What question did this study set out to answer?

April 5, 2026

Abstract 4376: Harnessing artificial intelligence to optimize neoantigen prediction and mRNA design for personalized cancer vaccine

Key Points

This research aims to improve neoantigen prediction and mRNA design for personalized cancer vaccines using AI.
Developed EchoNeo 1.0, a multimodal deep learning model for immunogenicity prediction.
Trained on diverse immunogenicity datasets and biological metrics.
Synthesized neoantigen mRNA vaccines formulated with lipid nanoparticles.
Evaluated vaccine safety and efficacy in mouse models.
EchoNeo demonstrated superior performance over traditional prediction tools.
In mice, the vaccine inhibited melanoma growth and showed no significant adverse effects.
Enhanced therapeutic effects were observed when the vaccine was combined with anti-PD-1 antibody.

Abstract

Abstract Background: Neoantigen-based mRNA personalized cancer vaccines (PCV) represent a promising frontier in cancer immunotherapy. However, current neoantigen prediction strategies primarily rely on MHC binding affinity, which leads to limited accuracy in immunogenicity and results in a high false-positive rate. The precise identification of neoantigen epitopes that are efficiently presented by MHC and elicit robust immunogenicity remains a central challenge hindering the clinical application of PCV. Here, we developed EchoNeo 1.0, a multimodal deep learning-driven pipeline that innovatively integrates immunogenicity prediction with mRNA sequence design to accelerate the of application of PCV. Methods: The core of pipeline is a multimodal deep learning model for immunogenicity prediction. Trained on publicly available databases (IEDB, TSNAdb v2.0, TESLA) and published immunogenicity data, the model integrates peptide/HLA pseudosequence features with multidimensional biological metrics to achieve accurate immunogenicity scoring. Neoantigen mRNA vaccines were synthesized and formulated with lipid nanoparticles (LNP), and evaluated in mouse models for safety (including toxicology and in vivo biodistribution) and therapeutic efficacy. Results: Approximately 3,700 peptides (8-11 amino acids) with confirmed experimental immunogenicity data were used to train and validate EchoNeo, which demonstrated superior performance in benchmark tasks against established prediction tools (e.g., DeepImmuno, IEDB Class I Immunogenicity tool) and showed excellent immunogenicity prediction accuracy on independent, clinically validated neoantigen datasets. In C57BL/6 mice intramuscularly administered with vaccine (at a maximum dose of 54 μg, approximately 2.7 mg/kg), no significant adverse changes were observed in body weight, body temperature, blood biochemistry (ALT, AST, DBIL, CREA, UREA, TG, TC, LDL, HDL, LDH, CK), or complete blood count (CBC, DIFF, and RET). In a melanoma model, the vaccine significantly inhibited tumor growth, and enhanced therapeutic effect was achieved when combined with anti-PD-1 antibody. Conclusion: Superior immunogenicity prediction accuracy was confirmed, as well as the favorable safety and potent antitumor efficacy in mouse models. This work represents a paradigm shift from conventional affinity-based prediction to an end-to-end framework that directly assesses the immunogenicity of neoantigen peptides, holding significant promise for accelerating the development of PCV. Citation Format: Yang Liu, Qingyi Mao, Wu Xinghan, Qiu Mantang. Harnessing artificial intelligence to optimize neoantigen prediction and mRNA design for personalized cancer vaccine abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4376.

Mark Helpful

Bookmark

Relay