Antibody-Diffusion-Properties

This repository contains the code to reproduce the results of the paper: "Guiding diffusion models for antibody sequence and structure co-design with developability properties".

Figure: (a) Illustration of an antibody, featuring the two heavy (in blue) and two light (in pink) chains. The variable regions in the enlarged area encompass the antigen-binding site including the six CDR loops. (b-d) Visualization of the generative diffusion process, showcasing the prior distributions for each modality and the designed CDR, for the (b) property-unconditioned mode, (c) property-aware prior approach, and (d) sampling by property approach.

Background

This repository builds upon (Luo et al. 2022) [DiffAb] and (Shan et al. 2022) [DDG Predictor]. To install the required packages, please refer to the documentation in the official repositories.

Datasets and model weights

Protein structures in the SAbDab dataset for training and testing can be downloaded here. Extract all_structures.zip into the data folder. The data folder contains a snapshot of the dataset index (sabdab_summary_all.tsv).

[DiffAb] model weights can be downloaded from either [Hugging Face] or [Google Drive]. Copy the files into the trained_models folder. The model weights for [DDG Predictor] can be found in diffab/tools/ddg/data/model.pt.

Property-guided design of antibodies

The config files are in the configs/test folder. To design the six CDRs separately, use the codesign_single model and config on the scripts design_pdb.py (one sample) and design_testset.py (full test set, 19 samples). The lists of options are in the scripts diffab/tools/runner/design_for_pdb.py and diffab/tools/runner/design_for_testset.py, respectively.

Property-unconditioned design

# Sample 7DK2_AB_C
python design_pdb.py ./data/examples/7DK2_AB_C.pdb \
  --config ./configs/test/codesign_single.yml

# Test set
for i in {0..18}; do
  python design_testset.py $i --config ./configs/test/codesign_single.yml
done

Guiding design: Property-aware prior

For hydropathy-aware prior with $b$, specify the option --prior_b as:

# Sample 7DK2_AB_C
python design_pdb.py ./data/examples/7DK2_AB_C.pdb \
  --config ./configs/test/codesign_single.yml --prior_b 0.8

# Test set
for i in {0..18}; do
  python design_testset.py $i \
    --config ./configs/test/codesign_single.yml --prior_b 0.8
done

Guiding design: Sampling by property

For sampling by property (ddG, hydropathy, or both), use the following config files:

Config file	Description
codesign_single_ddg.yml	Sequence-structure of one CDR, sampling by ddG.
codesign_single_hydro.yml	Sequence-structure of one CDR, sampling by hydropathy.
codesign_single_ddg_and_hydro.yml	Sequence-structure of one CDR, sampling by ddG and hydropathy.

Here, the extra options are: --sample_step_mode ("min", "max", or "softmax"), --sample_step_num (int), and --sample_step_period (int). For example, for sampling by ddG use:

# Sample 7DK2_AB_C
python design_pdb.py ./data/examples/7DK2_AB_C.pdb \
  --config ./configs/test/codesign_single_ddg.yml \
  --sample_step_mode min --sample_step_num 20 --sample_step_period 1

# Test set
for i in {0..18}; do
  python design_testset.py $i \
    --config ./configs/test/codesign_single_ddg.yml \
    --sample_step_mode min --sample_step_num 20 --sample_step_period 1
done

Evaluation

To compute the evaluation metrics: AAR, RMSD, Hydropathy Score, and Predicted ddG (the option --no_energy prevents the computation of Rosetta ddG) for all samples use:

python eval.py --no_energy --root results/codesign_single

Other options

For training details and more design options, such as "Fix-backbone Design", "Antibody Optimization", or "Antigen Only", please refer to the original implementation of [DiffAb].

References

If you find this repository useful in your research, please cite the following works.

@article{villegas2023guiding,
  title={Guiding diffusion models for antibody sequence and structure co-design with developability properties},
  author={Amelia Villegas-Morcillo and Jana M. Weber and Marcel J.T. Reinders},
  journal={NeurIPS 2023 Generative AI and Biology Workshop},
  year={2023},
  url={https://openreview.net/forum?id=bPcgbKDCUQ}
}

@inproceedings{luo2022antigenspecific,
  title={Antigen-Specific Antibody Design and Optimization with Diffusion-Based Generative Models for Protein Structures},
  author={Shitong Luo and Yufeng Su and Xingang Peng and Sheng Wang and Jian Peng and Jianzhu Ma},
  booktitle={Advances in Neural Information Processing Systems},
  editor={Alice H. Oh and Alekh Agarwal and Danielle Belgrave and Kyunghyun Cho},
  year={2022},
  url={https://openreview.net/forum?id=jSorGn2Tjg}
}

@article{shan2022deep,
  title={Deep learning guided optimization of human antibody against SARS-CoV-2 variants with broad neutralization},
  author={Shan, Sisi and Luo, Shitong and Yang, Ziqing and Hong, Junxian and Su, Yufeng and Ding, Fan and Fu, Lili and Li, Chenyu and Chen, Peng and Ma, Jianzhu and others},
  journal={Proceedings of the National Academy of Sciences},
  volume={119},
  number={11},
  pages={e2122954119},
  year={2022}
}