Synthetic long-peptide vaccines that present chosen neoepitopes directly, often with an adjuvant — the most clinically mature personalized-vaccine approach.
Updated results from the randomized phase II iNATURE trial, presented by FengMing P. Kong at ESMO 2026 , showed that combining a personalized neoantigen peptide vaccine with precision critical-lesion radiotherapy significantly improved progression-free survival in patients with h…
Neoadjuvant regimens often fail to eliminate micrometastases due to profound immunosuppression and the limited immunogenicity of residual tumor cells. In this work, a versatile lipid nanoplatform was engineered to orchestrate potent chemotherapy with in situ neoantigen vaccinatio…
A preprint describes NeoOmni Atlas, a pan-cancer T-cell epitope database built via proteogenomic reanalysis of 7,561 immunopeptidomics datasets (511 million raw spectra from 2,341 samples across 26 cancer types), yielding 872,414 eluted canonical and 198,961 non-canonical peptides across 18 molecular alteration classes, plus 759,386 unique ligands presented by 322 HLA alleles. A computational tumor-specific filtering framework identified 316,126 tumor-associated and 61,231 tumor-specific epitopes, with the tumor-specific set showing stronger predicted HLA class I A/B binding affinity, and 65 conserved tumor-specific epitopes recurrent across more than 10 tumors. Why it matters: mass-spectrometry-validated, non-canonical-inclusive target catalogs at this scale — including 369,781 novel ligands and 60 additional HLA alleles versus existing repositories — give vaccine developers and ML teams a ground-truth resource for target prioritization and model training.
A case report in locally recurrent osteosarcoma demonstrates an end-to-end immune monitoring workflow after personalized neoantigen vaccination: in vitro peptide stimulation plus IFN-gamma ELISPOT detected post-vaccination responses to 12 of 15 tested peptides, and 10x single-cell TCR sequencing yielded 26 candidate TCRs, of which 25 generated functional TCR-T cells. Only 4 were specific to vaccine peptides, and all 4 proved MHC class-II restricted — a key constraint for anyone planning to spin vaccine-discovered TCRs into conventional (class I-targeted) TCR-T assets. For the field, this validates single-cell TCR discovery as a pharmacodynamic readout of whether a vaccine actually elicited trackable antigen-specific clones, not just bulk ELISPOT signal.
[RECRUITING · PHASE1 · Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins] The purpose of this study is to determine whether the combination of a neoantigen vaccine targeting the DNAJB1-PRKACA fusion kinase in combination with DRP-104, Nivolumab and Ipilimumab is safe and…
Characterization of naturally processed MHC class I (MHC-I)-bound peptides is essential for defining allele-specific peptide-binding features and understanding the functional consequences of MHC-I polymorphism. However, immunopeptidome analyses of dog leukocyte antigen (DLA) clas…
HeteroTCR, a heterogeneous graph neural network for predicting peptide-TCR interaction, appears in Nature. Peptide-TCR prediction sits at the heart of neoantigen prioritization, separating predicted binders from antigens T cells actually recognize, so a graph-based method landing in a top journal is a relevant addition to the in-silico selection stack used across vaccine and TCR-therapy developers.
A preprint reports what the authors describe as the first application of ion trap CID and IRMPD fragmentation to HLA class I immunopeptidomics, testing four HLA-typed cell lines in technical triplicate against the canonical proteome and a Ribo-seq-derived cryptic ORF database on an Orbitrap Tribrid Apex MultiOmics. Because the peptide is itself the therapeutic target, ligands that HCD misses are targets never found, so complementary activation chemistries directly widen the addressable neoantigen pool. The catch for the AI stack: fragment-intensity predictors like AlphaPeptDeep are trained almost exclusively on HCD spectra and had to be fine-tuned per method, meaning wet-lab protocol changes propagate straight into model pipelines.
