Predicting HLA CD4 Immunogenicity in Human Populations.

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Predicting T cell immunogenicity is crucial for understanding T cell responses and applications.
• Human Leukocyte Antigen (HLA) binding affinity is a common predictor but limited by HLA variability and data availability.
• Alternative methods are needed to identify immunogenic peptides across diverse populations.

Purpose of the Study:

• To develop and validate a novel method for predicting CD4 T cell immunogenicity at the population level.
• To overcome limitations of HLA-based prediction methods, particularly the need for HLA typing data.

Main Methods:

• Neural networks were trained using dominant epitopes and negative peptides from published studies.
• An "immunogenicity score" was generated and compared to existing HLA class II binding prediction methods.
• The method was validated on independent datasets from 57 studies, encompassing over 2,000 peptides and 1,600 donors.

Main Results:

• The immunogenicity score achieved performance comparable to the 7-allele method (AUC 0.702 vs. 0.703).
• Combining the immunogenicity score with the 7-allele method significantly improved prediction accuracy (AUC 0.725).
• The tool is freely available on the Immune Epitope Database website.

Conclusions:

• The developed immunogenicity score effectively predicts CD4 T cell immunogenicity from protein sequences without requiring HLA typing.
• The method's efficacy was confirmed across various antigen sources, ethnicities, and epitope identification techniques.
• This tool offers a valuable resource for T cell epitope prediction in population-level studies.