Underlying data for "AlphaDesign: A de novo protein design framework based on AlphaFold"

This repository contains the underlying data for the manuscript "AlphaDesign: A de novo protein design framework based on AlphaFold".

code.tar.gz

Contains the source code for AlphaDesign under a CC BY NC SA 4.0 license, as well as the code for novobench under an Apache 2.0 license. It also includes readme files for both, and a directory of example inputs / expected outputs.

designs.tar.gz

This archive contains de novo designed protein structures, their corresponding designed sequences and AlphaFold / ESMfold-based scores for each designed sequence-structure pair. Scores are provided in the CSV format described below.

*_scores.csv file format

The *_scores.csv files contain values of relevant scores for quantifying the number of successful designs and selecting designs for experimental validation. They contain the following columns in order:

result: base name of the backbone PDB-file for this row, e.g. backbone_0.pdb

index: unique integer index of the designed sequence for this row.

sequence: designed sequence for this row.

sc_rmsd: self-consistent RMSD (scRMSD) between the designed backbone and the predicted structure of this row's sequence.

sc_tm: self-consistent TM-score (scTM) between designed and predicted structure.

plddt: mean pLDDT over all non-templated positions in the predicted structure.

pae: mean pAE of the predicted structure.

ipae: mean interface pAE of the predicted structure, if multiple chains are present. Otherwise inf.

mpae: minimum interface pAE of the predicted structure, if multiple chains are present. Otherwise inf.

subdirectories

The archive has the following directory structure:

├── binder_targets

│   ├── ...

├── binders

│   ├── pd1-100

│   ├── ...

├── complexes

│   ├── heterodimer-50-1.4

│   ├── ...

├── monomers

│   ├── camsol

│   ├── monomer-100-1.4

│   ├── ...

├── multistate

│   ├── confchange-tm0.1-50

│   └── ...

└── rf_diffusion

    ├── monomer-50

    └── ...

The following subdirectories of the archive (monomers, complexes, binders, multistate) contain files for each type and size of protein design. E.g. for de novo designed complexes:

complexes

├── heterodimer-50-1.4

│   ├── alpha_design

│   ├── alphafold_redesign_scores.csv

│   ├── esmfold_raw_scores.csv

│   ├── esmfold_redesign_scores.csv

│   ├── mpnn_redesign

│   └── redesign_template.yaml

├── …

Here, the alpha_design directory contains PDB-files of the designed structure candidates from the first step of the AlphaDesign pipeline. The adm_redesign directory contains a FASTA-file for each PDB-file in alpha_design. This FASTA-file contains the set of redesigned sequences generated using the ADM for this PDB-file. redesign_template.yaml contains the instructions used for redesigning the sequence for each PDB-file.

In addition, these subdirectories contain *_scores.csv files. For monomers and complexes, there are three of these:

esmfold_raw_scores.csv: scores for raw sequence-structure pairs from the first step of the AlphaDesign pipeline using ESMfold.

esmfold_redesign_scores.csv: scores for ADM-redesigned sequence-structure pairs using ESMfold.

alphafold_redesign_scores.csv: scores for ADM-redesigned sequence-structure pairs using AlphaFold.

For binders and multistate, only scores of ADM-redesigned sequence-structure pairs using AlphaFold are reported. Additionally, for multistate, scores are reported for all concurrently designed states of each protein:

alphafold_redesign_scores_state_X.pdb: scores using AlphaFold for designed state “X”. State 0 corresponds to the complex / bound state, and states 1 / 2 correspond to the two monomeric states for conformation-changing de novo designs. For bispecific binder designs, state 0 corresponds to the complex with the first target, state 1 to the second.

monomers additionally contains a camsol subdirectory:

monomers/camsol

├── camsol_intrinsic_raw_monomer_sequences.txt

├── camsol_intrinsic_redesigned_monomer_sequences.txt

├── raw_sequence.fasta

└── redesigned_sequence.fasta

This directory contains the raw sequences for all designed monomers from the first step of the AlphaDesign pipeline (raw_sequence.fasta), the corresponding best ADM-redesigned sequences (redesigned_sequence.fasta) and CamSol solubility scores for both (camsol_intrinsic_raw_monomer_sequences.txt; camsol_intrinsic_redesigned_monomer_sequences.txt).

