HeteroTCR: A heterogeneous graph neural network-based method for predicting peptide-TCR interaction

Zilan Yu,Mengnan Jiang,Xun Lan

DOI: https://doi.org/10.1038/s42003-024-06380-6

IF: 6.548

2024-06-05

Communications Biology

Abstract:Identifying interactions between T-cell receptors (TCRs) and immunogenic peptides holds profound implications across diverse research domains and clinical scenarios. Unsupervised clustering models (UCMs) cannot predict peptide-TCR binding directly, while supervised predictive models (SPMs) often face challenges in identifying antigens previously unencountered by the immune system or possessing limited TCR binding repertoires. Therefore, we propose HeteroTCR , an SPM based on Heterogeneous Graph Neural Network (GNN), to accurately predict peptide-TCR binding probabilities. HeteroTCR captures within-type (TCR-TCR or peptide-peptide) similarity information and between-type (peptide-TCR) interaction insights for predictions on unseen peptides and TCRs, surpassing limitations of existing SPMs. Our evaluation shows HeteroTCR outperforms state-of-the-art models on independent datasets. Ablation studies and visual interpretation underscore the Heterogeneous GNN module's critical role in enhancing HeteroTCR's performance by capturing pivotal binding process features. We further demonstrate the robustness and reliability of HeteroTCR through validation using single-cell datasets, aligning with the expectation that pMHC-TCR complexes with higher predicted binding probabilities correspond to increased binding fractions.

biology

What problem does this paper attempt to address?

The paper aims to address the problem of predicting interactions between T-cell receptors (TCRs) and immunopeptides, a research topic of significant importance in fields such as cancer immunotherapy, vaccine development, and personalized medicine. Existing prediction methods are mainly divided into two categories: Unsupervised Clustering Models (UCMs) and Supervised Prediction Models (SPMs). While UCMs can cluster TCR sequences to some extent, they cannot directly predict peptide-TCR binding; SPMs, on the other hand, can make predictions but perform poorly when dealing with unseen antigens or TCRs with limited binding spectra. To solve the above issues, the authors propose the HeteroTCR method, an SPM based on Heterogeneous Graph Neural Network (HGNN). HeteroTCR can utilize the same-type similarity (TCR-TCR or peptide-peptide) and cross-type interaction information (peptide-TCR) from TCR and peptide sequence information, thereby improving the prediction accuracy of the binding probability between unseen peptides and TCRs. Specifically, the model first extracts numerical embeddings of TCRs and peptides through a pre-trained Convolutional Neural Network (CNN) module, then uses the HGNN module to capture information between different types of nodes, and finally determines whether there is an interaction between TCRs and peptides through a Multi-Layer Perceptron (MLP). Evaluations on multiple independent datasets show that HeteroTCR outperforms existing advanced models in prediction performance. Additionally, ablation studies confirm the critical role of the HGNN module in enhancing overall performance. The model performs well under different data partitioning strategies, especially when dealing with unseen antigens and TCRs, demonstrating good generalization ability. This indicates that HeteroTCR has high practical value and potential in peptide-TCR interaction prediction.

HeteroTCR: A heterogeneous graph neural network-based method for predicting peptide-TCR interaction

GGNpTCR: A Generative Graph Structure Neural Network for Predicting Immunogenic Peptides for T-cell Immune Response

Hybrid gMLP model for interaction prediction of MHC-peptide and TCR

Enhancing TCR specificity predictions by combined pan- and peptide-specific training, loss-scaling, and sequence similarity integration

Ranking-based Convolutional Neural Network Models for Peptide-MHC Binding Prediction

Heterogeneous Graph Embedding Model for Predicting Interactions Between TF and Target Gene.

GTE: a graph learning framework for prediction of T-cell receptors and epitopes binding specificity

MIX-TPI: a flexible prediction framework for TCR–pMHC interactions based on multimodal representations

Neo-epitope identification by weakly-supervised peptide-TCR binding prediction

NetTCR: sequence-based prediction of TCR binding to peptide-MHC complexes using convolutional neural networks

Pretraining Transformers for TCR-pMHC Binding Prediction.

pan-MHC and cross-Species Prediction of T Cell Receptor-Antigen Binding

A unified cross-attention model for predicting antigen binding specificity to both HLA and TCR molecules

DTI-HETA: prediction of drug-target interactions based on GCN and GAT on heterogeneous graph

Heterogeneous Graph Attention Network for Drug-Target Interaction Prediction

NetTCR-2.1: Lessons and guidance on how to develop models for TCR specificity predictions

Assessing the Generalization Capabilities of TCR Binding Predictors via Peptide Distance Analysis

Pan-Peptide Meta Learning for T-cell Receptor–antigen Binding Recognition

Metapath-aggregated heterogeneous graph neural network for drug-target interaction prediction

Context-Aware Amino Acid Embedding Advances Analysis of TCR-Epitope Interactions

MATE-Pred: Multimodal Attention-based TCR-Epitope interaction Predictor