not support multi-line writing; it must be written in a single line. It can contain one or multiple ions, with different ions separated by commas. The quantity of each ion is indicated after a colon, and the number of each ion cannot exceed 50. An example of a file containing different ion information is as follows:

MG:2,ZN,CU:3

Result

The output result files are the top 5 ranked complex structures rank_1-5.cif and ranking_scores.csv, with the following information in the CSV:

Reference

Liu, L., Zhang, S., Xue, Y., Ye, X., Ye, X., Zhu, K., Li, Y., Li, Y., Zhao, W., Yu, H.,Wu, Z., Zhang, X., & Fang, X. (2024). Technical Report of HelixFold3 for Biomolecular Structure Prediction. DOI: arxiv-2408.16975

Patent BLAST

简介

针对抗体全长或者CDR区进行序列检索的模块。从专利中检索一条抗体可变区时，现有的BLAST程序（例如NCBI BLAST）通常是以全序列进行检索，但是对于抗体而言，功能主要取决于CDR，FR相对不重要，并且由于FR的通用性，许多不同抗体的FR是相同或高度同源的，而FR占序列的比重更高，就导致以抗体的可变区BLAST会得到很多FR相似但CDR不相似的序列。并且，专利申请时，除了保护可变区完整序列，很多情况也会对抗体CDR进行单独保护，以获得更大的保护范围，因此在抗体开发过程中，以CDR为目标进行同源序列检索就很有必要了。为此，唯信团队开发了该程序，可以从现有专利库中检索到与目标CDR最接近的序列。数据更新于：Dec 2024

• 序列来源：NCBI专利序列库
• 来自美国专利局USPTO提交的美国专利序列和通过INSDC合作包括的欧洲和日本专利序列
• 包含已授权专利中的权利要求和实施例中的全部序列
• 原始数据链接：https://ftp.ncbi.nlm.nih.gov/blast/db/pataa.tar.gz
• WeMol数据更新于：Dec 2024
• 数量：>700万个蛋白序列，其中14万条抗体CDR序列
• 检索原理：提取专利序列数据库中的抗体序列，使用Kabat规则识别CDR区，并将CDR1/2/3拼接成新的CDR序列，与目标抗体拼接后的CDRs进行比对，输出同源性最高的数条。

例如，输入序列L的完整序列，进行检索后，返回检索到同源性较高的序列的CDR，如下图所示。

如果需要查看某个检索到的序列的出处，可以根据检索的CDR的序列编号，从任务输出的log文件中找到对应的专利名，
例如序列ATJ10081.1来自于US专利9493553（SEQ ID为39），并且US专利9670274、9890209等多个专利中也出现了该CDR片段，他们的比对情况包括同源性也展示在后面，如下图所示。

根据唯信团队经验，通常CDR的保护范围精确到具体序列，即差异一个以上氨基酸，即视为不在专利的保护范围之内，但不排除存在等同侵权的风险，仅供参考。

参数说明

Antibody Sequence File

抗体序列文件， FASTA格式

Type

指定序列比对数据库类型：抗体全长（full）或者抗体CDR区域 (cdr)。
CDR区域数据库为专利保护抗体数据库。

结果说明

输出结果包括：

Patent BLAST

Introduction

A module for sequence retrieval of antibody full-length or CDR region. When retrieving an antibody variable region from a patent, existing BLAST programs (such as NCBI BLAST) usually search the whole sequence, but for antibodies, the function mainly depends on the CDR, FR is relatively not important, and due to the generality of FR, FR of many different antibodies is the same or highly homologous. However, FR accounts for a higher proportion of sequences, resulting in a lot of sequences with similar FR but different CDR by BLAST in the variable region of antibodies. Moreover, in addition to protecting the complete sequence of the variable region during patent application, in many cases, the antibody CDR will also be protected separately to obtain a wider range of protection, so it is necessary to search for homologous sequences with CDR as the target in the process of antibody development. To this end, the Vixon team developed the program, which can retrieve the closest sequence to the target CDR from the existing patent library. Data updated: Dec 2024

• Sequence Source: NCBI patent sequence database
• Includes US patent sequences submitted to the USPTO and European and Japanese patent sequences included through collaboration with INSDC
• Contains claims from granted patents and all sequences in the embodiments
• Original data link: https://ftp.ncbi.nlm.nih.gov/blast/db/pataa.tar.gz
• WeMol data updated: Dec 2024
• Quantity: >7 million protein sequences, including 140,000 antibody CDR sequences
• Search Principle: Extract antibody sequences from the patent sequence database, identify CDR regions using Kabat rules, concatenate CDR1/2/3 into a new CDR sequence, compare it with the concatenated CDRs of the target antibody, and output the top matching sequences based on homology.

For example, when inputting the complete sequence of antibody L for search, the returned CDR of the highly homologous sequences is shown in the image below.

If there is a need to check the source of a retrieved sequence, you can find the corresponding patent name based on the sequence number of the retrieved CDR from the log file output of the task. For example, sequence ATJ10081.1 is from US Patent 9493553 (SEQ ID 39), and the CDR fragment also appears in multiple patents such as US Patents 9670274, 9890209, etc., with their alignment details and homology shown as well, as depicted in the image below.

Based on the experience of the WeMol team, the protection range of CDRs is usually specified down to the specific sequence, meaning that a difference of one or more amino acids is considered outside the scope of patent protection. However, there may still be risks of equivalent infringement, so this information is for reference only.

Parameter Description

Antibody Sequence File

Antibody sequence file in FASTA format.

Type

Specifies the sequence alignment database type: antibody full-length (full) or antibody CDR region (cdr).
The CDR regional database is a patent protected antibody database.

Result Description

The output includes:

CIF2PDB

简介

CIF2PDB模块是基于BioPython将mmCIF文件转换成PDB文件。
单独化合物CIF转换部分存在问题。

参数说明

CIF File

mmCIF文件的结构。

Output PDB

输出PDB文件名称

结果说明

输出PDB文件，默认为output.pdb。

CIF2PDB

Introduction

The CIF2PDB module is based on BioPython to convert mmCIF files into PDB files.

Parameter

CIF File

mmCIF file of structure.

Output PDB

Output the PDB file name

Result

Output pdb file; default is output.pdb.

Target-Cyclic Peptide Complex Structure Prediction

简介

Target-Cyclic Peptide Complex Structure Prediction模块基于Alphafold2，用于预测首尾相接(Head-to-tail)环肽和靶点蛋白的复合物三维结构。
靶点-环肽复合物预测示例，展示了首尾相接酰胺键：

参数说明

Target Sequence File

靶标蛋白的序列文件，只支持输入一条链，不支持多条链，FASTA格式

Cyclic Peptide

环肽的序列，如：“ARDCPLVNPL”

结果说明

输出结果包括：

其中score.csv包括信息如下：

参考文献

Li Q, Vlachos E.N., Bryant P. Design of linear and cyclic peptide binders of different lengths only from a protein target sequence. bioRxiv. 2024. p. 2024.06.20.599739.

Target-Cyclic Peptide Complex Structure Prediction

Introduction

The Target-Cyclic Peptide Complex Structure Prediction module is based on AlphaFold2 and is used to predict the three-dimensional structure of head-to-tail cyclic peptides in complex with target proteins. An example of a target-cyclic peptide complex prediction is shown below, demonstrating the head-to-tail amide bond:

Parameter

Target Sequence File

The sequence file of the target protein, only supports the input of one strand, not multiple strands, FASTA format.

Cyclic Peptide

The sequence of the cyclic peptide, e.g., “ARDCPLVNPL”.

Hotspot

Residues on the target sequence that are binding sites for the cyclic peptide, numbered starting from 1. Specifying these sites can improve the accuracy and success rate of the design. Multiple sites are separated by commas, e.g., “23,45,67”.

Result

The output results include:

The score.csv includes the following information:

References

Li Q, Vlachos E.N., Bryant P. Design of linear and cyclic peptide binders of different lengths only from a protein target sequence. bioRxiv. 2024. p. 2024.06.20.599739.

Structure Prediction (Boltz-1)

简介

基于MIT（麻省理工学院）的Boltz-1算法的AF3 like结构预测模型。Boltz-1是一种开源深度学习模型，融合了模型架构、速度优化和数据处理方面的创新，在预测生物分子复合物的 3D结构方面达到了 AlphaFold3 级的准确度。Boltz-1 在一系列不同的基准测试中表现出与最先进的商业模型相当的性能，为结构生物学中可商业化使用的工具树立了新的标杆。

参数说明

Single Mode

Protein Sequence

蛋白的序列文件，FASTA格式，支持多条序列。
注意：多蛋白复合物结构预测，其氨基酸序列输入格式如下：

>protein1
AASJ...
>protein2
AASJ...
>peptide
ASDF...

