POPISK: T-cell reactivity prediction using support vector machines and string kernels.

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Predicting peptide immunogenicity is vital for vaccine design and understanding immune responses.
• T-cell reactivity prediction is challenging, with prior studies on T-cell receptor (TCR) recognition positions yielding inconsistent results.
• Large-scale analyses are needed to characterize sequence variations' effects on T-cell reactivity and develop accurate immunogenicity predictors.

Purpose of the Study:

• To develop a computational method for predicting peptide T-cell reactivity and identifying key recognition positions.
• To establish a comprehensive dataset for training and validating immunogenicity prediction models.
• To gain insights into the molecular mechanisms underlying TCR recognition and peptide immunogenicity.

Main Methods:

• Compiled a large immunogenicity dataset from three major immunology databases.
• Classified peptides by their associated MHC alleles to account for MHC restriction.
• Developed and applied the POPISK computational method, utilizing a support vector machine with a weighted degree string kernel, for T-cell reactivity prediction.

Main Results:

• Achieved a mean 10-fold cross-validation accuracy of 68% in predicting T-cell reactivity for HLA-A2-binding peptides.
• POPISK successfully predicted immunogenicity changes due to single-residue mutations, aligning with experimental findings.
• Identified important peptide recognition positions (4, 6, 8, and 9) and explored their relationship with MHC-peptide-TCR interactions.

Conclusions:

• The POPISK method accurately predicts immunogenicity and its changes upon single-residue modifications.
• Identified key peptide positions influencing T-cell reactivity, providing mechanistic insights.
• A web server for POPISK is available for public use, facilitating immunogenicity prediction.