Aug 2026· Global Health Care· 0 citations· 21 references
TL;DR
This work is the first to use a multi-modal foundation model for neoantigen vaccines, overcoming the tumor-type-specific constraints of conventional approaches and facilitating joint modelling of immunogenicity and clinical efficacy, thereby offering an AI decision engine for precision cancer vaccine design that is applicable to all cancer types.
Abstract
Neoantigen vaccines are a key component of personalised cancer immunotherapy; however, existing prediction techniques are primarily restricted to a single tumour type and have difficulty integrating multi-modal biological data, which leads to inadequate accuracy in immunogenicity evaluation and vaccine efficacy prediction. In order to predict neoantigen immunogenicity and customised vaccination clinical outcomes across various tumour types, this study attempts to develop and verify a universal multi-modal foundation model.Neoantigen peptide sequences, mass spectrometry-derived pHLA binding patterns, single-cell TCR repertoires, tumour transcriptomes, and clinical vaccination trial follow-up records were among the extensive paired data we gathered from 15 solid tumour types. We present the NeoVAX-FM multi-modal foundation model, which first embeds peptide sequences, pHLA complex 3D structures, and gene expression into a single semantic space using a contrastive language-image pretraining paradigm. It is then fine-tuned on downstream tasks to concurrently score immunogenicity and predict progression-free survival.The model was assessed in one prospective clinical trial and three external validation cohorts. NeoVAX-FM showed strong performance in melanoma, non-small cell lung cancer, and microsatellite stable colorectal cancer, with an average AUC of 0.94 for cross-tumor neoantigen immunogenicity prediction—a 12.3% increase over the best currently available techniques. Patients in the prospective vaccination cohort who were projected by the model to be “high responders” had a considerably higher median progression-free survival (HR = 0.28, p < 0.001), and the model was successful in identifying tumour microenvironment characteristics and universal TCR motifs that drive long-term responses.This work is the first to use a multi-modal foundation model for neoantigen vaccines, overcoming the tumor-type-specific constraints of conventional approaches and facilitating joint modelling of immunogenicity and clinical efficacy, thereby offering an AI decision engine for precision cancer vaccine design that is applicable to all cancer types.
Neoantigen vaccines have become a central direction in precision cancer immunotherapy because they aim to target tumor-specific peptide sequences generated by somatic alterations rather than self-antigens shared with normal tissues. This biological distinction reduces the barrier of central tolerance and creates a rational basis for individualized T-cell priming. The field has also changed technically. Tumor-normal sequencing, transcriptomic filtering, HLA typing, immunopeptidomics, and algorithmic prioritization now make it possible to move from a patient tumor sample to a ranked set of candidate vaccine targets within a clinically meaningful interval. Because durable vaccine responses frequently depend on CD4-positive T-cell help, we also give explicit attention to HLA class II prediction, which remains substantially less accurate than class I prediction. Among available delivery formats, mRNA platforms have become especially important because they can encode multiple patient-specific epitopes in a single product and can be redesigned rapidly as prediction and delivery methods improve, although synthetic long peptide, dendritic cell, viral vector, and DNA platforms retain specific advantages that we compare directly. This review re-examines neoantigen vaccines as a translational system rather than as a single technology. We first outline the biological basis of neoantigen recognition and classify the major antigen sources. We then discuss target discovery, HLA-restricted presentation, computational ranking, immunopeptidomic evidence, and functional validation. Next, we compare vaccine platforms, with particular emphasis on why personalized mRNA vaccines now dominate late-stage clinical development. Finally, we analyze current clinical evidence in melanoma, pancreatic ductal adenocarcinoma, renal cell carcinoma, glioblastoma, and other solid tumors-reporting primary efficacy endpoints, hazard ratios, patient numbers, and follow-up durations where available-and we identify the main barriers that still prevent broad clinical implementation. In our assessment-offered as an expert interpretation rather than as a conclusion derived from comparative or pooled analyses, because the supporting evidence still rests largely on single-arm and early-phase trials in heterogeneous tumor types-the strongest current signal supports use in adjuvant, perioperative, and minimal residual disease settings, usually in combination with checkpoint blockade or other immune-modifying strategies. Neoantigen vaccination is unlikely to become a universal standalone therapy. Its more realistic value is as a programmable immune-priming component within precision oncology.
Minglu Ge, Ning Wu· Cancer Treatment and Researc...· 0 citations
The current role of AI is summarized across the peptide cancer vaccine development pipeline, from neoantigen discovery and epitope prioritization to prediction of peptide–HLA binding, antigen presentation, and T-cell receptor recognition, and the application of modern computational frameworks.
P. Brlek, Jan Kolić, L. Bulić et al.· Frontiers in Genetics· 0 citations
Predicting immunogenicity remains a critical challenge in vaccine design, autoimmunity treatment, and pharmaceutical development. Current AI tools rely on limited data inputs, typically only class-I MHC-peptide sequences, missing crucial structural and biochemical information. We developed the ImmunoFoundation Model (IFM), a multimodal deep learning system that integrates not only peptide sequences, 3D molecular structures, and biochemical properties but also TCR-MHC-peptide (class-II and class-I) to achieve superior immunogenicity prediction and enable peptide optimization for therapeutic applications.
