{
  "abstract": "Background Personalized cancer vaccines depend on the accurate identification of immunogenic neoepitopes capable of eliciting strong T cell responses. However, widely used prediction tools such as NetMHCpan primarily focus on HLA binding affinity, which fails to account for the complex structural and dynamic factors essential for T cell recognition. Consequently, many predicted binders do not induce immune responses, limiting their translational value. We developed a rapid, physics-based screening platform that combines structural modeling and molecular dynamics (MD) simulations to improve neoepitope prioritization for personalized vaccine development. A total of 954 neoepitopes, derived from late-stage cancer patients and spanning four prevalent HLA class I alleles (HLA-A*02:01, HLA-A*24:02, HLA-A*02:06, and HLA-A*11:01), were evaluated against three clinically relevant TCRs.Methods Structural models of peptide-MHC (pMHC) in complex with T cell receptors (TCRs) were generated using TCRmodel2, 1 an AlphaFold2-based prediction tool. These models served as starting structures for short MD simulations, from which dynamic features of the pMHC, specifically at anchor and TCR-facing peptide residues, were extracted. Key descriptors included flexibility, solvent accessibility, and conformational convergence of TCR-facing residues, as well as the interaction profiles of anchor residues within the MHC class I binding groove. These features were integrated into a Consensus Scoring Function (CSF), producing immunogenicity scores on a 0–1 scale. The pipeline was first applied to the HLA-A*02:01 dataset to develop the CSF model (figure 1), followed by cross-validation using combinations of the remaining three HLA alleles and three TCRs each (figure 2, Step 1-2). CSF-based predictions were benchmarked against in-house ELISpot assays performed on neoepitopes derived from our patient cohort.Results The CSF-based platform outperformed NetMHCpan, 2 which predicted all 954 peptides as good binders without discriminating immunogenicity. We successfully completed 2,862 MD simulations within 5 days. Filtering based on CSF scores reduced the candidate pool by over 50%, while retaining 36–65% of experimentally validated immunogenic peptides across TCR-HLA combinations. In addition, Structural classifications (figure 2, Step 3) revealed structural features associated with immunogenicity. These findings support the role of conformational dynamics in T cell activation. Moving forward, we aim to rescue overlooked immunogenic candidates by identifying structural similarity between known immunogenic peptides and CSF-positive but ELISpot-negative epitopes.Conclusions This work highlights the value of incorporating structural dynamics into neoepitope screening pipelines for cancer vaccine design. Our model offers a generalizable, mechanistically grounded approach to prioritize neoepitopes with immunogenic potential. This platform complements existing sequence-based predictors and may significantly enhance the precision of personalized cancer immunotherapy.Acknowledgements The computations were performed using the following resources: 1. Research Center for Computational Science (RCCS), Okazaki, Japan, (Project ID: 25-IMS-C987). IMS-RCCS-etc-ja 2. TSUBAME 4.0 supercomputing system, provided by Science Tokyo through the HPCI System Research Project (Project ID: hp240181).References Reynisson B, Alvarez B, Paul S, Peters B, Nielsen M. NetMHCpan-4.1 and NetMHCIIpan-4.0: improved predictions of MHC antigen presentation by concurrent motif deconvolution and integration of MS MHC eluted ligand data. Nucleic Acids Research. 2023;51(W1):W569-W576. https://doi.org/10.1093/nar/gkad288 Hoof I, Peters B, Sidney J, Pedersen LE, Sette A, Lund O, Buus S, Nielsen M. NetMHCpan, a method for MHC class I binding prediction beyond humans. Immunogenetics. 2009;61:1–13. https://doi.org/10.1007/s00251-008-0341-zAbstract 1005 Figure 1Construction of the CSF model for neoepitope prioritization. Workflow for generating the Consensus Scoring Function (CSF) based on NetMHCpan filtering, TCRmodel2 structure prediction, and short MD simulations of pMHC (TCR removed). CSF integrates dynamic features of anchor and TCR-facing peptide residuesAbstract 1005 Figure 2CSF validation across HLA alleles and TCRs. Cross-validation of CSF-based predictions using ELISpot assay data across three TCRs and four HLA class I alleles. Clustering analysis reveals structural patterns linked to immunogenicity and enables rescue of overlooked epitope candidates",
  "authors": [
    {
      "affiliations": [
        "Laboratory of In Silico Design, Artificial Intelligence Center for Health and Biomedical Research, National Institutes of Biomedical Innovation, Health and Nutrition, Settsu, Osaka, Japan"
      ],
      "name": "Jelang M Dirgantara"
    },
    {
      "affiliations": [
        "Laboratory of Immunogenomics, Center for Intractable Diseases and ImmunoGenomics, National Institute of Biomedical Innovation, Health and Nutrition, Ibaraki, Osaka, Japan"
      ],
      "name": "Kazuma Kiyotani"
    },
    {
      "affiliations": [
        "Laboratory of Precision Immunology, Center for Intractable Diseases and ImmunoGenomics, National Institutes of Biomedical Innovation, Health and Nutrition, Ibaraki, Osaka, Japan"
      ],
      "name": "Takuto Nogimori"
    },
    {
      "affiliations": [
        "Laboratory of Precision Immunology, Center for Intractable Diseases and ImmunoGenomics, National Institutes of Biomedical Innovation, Health and Nutrition, Ibaraki, Osaka, Japan"
      ],
      "name": "Shokichi Takahama"
    },
    {
      "affiliations": [
        "Fukuoka General Cancer Clinic, Fukuoka, Fukuoka, Japan"
      ],
      "name": "Takashi Morisaki"
    },
    {
      "affiliations": [
        "Laboratory of Precision Immunology, Center for Intractable Diseases and ImmunoGenomics, National Institutes of Biomedical Innovation, Health and Nutrition, Ibaraki, Osaka, Japan"
      ],
      "name": "Takuya Yamamoto"
    },
    {
      "affiliations": [
        "Laboratory of In Silico Design, Artificial Intelligence Center for Health and Biomedical Research, National Institutes of Biomedical Innovation, Health and Nutrition, Settsu, Osaka, Japan"
      ],
      "name": "Suyong Re"
    }
  ],
  "title": "1005 Structure-based prioritization of immunogenic neoepitopes for personalized cancer vaccine design",
  "uid": "0f4ce6e5-423d-5ce4-bf93-5c79c1a62066"
}