Genome Medicine, Journal Year: 2023, Volume and Issue: 15(1)

Published: Sept. 13, 2023

Abstract Background T-cells play a crucial role in the adaptive immune system by triggering responses against cancer cells and pathogens, while maintaining tolerance self-antigens, which has sparked interest development of various T-cell-focused immunotherapies. However, identification antigens recognised is low-throughput laborious. To overcome some these limitations, computational methods for predicting CD8 + T-cell epitopes have emerged. Despite recent developments, most immunogenicity algorithms struggle to learn features peptide from small datasets, suffer HLA bias are unable reliably predict pathology-specific epitopes. Methods We developed TRAP (T-cell recognition potential HLA-I presented peptides), robust deep learning workflow MHC-I pathogenic self-peptides. uses transfer learning, architecture MHC binding information make context-specific predictions also detects low-confidence peptides that differ significantly those training datasets abstain making incorrect predictions. estimate with predictions, we further novel metric, RSAT (relative similarity autoantigens tumour-associated antigens), as complementary ‘dissimilarity self’ studies. Results was used identify glioblastoma patients well SARS-CoV-2 peptides, it outperformed other both settings. especially effective at extracting immunogenicity-associated properties restricted data emerging pathogens translating them onto related species, minimising loss likely imbalanced datasets. demonstrated metric termed able immunogenic lengths species. implementation available at: https://github.com/ChloeHJ/TRAP . Conclusions This study presents accurately foster better understanding antigen-specific response clinical therapeutics.

Language: Английский