The binder_targets subdirectory contains PDB structures of all the target proteins used for binder design in this work. The RcaT structures here (RcaTSen2_active_domain.pdb and RcaTEco1_active_domain.pdb) contain contiguous cropped structures around the putative active site of these RcaT homologs.

Finally, the rf_diffusion directory and its subdirectories contain structures and sequences generated using RFdiffusion and ProteinMPNN for the comparison between RFdiffusion and AlphaDesign. There are subdirectories for the following types of designs:

monomer-<50, 100, 200, 300>: contain monomer designs with 50 to 300 amino acids

homomer-<2, 3, 4>: contain homooligomers with 50 amino acid monomers and 2 - 4 subunits.

heterodimer: contains heterodimers with 50 amino acid monomers

Each subdirectory contains an alphafold_scores.csv file with AlphaFold scores (as above) for each design; a design subdirectory with designed protein backbones in PDB format; a protein_mpnn directory with designed amino acid sequences in FASTA format.

MD_input.tar.gz

Input files for preparing ensemble all-atom molecular dynamics (MD) simulations of a subset of designed RcaT-Sen2 binder complexes in this work. The directory has the following structure:

├── RcaT_bispecifics

│   └── uncropped

│       └── ... # designed systems

├── RcaT_conf_change

│   └── cropped

│       └── ... # designed systems

├── RcaT_Sen2

│   └── ... # designed systems

└── README

With each subdirectory containing MD input files for a specific class of designed RcaT-Sen2 binders:

RcaT_bispecifics: bispecific binders for RcaT-Sen2 and RcaT-Eco1

RcaT_conf_change: RcaT-Sen2 binders designed to change conformation upon binding

RcaT_Sen2: monospecific binders to RcaT-Sen2

Each of the designed_systems subdirectories has the following contents:

<design>

├── build

│   ├── <design>.pdb

│   └── build.tleap

├── ensemble

│   └── out

│     └── 1

│         └── prod

└── eq

├── out_eq1

│   └── ref-min-10.in

├── ...

│   └── ...

└── out_eq11

    └── ref-equil-NPT.in

These files have the following function:

build/<design.pdb>: PDB-format structure of the designed monomer or target-binder complex.

build/build.tleap: leap file for solvating the structure, adding ions and producing amber parameters and coordinates in the form of .crd, .pdb and .prmtop files.

ensemble/out/1/prod: amber input file for running a single replica production trajectory of an ensemble simulation. To produce additional replicas (2….N), make copies of this directory in the same path.

eq/out_eq<N>/<equilibration>.in: amber input files for each separate equilibration step (11 in total) performed before running production. These equilibration steps are described in detail in the methods section of this work.

MD_outputs.tar.gz

Time-series statistics extracted from all-atom explicit solvent molecular dynamics (MD) runs. This directory has the following structure:

└── <design type>

    └── <target - # amino acids>

        └── <design>

            ├── ensemble_timeseries

            │   └── <(1 - 50)>.dat

            └── prodigy_ensemble

                └── <(1 - 50)>

                    ├── mean.dat

                                           └── time.dat

Each designed binder for which we ran ensemble MD has two directories associated with it:

ensemble_timeseries: This directory contains statistics about each time-step in each replica trajectory (e.g. 1.dat for the first replica trajectory in the ensemble). The data are given in fixed-width format with the following columns:

global RMSD, global RMSF, monomer 1 RMSD, monomer 2 RMSD, monomer 1 RMSF, monomer 2 RMSF, global intra-chain contacts, global inter-chain contacts, global total number of contacts, monomer 1 intra-chain contacts, monomer 2 intra-chain contacts, monomer 1 - monomer 2 interface contacts

prodigy_ensemble: This directory contains per-replica means of Prodigy IC values and predicted binding affinity (mean.dat) and per-replica per-time-step outputs from PRODIGY (time.dat).