DNA Sequence

DNA核酸的序列文件，FASTA格式，支持多条序列。

RNA Sequence

RNA核酸分子的序列文件，FASTA格式，支持多条序列。

备注：当前24GB的GPU显存支持计算的残基/碱基数量在1000个左右。

Ligand

文本文件包含小分子信息，TXT格式。支持SMILES或 CCD Code（化学组分词典编号）。如果使用SMILES格式，每行应包含一个小分子；如果使用CCD Code，每行可以包含一个或多个小分子，使用逗号分隔，并加上CCD前缀。示例如下：

CC(=O)OC1C[NH+]2CCC1CC2
CCD,ATP,HY3,P1L
CCD,MG

注意：不适用于配体蛋白或多肽的氨基酸序列格式输入。

Modification

包含翻译后修饰（PTM）信息的文本文件，TXT格式。每行放置一个PTM信息，每个PTM信息由三部分组成：

• 发生PTM序列的顺序编号
• PTM类型的CCD编号
• 发生PTM的残基位置编号
  三部分由逗号分隔，例如：1,HY3,1 表示第一条序列的第一个残基，发生了类型为HY3（CCD编号，为3-羟基脯氨酸，为脯氨酸的羟基化）的PTM
  备注：
• 序列的顺序编号，是依次按上述参数Protein、DNA、RNA中的序列顺序与数量，从1开始进行编号，例如：当有2条蛋白序列，1条DNA序列，1条RNA序列时，各序列对应的编号为：第一条蛋白序列编号为1，第二条蛋白序列编号为2，DNA序列编号为3，RNA序列编号为4
• CCD的介绍参考https://www.wwpdb.org/data/ccd 编号查询网址为https://www.ebi.ac.uk/pdbe-srv/pdbechem/
  包含多个PTM信息的文件内容示例如下：

1,HY3,1
1,P1L,5
2,HY3,3

Covalent Bond

共价键信息的文本文件，TXT格式。每行放置一个共价键信息，每个共价键信息包含两个原子信息，每个原子信息由三部分组成：

• 原子所在序列或小分子的顺序编号（编号规则在Modification中定义的序列编号规则基础上，在最后加入小分子的顺序即可）
• 原子所在残基的位置编号（如残基为小分子时，编号为1）
• 原子的标准名称（CCD中定义）
  三部分由逗号分隔，例如：3,1,CA表示第三个实体（序列或小分子）中的第一个残基（或小分子）的CA原子
  一个共价键是由两个原子信息组成，原子间用分号分隔，如：1,1,CA;2,1,CA
  表示一个共价键，该共价键由两个原子组成，第一个原子为1,1,CA，第二个原子为2,1,CA
  包含多个共价键信息的文件内容示例如下：

1,1,CA;2,1,CA
1,1,CA;3,1,CHA

Pocket

结合位点类型限制信息的文本文件，TXT格式。每行放置一个结合位点信息，每个结合位点信息由两部分组成：

• Binder的顺序编号（与共价键定义中的序列或小分子的顺序编号一致），Binder可以是小分子，蛋白/核酸序列的任意一种，目前一个结合位点只支持定义一条Binder（即一个编号）
• 结合位点的残基信息，每个残基信息由其所在序列编号与残基位置编号组成，逗号分隔，如：1,25 表示第一条序列中的第25个残基；可以定义多个残基信息，由英文分号“;”进行分隔，如1,25;1,27;1,32;1,38表示第一条序列中的第25/27/32/38号残基形成结合位点
  上述两部信息之间也用英文分号“;”进行分隔，例如：2;1,55;1,62;1,91;1,92;1,99;1,110表示第二个实体（序列或小分子）作为Binder，与第一条序列的第55/62/91/92/99/110号残基所形成的结合位点进行结合。
  包含多个结合位点信息的文件内容示例如下：

2;1,55;1,62;1,91;1,92;1,99;1,110
3;1,25;1,27;1,32;1,38

Domain

定义的残基区域信息。模块将输出区域中所有残基平均的pLDDT数值。一个残基区域由序列顺序编号与残基组合编号组成：

• 序列顺序编号（同Modification参数中的定义），值为1时，可省略（即默认为1）
• 残基组合编号，使用残基位置编号，多个残基用逗号分隔，指定残基范围用横杠符号。如：“3,10,24-30”表示目标序列上的第3、第10与第24至30号残基。
  例如：1:24,28,32-40 表示第一条序列中的第24/28/32至40号残基所组成的区域，因为是第一条序列，数值1可以省略，等同于24,28,32-40 ，该区域的所有残基的平均pLDDT值将输出到结果文件中。

残基区域支持定义多个，每个残基区域之间用英文“;”分隔，例如：
1:24,28,32-40;2:15,23,50-60表示定义了两个区域，区域一为第一条序列的第24/28/32至40号残基，区域二为第二条序列的第15/23/50至60号残基。两个区域各自的残基平均pLDDT值，将输出到结果文件中。

Format

输出结构的格式，支持PDB或CIF格式，默认为PDB格式。

Batch Mode

Protein Sequence

蛋白的序列文件，FASTA格式，支持多条序列。
每一条记录代表一个待预测的结构，每条记录的名称要唯一不能重复。一条记录中有多条链时，通过英文冒号（:）相连，文件内容示例如下：

>1
EVQLVESGGGVVQPGRSLRLS:NFMLTQPHSVSVL:ELFPQWHLPIKIAAIIAS
>2
YYCAKDRRDMGYFQHWGQGTLVTVSSQWYQQRPGSAPTTVIYEDNQRPSGTPPTFMIAVFLPIVVLIFKSILFL

表示有两个待预测的结构，第一条记录的名称为1，有三条蛋白链，用:进行分隔。第二条记录的名称为2，为单链。

DNA Sequence

DNA核酸的序列文件，FASTA格式，支持多条序列。
每一条序列记录代表一个待预测的结构，每条记录的名称要唯一不能重复（可与Protein参数中的记录名称一致，表示该记录的DNA序列与Protein序列归属于同一结构）。一条记录中有多条链时，通过英文冒号（:）相连，文件内容示例如下：

>dna
GACCTCT:CCTAGCT
>1
CCTAGCT

表示有两条记录，第一条的名称为dna，有两条DNA链，用:进行分隔，因为该名称不存在与Protein示例记录中，属于新结构。第二条的名称为1，有一条DNA链，因为该名称存在于Protein示例记录中，则表示同属一个结构（该结构同时包含Protein序列和该DNA序列）。

RNA Sequence

RNA核酸分子的序列文件，FASTA格式，支持多条序列。
每一条序列记录代表一个待预测的结构，每条记录的名称要唯一不能重复（可与DNA或Protein参数中的记录名称一致，表示该记录的RNA序列与上面的DNA序列或Protein序列归属于同一结构）。一条记录中有多条链时，通过英文冒号（:）相连，文件内容示例如下：

>1
AGCU
>rna
AGGCU:UGAUC

表示有两条记录，第一条的名称为1，为单链，因为该名称存在于DNA及Protein示例记录中，表示同属一个结构（该结构同时包含了Protein序列、DNA序列及该RNA序列）。第二条的名称为rna，有两条RNA链，用:进行分隔，因为该名称不存在于DNA或Protein示例记录中，属于新结构。

Ligand

文本文件包含小分子信息，TXT格式。支持SMILES或 CCD Code（化学组分词典编号）。如果使用SMILES格式，每行应包含一个小分子；如果使用CCD Code，每行可以包含一个或多个小分子，使用逗号分隔。
每行代表一个待预测的结构，每行可放置多个ligand，且以唯一不重复的名称开头（该名称可与上述RNA，DNA或Protein参数中的记录名称一致，表示该行的所有ligands，与上述的RNA或DNA或Protein序列归属于同一结构），名称与所有ligands都以英文冒号（:）分隔。文件内容示例如下：

1:CC(=O)OC1C[NH+]2CCC1CC2:CCD,ATP,HY3
lig:CC1CC(CN2NC[N+](=C3CCC(OC(F)(F)F)CC3)C2=O)CC(C)C1OC(C)(C)C(=O)[O-]

表示有两条记录，第一条的名称为1，有三个ligand（一个SMILES，两个CCD codes），因为该名称存在于上述的RNA或DNA或Protein示例记录中，表示同属一个结构。第二条的名称为lig，有一个ligand（为SMILES），因为该名称不存在上述的RNA或DNA或Protein示例记录中，属于新结构。

Modification

包含翻译后修饰（PTM）信息的文本文件，TXT格式。每个PTM的信息与Single模式中一致（参考Single模式中的定义）。
每行定义一个结构的所有PTM信息，且以唯一名称开头（该名称必须存在于前述的Protein或DNA或RNA记录中），都以英文冒号（:）分隔。文件内容示例如下：

1:1,HY3,1:1,P1L,5:2,HY3,3
2:1,HY3,1:2,HY3,3

表示前述名称为1的结构中（Protein或DNA或RNA），有三个PTM。名称为2的结构中，有两个PTM。

Covalent Bond

共价键信息的文本文件，TXT格式。每个共价键的信息与Single模式中一致（参考Single模式中的定义）。
Batch模式下，每行定义一个结构的所有共价键，且以唯一名称开头（该名称必须存在于前述的Protein或DNA或RNA记录中），都以英文冒号（:）分隔。文件内容示例如下：

1:1,1,CA;2,1,CA:1,1,CA;3,1,CHA
2:1,1,CA;3,1,CHA

表示前述名称为1的结构中（Protein或DNA或RNA），有两个共价键。名称为2的结构中，有一个共价键。

Pocket

结合位点类型限制信息的文本文件，TXT格式。每个结合位点信息的定义与Single模式中一致（参考Single模式中的定义）。
Batch模式下，每行定义一个结构的所有结合位点限制信息，且以唯一名称开头（该名称必须存在于前述的Protein或DNA或RNA记录中），都以英文冒号（:）分隔。文件内容示例如下：

1:2;1,55;1,62;1,91;1,92;1,99;1,110
2:1;2,15;2,17;2,18;2,56:1;3,76;3,78;3,96

表示前述名称为1的结构中（Protein或DNA或RNA），有一个结合位点限制。名称为2的结构中，有两个结合位点限制。

Format

输出结构的格式，支持PDB或CIF格式，默认为PDB格式。

结果说明

输出结果文件为排名前5的复合物结构rank_1-5.cif和pred_scores_boltz.csv，csv中包含信息如下：

final_results.tar.gz文件为Batch模式下额外生成一个所有预测结果的打包文件

参考文献

Boltz-1 Democratizing Biomolecular Interaction Modeling. Jeremy Wohlwend, Gabriele Corso, Saro Passaro, Mateo Reveiz, Ken Leidal, Wojtek Swiderski, Tally Portnoi, Itamar Chinn, Jacob Silterra, Tommi Jaakkola, Regina Barzilay, bioRxiv 2024.11.19.624167;

Structure Prediction (Boltz-1)

Introduction

Developed based on the Boltz-1 model, Boltz-1 is an open-source deep learning model that integrates innovations in model architecture, speed optimization, and data processing. It achieves AlphaFold3-level accuracy in predicting the 3D structures of biomolecular complexes. Boltz-1 demonstrates performance comparable to state-of-the-art commercial models across a range of benchmarks, setting a new standard for commercially usable tools in structural biology.

