Progress of Deep Learning Prediction of CD8+ T-Cell Epitopes

CD8+ T-cell epitopes are special peptide fragments produced by the catabolism of antigenic proteins. After being presented on the surface of antigen-presenting cells in association with Major Histocompatibility Complex-I (MHC-I) molecules, CD8+ T-cell epitopes can be specifically recognized by T-cell receptor (TCR) expressed on T cells, thereby initiating an antigen-specific cytotoxic response. Identifying CD8+ T-cell epitopes by experimental methods is both costly and time-consuming. By comparison, computational prediction can reduce experimental costs and significantly improve epitope discovery efficiency. Therefore, the prediction of CD8+ T-cell epitopes has always been a central topic in vaccine design and immunotherapy. With the breakthrough achievements of deep learning techniques, significant progress has been made in deep learning prediction of CD8+ T-cell epitopes in the past two decades. Herein, we provide a comprehensive review of recent advances in protein language model-based epitope encoding schemes and deep learning models for predicting MHC-I binding affinities, TCR-peptide reactivities, and pMHC-TCR binding affinities.