Antibody-SGM is a score-based generative modeling for de novo antibody heavychain design. This repository contains the codebase for [Antibody-SGM: Antigen-Specific Joint Design of Antibody Sequence and Structure using Diffusion Models]
To install AB-SGM and the necessary dependancies run the following commands, installation should take less than 15 minutes. We recommend using the conda environment supplied in this repository.
- Clone repository ` git clone https://github.com/xxiexuezhi/ABSGM.git ' or download the zip files and extract all.
- Install conda environment
create -f ab_env.yaml - Activate conda environment
conda activate ab_env
To run AB-SGM, download and extract the model parameters,Saved weights
The code is tested on Python 3.8.17 and it needs pytorch gpu version for both training and sampling. We are using a GPU for model training and 6D coordinate sampling, and CPU batch processing for Rosetta. More specifically, we used a single NVIDIA V100 for training and inference, and at least 2 core CPU/8GB RAM per Rosetta job. Please refer to the shell script inside github for detailed configurations.
All related codes are in CDR_inpainting_conditional_generations directory. Generation of structures is achieved by first sampling 6D coordinates from the model, and running Rosetta.
Pleese use python sampling_6d.py to generate 6d coodinates and sequences. For instance,
python sampling_6d.py ./configs/inpainting_ch6.yml ../saved_weights/h3_inpaint.pth --pkl proteindataset_benchmark_half_12testset_h3.pkl --chain A --index 0 --tag singlecdr_inpaint_h3
The descriptions of each parameter are as below:
-
./configs/inpainting_ch6.yml is the config files. containing hyperparamters like batch size, data dimensions etc.
-
../saved_weights/h3_inpaint.pth This is to load the saved weight.
-
--pkl proteindataset_benchmark_half_12testset_h3.pkl is the pickle file contationing the all encoded data with H3 regions indicated to be masked out.
-
--index 0 refers to the index number inside the proteindataset_benchmark_half_12testset_h3.pkl to generated. this index number would match the generated file number. for example, the generated file is named as samples_index.pkl.
-
--tag singlecdr_inpaint_h3 is the generaetd folder name. the code would create this folder under current directory and stored all the generated data inside.
The encdoed examples are uploaded. you could find from
- H1 inpaint encoded examples. this link
- H2 inpaint encoded examples.this link
- H3 inpaint encoded examples. this link
To encode your given antigen-antibody complex structures, please refer to the readme inside the CDR_inpainting_conditional_generations/encoding/ .
6D coordinate sampling should ~1 minute per sample on a normal GPU, and Rosetta minimization should take a maximum of 3 hour per iteration depending on the size of the selected H1,H2, or h3 region for design.
To generate conditional samples, you have to first enocde epitope-antibody complex using the encoding file, then provide the encoded matrix incuding H1,H2, or H3 the regions to be masked during inference. For convenience, I made a pandas data frame stored all the related antigen information called . Please noted, this information should be stored indside matrix as the "ss_helices: ". For instance, to mask and re-generated H3 regions, you can run the following command:
First, please cd into the "CDR_inpainting_conditional_generations directory".
Conditional generation additionally requires the epitope-antibody complex with the designed CDR been masked out.
Pleese use python 'convert_6d_seq_to_pdb.py' to convert the 6d coodinates and sequnces pairs into the pdb uing Rosetta. For instance,
python convert_6d_seq_to_pdb.py ../proteinsgm_singlecdr_inpaint_retrain/singlecdr_inpaint_h3/samples_${SLURM_ARRAY_TASK_ID}.pkl half_h3_single_heavy_benckmark_pdb/${files[${SLURM_ARRAY_TASK_ID}]} ${SLURM_ARRAY_TASK_ID} 3
- ../proteinsgm_singlecdr_inpaint_retrain/singlecdr_inpaint_h3/samples_${SLURM_ARRAY_TASK_ID}.pkl . the generated 6d files locations.
- half_h3_single_heavy_benckmark_pdb/${files[${SLURM_ARRAY_TASK_ID}]} Corrsponding pdb file for the antibody.
- SLURM_ARRAY_TASK_ID the index of the generated struture inside the pkl file.
-
- cdr h3 regions.
The Rosetta protocol saves all iterations and intermediate structures to subdirectories in outPath, including structures before FastDesign and before relaxation. The default number of iterations is 3, and the final minimized structure can be found under outPath/.../best_run/final_structure.pdb.
Raw antigen-antibody complex dataset for CDR condtional generations can be downloaded here
The 'sabdab_downloader.py' is also provided for downloading.
The encoded dataet for antigen-antibody complex could be found here.
python single_cdr_train_cdr_ch6_with_epitope2.py configs/inpainting_ch6.yml --pkl H_chain_only_mask_only_H${SLURM_ARRAY_TASK_ID}_protein_dataset_dataset3_l_with_epitope_match_cdr_4k_all_matching_info_add_after_masking_and_epitope_info_no_padding_midptr_64_Mar10_2023.pkl
- configs/inpainting_ch6.yml config file.
- H_chain_only_mask_only_H${SLURM_ARRAY_TASK_ID}_protein_dataset_dataset3_l_with_epitope_match_cdr_4k_all_matching_info_add_after_masking_and_epitope_info_no_padding_midptr_64_Mar10_2023.pkl . training dataset.
#Please see required gpu, cpu, memory a below:
#SBATCH --gres=gpu:v100:1 # Number of GPU(s) per node
#SBATCH --cpus-per-task=12 # CPU cores/threads
#SBATCH --mem=60000M # memory per node
#SBATCH --time=0-23:05
Antibody-SGM: Antigen-Specific Joint Design of Antibody Sequence and Structure using Diffusion Models
Install conda environment conda env create -f ab_env.yaml
Activate conda environment conda activate ab_env
Saved weights could be access here: https://drive.google.com/drive/folders/1w1yPn3rYz04p9eejsr15bJN6K7kVzSAg?usp=sharing
AB dataset could be access here: https://opig.stats.ox.ac.uk/webapps/newsabdab/sabdab/archive/all/