Parameter

Single Mode

Protein Sequence

The sequence file of proteins in FASTA format, supporting multiple sequences.
Note: For multi-protein complex structure prediction, the amino acid sequence input format is as follows:

>protein1
AASJ...
>protein2
AASJ...
>peptide
ASDF...

DNA Sequence

The sequence file of DNA nucleic acids in FASTA format, supporting multiple sequences.

RNA Sequence

The sequence file of RNA nucleic acid molecules in FASTA format, supporting multiple sequences.

Ligand

A text file containing small molecule information in TXT format. It supports SMILES or CCD Code (Chemical Component Dictionary number). If using the SMILES format, each line should contain one small molecule; if using the CCD Code, each line can contain one or more small molecules, separated by commas and prefixed with CCD. An example is as follows:

CC(=O)OC1C[NH+]2CCC1CC2
CCD,ATP,HY3,P1L
CCD,MG

Note: Not applicable for amino acid sequence format input of ligand proteins or polypeptides.

Modification

A text file containing post-translational modification (PTM) information in TXT format. Each line contains one PTM information entry, consisting of three parts:

• Sequence order number where the PTM occurs
• CCD number of the PTM type
• Residue position number where the PTM occurs
  The three parts are separated by commas. For example, 1,HY3,1 indicates that the first residue of the first sequence undergoes a PTM of type HY3 (CCD number, which is 3-hydroxyproline, a hydroxylation of proline).

Note:

• The sequence order number is numbered sequentially according to the order and number of sequences in the parameters Protein, DNA, and RNA, starting from 1. For example, if there are 2 protein sequences, 1 DNA sequence, and 1 RNA sequence, the sequence numbers are: the first protein sequence is numbered 1, the second protein sequence is numbered 2, the DNA sequence is numbered 3, and the RNA sequence is numbered 4.
• For CCD introduction, refer to https://www.wwpdb.org/data/ccd. The CCD number lookup website is https://www.ebi.ac.uk/pdbe-srv/pdbechem/.

An example of a file containing multiple PTM information entries is as follows:

1,HY3,1
1,P1L,5
2,HY3,3

Covalent Bond

A text file containing covalent bond information in TXT format. Each line contains one covalent bond information entry, and each entry includes two atom information entries, each consisting of three parts:

• Sequence or small molecule order number (following the sequence numbering rule defined in Modification, with small molecule order added at the end)
• Position number of the residue where the atom is located (if the residue is a small molecule, the number is 1)
• Standard name of the atom (as defined in CCD)

The three parts are separated by commas. For example, 3,1,CA indicates the CA atom of the first residue (or small molecule) in the third entity (sequence or small molecule).

A covalent bond consists of two atom information entries, separated by a semicolon, such as 1,1,CA;2,1,CA, indicating a covalent bond composed of two atoms: the first atom is 1,1,CA, and the second atom is 2,1,CA.

An example of a file containing multiple covalent bond information entries is as follows:

1,1,CA;2,1,CA
1,1,CA;3,1,CHA

Pocket

A text file with pocket type restriction information, in TXT format. Each line contains the information of one pocket, which is composed of two parts:
The sequential number of the Binder (consistent with the sequential number of the sequence or small molecule in the covalent bond definition), the Binder can be any one of small molecules, protein/nucleic acid sequences, and currently, only one Binder (i.e., one number) is supported for a pocket.
The residue information of the pocket, each residue information consists of the sequence number where it is located and the residue position number, separated by a comma, such as: 1,25 indicates the 25th residue in the first sequence; multiple residue information can be defined, separated by an English semicolon “;”, for example, 1,25;1,27;1,32;1,38 indicates that the 25th, 27th, 32nd, and 38th residues in the first sequence form the pocket.
The above two pieces of information are also separated by an English semicolon “;”. For example: 2;1,55;1,62;1,91;1,92;1,99;1,110 indicates that the second entity (sequence or small molecule) as a Binder, binds to the pocket formed by the 55th, 62nd, 91st, 92nd, 99th, and 110th residues in the first sequence.

An example of a file content containing multiple pockets information is as follows:

2;1,55;1,62;1,91;1,92;1,99;1,110
3;1,25;1,27;1,32;1,38

Domain

The defined residue region information. The module will output the average pLDDT value of all residues in the region. A residue region is composed of sequence order numbers and residue combination numbers:
Sequence order numbers (as defined in the Modification parameter), the value 1 can be omitted (i.e., defaulting to 1).
Residue combination numbers, using residue position numbers, with multiple residues separated by commas and specified residue ranges indicated by hyphen symbols. For example, “3,10,24-30” indicates the 3rd, 10th, and 24th to 30th residues on the target sequence.
For example: 1:24,28,32-40 indicates the region composed of the 24th, 28th, and 32nd to 40th residues in the first sequence. Since it is the first sequence, the number 1 can be omitted, equivalent to 24,28,32-40. The average pLDDT value of all residues in this region will be output to the result file.
Multiple residue regions are supported, with each residue region separated by an English semicolon “;”. For example: 1:24,28,32-40;2:15,23,50-60 defines two regions. Region one consists of the 24th, 28th, and 32nd to 40th residues in the first sequence, and region two consists of the 15th, 23rd, and 50th to 60th residues in the second sequence. The average pLDDT values of the residues in each of the two regions will be output to the result file.

Format

The output structure format supports PDB or CIF, with PDB format as the default.

Batch Mode

Protein Sequence

The protein sequence file in FASTA format, supporting multiple sequences. Each record represents a structure to be predicted, and each record name must be unique. If a record contains multiple chains, they should be connected by a colon (:). Example content:

>1
EVQLVESGGGVVQPGRSLRLS:NFMLTQPHSVSVL:ELFPQWHLPIKIAAIIAS
>2
YYCAKDRRDMGYFQHWGQGTLVTVSSQWYQQRPGSAPTTVIYEDNQRPSGTPPTFMIAVFLPIVVLIFKSILFL

This indicates two structures to be predicted, with the first record named 1 containing three protein chains separated by colons. The second record is named 2 and contains a single chain.

DNA Sequence

The DNA sequence file in FASTA format, supporting multiple sequences. Each sequence record represents a structure to be predicted, and each record name must be unique. (It can match the record names in the Protein parameter, indicating that the DNA sequence belongs to the same structure as the Protein sequence.) If a record contains multiple chains, they should be connected by a colon (:). Example content:

>dna
GACCTCT:CCTAGCT
>1
CCTAGCT

This indicates two records, with the first named dna containing two DNA chains separated by a colon. Since this name does not appear in the Protein example records, it represents a new structure. The second record is named 1, containing one DNA chain, and since this name exists in the Protein example records, it indicates that they belong to the same structure (which contains both Protein and DNA sequences).

RNA Sequence

The RNA sequence file in FASTA format, supporting multiple sequences. Each sequence record represents a structure to be predicted, and each record name must be unique. (It can match the record names in the DNA or Protein parameters, indicating that the RNA sequence belongs to the same structure.) If a record contains multiple chains, they should be connected by a colon (:). Example content:

>1
AGCU
>rna
AGGCU:UGAUC

This indicates two records, with the first named 1, which is a single chain. Since this name exists in the DNA and Protein example records, it indicates that they belong to the same structure (which includes Protein, DNA, and this RNA sequence). The second record is named rna, containing two RNA chains separated by a colon. Since this name does not appear in the DNA or Protein example records, it represents a new structure.

Ligand

A text file containing information on small molecules in TXT format. It supports either SMILES or CCD Code. If using SMILES format, each line should contain one small molecule; if using CCD Code, each line can contain one or more small molecules, separated by commas. Each line represents a structure to be predicted, and each line must start with a unique name (this name can match those in the RNA, DNA, or Protein parameters, indicating that all ligands in that line belong to the same structure). The name and all ligands are separated by a colon (:). Example content:

1:CC(=O)OC1C[NH+]2CCC1CC2:CCD,ATP,HY3
lig:CC1CC(CN2NC[N+](=C3CCC(OC(F)(F)F)CC3)C2=O)CC(C)C1OC(C)(C)C(=O)[O-]

This indicates two records, with the first named 1, containing three ligands (one SMILES and two CCD codes). Since this name exists in the RNA, DNA, or Protein example records, it indicates that they belong to the same structure. The second record is named lig, containing one ligand (in SMILES format). Since this name does not appear in the RNA, DNA, or Protein example records, it represents a new structure.

Modification

A text file containing post-translational modification (PTM) information in TXT format. Each PTM entry is consistent with that in Single mode (refer to the definitions in Single mode). Each line defines all PTM information for a structure, starting with a unique name (this name must exist in the previously defined Protein, DNA, or RNA records), separated by colons (:). Example content:

1:1,HY3,1:1,P1L,5:2,HY3,3
2:1,HY3,1:2,HY3,3

This indicates that the structure named 1 (Protein, DNA, or RNA) has three PTMs, while the structure named 2 has two PTMs.

Covalent Bond

A text file containing covalent bond information in TXT format. Each covalent bond entry is consistent with that in Single mode (refer to the definitions in Single mode). In Batch mode, each line defines all covalent bonds for a structure, starting with a unique name (this name must exist in the previously defined Protein, DNA, or RNA records), separated by colons (:). Example content:

1:1,1,CA;2,1,CA:1,1,CA;3,1,CHA
2:1,1,CA;3,1,CHA

This indicates that the structure named 1 (Protein, DNA, or RNA) has two covalent bonds, while the structure named 2 has one covalent bond.