A biorxiv preprint introduces TWISTAR, a platform that deorphanizes single-cell-derived TCRs by screening them against tumor transcriptomes, validating actual TCR engagement so only functionally immunogenic antigens are recovered. Applied to melanoma, breast, and hepatocellular carcinoma patients responding to immune checkpoint blockade, it shows tumor-reactive TCRs recognize targets beyond neoantigens—including peptides from unannotated coding sequences and novel transcripts—many shared across patients and cancer types, with multiple clones converging on a few immune-dominant epitopes. Why it matters: shared, immune-dominant antigens are the prerequisite for off-the-shelf or pan-cancer vaccines, and the finding that the immunogenic landscape is broader than the neoantigen set expands the addressable target space. The work is method-stage and retrospective on ICB responders; no prospective vaccine validation is reported, so watch for independent replication and whether the platform is productized by an antigen-discovery player.
Beijing Tiantan Hospital is running a single-center, open-label phase 1 of ZSNeo-DC1.1, an autologous, genetically unmodified dendritic cell vaccine loaded with multiple tumor neoantigen peptides, in roughly 15 adults with recurrent or progressive WHO grade III-IV glioma after reoperation. The protocol uses a fixed 1×10^7-cell dose, two immunization schedules, and a DLT window running to 21 days after the third injection, starting with a six-subject dose confirmation stage. It adds a China-based safety and preliminary efficacy datapoint for personalized DC vaccines in recurrent high-grade glioma post-standard treatment.
PREpiBind benchmarks ten protein representations inside a single joint-attention architecture for peptide-MHC class II binding prediction under fixed data splits; ESM3 Large reached ROC-AUC 0.927±0.002 on the qualitative dataset. PLM-based variants exceeded reference methods on qualitative and mass spectrometry benchmarks, but NetMHCIIpan-4.3 retained the edge on thresholded IC50 via its affinity head, and PLM advantages narrowed in allele-wise and held-out-MHC evaluations. The takeaway for vaccine target selection: no single representation wins everywhere — benchmark against your specific allele and data-type regime.
A preprint integrating immunopeptidomics and transcriptomics across all four medulloblastoma subgroups (Group 3, Group 4, WNT, SHH) mapped the non-canonical HLA class I ligandome: 2,298 non-canonical peptides across 21 samples, of which 598 were cancer-restricted after filtering against normal-tissue immunopeptidomic and ribosome profiling data. These derive from retained introns, out-of-frame open reading frames, and lncRNAs rather than mutations — significant because medulloblastoma's low mutational burden limits mutation-derived neoantigens. Some cancer-restricted candidates are shared across patients, pointing to targets usable beyond single-patient personalization.
Peptide-HLA binding prediction is a critical step in neoantigen identification for personalized cancer immunotherapy and holds significant clinical value. However, the training data available for many HLA alleles are extremely limited, which severely constrains the performance of…
The paper describes a polymer-based polyvalent peptide and adjuvant platform, SPPA, designed to address limitations of neoantigen peptide vaccines such as poor antigen stability, inefficient delivery, and inadequate immune activation. SPPA co-delivers lipid-conjugated tumor-specific peptides with a TLR7/8 agonist, 3M-052, and a STING agonist, 2'3'-cGAMP, to improve peptide encapsulation, sustained release, and targeted delivery to antigen-presenting cells. In vitro, SPPA upregulated pro-inflammatory genes and cytokine secretion and showed effective cellular uptake and lymphatic trafficking. In vivo, SPPA alone or with anti-PD-1 antibody elicited cytotoxic T lymphocyte responses and inhibited tumor growth in four aggressive syngeneic mouse models, including triple-negative breast cancer.
The study argues that label noise in TCR-pMHC specificity datasets is a major driver of weak performance for unseen peptide prediction, where structural modeling has so far shown the most predictive power. Using an AlphaFold3-based pipeline adapted for TCR-pMHC structural modeling, the authors report state-of-the-art specificity prediction that outperforms AlphaFold2.3-based and sequence-based methods and performs at par with the leading Immrep2025 competition submission. A cluster-based denoising step that removes mislabeled points from a large specificity dataset increased binder ranking accuracy by more than 70% relative to the full dataset, implying that data quality improvements could materially affect TCR-based immunotherapy and vaccine design.
