Gene and protein sequence features augment HLA class I ligand predictions

Area of Science:

• Immunology
• Proteomics
• Bioinformatics

Background:

• T cell-based immunotherapies rely on malignant tissues presenting targetable human leukocyte antigen (HLA) class I ligands.
• While peptide factors like HLA affinity and proteasomal processing are known, the impact of gene and protein sequence features on epitope presentation is less understood.

Purpose of the Study:

• To systematically evaluate the contribution of 7,135 gene and protein sequence features to HLA ligand sampling.
• To determine if these sequence features can improve predictions of HLA ligand presentation.

Main Methods:

• Performed HLA ligandome mass spectrometry.
• Analyzed 7,135 gene and protein sequence features.
• Integrated sequence features into a machine learning model.

Main Results:

• Identified predicted modifiers of mRNA and protein abundance/turnover, including mRNA methylation and protein ubiquitination sites, as informative for HLA ligand presence.
• Demonstrated that integrating these sequence features into machine learning models significantly augmented HLA ligand predictions.
• Showed that sequence feature integration achieved prediction accuracy comparable to experimental gene expression data.

Conclusions:

• Gene and protein sequence features are valuable, previously underappreciated determinants of HLA ligand presentation.
• Incorporating "hard-coded" sequence information enhances the predictive power for HLA ligand presentation, crucial for immunotherapy development.