Pocket

A text file containing pockets information in TXT format. Each pocket is consistent with that in Single mode (refer to the definitions in Single mode). In Batch mode, each line defines all pockets for a structure, starting with a unique name (this name must exist in the previously defined Protein, DNA, or RNA records), separated by colons (:). Example content:

1:2;1,55;1,62;1,91;1,92;1,99;1,110
2:1;2,15;2,17;2,18;2,56:1;3,76;3,78;3,96

This indicates that the structure named 1 (Protein, DNA, or RNA) has one pocket, while the structure named 2 has two pockets.

Format

The output structure format supports PDB or CIF, with PDB format as the default.

Result

The output result files are the top 5 ranked complex structures rank_1-5.cif and pred_scores_boltz.csv, with the following information in the CSV:

final_results.tar.gz, An additional compressed file containing all predicted results generated in Batch mode.

Reference

Boltz-1 Democratizing Biomolecular Interaction Modeling. Jeremy Wohlwend, Gabriele Corso, Saro Passaro, Mateo Reveiz, Ken Leidal, Wojtek Swiderski, Tally Portnoi, Itamar Chinn, Jacob Silterra, Tommi Jaakkola, Regina Barzilay, bioRxiv 2024.11.19.624167;

Structure Prediction (Chai-1)

简介

基于Chai Discovery, Inc.（OpenAI投资）的Chai-1算法的AF3 like结构预测模型。Chai-1是一种用于分子结构预测的多模态基础模型，在各种基准测试中均表现出色，可以预测包括蛋白质、小分子、DNA、RNA、糖基化等。

参数说明

Protein Sequence

蛋白的序列文件，FASTA格式，支持多条序列。
注意：多蛋白复合物结构预测，其氨基酸序列输入格式如下：

>protein1
AASJ...
>protein2
AASJ...
>peptide
ASDF...

DNA Sequence

DNA核酸的序列文件，FASTA格式，支持多条序列。

RNA Sequence

RNA核酸分子的序列文件，FASTA格式，支持多条序列。

备注：当前24GB的GPU显存能计算的残基/碱基数量在1000个左右。

在Protein、DNA、RNA序列中，都支持残基或碱基的修饰，用CCD进行定义，CCD的介绍参考https://www.wwpdb.org/data/ccd 编号查询网址为https://www.ebi.ac.uk/pdbe-srv/pdbechem/
定义残基或碱基修饰时，直接在序列中用英文括号‘()’包含CCD code即可，示例如下：

>seq
(ACE)GQLEEIAK

表示在序列的N端发生了乙酰化；

>seq
AGSHSMRYFSTSVSR(HY3)GRGEPRFIAVG

表示序列中的残基P发生了羟基化修饰，变成HY3（CCD code）

Ligand

文本文件包含小分子的结构信息，用SMILES格式，支持多个小分子，每行放置一个，示例如下：

CC(=O)OC1C[NH+]2CCC1CC2
[Mg+2]

注意：不适用于配体蛋白或多肽的氨基酸序列格式输入。

Restraints

包含残基间距离限制信息的文本文件。距离限制的类型有两种：两个残基间的距离限制，一个残基与一条链之间的距离限制。

两个残基间的距离限制的定义由五部分组成：

• 残基1所在序列的顺序编号（序列的顺序编号，是依次按上述参数Protein、DNA、RNA中的序列顺序与数量，从1开始进行编号，例如：当有2条蛋白序列，1条DNA序列，1条RNA序列时，各序列对应的编号为：第一条蛋白序列编号为1，第二条蛋白序列编号为2，DNA序列编号为3，RNA序列编号为4）
• 残基1的符号及位置编号（如：R84表示84号残基R）
• 残基2所在序列的顺序编号
• 残基2的符号及位置编号
• 残基间的最大距离（单位为埃）

五部分由逗号分隔，例如：1,R84,3,G7,10.0
表示第1条序列中的84号残基R，与第3条序列中的7号残基G，之间的最大距离为10.0埃。

一个残基与一条链之间的距离限制表示该残基与链中任意一个残基的距离满足限制即可。其定义方式与上述类似，差异在于，残基1与残基2的符号及位置编号，其中一个需设置为0（不可同时为0），例如：1,R84,3,0,10.0
表示第1条序列中的84号残基R，与第3条链的任意一个残基/碱基的最大距离为10.0埃即可。

支持放置多个距离限制，每行放置一个即可，包含多个距离限制信息的文件内容示例如下：

1,H189,3,L4,8.0
1,R84,3,0,10.0

结果说明

输出结果文件为排名前5的复合物结构rank_1-5.cif和pred_scores_chai1.csv，csv中包含信息如下：

参考文献

Chai-1: Decoding the molecular interactions of life. Chai Discovery team, Jacques Boitreaud, Jack Dent, Matthew McPartlon, Joshua Meier, Vinicius Reis, Alex Rogozhonikov, Kevin Wu
bioRxiv 2024.10.10.615955

Structure Prediction (Chai-1)

Introduction

Based on Chai-1 structure prediction model implementation. Chai-1 is a multimodal basis model for molecular structure prediction that performs well on various benchmarks and can predict including proteins, small molecules, DNA, RNA, glycosylation, and more.

Parameter

Protein Sequence

The sequence file of proteins in FASTA format, supporting multiple sequences.
Note: For multi-protein complex structure prediction, the amino acid sequence input format is as follows:

>protein1
AASJ...
>protein2
AASJ...
>peptide
ASDF...

DNA Sequence

The sequence file of DNA nucleic acids in FASTA format, supporting multiple sequences.

RNA Sequence

The sequence file of RNA nucleic acid molecules in FASTA format, supporting multiple sequences.
** Note: Current 24GB GPU memory can calculate around 1000 residues/bases. **
In Protein, DNA, RNA sequences, all support the modification of residues or bases, which are defined by CCD, The introduction of the CCD reference https://www.wwpdb.org/data/ccd Number query url for https://www.ebi.ac.uk/pdbe-srv/pdbechem/
To define a residue or base modification, simply include the CCD code in parentheses’ () 'in the sequence, as shown in the following example:

>seq
(ACE)GQLEEIAK

Indicates acetylation at the N-terminus of the sequence;

>seq
AGSHSMRYFSTSVSR(HY3)GRGEPRFIAVG

Indicates that residue P in the sequence is hydroxylated and becomes HY3 (CCD code).

Ligand

The text file contains structural information about small molecules, in SMILES format, supporting multiple small molecules, one per line, as shown in the following example:

CC(=O)OC1C[NH+]2CCC1CC2
[Mg+2]

Note: Not applicable for amino acid sequence format input of ligand proteins or polypeptides.

Restraints

• Sequence number of the sequence in which residue 1 is located (The sequence number of the sequence is numbered from 1 according to the sequence order and quantity in the above parameters Protein, DNA and RNA in turn. For example, when there are 2 protein sequences, 1 DNA sequence and 1 RNA sequence, the corresponding number of each sequence is: The first protein sequence is numbered 1, the second protein sequence is numbered 2, the DNA sequence is numbered 3, and the RNA sequence is numbered 4)
• Symbol and position number of residue 1 (e.g. R84 for residue 84 R)
  -The sequence number of the sequence in which residue 2 is located
  -Symbol and position number of residue 2
• Maximum distance between residues (in angstroms)

The five parts are separated by commas, for example: 1,R84,3,G7,10.0
Denote residue 84 R in the first sequence, and residue 7 G in the third sequence, with a maximum distance of 10.0 angstroms.

** The distance limit between a residue and a chain ** means that the distance between the residue and any residue in the chain satisfies the limit. It is defined in the same way as above, except that the symbol and position number of residue 1 and residue 2 need to be set to 0 (not both), e.g. 1,R84,3,0,10.0
Denotes residue 84 R in the first sequence, and a maximum distance of 10.0 angstroms from any residue/base of the third strand is sufficient.

Multiple distance limits are supported, one per line, and an example file containing multiple distance limits is as follows:

1,H189,3,L4,8.0
1,R84,3,0,10.0

Result

The output files are the top 5 complex structures rank_1-5.cif and pred_scores_chai1.csv, which contain the following information:

ADMET Prediction (v2)

简介

ADMET Prediction (v2)是一个基于机器学习的小分子ADMET性质预测模块。能快速批量预测小分子的ADMET性质，支持图注意力神经网络模型（GNN）、轻量梯度提升树模型(LGBM)、随机森林模型（RF）、梯度提升树模型(XGBT)4种常见高效的机器学习算法，分子特征支持分子指纹(Morgan FP)以及分子描述符(Descriptors)两种方法，能对小分子化合物库进行快速批量预测。模块支持27种ADMET性质，其中7种回归模型，20种分类模型。不同机器学习方法以及分子特征化方法预测性能如下：


模块自动选择最理想的机器学习算法和分子特征化方法的组合进行预测。

参数说明

Small Molecules

待预测的小分子文件，SDF格式。

Properties

ADMET预测列表，ADMET性质见结果说明部分。

Predicted Results

输出的预测结果文件，默认为predicted_results.csv

结果说明

输出结果中，如果是分类模型，输出0或1分类。如果是回归模型，预测出实际值。
ADMET性质信息如下：

ADMET Prediction (v2)

Introduction

ADMET Prediction (v2) is a machine learning-based module for predicting the ADMET properties of small molecules. It enables rapid batch predictions of ADMET properties and supports four common and efficient machine learning algorithms: Graph Attention Neural Network (GAT), Light Gradient Boosting Machine (LightGBM), Random Forest, and Gradient Boosting Machine (GBM). The module supports two methods for molecular feature representation: molecular fingerprints and molecular descriptors, allowing for quick batch predictions on libraries of small molecule compounds. It supports 27 ADMET properties, including 7 regression models and 20 classification models. Users can select the ideal machine learning algorithm and molecular characterization method based on the predictive performance data provided in the documentation. The predictive performance of different machine learning methods and molecular characterization methods is as follows:


The module selects the ideal machine learning algorithm and molecular characterization method automaticaly based on the predictive performance data provided in the documentation.

