Abstract 3649: Artificial intelligence accelerates shared neoantigen peptide cancer vaccine development

Background: Many cancers remain intractable despite the recent breakthroughs in immune checkpoint inhibitors and gene and cell therapies, making the pursuit of novel, efficacious, and safe treatment options critical. Recently, there is a renewed interest in cancer vaccines as a promising cancer immunotherapy approach. Nevertheless, precisely predicting the immune response elicited by cancer vaccines continues to be a significantly hurdle. Methods: From published literature and available datasets, we collected over 17 million peptide entries ranging from 8 to 11 amino acids, and approximately 4 million peptide entries ranging from 13 to 21 amino acids to train models predicts the probability of a given peptide to be presented by MHC-I or MHC-II, respectively. Those peptides were restricted to 150 unique MHC-I alleles and 19 unique MHC-II alleles. Overall, both MHC-I and MHC-II presentation models were constructed with about 30 million parameters. In addition, we developed algorithms to specifically introduce single point mutations to further enhance binding to MHCs, and in the meanwhile, not influence cognate recognition. To validate the model prediction accuracy, automated peptide synthesis and surface plasmon resonance (SPR) technologies were applied. We then examined the presentation of designed peptides using MHC allele-engineered cell line, and further validated the peptide immunogenicity in mice by assessing T cell activation and tumor cell killing capabilities. Results: Our advanced AI models has outperformed the state-of-the-art prediction algorithms, including NetMHCpan and MixMHCpred, in predicting antigen presentation for both MHC-I and MHC-II. To validate our model accuracy, 28 novel peptides based on mutated TP53 protein were predicted and their binding to corresponding HLA alleles were measured using SPR. >80% of the peptides displayed stronger binding affinities than the positive controls, demonstrating a significantly improved algorithm accuracy for peptide neoepitope design. In the cellular presentation study, one out of six designed neoepitope peptides were validated by mass spectrometry to be presented by HLA-A*02:01 expressing cell line. Importantly, to further evaluate the immunogenicity, we immunized mice with several mouse-derived neoepitope peptides. Ex vivo ELISpot assay showed that >20% mouse-derived neoepitope peptides could elicit strong immune responses.Moreover, the activated T cells demonstrated potent (>50%) tumor cell killing capability in the ex vivo cellular experiment. Conclusions: We crafted sophisticated AI algorithms capable of rapidly designing shared neoantigen T cell epitopes that exhibit strong immunogenicity. We believe that our AI technology has substantial promise for propelling the development of ready-made cancer vaccines, and will eventually improve patient outcomes when successfully applied in clinical settings. Citation Format: Genwei Zhang, Jiewen Du, Xiangrui Gao, Zhenghui Wang, Qingxia Zhang, Tongren Liu, Dong Chen, Chi Han Samson Li, Kin Tak Chan, Hong Wang, Melvin Toh, Lipeng Lai. Artificial intelligence accelerates shared neoantigen peptide cancer vaccine development [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3649.