Identifying T cell antigen at the atom level with graph convolutional network

Precise identification of T cell antigen is crucial for the development of cancer mRNA vaccine. However, existing computational methods identify the interaction between antigen and Human Leukocyte Antigens (HLA) or T cell receptor (TCR) only at the sequence or residue level, which fails to capture atom-level binding patterns. In this study, we innovatively transformed each residue sequence into a topological graph, in which each node corresponds to an atom and each edge corresponds to chemical bond, and then proposed a graph convolutional network framework, called deepAntigen, to identify the interactions between antigens and TCR/HLA at the atom level. Compared to the current state-of-the-art methods, deepAntigen achieves the best performance in identifying antigens presented by HLA and recognized by TCR. Importantly, deepAntigen can discovery antigen-specific TCR motifs and capture the mutation effect on T cell immune response, facilitating to decipher underlying binding mechanism. Overall, we provide a novel method named deepAntigen for accurately identifying T cell antigens, which will contribute to the development of personalized neoantigen-targeted immunotherapies for cancer patients.

How to install deepAntigen

To install deepAntigen, make sure you have installed PyTorch and PyTorch Geometric. If you need more details on the dependences, look at the environment.yml file.

• set up conda environment for deepAntigen

conda create -n deepAntigen-env python=3.8

• install deepAntigen from shell

pip install deepAntigen
pip install torch==1.9.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html
pip install torch-cluster==1.5.9 torch-scatter==2.0.7 torch-sparse==0.6.12 torch-spline-conv==1.2.1 -f https://data.pyg.org/whl/torch-1.9.0%2Bcu111.html
pip install torch-geometric==2.4.0

How to import deepAntigen

Using deepAntigen to achieve different tasks, please import corresponding module to your jupyter notebooks or scripts.

If you want to predict antigen-HLAI binding at the sequence level,

from deepantigen.antigenHLAI import run_antigenHLAI_seq

If you want to predict atom-level contact between antigen and HLAI,

from deepantigen.antigenHLAI import run_antigenHLAI_atom

If you want to predict antigen-HLAII binding at the sequence level,

from deepantigen.antigenHLAII import run_antigenHLAII_seq

If you want to predict atom-level contact between antigen and HLAII,

from deepantigen.antigenHLAII import run_antigenHLAII_atom

If you want to predict antigen-TCR binding at the sequence level,

from deepantigen.antigenTCR import run_antigenTCR_seq

If you want to predict atom-level contact between antigen and TCR,

from deepantigen.antigenTCR import run_antigenTCR_atom

How to run deepAntigen for antigen-HLAI binding prediction

For sequence-level prediction, please prepare your antigen-HLAI data and place them in a .csv file format similar to the test_antigenHLAI/sequence/test.csv provided. The column 'label' is optional.

df = run_antigenHLAI_seq.Inference(path)

The returned DataFrame, df, is prediction results of deepAntigen, which includes the binding probability for each antigen-HLAI pair. For atom-level prediction, please prepare your antigen-HLAI data and place them in a .csv file format similar to the test_antigenHLAI/crystal_structure/sample.csv provided.

peptide_atoms, HLAI_atoms, contact_maps = run_antigenHLAI_atom.Inference(path)

The returned three lists correspond top-k atoms of the peptide, top-k atoms of the HLAI and atom-level contact probability. Each element in peptide_atoms orHLAI_atoms is a list with length of k. Each element in contact_maps is a k*k DataFrame.

If you want to train deepAntigen with your own antigen-HLAI binding data, please reference the detailed Documentaion about deepAntigen.

How to run deepAntigen for antigen-HLAII binding prediction

For sequence-level prediction, please prepare your antigen-HLAII data and place them in a .csv file format similar to the test_antigenHLAII/sequence/test.csv provided. The column 'label' is optional.

df = run_antigenHLAII_seq.Inference(path)

The returned DataFrame, df, is prediction results of deepAntigen, which includes the binding probability for each antigen-HLAII pair. For atom-level prediction, please prepare your antigen-HLAII data and place them in a .csv file format similar to the test_antigenHLAII/crystal_structure/sample.csv provided.

peptide_atoms, HLAII_atoms, contact_maps = run_antigenHLAII_atom.Inference(path)

The returned three lists correspond top-k atoms of the peptide, top-k atoms of the HLAII and atom-level contact probability. Each element in peptide_atoms or HLAII_atoms is a list with length of k. Each element in contact_maps is a k*k DataFrame.

If you want to train deepAntigen with your own antigen-HLAII binding data, please reference the detailed Documentaion about deepAntigen.

How to run deepAntigen for antigen-TCR binding prediction

For sequence-level prediction, please prepare your antigen-TCR data and place them in a .csv file format similar to the test_antigenTCR/sequence/test.csv provided. The column 'label' is optional.

df = run_antigenTCR_seq.Inference(path)

The returned DataFrame, df, is prediction results of deepAntigen, which includes the binding probability for each antigen-TCR pair. For atom-level prediction, please prepare your antigen-TCR data and place them in a .csv file format similar to the test_antigenTCR/crystal_structure/sample.csv provided.

peptide_atoms, TCR_atoms, contact_maps = run_antigenTCR_atom.Inference(path)

The returned three lists correspond top-k atoms of the peptide, top-k atoms of the TCR and atom-level contact probability. Each element in peptide_atoms or TCR_atoms is a list with length of k. Each element in contact_maps is a k*k DataFrame.

If you want to train deepAntigen with your own antigen-TCR binding data, please reference the detailed Documentaion about deepAntigen.

Documentation

See detailed documentation and examples at https://deepAntigen.readthedocs.io/en/latest/index.html.

Contact

Feel free to submit an issue or contact us at quejinhao2021@163.com for problems about the package.