Parameters

Small Molecules

Small molecular structure file in SDF format

Properties

ADMET properties. Details can be seen in results.

Predicted Results

Output prediction results file name with default predicted_results.csv

Results

In the output results, if it is a classification model, the output will be a classification of 0 or 1. The predicted output will be the actual value if it is a regression model. The endpoint descriptions are as follows:

Structure Prediction (WFold)

简介

Structure Prediction模块是基于最新的生物分子结构预测模型，进行各类生物分子的结构预测。

参数说明

Single Mode

Protein Sequence

蛋白的序列文件，FASTA格式，支持多条序列。
注意：多蛋白复合物结构预测，其氨基酸序列输入格式如下：

>protein1
AASJ...
>protein2
AASJ...
>peptide
ASDF...

DNA Sequence

DNA核酸的序列文件，FASTA格式，支持多条序列。

RNA Sequence

RNA核酸分子的序列文件，FASTA格式，支持多条序列。

Ligand

文本文件包含小分子信息，TXT格式。支持SMILES或 CCD Code（化学组分词典编号）。如果使用SMILES格式，每行应包含一个小分子；如果使用CCD Code，每行可以包含一个或多个小分子，使用逗号分隔，并加上CCD前缀。示例如下：

CC(=O)OC1C[NH+]2CCC1CC2
CCD,ATP,HY3,P1L
CCD,MG

注意：不适用于配体蛋白或多肽的氨基酸序列格式输入。

Modification

包含翻译后修饰（PTM）信息的文本文件，TXT格式。每行放置一个PTM信息，每个PTM信息由三部分组成：

• 发生PTM序列的顺序编号
• PTM类型的CCD编号
• 发生PTM的残基位置编号
  三部分由逗号分隔，例如：1,HY3,1 表示第一条序列的第一个残基，发生了类型为HY3（CCD编号，为3-羟基脯氨酸，为脯氨酸的羟基化）的PTM
  备注：
• 序列的顺序编号，是依次按上述参数Protein、DNA、RNA中的序列顺序与数量，从1开始进行编号，例如：当有2条蛋白序列，1条DNA序列，1条RNA序列时，各序列对应的编号为：第一条蛋白序列编号为1，第二条蛋白序列编号为2，DNA序列编号为3，RNA序列编号为4
• CCD的介绍参考https://www.wwpdb.org/data/ccd 编号查询网址为https://www.ebi.ac.uk/pdbe-srv/pdbechem/
  包含多个PTM信息的文件内容示例如下：

1,HY3,1
1,P1L,5
2,HY3,3

Covalent Bond

共价键信息的文本文件，TXT格式。每行放置一个共价键信息，每个共价键信息包含两个原子信息，每个原子信息由三部分组成：

• 原子所在序列或小分子的顺序编号（编号规则在Modification中定义的序列编号规则基础上，在最后加入小分子的顺序即可）
• 原子所在残基的位置编号（如残基为小分子时，编号为1）
• 原子的标准名称（CCD中定义）
  三部分由逗号分隔，例如：3,1,CA表示第三个实体（序列或小分子）中的第一个残基（或小分子）的CA原子
  一个共价键是由两个原子信息组成，原子间用分号分隔，如：1,1,CA;2,1,CA
  表示一个共价键，该共价键由两个原子组成，第一个原子为1,1,CA，第二个原子为2,1,CA
  包含多个共价键信息的文件内容示例如下：

1,1,CA;2,1,CA
1,1,CA;3,1,CHA

Batch Mode

Protein Sequence

蛋白的序列文件，FASTA格式，支持多条序列。
每一条记录代表一个待预测的结构，每条记录的名称要唯一不能重复。一条记录中有多条链时，通过英文冒号（:）相连，文件内容示例如下：

>1
EVQLVESGGGVVQPGRSLRLS:NFMLTQPHSVSVL:ELFPQWHLPIKIAAIIAS
>2
YYCAKDRRDMGYFQHWGQGTLVTVSSQWYQQRPGSAPTTVIYEDNQRPSGTPPTFMIAVFLPIVVLIFKSILFL

表示有两个待预测的结构，第一条记录的名称为1，有三条蛋白链，用:进行分隔。第二条记录的名称为2，为单链。

DNA Sequence

DNA核酸的序列文件，FASTA格式，支持多条序列。
每一条序列记录代表一个待预测的结构，每条记录的名称要唯一不能重复（可与Protein参数中的记录名称一致，表示该记录的DNA序列与Protein序列归属于同一结构）。一条记录中有多条链时，通过英文冒号（:）相连，文件内容示例如下：

>dna
GACCTCT:CCTAGCT
>1
CCTAGCT

表示有两条记录，第一条的名称为dna，有两条DNA链，用:进行分隔，因为该名称不存在与Protein示例记录中，属于新结构。第二条的名称为1，有一条DNA链，因为该名称存在于Protein示例记录中，则表示同属一个结构（该结构同时包含Protein序列和该DNA序列）。

RNA Sequence

RNA核酸分子的序列文件，FASTA格式，支持多条序列。
每一条序列记录代表一个待预测的结构，每条记录的名称要唯一不能重复（可与DNA或Protein参数中的记录名称一致，表示该记录的RNA序列与上面的DNA序列或Protein序列归属于同一结构）。一条记录中有多条链时，通过英文冒号（:）相连，文件内容示例如下：

>1
AGCU
>rna
AGGCU:UGAUC

表示有两条记录，第一条的名称为1，为单链，因为该名称存在于DNA及Protein示例记录中，表示同属一个结构（该结构同时包含了Protein序列、DNA序列及该RNA序列）。第二条的名称为rna，有两条RNA链，用:进行分隔，因为该名称不存在于DNA或Protein示例记录中，属于新结构。

Ligand

文本文件包含小分子信息，TXT格式。支持SMILES或 CCD Code（化学组分词典编号）。如果使用SMILES格式，每行应包含一个小分子；如果使用CCD Code，每行可以包含一个或多个小分子，使用逗号分隔。
每行代表一个待预测的结构，每行可放置多个ligand，且以唯一不重复的名称开头（该名称可与上述RNA，DNA或Protein参数中的记录名称一致，表示该行的所有ligands，与上述的RNA或DNA或Protein序列归属于同一结构），名称与所有ligands都以英文冒号（:）分隔。文件内容示例如下：

1:CC(=O)OC1C[NH+]2CCC1CC2:CCD,ATP,HY3
lig:CC1CC(CN2NC[N+](=C3CCC(OC(F)(F)F)CC3)C2=O)CC(C)C1OC(C)(C)C(=O)[O-]

表示有两条记录，第一条的名称为1，有三个ligand（一个SMILES，两个CCD codes），因为该名称存在于上述的RNA或DNA或Protein示例记录中，表示同属一个结构。第二条的名称为lig，有一个ligand（为SMILES），因为该名称不存在上述的RNA或DNA或Protein示例记录中，属于新结构。

Modification

包含翻译后修饰（PTM）信息的文本文件，TXT格式。每个PTM的信息与Single模式中一致（参考Single模式中的定义）。
每行定义一个结构的所有PTM信息，且以唯一名称开头（该名称必须存在于前述的Protein或DNA或RNA记录中），都以英文冒号（:）分隔。文件内容示例如下：

1:1,HY3,1:1,P1L,5:2,HY3,3
2:1,HY3,1:2,HY3,3

表示前述名称为1的结构中（Protein或DNA或RNA），有三个PTM。名称为2的结构中，有两个PTM。

Covalent Bond

共价键信息的文本文件，TXT格式。每个共价键的信息与Single模式中一致（参考Single模式中的定义）。
Batch模式下，每行定义一个结构的所有共价键，且以唯一名称开头（该名称必须存在于前述的Protein或DNA或RNA记录中），都以英文冒号（:）分隔。文件内容示例如下：

1:1,1,CA;2,1,CA:1,1,CA;3,1,CHA
2:1,1,CA;3,1,CHA

表示前述名称为1的结构中（Protein或DNA或RNA），有两个共价键。名称为2的结构中，有一个共价键。

Enhanced Mode

该模式下，会默认使用1000个随机种子，每个随机种子进行5个结构采样，共进行5000个结构的大批量采样，并从中选择评分靠前的多个预测结构，最终获得更高精度的预测结构。该模式特别适用于抗原-抗体复合物结构的高精度预测，有研究表明该模式下抗体-抗原复合物结构预测准确性提升60%。该模式的输入参数与Single Mode一致，一次运行时间约10~20小时。
备注：Enhanced Mode模式下预测的氨基酸序列不能超过450AA。

结果说明

输出结果文件为排名前5的复合物结构rank_1-5.pdb和pred_scores.csv，csv中包含信息如下：

Structure Prediction (WFold)

Introduction

The Structure Prediction module is designed to predict the structures of various biomolecules based on the latest biomolecular structure prediction models.

Parameter

Single Mode

Protein Sequence

The protein sequence file in FASTA format, supporting multiple sequences.
Note: For multi-protein complex structure prediction, the amino acid sequence input format is as follows:

>protein1
AASJ...
>protein2
AASJ...
>peptide
ASDF...

DNA Sequence

The DNA sequence file in FASTA format, supporting multiple sequences.

RNA Sequence

The RNA sequence file in FASTA format, supporting multiple sequences.

Ligand

A text file containing information on small molecules in TXT format. It supports either SMILES or CCD Code (Chemical Component Dictionary number). If using SMILES format, each line should contain one small molecule; if using CCD Code, each line can contain one or more small molecules, separated by commas and prefixed with CCD. Example:

CC(=O)OC1C[NH+]2CCC1CC2
CCD,ATP,HY3,P1L
CCD,MG

Note: Not applicable for amino acid sequence format input of ligand proteins or polypeptides.

