A robust deep learning workflow to predict CD8 + T-cell epitopes.

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• T-cells are crucial for adaptive immunity against cancer and pathogens.
• Identifying T-cell antigens is challenging and low-throughput.
• Existing computational methods for CD8+ T-cell epitope prediction have limitations, including HLA bias and poor performance on small datasets.

Purpose of the Study:

• To develop a robust deep learning workflow (TRAP) for predicting CD8+ T-cell epitopes.
• To improve the accuracy and efficiency of T-cell epitope identification in both pathogenic and cancer settings.
• To introduce a novel metric (RSAT) for estimating immunogenicity of pathogenic peptides.

Main Methods:

• Developed TRAP, a deep learning workflow utilizing transfer learning and MHC binding information.
• TRAP predicts CD8+ T-cell epitopes from MHC-I presented pathogenic and self-peptides.
• Introduced RSAT metric to estimate immunogenicity for low-confidence predictions.

Main Results:

• TRAP outperformed existing algorithms in predicting epitopes from glioblastoma and SARS-CoV-2.
• TRAP effectively extracts immunogenicity features from limited and imbalanced datasets.
• The RSAT metric accurately estimated immunogenicity of pathogenic peptides across various lengths and species.

Conclusions:

• TRAP offers a novel computational approach for accurate CD8+ T-cell epitope prediction.
• This workflow enhances understanding of antigen-specific T-cell responses.
• TRAP facilitates the development of effective T-cell-based immunotherapies.