PHLA-SiNet: A novel peptide-HLA binding prediction model using heterogeneous Siamese neural networks.

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Predicting peptide-HLA binding is critical for immunotherapy development.
• Current models struggle with peptide length variability, HLA similarity, and limited negative data.

Purpose of the Study:

• To develop an efficient and accurate pipeline for peptide-HLA binding prediction.
• To address limitations of existing prediction models in immunotherapy.

Main Methods:

• PHLA-SiNet pipeline combines ESM-Pep (peptide representation from language model) and IC-HLA (allele-specific HLA representation).
• Utilizes a Siamese neural network (SiNet) to align peptide and HLA embeddings.
• Validated across benchmarks, including comparisons with leading predictors and diverse HLA types.

Main Results:

• PHLA-SiNet demonstrates strong predictive performance, prioritizing sensitivity for immunotherapy.
• Computational validation and clinical data analysis show significant predictive potential.
• Identified HLA-B08:01 as a potential restriction element for future investigation.

Conclusions:

• PHLA-SiNet offers a generalizable, interpretable, and resource-efficient solution for peptide-HLA binding prediction.
• The pipeline shows promise for accelerating personalized immunotherapy development.
• Highlights the potential of AI-driven approaches in advancing cancer treatment.