Modification

A text file containing post-translational modification (PTM) information in TXT format. Each line contains one PTM entry, which consists of three parts:

• The order number of the sequence where the PTM occurs.
• The CCD number for the type of PTM.
• The position number of the residue where the PTM occurs.

These three parts are separated by commas. For example, 1,HY3,1 indicates that the first residue of the first sequence has undergone a PTM of type HY3 (CCD number for 3-hydroxyproline, which is hydroxylation of proline).

Remarks:

• The sequence order number is assigned sequentially based on the order and number of sequences in the Protein, DNA, and RNA parameters, starting from 1. For example, if there are 2 protein sequences, 1 DNA sequence, and 1 RNA sequence, the corresponding numbers would be: the first protein sequence is 1, the second protein sequence is 2, the DNA sequence is 3, and the RNA sequence is 4.
• For information on CCD, refer to https://www.wwpdb.org/data/ccd. The number query website is https://www.ebi.ac.uk/pdbe-srv/pdbechem/.

Example content of a file with multiple PTM entries:

1,HY3,1
1,P1L,5
2,HY3,3

Covalent Bond

A text file containing covalent bond information in TXT format. Each line contains one covalent bond entry, which consists of two atom entries. Each atom entry consists of three parts:

• The order number of the sequence or small molecule (the numbering rule is based on the sequence numbering defined in Modification, with the addition of the order of the small molecule).
• The position number of the residue where the atom is located (if the residue is a small molecule, the number is 1).
• The standard name of the atom (as defined in CCD).

These three parts are separated by commas. For example, 3,1,CA indicates the CA atom of the first residue (or small molecule) in the third entity (sequence or small molecule).

A covalent bond is composed of two atom entries, separated by a semicolon, such as 1,1,CA;2,1,CA, indicating that this covalent bond consists of the first atom 1,1,CA and the second atom 2,1,CA.

Example content of a file with multiple covalent bond entries:

1,1,CA;2,1,CA
1,1,CA;3,1,CHA

Batch Mode

Protein Sequence

The protein sequence file in FASTA format, supporting multiple sequences. Each record represents a structure to be predicted, and each record name must be unique. If a record contains multiple chains, they should be connected by a colon (:). Example content:

>1
EVQLVESGGGVVQPGRSLRLS:NFMLTQPHSVSVL:ELFPQWHLPIKIAAIIAS
>2
YYCAKDRRDMGYFQHWGQGTLVTVSSQWYQQRPGSAPTTVIYEDNQRPSGTPPTFMIAVFLPIVVLIFKSILFL

This indicates two structures to be predicted, with the first record named 1 containing three protein chains separated by colons. The second record is named 2 and contains a single chain.

DNA Sequence

The DNA sequence file in FASTA format, supporting multiple sequences. Each sequence record represents a structure to be predicted, and each record name must be unique. (It can match the record names in the Protein parameter, indicating that the DNA sequence belongs to the same structure as the Protein sequence.) If a record contains multiple chains, they should be connected by a colon (:). Example content:

>dna
GACCTCT:CCTAGCT
>1
CCTAGCT

This indicates two records, with the first named dna containing two DNA chains separated by a colon. Since this name does not appear in the Protein example records, it represents a new structure. The second record is named 1, containing one DNA chain, and since this name exists in the Protein example records, it indicates that they belong to the same structure (which contains both Protein and DNA sequences).

RNA Sequence

The RNA sequence file in FASTA format, supporting multiple sequences. Each sequence record represents a structure to be predicted, and each record name must be unique. (It can match the record names in the DNA or Protein parameters, indicating that the RNA sequence belongs to the same structure.) If a record contains multiple chains, they should be connected by a colon (:). Example content:

>1
AGCU
>rna
AGGCU:UGAUC

This indicates two records, with the first named 1, which is a single chain. Since this name exists in the DNA and Protein example records, it indicates that they belong to the same structure (which includes Protein, DNA, and this RNA sequence). The second record is named rna, containing two RNA chains separated by a colon. Since this name does not appear in the DNA or Protein example records, it represents a new structure.

Ligand

A text file containing information on small molecules in TXT format. It supports either SMILES or CCD Code. If using SMILES format, each line should contain one small molecule; if using CCD Code, each line can contain one or more small molecules, separated by commas. Each line represents a structure to be predicted, and each line must start with a unique name (this name can match those in the RNA, DNA, or Protein parameters, indicating that all ligands in that line belong to the same structure). The name and all ligands are separated by a colon (:). Example content:

1:CC(=O)OC1C[NH+]2CCC1CC2:CCD,ATP,HY3
lig:CC1CC(CN2NC[N+](=C3CCC(OC(F)(F)F)CC3)C2=O)CC(C)C1OC(C)(C)C(=O)[O-]

This indicates two records, with the first named 1, containing three ligands (one SMILES and two CCD codes). Since this name exists in the RNA, DNA, or Protein example records, it indicates that they belong to the same structure. The second record is named lig, containing one ligand (in SMILES format). Since this name does not appear in the RNA, DNA, or Protein example records, it represents a new structure.

Modification

A text file containing post-translational modification (PTM) information in TXT format. Each PTM entry is consistent with that in Single mode (refer to the definitions in Single mode). Each line defines all PTM information for a structure, starting with a unique name (this name must exist in the previously defined Protein, DNA, or RNA records), separated by colons (:). Example content:

1:1,HY3,1:1,P1L,5:2,HY3,3
2:1,HY3,1:2,HY3,3

This indicates that the structure named 1 (Protein, DNA, or RNA) has three PTMs, while the structure named 2 has two PTMs.

Covalent Bond

A text file containing covalent bond information in TXT format. Each covalent bond entry is consistent with that in Single mode (refer to the definitions in Single mode). In Batch mode, each line defines all covalent bonds for a structure, starting with a unique name (this name must exist in the previously defined Protein, DNA, or RNA records), separated by colons (:). Example content:

1:1,1,CA;2,1,CA:1,1,CA;3,1,CHA
2:1,1,CA;3,1,CHA

This indicates that the structure named 1 (Protein, DNA, or RNA) has two covalent bonds, while the structure named 2 has one covalent bond.

Enhanced Mode

In this mode, a default of 1000 random seeds will be used, with each seed conducting 5 structural samplings, totaling 5000 structures for large-scale sampling. From these, multiple predicted structures with high scores will be selected to ultimately obtain a more accurate predicted structure. This mode is particularly suitable for high-precision prediction of antigen-antibody complex structures, and studies have shown that the accuracy of antibody-antigen complex structure prediction can be increased by 60% in this mode. The input parameters for this mode are consistent with those in Single Mode, and the runtime for one session is approximately 10 to 20 hours.
Note: In Enhanced Mode, the predicted amino acid sequence cannot exceed 450 amino acids.

Result

The output result files include the top 5 ranked complex structures rank_1-5.pdb and ranking_scores.csv, which contains the following information:

Evaluate Nucleic Acid (AlphaRNA)

简介

Evaluate Nucleic Acid (AlphaRNA)模块用于评估核酸序列的其表达量和半衰期、抗体滴度等。支持人、小鼠、大鼠、猪等种属。

参数说明

Nucleic Acid Sequence

核酸序列，必须为3的倍数，否则截断尾部序列以达到3的倍数序列，比如：GGAGCCGAGGCGGGCCGATTCACGATCGGTTCGCAAAAACTGTTTGGG

Specis

序列所属物种，Homo_Sapiens、Mamalian、Pig、Rat。

结果

输出结果文件为result.csv，包含信息如下：

Evaluate Nucleic Acid (AlphaRNA)

Introduction

The Evaluate Nucleic Acid (AlphaRNA) module is used to assess the expression levels, half-lives, antibody titers, and other characteristics of nucleic acid sequences.

Parameter

Nucleic Acid Sequence

The nucleic acid sequence must be a multiple of three; otherwise, the tail of the sequence will be truncated to achieve a length that is a multiple of three. For example: GGAGCCGAGGCGGGCCGATTCACGATCGGTTCGCAAAAACTGTTTGGG.

Species

The species to which the sequence belongs, such as Homo_Sapiens, Mammalian, Pig, or Rat.

Results

The output result file is result.csv, which contains the following information:

Back Mutation Grouping v2.4

简介

Back Mutation Grouping是抗体人源化设计流程中分组模块，根据Mutation Score模块输出的回复突变评分表对回复突变进行分组，并返回突变后的序列。

参数说明

Grafted Chain

抗体CDR区嫁接后序列文件，FASTA格式，由Grafting模块生成

Raw Chain

抗体序列文件，FASTA格式

Mutation Score

人源化突变评分文件，CSV格式，由Mutation Score模块生成

Output File

指定输出的突变序列文件名称，FASTA格式

Cutoff

打分分组的截断值，逗号分割，例如：2,5,10表示将氨基酸突变评分大于10的为一组，5~10的氨基酸为一组，小于2的氨基酸分为一组。

Output Policy

指定输出的回复突变的文件

结果说明

根据不同截断值得到突变分组结果文件mutate_policy.json。

Back Mutation Grouping v2.4

Introduction

Back Mutation Grouping is a grouping module in the humanization design process of antibodies, which groups back mutations based on the mutation score table output by the Mutation Score module and returns the mutated sequences.

Parameters

Grafted Chain

Sequence file of the antibody CDR region after grafting, in FASTA format, generated by the Grafting module.

Raw Chain

Sequence file of the antibody, in FASTA format.

Mutation Score

Humanization mutation score file, in CSV format, generated by the Mutation Score module.

Output File

Specify the name of the output mutation sequence file, in FASTA format.

