Predicting peptide presentation by major histocompatibility complex class I: an improved machine learning approach to...

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Accurate identification of peptides presented by major histocompatibility complex class I (MHC-I) is crucial for advancing immunopeptidomics.
• Existing tools for MHC-I peptide prediction often rely on less biologically relevant chemical affinity data or underutilize machine learning.
• There is a need for improved, biologically relevant methods to predict MHC-I peptide presentation.

Purpose of the Study:

• To develop and validate a novel machine learning approach for predicting MHC-I presented peptides.
• To leverage a comprehensive database of mass spectrometry-identified peptides for training predictive models.
• To assess the performance of the developed method against existing state-of-the-art tools.

Main Methods:

• Assembled a large, publicly available database of human peptides identified via mass spectrometry from the MHC-I immunopeptidome.
• Trained random forest classifiers (ForestMHC) using this curated dataset to predict MHC-I peptide presentation.
• Compared the performance of ForestMHC against established prediction algorithms (NetMHC, NetMHCpan, MixMHCpred) and neural network models.

Main Results:

• ForestMHC demonstrated superior prediction accuracy compared to NetMHC and NetMHCpan on test sets and novel ovarian carcinoma cell line data.
• Random forest scores showed a monotonic correlation with chemical binding affinities, highlighting biological relevance.
• Analysis revealed the critical importance of specific peptide anchor positions for MHC-I presentation, and gene expression was confirmed to partially influence presentation.

Conclusions:

• ForestMHC represents a significant advancement in predicting peptides presented by MHC-I, outperforming current methods.
• The study highlights the efficacy of random forest approaches and the value of mass spectrometry-based data for MHC-I prediction.
• ForestMHC has broad potential applications in basic immunology, vaccine design, and cancer immunotherapy, and is publicly available.