S3AI

The official code repository of "Interpretable antibody-antigen interaction prediction by introducing route and priors guidance".

Table of Contents

• Overview
• Installation
• Data
• Model inference
• Model training
• Citation
• License

Overview

With the application of personalized and precision medicine, more precise and efficient antibody drug development technology is urgently needed. Identification of antibody-antigen interactions is key to antibody engineering. The time-consuming and expensive nature of wet-lab experiments calls for efficient computational methods. Previous deep-learning-based computing methods for antibody-antigen interaction prediction are distinctly divided into two categories: structure-based and sequence-based. Taking into account the non-overlapping advantage of these two major categories, we propose an interpretable antibody-antigen interaction prediction method, S3AI, that bridges structures to sequences through structural information distillation. Furthermore, non-covalent interactions are modeled explicitly to guide neural networks in understanding the underlying patterns in antigen-antibody docking. Supported by the two innovative designs mentioned above, S3AI significantly and comprehensively surpasses the state-of-the-art models. S3AI maintains excellent robustness when predicting unknown antibody-antigen pairs, surpassing specialized prediction methods designed for out-of-distribution generalization in fair comparisons. More importantly, S3AI captures the universal pattern of antibody-antigen interactions, which not only identifies the CDRs responsible for specific binding to the antigen but also unearthed the importance of CDR-H3 for the interaction. The implicit introduction of knowledge of structure modality and the explicit modeling of chemical constraints build a 'sequence-to-function' route, thereby facilitating S3AI's understanding of complex molecular interactions through providing route and priors guidance. S3AI, which does not require structure input, is suitable for large-scale, parallelized antibody optimization and screening while outperforming state-of-the-art prediction methods. It helps to quickly and accurately identify potential candidates in the vast antibody space, thereby accelerating the development process of antibody drugs.

Hardware requirements

The experiments are tested and conducted on one Tesla V100 (32GB).

Installation

We highly recommand that you use Anaconda for Installation

conda create -n S3AI
conda activate S3AI
pip install -r requirements.txt
pip install fair-esm pyaml==21.10.1

Data

The SARS-CoV-2 IC50 data is in the data folder.

• data/updated_processed_data.csv is the paired Ab-Ag data.
• data/Ag_sequence.csv is the Ag sequence data.

Model inference

Download the checkpoint of S3AI and modify the paths in the code.

Content	Link
Checkpoint on SARS-CoV-2	link
Checkpoint on HIV cls	link
Checkpoint on HIV reg	link

To test S3AI on SARS-CoV-2 IC50 test data, please run

python main.py --config=configs/test_on_sarscov2.yml

To test S3AI on HIV test data for classification, please run

python main.py --config=configs/test_on_HIV_cls.yml

To test S3AI on HIV test data for regression, please run

python main.py --config=configs/test_on_HIV_reg.yml

Model training

To train S3AI on downstream tasks from scratch, please run

python main.py --config=configs/train_on_sarscov2.yml
python main.py --config=configs/train_on_HIV_cls.yml
python main.py --config=configs/train_on_HIV_reg.yml

Attribution evaluation

Attribution of entire CDRs and FRs

To evaluate the attribution of the entire CDRs and FRs to the prediction output on test data, please run:

python attribution_entire.py --config=configs/test_attribution.yml

File Description

You will save a file in the format shapley_phi_st{args.sample_st}_samplenum{args.sample_num}.npz, where sample_st indicates the starting index of the stored samples in the training set, and sample_num indicates the number of samples for which attributions are computed.

This file stores the following elements:

• v_N: Predicted output of the network on each sample
• phi: Attribution value of each FR (Framework Region) and the entire CDRs (Complementarity-determining regions)
• player: Partition of the variables. For each variable, it saves a tuple $(l, r)$, where the tuple represents the position of the corresponding FR on the antibody sequence. Specifically, $(-1, -1)$ indicates the entire CDRs.
• CDR_H: Saves three tuples $(l, r)$ representing the positions of CDR-H1, CDR-H2, and CDR-H3 on the antibody sequence.
• CDR_L: Saves three tuples $(l, r)$ representing the positions of CDR-L1, CDR-L2, and CDR-L3 on the antibody sequence.
• ic50: True IC50 values of each sample.
• cls_label: True classification results of each sample.

Attribution of each CDR

To evaluate the attribution of each CDR on test data, please run:

python attribution_eachCDR.py --config=configs/test_attribution.yml

File Description

You will save a file in the format inner_phi.npz. This file stores the following elements:

• v_N: attribution value of the entire CDRs under different masks
• phi: Attribution value of each CDR (Complementarity-determining region)
• player: Partition of the variables. For each variable, it saves a tuple $(l, r)$, where the tuple represents the position of the corresponding CDR on the antibody sequence.
• CDR_H: Saves three tuples $(l, r)$ representing the positions of CDR-H1, CDR-H2, and CDR-H3 on the antibody sequence.
• CDR_L: Saves three tuples $(l, r)$ representing the positions of CDR-L1, CDR-L2, and CDR-L3 on the antibody sequence.

Citation

If you find this code or the models useful in your research, please cite:

@article{liu2024interpretable,
  title={Interpretable antibody-antigen interaction prediction by bridging structure to sequence},
  author={Liu, Yutian and Nie, Zhiwei and Chen, Jie and Zheng, Xinhao and Fu, Jie and Liu, Zhihong and Liu, Xudong and Xu, Fan and Huang, Xiansong and Zhang, Wen-Bin and others},
  journal={bioRxiv},
  pages={2024--03},
  year={2024},
  publisher={Cold Spring Harbor Laboratory}
}

License

This project is licensed under the MIT License.