Cutoff

Cutoff values for scoring grouping, separated by commas. For example, 2,5,10 indicates grouping amino acid mutations with scores greater than 10 in one group, amino acids with scores between 5 and 10 in another group, and amino acids with scores less than 2 in a separate group.

Output Policy

Specify the file for the output of back mutations.

Results

The mutation grouping results file, mutate_policy.json, is generated based on different cutoff values.

Template-guided structure prediction

简介

基于自定义的蛋白结构模板，采用colabfold进行蛋白结构预测。

参数说明

Protein Sequence

蛋白的序列文件，FASTA格式

Template Structure

蛋白的模板结构，PDB格式

结果说明

其中scores.csv包含如下信息：

参考文献

Mirdita M, Schütze K, Moriwaki Y, Heo L, Ovchinnikov S, Steinegger M. ColabFold: making protein folding accessible to all. Nat Methods. 2022 Jun;19(6):679-682.

Template-guided Structure Prediction

Introduction

Protein structure prediction is performed using ColabFold based on a custom protein structure template.

Parameter

Protein Sequence

The sequence file of the protein in FASTA format.

Template Structure

The template structure of the protein in PDB format.

Result Description

The scores.csv file contains the following information:

References

Mirdita M, Schütze K, Moriwaki Y, Heo L, Ovchinnikov S, Steinegger M. ColabFold: making protein folding accessible to all. Nat Methods. 2022 Jun;19(6):679-682.

TCR-pMHC Complex Structure Prediction

简介

细胞免疫系统是人体免疫的重要组成部分，它使用 T 细胞受体 (TCR) 识别由主要组织相容性复合体 (MHC) 蛋白呈递的肽形式的抗原蛋白。准确定义TCR的结构基础及其与肽-MHC的结合可以为正常和异常免疫提供重要见解，并有助于指导疫苗和免疫疗法的设计。鉴于实验确定的TCR-肽-MHC结构数量有限，而每个个体内的TCR以及抗原靶标数量巨大，因此需要准确的建模方法。该模块基于TCRmodel2实现，TCRmodel2在AlphaFold基础上对TCR-肽-MHC复合物建模做了优化，与原生AlphaFold和其他基于基准测试的TCR-肽-MHC复合物建模方法相比，其准确度相似或更高，可在30分钟内完成复合物结构预测。

参数说明

TCR α

TCR α链的序列，如：AQEVTQIPAALSVPEGENLVLNCSFTDSAIYNLQWFRQDPGKGLTSLLLIQSSQREQTSGRLNASLDKSSGRSTLYIAASQPGDSATYLCAVTNQAGTALIFGKGTTLSVSS

TCR β

TCR β链的序列，如：NAGVTQTPKFQVLKTGQSMTLQCSQDMNHEYMSWYRQDPGMGLRLIHYSVGAGITDQGEVPNGYNVSRSTTEDFPLRLLSAAPSQTSVYFCASSYSIRGSRGEQFFGPGTRLTVL

Peptide Sequence

多肽序列，如：LAWEWWRTVAL
注：输入的多肽序列长度需要符合相应要求，如下：
I型TCR-pMHC复合物中，多肽的序列长度在8-15之间；
II型TCR-pMHC复合物中，多肽的长度为11。

MHC(I or II α)

MHC-I型序列或MHC-II α链序列。
当预测I型TCR-pMHC复合物时，输入MHC-I型序列，如：SHSLKYFHTSVSRPGRGEPRFISVGYVDDTQFVRFDNDAASPRMVPRAPWMEQEGSEYWDRETRSARDTAQIFRVNLRTLRGYYNQSEAGSHTLQWMHGCELGPDGRFLRGYEQFAYDGKDYLTLNEDLRSWTAVDTAAQISEQKSNDASEAEHQRAYLEDTCVEWLHKYLEKGKETLLH
当预测II型TCR-pMHC复合物时，输入MHC-II α链序列，如：IKADHVSTYAAFVQTHRPTGEFMFEFDEDEMFYVDLDKKETVWHLEEFGQAFSFEAQGGLANIAILNNNLNTLIQRSNHTQAT

MHC II β

MHC-II β链序列，当预测II型TCR-pMHC复合物时才需要输入，如：PENYLFQGRQECYAFNGTQRFLERYIYNREEFARFDSDVGEFRAVTELGRPAAEYWNSQKDILEEKRAVPDRMCRHNYELGGPMTLQR

结果说明

输出结果包括：

其中scores.csv包含如下信息：

参考文献

Yin R, Ribeiro-Filho HV, Lin V, Gowthaman R, Cheung M, Pierce BG. TCRmodel2: high-resolution modeling of T cell receptor recognition using deep learning. Nucleic Acids Res. 2023 Jul 5;51(W1):W569-W576.

TCR-pMHC Complex Structure Prediction

Introduction

The cellular immune system is a crucial component of the human immune response, utilizing T cell receptors (TCRs) to recognize peptide-form antigens presented by major histocompatibility complex (MHC) proteins. Accurately defining the structural basis of TCRs and their binding to peptide-MHC complexes can provide important insights into both normal and abnormal immune responses and assist in guiding the design of vaccines and immunotherapies. Given the limited number of experimentally determined TCR-peptide-MHC structures and the vast number of TCRs and antigen targets within each individual, accurate modeling methods are needed. This module is based on TCRmodel2, which optimizes TCR-peptide-MHC complex modeling on the foundation of AlphaFold. It achieves comparable or higher accuracy than native AlphaFold and other benchmark-based TCR-peptide-MHC modeling methods, completing complex structure predictions within 30 minutes.

Parameter

TCR α

The sequence of the TCR α chain, for example: AQEVTQIPAALSVPEGENLVLNCSFTDSAIYNLQWFRQDPGKGLTSLLLIQSSQREQTSGRLNASLDKSSGRSTLYIAASQPGDSATYLCAVTNQAGTALIFGKGTTLSVSS

TCR β

The sequence of the TCR β chain, for example: NAGVTQTPKFQVLKTGQSMTLQCSQDMNHEYMSWYRQDPGMGLRLIHYSVGAGITDQGEVPNGYNVSRSTTEDFPLRLLSAAPSQTSVYFCASSYSIRGSRGEQFFGPGTRLTVL

Peptide Sequence

The peptide sequence, for example: LAWEWWRTVAL.
Note: The length of the input peptide sequence must meet the following requirements:
For Class I TCR-pMHC complexes, the peptide sequence length should be between 8-15;
For Class II TCR-pMHC complexes, the peptide length is 11.

MHC (I or II α)

The MHC-I sequence or MHC-II α chain sequence.
When predicting Class I TCR-pMHC complexes, input the MHC-I sequence, for example: SHSLKYFHTSVSRPGRGEPRFISVGYVDDTQFVRFDNDAASPRMVPRAPWMEQEGSEYWDRETRSARDTAQIFRVNLRTLRGYYNQSEAGSHTLQWMHGCELGPDGRFLRGYEQFAYDGKDYLTLNEDLRSWTAVDTAAQISEQKSNDASEAEHQRAYLEDTCVEWLHKYLEKGKETLLH.
When predicting Class II TCR-pMHC complexes, input the MHC-II α chain sequence, for example: IKADHVSTYAAFVQTHRPTGEFMFEFDEDEMFYVDLDKKETVWHLEEFGQAFSFEAQGGLANIAILNNNLNTLIQRSNHTQAT.

MHC II β

The MHC-II β chain sequence, which is required only when predicting Class II TCR-pMHC complexes, for example: PENYLFQGRQECYAFNGTQRFLERYIYNREEFARFDSDVGEFRAVTELGRPAAEYWNSQKDILEEKRAVPDRMCRHNYELGGPMTLQR.

Result

The output results include:

The scores.csv contains the following information:

References

Yin R, Ribeiro-Filho HV, Lin V, Gowthaman R, Cheung M, Pierce BG. TCRmodel2: high-resolution modeling of T cell receptor recognition using deep learning. Nucleic Acids Res. 2023 Jul 5;51(W1):W569-W576.

Alanine Scan (MMPBSA)

简介

Alanine Scan (MMPBSA)是计算丙氨酸突变后的结合自由能，并且提供能量分解数据、结合常数（Ka）、抑制剂常数（Ki）。熵的计算采用的是张增辉教授的相互作用熵的方法，该方法直接从分子动力学模拟计算结合自由能的熵组分（相互作用熵或-TΔS），但是需要选取结构稳定部分进行计算。
本模块提供四种计算方法，其中 Trajectory 是针对进行分子动力学模拟轨迹后计算其受配体之间的结合自由能；One Structure 是针对一帧pdb结构或者对接得到的PDB结构计算受配体之间的结合自由能，MMPBSA of One Structure 计算流程中可以直接输入PDB进行计算。
Index Name 是输入受配体组别名称；Custom Name 则是输入受配体的在PDB中的残基编号。

参数说明

Trajectory方法

Path File

MD模拟后得到的路径文件，可以在GMX MD Run (GMX2023)模块或者AlphaAutoMD模块中获取。

Receptor Name

受体名称，可以为Protein、DNA、RNA。

Ligand Name

配体名称，可以为Protein、DNA、RNA。如果为小分子，填写其在PDB中的名称。如果体系中除了蛋白以外为配体（包括小分子）可用Other表示。

Mutation Residue

突变扫描为丙氨酸（ALA）的氨基酸位置。格式为res1:res2:res3:res4，其中“res1-res4”数字为残基编号。

Force File

丙氨酸扫描时使用的力场。

Start Time (ps)

起始帧时间，单位ps。最好选取RMSD稳定时区进行计算，以消除结构不稳定时导致的整体熵偏大。

End Time (ps)

结束帧时间，单位ps。最好选取RMSD稳定时区进行计算，以消除结构不稳定时导致的整体熵偏大。

Skip Time (ps)

