BiLETCR: An Efficient PMHC-TCR Combined Forecasting Method

This paper introduces a method incorporating a multi-attention mechanism with a Bidirectional Long Short-Term Memory Network (BiLSTM) structure, aimed at predicting the binding probability between peptides and major histocompatibility complex (pMHC) with T-cell receptors (TCR), a critical aspect of cancer immunotherapy. The method specifically targets the prediction of binding specificity between neoantigens and TCR within Class I MHC complexes. By analyzing TCR sequences, antigens and Class I MHC alleles, this study not only achieves accurate prediction of pairing but also significantly enhances computational efficiency. We employed one-hot encoding and acceleration techniques based on the TRUST algorithm to encode antigen, MHC and TCR sequences, thereby extracting essential features to construct an efficient predictive model. Experimental results demonstrate the method’s efficacy in predicting the binding probability between antigens, MHC molecules and TCR, showcasing its potential for application.