本仓库包含论文手稿《AlphaDesign：一款基于AlphaFold的从头蛋白质设计（de novo protein design）框架》的配套原始数据。

code.tar.gz

该压缩包包含遵循CC BY NC SA 4.0许可协议的AlphaDesign源代码，以及遵循Apache 2.0许可协议的novobench源代码，同时附带二者的README文件，以及示例输入/预期输出目录。

designs.tar.gz

该归档文件包含从头蛋白质设计的蛋白质结构、对应的设计序列，以及每一对设计序列-结构对基于AlphaFold和ESMfold计算得到的评分。评分以如下所述的CSV格式提供。

*_scores.csv 文件格式

*_scores.csv文件包含用于量化成功设计数量、筛选用于实验验证的设计的相关评分值，按顺序包含以下列：

result: 当前行对应骨架PDB（Protein Data Bank）文件的基础名称，例如backbone_0.pdb

index: 当前行设计序列的唯一整数索引

sequence: 当前行的设计序列

sc_rmsd: 设计骨架与当前行序列的预测结构之间的自洽均方根偏差（self-consistent RMSD, scRMSD）

sc_tm: 设计结构与预测结构之间的自洽TM评分（self-consistent TM-score, scTM）

plddt: 预测结构中所有非模板化残基的平均pLDDT值

pae: 预测结构的平均pAE值

ipae: 若存在多链，则为预测结构的平均界面pAE值；否则为无穷大（inf）

mpae: 若存在多链，则为预测结构的最小界面pAE值；否则为无穷大（inf）

子目录结构

该归档文件的目录结构如下：
├── binder_targets
│   ├── ...
├── binders
│   ├── pd1-100
│   ├── ...
├── complexes
│   ├── heterodimer-50-1.4
│   ├── ...
├── monomers
│   ├── camsol
│   ├── monomer-100-1.4
│   ├── ...
├── multistate
│   ├── confchange-tm0.1-50
│   └── ...
└── rf_diffusion
    ├── monomer-50
    └── ...

归档文件的以下子目录（monomers、complexes、binders、multistate）包含对应各类别和尺寸的蛋白质设计文件。例如，针对从头设计的复合物：

complexes
├── heterodimer-50-1.4
│   ├── alpha_design
│   ├── alphafold_redesign_scores.csv
│   ├── esmfold_raw_scores.csv
│   ├── esmfold_redesign_scores.csv
│   ├── mpnn_redesign
│   └── redesign_template.yaml
├── …

其中，alpha_design目录包含AlphaDesign流程第一步得到的设计结构候选的PDB文件。ADM重新设计目录包含alpha_design中每个PDB文件对应的FASTA文件，该FASTA文件包含针对该PDB文件使用ADM生成的一组重新设计序列。redesign_template.yaml包含为每个PDB文件重新设计序列时所用的指令。

此类子目录还包含*_scores.csv文件。对于monomers和complexes，共包含三类此类文件：

esmfold_raw_scores.csv：AlphaDesign流程第一步中使用ESMfold对原始序列-结构对计算得到的评分。

esmfold_redesign_scores.csv：使用ESMfold对ADM重新设计的序列-结构对计算得到的评分。

alphafold_redesign_scores.csv：使用AlphaFold对ADM重新设计的序列-结构对计算得到的评分。

对于binders和multistate，仅报告使用AlphaFold对ADM重新设计的序列-结构对得到的评分。此外，针对multistate，还会报告每种蛋白质所有同时设计的构象的评分：

alphafold_redesign_scores_state_X.pdb：针对设计构象“X”使用AlphaFold计算得到的评分。状态0对应复合物/结合态，状态1和2对应构象变化型从头设计的两种单体态。对于双特异性结合体设计，状态0对应与第一个靶标的复合物，状态1对应与第二个靶标的复合物。

monomers子目录还包含一个camsol子目录，其结构如下：

monomers/camsol
├── camsol_intrinsic_raw_monomer_sequences.txt
├── camsol_intrinsic_redesigned_monomer_sequences.txt
├── raw_sequence.fasta
└── redesigned_sequence.fasta

