Abstract
Identifying neoepitopes that elicit an adaptive immune response is a major bottleneck to developing personalized cancer vaccines. Experimental validation of candidate neoepitopes is extremely resource intensive, and the vast majority of candidates are non-immunogenic, making their identification a needle-in-a-haystack problem. To address this challenge, we present computational methods for predicting MHC-I epitopes and identifying immunogenic neoepitopes with high precision. The BigMHC method comprises an ensemble of seven pan-allelic deep neural networks trained on peptide-MHC eluted ligand data from mass spectrometry assays and transfer learned on data from assays of antigen-specific immune response. Compared with four state-of-the-art classifiers, BigMHC significantly improves the prediction of epitope presentation on a test set of 45,409 MHC ligands among 900,592 random negatives (AUROC=0.9733, AUPRC=0.8779). After transfer learning on immunogenicity data, BigMHC yields significantly higher precision than seven state-of-the-art models in identifying immunogenic neoepitopes, making BigMHC effective in clinical settings. All data and code are freely available at https://github.com/KarchinLab/bigmhc.
Competing Interest Statement
Under a license agreement between Genentech and the Johns Hopkins University, X.M.S., and R.K., and the University are entitled to royalty distributions related to MHCnuggets technology discussed in this publication. This arrangement has been reviewed and approved by the Johns Hopkins University in accordance with its conflict-of-interest policies. V.A. receives research funding to her institution from Astra Zeneca and has received research funding to her institution from Bristol Myers Squibb over the past five years. The remaining authors have declared no conflicts of interest.
Footnotes
Updated BigMHC architecture, training procedure, and immunogenicity evaluation protocol. Updated the conceptual framework to include more endogenous peptide processing. The revised experimental procedure and architecture are illustrated in the revised Fig. 1 and Fig. 2. Included more fine-grain analysis of presentation prediction performance and added statistical tests to establish significance in Fig. 3. Updated immunogenicity test data to not use negative EL background to avoid false negative bias from mass spectrometry data; the immunogenicity positives and negatives are validated with immunoassays. Included comparisons of immunogenicity prediction to additional models: MixMHCpred-2.2, PRIME-2.0, and HLAthena. Included both types of outputs of prior models: percentile ranks and raw scores. Expanded discussion and explored many limitations of the proposed method. Added identifiers for the three clinical trials in which BigMHC is being used.
