Abstract:Abstract This research was aimed at providing a machine-learning approach to predict binding affinity of MHC molecules and mutation-derived neoepitopes. Immunogenic short peptides derived from somatic mutations, generally refered to neoepitopes, play a vital role in immunotherapy against tumor cell. The binding of neoepitopes to MHC molecule is essential to form the complexes to elicit T cell-mediated immune response. Various tools have been developed to predict this binding in silico, for example, NetMHC, NetMHCpan, MHCflurry and Pickpocket. Almost all of these tools are based on machine-learning strategy and using single model to process sequence features. However, these predictors were proved to be inconsistent with experiments result under certain circumstances due to the lack of training data and model adaptability. We have recently developed a hybrid-model predictor, iNeo-PRED, that integrates several algorithms, including random forest (RF), convolutional neural network (CNN) and gradient boosting decision tree (GBDT). iNeo-PRED is trained on a large dataset that consists epitopes from trustworthy database such as IEDB, MHCDB, TANTIGEN, in-house peptidome datasets generated by mass spectrometry (MS) profiling of HLA ligands, and publicly available affinity-confirmed neoepitopes collected from scientific articles in the past 10 years. Totally, 568,461 9mer epitopes covering 163 HLA alleles were acquired for model training (110,895 for IC50 and 357,566 for MS). Several methods includes blosum50 matrix coding, one-hot coding and natural language coding were applied for peptide sequence transformation. The benchmark data provided by IEDB were used to test the performance of iNeo-PRED. The average AUC across all HLA was 0.71 and average F1 achieved was 0.73. During our clinical trial of neoantigen peptide vaccine for treating late stage patients with solid tumors, iNeo-PRED was implemented to predict epitopes, which were then incorperated into long peptides that were then synthesized with a length range from 15 to 30 amino acids. The most patients who received vaccination were benefited from the treatment. The disease control rate (percentage of patients whose best response was not PD) was 75% and the progression free survival was 4.6 months (95% confidence interval, 2.5 to 5.2). Ex vivo responses via ELISpot assays using post-vaccination peripheral blood cells demonstrated that over 80% of long peptides elicited immune response. These results indicate that iNeo-PRED is able to identify potent MHC class I neoepitopes and could be beneficial for cancer immunotherapy. Citation Format: Fan Mo, Rongchang Chen, Jian Wu, Qiang Yang, Yong Fang, Wei Dong, Yingqiang Sun, Kui Wang, Shuqing Chen. iNeo-PRED: A hybrid-model predictor for MHC class I neoantigens [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 4422.

PepCleav: a classification method for evaluating the cleavability of peptide fragments presented by MHC class I

Pep2Vec: An Interpretable Model for Peptide-MHC Presentation Prediction and Contaminant Identification in Ligandome Datasets

What cleaves? Is proteasomal cleavage prediction reaching a ceiling?

Improved proteasomal cleavage prediction with positive-unlabeled learning

Ranking-based Convolutional Neural Network Models for Peptide-MHC Binding Prediction

PepFormer: End-to-End Transformer-Based Siamese Network to Predict and Enhance Peptide Detectability Based on Sequence Only

E-CLEAP: An ensemble learning model for efficient and accurate identification of antimicrobial peptides

Abstract 4422: Ineo-Pred: A Hybrid-Model Predictor for MHC Class I Neoantigens

dbPepNeo: a manually curated database for human tumor neoantigen peptides

APE-Gen2.0: Expanding Rapid Class I Peptide–Major Histocompatibility Complex Modeling to Post-Translational Modifications and Noncanonical Peptide Geometries

ACP-MLC: A two-level prediction engine for identification of anticancer peptides and multi-label classification of their functional types

ACME: pan-specific peptide-MHC class I binding prediction through attention-based deep neural networks.

PepQuery Enables Fast, Accurate, and Convenient Proteomic Validation of Novel Genomic Alterations

PepCNN deep learning tool for predicting peptide binding residues in proteins using sequence, structural, and language model features

Identification and Analysis of the Cleavage Site in a Signal Peptide Using SMOTE, Dagging, and Feature Selection Methods.

NIEluter: Predicting peptides eluted from HLA class I molecules

ConvNeXt-MHC: improving MHC-peptide affinity prediction by structure-derived degenerate coding and the ConvNeXt model

PepAnalyzer: predicting peptide properties using its sequence

DeepNeuropePred: A robust and universal tool to predict cleavage sites from neuropeptide precursors by protein language model

DapPep: Domain Adaptive Peptide-agnostic Learning for Universal T-cell Receptor-antigen Binding Affinity Prediction

ACP-CapsPred: an explainable computational framework for identification and functional prediction of anticancer peptides based on capsule network