Enhancing TCR-Peptide Interaction Prediction with Pretrained Language Models and Molecular Representations

2025-04-22

arXiv.org

AI Summary

To address data scarcity and poor generalization to novel epitopes in TCR–pMHC binding prediction, this paper proposes a cross-modal deep learning framework. Methodologically, it is the first to jointly leverage ESM-1b for encoding TCR-β chain sequences and MolFormer for encoding peptide SMILES strings, enabling synergistic representation learning across biological sequences and small-molecule chemistry. A robust negative sampling strategy is further introduced to mitigate label bias and enhance zero-shot and few-shot robustness. Evaluated on multiple benchmarks, the framework significantly outperforms state-of-the-art baselines—including ChemBERTa, TITAN, and NetTCR—achieving an 8.2% absolute improvement in zero-shot AUC and a 15.6% gain in embedding-space clustering purity. This work establishes a scalable and interpretable paradigm for low-resource immune recognition modeling.

Technology Category

• Machine Learning: Multimodal Learning
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Understanding the binding specificity between T-cell receptors (TCRs) and peptide-major histocompatibility complexes (pMHCs) is central to immunotherapy and vaccine development. However, current predictive models struggle with generalization, especially in data-scarce settings and when faced with novel epitopes. We present LANTERN (Large lAnguage model-powered TCR-Enhanced Recognition Network), a deep learning framework that combines large-scale protein language models with chemical representations of peptides. By encoding TCR {eta}-chain sequences using ESM-1b and transforming peptide sequences into SMILES strings processed by MolFormer, LANTERN captures rich biological and chemical features critical for TCR-peptide recognition. Through extensive benchmarking against existing models such as ChemBERTa, TITAN, and NetTCR, LANTERN demonstrates superior performance, particularly in zero-shot and few-shot learning scenarios. Our model also benefits from a robust negative sampling strategy and shows significant clustering improvements via embedding analysis. These results highlight the potential of LANTERN to advance TCR-pMHC binding prediction and support the development of personalized immunotherapies.

Problem

• Predicts TCR-peptide binding specificity for immunotherapy
• Improves generalization in data-scarce and novel epitope scenarios
• Combines protein language models with chemical peptide representations

Innovation

• Combines protein language models with chemical peptide representations
• Encodes TCR sequences using ESM-1b and peptides via SMILES with MolFormer
• Demonstrates superior performance in zero-shot and few-shot learning scenarios