TripHLApan: predicting HLA molecules binding peptides based on triple coding matrix and transfer learning.

Area of Science:

• Immunoinformatics
• Computational Biology
• Vaccine Development

Background:

• Accurate prediction of human leukocyte antigen (HLA) and peptide binding is essential for designing effective tumor vaccines.
• Current computational models face challenges in precisely predicting these interactions.

Purpose of the Study:

• To develop an advanced computational model, TripHLApan, for accurate prediction of HLA-peptide binding.
• To improve the accuracy and scalability of HLA-peptide binding predictions for tumor vaccine synthesis.

Main Methods:

• Integration of a triple coding matrix, BiGRU (Bidirectional Gated Recurrent Unit) with Attention mechanisms, and transfer learning.
• Development of novel preprocessing and coding strategies based on identified HLA-peptide interaction sites and binding motifs.
• Utilizing sequence-level binding information through attention modules and BiGRU layers.

Main Results:

• TripHLApan demonstrates strong predictive performance across various conditions and sample ratios, outperforming existing optimal models.
• The model shows superiority and scalability, validated on recent datasets and melanoma patient samples.
• Identified key interaction regions between HLA molecules and peptides and their correlation with encoding and binding motifs.

Conclusions:

• TripHLApan is a powerful and scalable tool for predicting HLA-I and HLA-II molecular peptide binding, significantly aiding tumor vaccine development.
• The model's approach enhances understanding of HLA-peptide interactions and improves prediction accuracy.
• TripHLApan is publicly available for broader research application.