---

English | 简体中文

This package provides an implementation of IgGM inference, which can design the overall structure based on a given framework region sequence, as well as tools for CDR region sequences, and can design corresponding antibodies against specific epitopes.

We also provide:

The test set (SAb-23H2) constructed in the paper includes the latest antigen-antibody complexes published since the second half of 2023, with rigorous deduplication applied.

Any publication that discloses findings arising from using this source code or the model parameters should cite the IgGM paper.

Please also refer to the Supplementary Information for a detailed description of the method.

If you have any questions, please contact the IgGM team at fandiwu@tencent.com.

For business partnership opportunities, please contact leslielwang@tencent.com.

OverView

Main Results(SAb-23H2-Ab)

Model	AAR-CDR-H3	RMSD-CDR-H3	DockQ
DiffAb(IgFold)	0.214	2.358	0.022
DiffAb(AlphaFold 3)	0.226	2.300	0.208
MEAN(IgFold)	0.248	2.741	0.022
MEAN(AlphaFold 3)	0.246	2.646	0.207
dyMEAN	0.294	2.454	0.079
IgGM	0.360	2.131	0.246

Installation

1. Clone the package

git clone https://github.com/TencentAI4S/IgGM.git
cd IgGM

2. Prepare the environment

conda env create -n IgGM -f environment.yaml
conda activate IgGM
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.0.1+cu117.html

3. Download pre-trained weights under params directory
   • Zenodo

Note: If you download the weights in the folder ./checkpoints, you can proceed directly with the following steps.

Dataset

Test set we construct in our paper

• Zenodo (download from zenodo)

This folder contains files related to the dataset.

• SAb-23-H2-Ab

  • prot_ids.txt: Text file containing protein IDs.
  • fasta.files.native: Original fasta file (H represents heavy chain, L represents light chain, A represents antigen, corresponding to the last three letters in the ID).
  • pdb.files.native: Original pdb structure file (H represents heavy chain, L represents light chain, A represents antigen, corresponding to the last three letters in the ID).
  • fasta.files.design: Contains the masked CDR area fasta file, where X represents mask.

• SAb-23-H2-Nano

  • prot_ids.txt: Text file containing protein IDs.
  • fasta.files.native: Original fasta file (H represents antibody, NA represents absence, and A represents antigen, which correspond to the last three letters in the ID.).
  • pdb.files.native: Original pdb structure file (H represents antibody, NA represents absence, and A represents antigen, which correspond to the last three letters in the ID.).
  • fasta.files.design: Contains the masked CDR area fasta file, where X represents mask.

Quick Start

You can use a fasta file (--fasta) and antigen's pdb file(--pdb). The epitope information can be provided by specifying the residue numbers in the antigen pdb file.

Example 1: predicting the structure of an antibody & nanobody against a given antigen and binding epitopes using IgGM

• If a complex structure is available in the PDB, the command will automatically generate epitope information. You can delete the epitope information from the command(--epitope) if needed.

# antibody
python design.py --fasta examples/fasta.files.native/8iv5_A_B_G.fasta --antigen examples/pdb.files.native/8iv5_A_B_G.pdb --epitope 126 127 129 145 146 147 148 149 150 155 156 157 158 160 161 162 163 164

# nanobody
python design.py --fasta examples/fasta.files.native/8q94_C_NA_A.fasta --antigen examples/pdb.files.native/8q94_C_NA_A.pdb --epitope 41 42 43 44 45 46 49 50 70 71 73 74

Example 2: redesign the sequence of an antibody & nanobody CDR H3 loops against a given antigen using IgGM, and predict the overall structure.

# antibody
python design.py --fasta examples/fasta.files.design/8hpu_M_N_A/8hpu_M_N_A_CDR_H3.fasta --antigen examples/pdb.files.native/8hpu_M_N_A.pdb

# nanobody
python design.py --fasta examples/fasta.files.design/8q95_B_NA_A/8q95_B_NA_A_CDR_3.fasta --antigen examples/pdb.files.native/8q95_B_NA_A.pdb

Example 3: design the sequence of an antibody & nanobody CDR loops against a given antigen using IgGM, and predict the overall structure.

# antibody
python design.py --fasta examples/fasta.files.design/8hpu_M_N_A/8hpu_M_N_A_CDR_All.fasta --antigen examples/pdb.files.native/8hpu_M_N_A.pdb

# nanobody
python design.py --fasta examples/fasta.files.design/8q95_B_NA_A/8q95_B_NA_A_CDR_All.fasta --antigen examples/pdb.files.native/8q95_B_NA_A.pdb

You can specify other regions for design; more examples can be explored in the examples folder.

Example 4: Design the sequences of antibody and nanobody CDR loops based on given antigens and binding epitopes, without the need to provide complex structural information, and predict the overall structure.

• It is possible to design antibodies targeting a new epitope.

# antibody
python design.py --fasta examples/fasta.files.design/8hpu_M_N_A/8hpu_M_N_A_CDR_All.fasta --antigen examples/pdb.files.native/8hpu_M_N_A.pdb --epitope 126 127 129 145 146 147 148 149 150 155 156 157 158 160 161 162 163 164

# nanobody
python design.py --fasta examples/fasta.files.design/8q95_B_NA_A/8q95_B_NA_A_CDR_All.fasta --antigen examples/pdb.files.native/8q95_B_NA_A.pdb --epitope 41 42 43 44 45 46 49 50 70 71 73 74

For a completely new antigen, you can specify epitopes to design antibodies that can bind to those epitopes.

Citing IgGM

If you use IgGM in your research, please cite our paper

@article{Wang2024IgGM,
	author = {Wang, Rubo and Wu, Fandi and Gao, Xingyu and Wu, Jiaxiang and Zhao, Peilin and Yao, Jianhua},
	title = {IgGM: A Generative Model for Functional Antibody and Nanobody Design},
	elocation-id = {2024.09.19.613838},
	year = {2024},
	doi = {10.1101/2024.09.19.613838},
	publisher = {Cold Spring Harbor Laboratory}
}