Advances in HLA ligandome analyses are revealing immunogenic peptides from noncoding regions, expanding the neoantigen pool beyond traditional whole-exome sequencing. This shift is crucial for identifying clinically actionable targets for cancer vaccines and adoptive cell therapies in patients who do not respond to immune checkpoint inhibitors.
Researchers present RyboCypher, an AI-assisted proteogenomics platform that identified ~80,000 cryptic peptides (including ~10,000 cancer-associated) from 2,229 patient samples. By mapping the 'dark transcriptome' to unannotated proteins, the study establishes the CypherAtlas, offering a vast new reservoir of potential neoantigen targets for first-in-class therapeutic discovery.
MHChron introduces a unified peptide-MHC binding prediction framework built on a diverse dataset covering 214 class I and 98 class II alleles, with peptides ranging from 8 to 36 residues. The study emphasizes rigorous data curation and leakage-controlled splits to ensure robust generalization. Both sequence-based and structure-aware models trained on this balanced dataset outperformed state-of-the-art predictors, particularly in extrapolating to unseen alleles.
Volume 17 - 2026 | https://doi.org/10.3389/fgene.2026.1875066 1 Introduction and immunological background 2 AI-enabled neoantigen and epitope discovery 3 AI for HLA binding, antigen processing, and immunogenicity prediction 4 Personalization, clinical translation, and current evi…
Researchers have applied Quantum Convolutional Neural Networks (QCNNs) to neoantigen prediction, addressing limitations of classical models in handling noisy, small datasets. Using a 46-qubit hardware experiment with noise mitigation techniques, the study achieved a 6% increase in classification accuracy for MHC binding and immunogenicity prediction compared to classical baselines. This work demonstrates the potential of quantum computing to enhance precision in neoantigen identification.
An analysis of 352 patients receiving personalized peptide-pulsed dendritic cell vaccines reveals that only 10-20% of selected peptides induce detectable T-cell responses. The study evaluated 2,317 short peptides from single nucleotide variants using IFN-γ ELISPOT assays to define immunogenicity thresholds. These findings underscore the current limitations of prediction algorithms and the need for improved biomarkers to select clinically relevant neoantigens.
Scientists developed a carrier-free nanovaccine for hepatocellular carcinoma by self-assembling fluorinated CpG with HCC neoantigen peptides. The resulting 69 nm nanoparticles demonstrated high loading efficiency and enhanced dendritic cell uptake, improving antigen cross-presentation by 2.05-fold in vitro. In vivo, the formulation targeted lymph nodes to orchestrate both innate and adaptive immune responses, offering a biosafe alternative to traditional adjuvant-loaded nanovaccines.
The DRIVE database addresses the lack of resources linking drug-induced transcriptomic changes to neoantigen potential by integrating 3,911 samples across 278 drugs and 272 cell lines. Using LLMs for metadata curation, it quantifies differential splicing events and predicts resulting HLA-binding peptides. This resource enables the systematic identification of immunogenic neoantigens derived from drug-induced aberrant splicing, offering a new avenue for vaccine target discovery in pharmacotherapy contexts.
Researchers have identified limitations in current sequence-based approaches for predicting MHC-peptide interactions and introduced a structure-based method using geometric deep learning (GDL). By employing self-supervised learning on 3D structures without exposure to binding affinity data, the model shows improved generalizability across unseen MHC alleles. This suggests a viable path toward more robust computational tools for neoantigen identification in cancer immunotherapy.
T cells are a key part of the adaptive immune system. Using their surface-bound T cell antigen receptors (TCRs), these cells scan peptides and other antigens presented to them by major histocompatibility complex molecules (MHCs) on the surface of cells, searching for abnormalitie…
Researchers benchmarked AlphaFold2, AlphaFold3, and related deep learning methods for modeling antibody-protein, antibody-peptide, and TCR-pMHC recognition. Results indicate that increased sampling and AlphaFold3 generally outperform default AlphaFold2 settings, though predictive accuracy varies significantly by interface class. The authors note that antibody-peptide complexes remain particularly difficult to model accurately and propose model pooling as a potential solution to leverage method complementarity.