High-throughput prediction of MHC Class I and Class II neoantigens with MHCnuggets

Abstract

Running title: High-throughput prediction of neoantigens with MHCnuggets Potential Conflicts of Interest: V.A receives research funding from Bristol-Myers Abstract Computational prediction of binding between neoantigen peptides and major histocompatibility complex (MHC) proteins is an emerging biomarker for predicting patient response to cancer immunotherapy. Current neoantigen predictors focus on in silico estimation of MHC binding affinity and are limited by low positive predictive value for actual peptide presentation, inadequate support for rare MHC alleles and poor scalability to high-throughput data sets. To address these limitations, we developed MHCnuggets, a deep neural network method to predict peptide-MHC binding. MHCnuggets is the only method to handle binding prediction for common or rare alleles of MHC Class I or II, with a single neural network architecture. Using a long shortterm memory network (LSTM), MHCnuggets accepts peptides of variable length and is capable of faster performance than other methods. When compared to methods that integrate binding affinity and HLAp data from mass spectrometry, MHCnuggets yields a fourfold increase in positive predictive value on independent MHC-bound peptide (HLAp) data. We applied MHCnuggets to 26 cancer types in TCGA, processing 52.6 million allele-peptide comparisons in under 2.3 hours, yielding 103,587 candidate immunogenic missense mutations (IMMs). IMM hotspots occurred in 36 genes, including 22 driver genes. Predicted IMM load was significantly associated with increased immune cell infiltration (p<2e-16) including CD8+ T cells. Notably, only 0.15% of predicted immunogenic missense mutations were observed in >2 patients, with 65% of these derived from driver mutations. Our results provide a new method for neoantigen prediction with high performance characteristics and demonstrate its utility in large data sets across human cancers. SynopsisWe developed a new in silico predictor of Major Histocompatibility Complex (MHC) ligand binding and demonstrated its utility to assess potential neoantigens and immunogenic missense mutations (IMMs) in 6613 TCGA patients.

Burden of tumor mutations, neoepitopes, and other variants are weak predictors of cancer immunotherapy response and overall survival

“…HLA type prediction and related analyses MHC Class I alleles for each patient were predicted from tumor WES reads using Optitype (v1.0) [40], and MHC Class II alleles for each patient were predicted from tumor WES reads using seq2hla (v2.2) [41]. For each neoepitope sequence predicted from phased variants (see above), a patient's predicted MHC Class I and MHC Class II alleles were used for binding affinity predictions with MHCnuggets (v2.1) [42]. Neoepitopes were counted toward a patient's neoepitope burden if they bound at least one of a patient's MHC alleles with high affinity (≤ 500 nM).…”

Discordant results among MHC binding affinity prediction tools

“…MHC class I binding affinity predictions were performed for the peptides generated from the kmerization process above using 4 tools: netMHCpan v4.1 (23), HLAthena v1.0 (27), MHCflurry v2.0 (25), and MHCnuggets v2.3 (51). netMHCpan was run with default options with the ‘-l’ option to specify peptides of lengths 8-12.…”

“…We additionally generated a set of 1 million random peptides of length 8-12 drawn uniformly at random. Peptide sets had negligible overlap (<1% shared between human vs viral vs random peptides).Peptide-MHC class I binding affinity predictionsMHC class I binding affinity predictions were performed for the peptides generated from the kmerization process above using 4 tools: netMHCpan v4.1(23), HLAthena v1.0(27), MHCflurry v2.0(25), and MHCnuggets v2.3(51). netMHCpan was run with default options with the '-l' option to specify peptides of lengths 8-12.…”

pVACtools: A Computational Toolkit to Identify and Visualize Cancer Neoantigens

“…The current MHC Class I algorithms supported by pVACseq are NetMHCpan (21), NetMHC (7,21), NetMHCcons (22), PickPocket (23), SMM (24), SMMPMBEC (25), MHCflurry (26), and MHCnuggets (27). The four MHC Class II algorithms that are supported are NetMHCIIpan (28), SMMalign (29), NNalign (30), and MHCnuggets (31). For the demonstration analysis, we limited our prediction to only MHC Class I alleles due to availability of HLA typing information from TCIA, though binding predictions for Class II alleles can also be generated using pVACtools.…”