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.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.