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

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• CD8+ T-cell epitopes are peptide fragments crucial for initiating antigen-specific cytotoxic responses.
• Experimental identification of these epitopes is costly and time-consuming.
• Computational prediction offers a more efficient alternative for epitope discovery.

Purpose of the Study:

• To review recent advances in computational prediction of CD8+ T-cell epitopes.
• To focus on protein language models and deep learning approaches.
• To cover predictions of MHC-I binding, TCR-peptide reactivity, and pMHC-TCR binding affinities.

Main Methods:

• Review of recent literature on deep learning and protein language models for epitope prediction.
• Analysis of methods for encoding protein sequences.
• Examination of deep learning models for predicting MHC-I binding affinities, TCR-peptide reactivities, and pMHC-TCR binding affinities.

Main Results:

• Deep learning techniques have significantly improved CD8+ T-cell epitope prediction over the past two decades.
• Protein language models offer advanced epitope encoding schemes.
• Current models show progress in predicting MHC-I binding, TCR-peptide reactivities, and pMHC-TCR binding affinities.

Conclusions:

• Computational prediction, especially using deep learning and protein language models, is vital for efficient CD8+ T-cell epitope discovery.
• These advancements are critical for accelerating vaccine design and immunotherapy development.
• Further research in deep learning-based prediction holds significant promise for the field.