间隔时间，单位ps。

Index File

索引文件，ndx格式。当体系中存在膜结构的时候可以通过Membrane Solvation模块获取index.ndx文件。膜模拟中的受体为receptor，配体为ligand，膜为membrane。只能是AMBER力场下构建的膜体系才能进行计算。

Custom Receptor

定义受体组别进行结合能计算。组别中填写的为蛋白氨基酸或者核酸碱基的序号。例如1-213或者1-211,212-213。蛋白氨基酸或者核酸碱基序号从1开始重新编号，与初始pdb氨基酸编号无关。

Custom Ligand

定义配体组别进行结合能计算。组别中填写的为蛋白氨基酸或者核酸碱基的序号。例如1-213或者1-211,212-213。蛋白氨基酸或者核酸碱基序号从1开始重新编号，与初始pdb氨基酸编号无关。

One Structure方法

System Topology

拓扑文件，由MD Solvation模块或者Membrane Solvation模块得到。

System GRO

结构文件，.gro格式，由MD Solvation模块或者Membrane Solvation模块得到。

System ITP

体系参数压缩文件，tar.gz格式。由MD Solvation模块或者Membrane Solvation模块得到。

结果说明

输出结果包括：

参考文献

GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, E. Lindahl, SoftwareX, 1, (2015), 19-25.
Duan L, Liu X, Zhang JZ. Interaction Entropy: A New Paradigm for Highly Efficient and Reliable Computation of Protein-Ligand Binding Free Energy. J Am Chem Soc. 2016 May 4;138(17):5722-8.
https://github.com/Electrostatics/apbs/blob/main/docs/index.rst

Alanine Scan (MMPBSA)

Introduction

Alanine Scan (MMPBSA) calculates the binding free energy after alanine mutations and provides energy decomposition data, binding constants (Ka), and inhibitor constants (Ki). The entropy calculation uses the method of interaction entropy proposed by Professor Zhang Zenghui, which directly calculates the entropy component of the binding free energy (interaction entropy or -TΔS) from molecular dynamics simulations, but requires selecting stable structural regions for calculation. This module provides four calculation methods. Among them, Trajectory calculates the binding free energy between the receptor and ligand after molecular dynamics simulation trajectories; One Structure calculates the binding free energy between the receptor and ligand for a single frame PDB structure or a docked PDB structure; MMPBSA of One Structure allows direct input of a PDB for calculation. Index Name is the input receptor-ligand group name, while Custom Name is the input residue numbers of the receptor-ligand in the PDB.

Parameter Description

Trajectory Method

Path File

Path file obtained after MD simulation, available in the GMX MD Run (GMX2023) module or AlphaAutoMD module.

Receptor Name

Name of the receptor, can be Protein, DNA, or RNA.

Ligand Name

Name of the ligand, can be Protein, DNA, or RNA. If it is a small molecule, enter its name in the PDB. Use ‘Other’ if the system contains a ligand (including small molecules) other than proteins.

Mutation Residue

Amino acid positions where mutations to alanine (ALA) are scanned. The format is res1:res2:res3:res4, where “res1-res4” are residue numbers.

Force File

Force field used for alanine scanning.

Start Time (ps)

Start frame time in ps. It is best to select a stable RMSD region for calculation to eliminate large overall entropy caused by structural instability.

End Time (ps)

End frame time in ps. It is best to select a stable RMSD region for calculation to eliminate large overall entropy caused by structural instability.

Skip Time (ps)

Time interval in ps.

Index File

Index file in ndx format. When there is a membrane structure in the system, you can obtain the index.ndx file through the Membrane Solvation module. In membrane simulations, the receptor is ‘receptor’, the ligand is ‘ligand’, and the membrane is ‘membrane’. Only membrane systems built under the AMBER force field can be calculated.

Custom Receptor

Define receptor groups for binding energy calculation. Enter the sequence numbers of protein amino acids or nucleic acid bases in the group. For example, 1-213 or 1-211,212-213. Protein amino acid or nucleic acid base numbers start from 1 and are independent of the initial pdb amino acid numbering.

Custom Ligand

Define ligand groups for binding energy calculation. Enter the sequence numbers of protein amino acids or nucleic acid bases in the group. For example, 1-213 or 1-211,212-213. Protein amino acid or nucleic acid base numbers start from 1 and are independent of the initial pdb amino acid numbering.

One Structure Method

System Topology

Topology file obtained from the MD Solvation module or Membrane Solvation module.

System GRO

Structure file in .gro format obtained from the MD Solvation module or Membrane Solvation module.

System ITP

System parameter compression file in tar.gz format. Obtained from the MD Solvation module or Membrane Solvation module.

Result Description

The output includes:

References

GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, E. Lindahl, SoftwareX, 1, (2015), 19-25.
Duan L, Liu X, Zhang JZ. Interaction Entropy: A New Paradigm for Highly Efficient and Reliable Computation of Protein-Ligand Binding Free Energy. J Am Chem Soc. 2016 May 4;138(17):5722-8.
https://github.com/Electrostatics/apbs/blob/main/docs/index.rst

Human Germline BLAST

简介

通过序列比对在人类生殖系数据库中搜索与目标抗体序列最接近的同源模板，输出对应的模板序列以及序列一致性信息。

参数说明

Sequence String模式

Input Sequence

抗体的序列（纯序列信息，非FASTA格式文件）。

Type

抗体编号类型：kabat、chothia、imgt。

TopHits

输出同源性最高的n条序列。

Fasta File模式

FASTA File

抗体的序列文件，FASTA格式。

Type

抗体编号类型：kabat、chothia、imgt。

TopHits

输出同源性最高的n条序列。

结果说明

相关内容

抗体常用的germline模板：

临床后期及已上市抗体的germline配对分布情况（统计自Adimab数据集）：

Human Germline BLAST

Introduction

This module performs sequence alignment to search for the closest homologous template in the human germline database for a given target antibody sequence. It outputs the corresponding template sequence along with sequence similarity information.

Parameter Description

Sequence String Mode

Input Sequence

The antibody sequence (pure sequence information, not in FASTA format).

Type

Type of antibody numbering: kabat, chothia, imgt.

TopHits

Number of top hits to output.

Fasta File Mode

FASTA File

Antibody sequence file in FASTA format.

Type

Type of antibody numbering: kabat, chothia, imgt.

TopHits

Number of top hits to output.

Result Description

Related Content

Commonly used germline templates for antibodies:

Distribution of germline pairing for late-stage and marketed antibodies (statistics from the Adimab dataset):

Humanization Report

简介

Humanization Report是抗体人源化设计报告生成模块，用于生成最终的抗体人源化设计报告以及相应的专利实施例段落。

参数说明

Graft Policy

Grafting模块生成的Graft Policy文件。

Mutate Policy

Back Mutation Grouping模块生成的Policy文件。

结果说明

输出结果包括：

其中batch_registration_template.xlsx包含如下信息：

其中hotspot_summary.xlsx包含如下信息：

Humanization Report

Introduction

The Humanization Report is a module for generating reports on antibody humanization design, including the final antibody humanization design report and corresponding paragraphs for patent implementation examples.

Parameter Description

Graft Policy

The Graft Policy file generated by the Grafting module.

Mutate Policy

The Policy file generated by the Back Mutation Grouping module.

Result Description

The output results include:

The batch_registration_template.xlsx file contains the following information:

The hotspot_summary.xlsx file contains the following information:

Back Mutation Grouping v2.3

简介

Back Mutation Grouping是抗体人源化设计流程中分组模块，根据Mutation Score模块输出的回复突变评分表对回复突变进行分组，并返回突变后的序列。

参数说明

Grafted Chain

抗体CDR区嫁接后序列文件，FASTA格式，由Grafting模块生成

Raw Chain

抗体序列文件，FASTA格式

Mutation Score

人源化突变评分文件，CSV格式，由Mutation Score模块生成

Output File

指定输出的突变序列文件名称，FASTA格式

Cutoff

打分分组的截断值，逗号分割，例如：2,5,10表示将氨基酸突变评分大于10的为一组，5~10的氨基酸为一组，小于2的氨基酸分为一组。

Output Policy

指定输出的回复突变的文件

结果说明

根据不同截断值得到突变分组结果文件mutate_policy.json。

Back Mutation Grouping v2.3

Introduction

Back Mutation Grouping is a grouping module in the humanization design process of antibodies, which groups back mutations based on the mutation score table output by the Mutation Score module and returns the mutated sequences.

Parameters

Grafted Chain

Sequence file of the antibody CDR region after grafting, in FASTA format, generated by the Grafting module.

Raw Chain

Sequence file of the antibody, in FASTA format.

Mutation Score

Humanization mutation score file, in CSV format, generated by the Mutation Score module.

Output File

Specify the name of the output mutation sequence file, in FASTA format.

Cutoff

Cutoff values for scoring grouping, separated by commas. For example, 2,5,10 indicates grouping amino acid mutations with scores greater than 10 in one group, amino acids with scores between 5 and 10 in another group, and amino acids with scores less than 2 in a separate group.

Output Policy

Specify the file for the output of back mutations.

Results

The mutation grouping results file, mutate_policy.json, is generated based on different cutoff values.

Mutation Score

简介

Mutation Score是抗体人源化设计中核心模块，是一个基于结构的自动化评分模块。该模块基于抗体的结构信息以及CDR嫁接后的序列信息，对移植抗体（graft）后的FR区的每个氨基酸在替换前后的变化程度进行量化评分。评分越高，说明CDR嫁接过程中氨基酸的替换对CDR区的构象可能影响较大，需要进行回复突变。模块输出每个氨基酸的打分值，用于抗体人源化设计流程中后续的分组和人源化抗体序列的生成。

参数说明

Sequence File

抗体Fv区序列文件，FASTA格式。

Model File

抗体结构文件，PDB格式。

Grafted Sequence

抗体CDR区Graft后的序列文件，FASTA格式。

Output Score

指定输出打分文件的名称，CSV格式。