IFM comprises three modules: (1) ESM3 transformer for embedding antigen-peptide, MHC, and TCR sequences from vast biomedical data; (2) geometric scattering transformer networks to capture molecular structure from AlphaFold3-predicted peptide-MHC complexes; (3) Autoencoder for biochemical property embedding including surface area and thermal stability. Cross-modal attention layers integrate these representations. Training utilized IEDB, VDJdb, McPAS-TCR, and TCR3d datasets totaling >300,000 samples across MHC class I and II.
The preliminary model achieved state-of-the-art performance on CEDAR cancer neoepitope datasets. Attention mechanism analysis revealed structural motifs influencing immunogenicity, distinguishing between KRAS G12V and G12D mutants. The model successfully predicted vaccine cassette immunogenicity and identified key peptide-MHC interaction sites. Current IFM development shows improved multimodal integration with enhanced predictive accuracy across viral and cancer peptide immunogenicity tasks.
IFM represents a paradigm shift in immunogenicity prediction by comprehensively modeling the complex antigen-MHC-TCR interaction system. Its generative capabilities enable peptide optimization for cancer vaccines and personalized immunotherapy, with potential applications in autoimmunity treatment and biologics development.
Yale Colton Center for Autoimmunity
Computational and Systems Immunology (COMP)
Smita Krishnaswamy, J. Rocha, Hiren Madhu et al.· Journal of Immunology· 0 citations
Melanoma, a highly aggressive and therapy-resistant skin cancer, is increasingly treated with immunotherapy, driven by advances in molecular biology and cancer immunology. This study aimed to develop an integrated computational method that combines deep learning based somatic mutation detection, MHC-binding prediction, and reverse vaccinology to identify melanoma neoantigens and design an epitope-based vaccine construct. Paired tumor-normal whole-genome sequencing data were investigated using deep learning-based variant callers, to identify somatic mutations. RNA-seq data were used to validate transcript expression and extract mutant coding sequences. MHC class I binding affinity was evaluated using the network-based deep learning model, and high-affinity binders were assessed using Immunoinformatic filters. Multiple reverse vaccinology filters were then utilized to identify potential neoantigens. A total of 4,050 mutant epitopes were initially predicted, from which nine epitopes met all immunoinformatic selection criteria and were used to construct the multi-epitope vaccine (MEVC). These epitopes were linked using AAY linkers, while a TLR-4 agonist adjuvant was additionally attached via an EAAAK linker to enhance the immunogenicity of the vaccine construct. The final MEVC comprised 145 amino acids, with stable physicochemical properties, signifying improved cellular uptake and immune interaction. Structural modeling, and molecular dynamics simulations of 100-ns confirmed favorable stability and interaction between the vaccine construct and TLR-4. Furthermore, Immune simulation (C-IMMSIM) showed a balanced humoral and cellular immune response. Overall, the results suggest that the proposed MEVC is stable and immunogenic; however, experimental and preclinical validation is required to confirm these findings.
Saba Ismail, Devin Atkin, Khaled Barakat. A Genome-to-Vaccine: An Integrated Deep learning-Driven Neoantigen Identification and Multi-Epitope Vaccine Construction for Personalized Melanoma Immunotherapy [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Breaking Barriers in the Fight against Rare Cancers; 2026 Jul 18-20; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2026;86(14_Suppl):Abstract nr B055.
S. Ismail, Devin Atkin, K. Barakat· Cancer Research· 0 citations
Understanding vaccine durability is key to designing immunizations with long-term efficacy. Leveraging the Immune Signatures Data Resource, a compendium of transcriptomic and immunological responses from 1405 healthy adults (18+ years) across 24 vaccines, we investigated shared immune mechanisms underlying durable antibody responses. The dataset spans live (yellow fever, smallpox), recombinant viral-vector (Ebola), inactivated (influenza), and glycoconjugate (pneumococcal) vaccines.
Data preprocessing included imputation, normalization, and alignment across post-vaccination time points. We applied advanced machine learning (ML) frameworks to predict antibody immunogenicity and durability. Feature selection for high-dimensional, low-sample-size multi-omics datasets was performed using HSIC Lasso to identify predictors of antibody responses. Selected features served as input to ensemble, regularized regression, and gradient-boosting models (e.g. DT, RF, LASSO, XGB, CatBoost). We compared single-target and multi-output approaches, evaluating stacked, chained, and wrapper-based strategies, and implemented multi-layer neural networks to capture complex relationships among immune features.
Post-vaccination time points explained ∼15% of the total variance, indicating shared immune kinetics across vaccine types, while age and sex contributed minimally. Gradient-boosting and multi-output modeling approaches achieved the highest predictive accuracy across vaccines, highlighting the value of integrating correlated outcomes. Neural network models similarly captured complex, nonlinear immune signatures, albeit with reduced explainability.
Conserved transcriptional modules, particularly interferon-signaling and plasmablast-related pathways, emerged as strong predictors of antibody durability. This integrative ML framework enables identification of key immune signatures critical for developing vaccines with durable responses, advancing data-driven strategies for systems vaccinology.
n/a
Computational and Systems Immunology (COMP)
Marco Sanna, Jing Chen, Paolo Palma et al.· Journal of Immunology· 0 citations
What if pathology foundation models could do more with less? GigaPath-Flash and GigaTIME-Flash cut computational demands while maintaining strong performance, opening the door to larger studies and broader exploration. The post GigaPath-Flash and GigaTIME-Flash: Toward population-scale discovery with efficient pathology foundation models appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduAug 31, 2026
With millions of users across the world, Julia has been used to conduct cutting-edge research and to design new drugs, jet engines, heat pumps, and more.