TransHLA: a Hybrid Transformer model for HLA-presented epitope detection

Area of Science:

• Immunoinformatics
• Computational Biology
• Vaccinology

Background:

• Accurate prediction of epitope presentation on human leukocyte antigen (HLA) molecules is vital for vaccine development and immunotherapy.
• Current prediction tools often lack universality and comprehensive HLA site analysis, limiting peptide filtering.

Purpose of the Study:

• To develop a universal tool for epitope prediction across all HLA alleles.
• To improve the efficiency of filtering invalid peptide segments for vaccine design.

Main Methods:

• Introduction of TransHLA, integrating Transformer and Residue CNN architectures.
• Utilization of the ESM2 large language model for sequence and structure embeddings.
• Validation on IEDB, CEDAR, and VDJdb datasets.

Main Results:

• TransHLA achieved 84.72% accuracy and 91.95% AUC for HLA class I on IEDB data.
• TransHLA achieved 79.94% accuracy and 88.14% AUC for HLA class II on IEDB data.
• Demonstrated superior specificity and sensitivity in identifying immunogenic epitopes and neoepitopes compared to existing models.

Conclusions:

• TransHLA significantly advances vaccine design and immunotherapy by enabling efficient identification of broadly reactive peptides.
• The developed tool and associated resources are publicly available for research use.