该目录包含AlphaDesign流程第一步中所有设计的单体的原始序列（raw_sequence.fasta）、对应的最优ADM重新设计序列（redesigned_sequence.fasta），以及二者的CamSol溶解度评分（分别存储于camsol_intrinsic_raw_monomer_sequences.txt和camsol_intrinsic_redesigned_monomer_sequences.txt）。

binder_targets子目录包含本研究中用于结合体设计的所有靶标蛋白质的PDB结构。此处的RcaT结构（RcaTSen2_active_domain.pdb和RcaTEco1_active_domain.pdb）包含这些RcaT同源蛋白推定活性位点周围的连续截短结构。

最后，rf_diffusion目录及其子目录包含使用RFdiffusion和ProteinMPNN生成的结构与序列，用于对比RFdiffusion与AlphaDesign的性能。包含以下类型设计的子目录：

monomer-<50, 100, 200, 300>：包含氨基酸长度为50至300的单体设计

homomer-<2, 3, 4>：包含由50个氨基酸残基的单体构成、亚基数为2-4的同源寡聚体设计

heterodimer：包含由50个氨基酸残基的单体构成的异二聚体设计

每个子目录包含：一个alphafold_scores.csv文件，内含每项设计的AlphaFold评分（格式如前所述）；一个design子目录，内含PDB格式的设计蛋白质骨架；一个protein_mpnn目录，内含FASTA格式的设计氨基酸序列。

MD_input.tar.gz

该压缩包包含为本研究中部分设计的RcaT-Sen2结合体复合物制备集合式全原子分子动力学（Molecular Dynamics, MD）模拟所需的输入文件。其目录结构如下：

├── RcaT_bispecifics
│   └── uncropped
│       └── ... # 设计的体系
├── RcaT_conf_change
│   └── cropped
│       └── ... # 设计的体系
├── RcaT_Sen2
│   └── ... # 设计的体系
└── README

每个子目录对应一类特定的设计RcaT-Sen2结合体：

RcaT_bispecifics：针对RcaT-Sen2和RcaT-Eco1的双特异性结合体

RcaT_conf_change：设计为结合后发生构象变化的RcaT-Sen2结合体

RcaT_Sen2：针对RcaT-Sen2的单特异性结合体

每个设计体系子目录的内容如下：

<design>
├── build
│   ├── <design>.pdb
│   └── build.tleap
├── ensemble
│   └── out
│       └── 1
│           └── prod
└── eq
├── out_eq1
│   └── ref-min-10.in
├── ...
│   └── ...
└── out_eq11
    └── ref-equil-NPT.in

这些文件的功能如下：

build/<design>.pdb：设计的单体或靶标-结合体复合物的PDB格式结构

build/build.tleap：用于溶剂化结构、添加离子并生成Amber参数与坐标文件（格式为.crd、.pdb和.prmtop）的leap脚本文件

ensemble/out/1/prod：用于运行集合式模拟中单副本生产轨迹的Amber输入文件。如需生成额外副本（2…N），只需在相同路径下复制该目录即可

eq/out_eq<N>/<equilibration>.in：运行生产模拟前执行的11个独立平衡步骤对应的Amber输入文件。这些平衡步骤的详细说明见于本研究的方法部分。

MD_outputs.tar.gz

该压缩包包含从全原子显式溶剂分子动力学（MD）运行中提取的时间序列统计数据。其目录结构如下：

└── <design type>
    └── <target - # amino acids>
        └── <design>
            ├── ensemble_timeseries
            │   └── <(1 - 50)>.dat
            └── prodigy_ensemble
                └── <(1 - 50)>
                    ├── mean.dat
                                           └── time.dat

每个运行了集合式MD的设计结合体对应两个关联目录：

ensemble_timeseries：该目录包含每个副本轨迹各时间步的统计数据（例如，1.dat对应集合中的第一个副本轨迹）。数据以固定宽度格式存储，各列依次为：全局均方根偏差、全局均方根波动、单体1均方根偏差、单体2均方根偏差、单体1均方根波动、单体2均方根波动、全局链内接触数、全局链间接触数、全局总接触数、单体1链内接触数、单体2链内接触数、单体1-单体2界面接触数

prodigy_ensemble：该目录包含PRODIGY IC值和预测结合亲和力的每副本均值（mean.dat），以及PRODIGY输出的每副本每时间步数据（time.dat）。