Deeper, hand-written takes — what the daily brief adds up to over time, read against the data.
A neoag reader's map of ASCO 2026: the three personalized-vaccine readouts that matter (intismeran's five-year melanoma update, NeoVax in glioblastoma, NOUS-209 in Lynch syndrome), the arrival of neoantigen-reactive TCR-T as a real category, the competing modalities (bispecifics, ADCs, ctDNA-guided therapy, pan-RAS), and the readouts that were notably absent.
Neoantigen vaccines and neoantigen-reactive TCR-T cells solve the same upstream problem — read a tumor's mutations, find which mutated peptides are presented on the patient's HLA, pick the targets — and diverge only in delivery: teach the patient's T cells (vaccine) or manufacture T cells that already recognize the target (cell therapy). ASCO 2026's TP53 R175H data make the contrast concrete. Here is the trade-off, setting by setting. · 2026-06-04
A maintained field guide to the neoantigen-prediction toolchain: the end-to-end open-source pipelines (pVACtools, Seq2Neo, NeoPredPipe, pTuneos, OmniNeo and more), the peptide–HLA presentation predictors beneath them (NetMHCpan, MHCflurry, MixMHCpred), and the immunogenicity frontier where the hard, unsolved problem lives. · 2026-05-30
NetTCR, ERGO-II, TITAN, STAPLER, MixTCRpred, TABR-BERT, TULIP, tcrLM, pMTnet, epiTCR, TCR-ESM and the structure-based newcomers — a neutral, sourced field guide to TCR–pMHC binding prediction. · 2026-05-30
NetMHCpan-4.1 and MHCflurry 2.0 both predict whether a peptide will be presented by an HLA molecule, but they differ in licensing, scriptability, allele coverage, and class II support. Neither is uniformly better. Here is how to choose, and the caveat that matters more than the choice. · 2026-05-30
HLA binding is largely a solved problem. Antigen presentation is mostly solved. Immunogenicity — whether a presented peptide actually triggers a T-cell response — is not. We unpack why the prediction pipeline breaks at the last step, and how the field is trying to fix it. · 2026-05-30
Protein and biology foundation models — the ESM lineage, AlphaFold2/3, and BERT-style sequence models — are being aimed at the hardest parts of neoantigen vaccine design: pMHC binding, TCR recognition, and de novo peptide generation. A sharp look at what is demonstrated, what is promised, and the under-appreciated HLA-equity dividend. · 2026-05-30
Personalized and off-the-shelf neoantigen vaccines make opposite bets on precision versus scale. We map the tradeoffs in manufacturing, turnaround, cost, applicability and the prevention angle — and where AI fits each — citing the lead programs from Moderna/Merck, BioNTech, Nouscom, Elicio and others. · 2026-05-30
The companies, platforms, and catalysts in neoantigen cancer vaccines as of mid-2026 — personalized mRNA leaders, AI-native mid-caps, and off-the-shelf shared-antigen approaches, with endpoints and risks called straight. · 2026-05-30
Every patient receives a uniquely sequenced vaccine, which collides with a regulatory framework built around a single defined product. FDA's answer — validate the process, not the molecule — plus the milasen N-of-1 precedent, the Platform Technology Designation, the designations real programs hold, and the open questions a first approval would settle. · 2026-05-30
Eight computational phases turn a patient's tumor mutations into a vaccine: HLA typing, variant calling, pMHC binding, immunogenicity, construct assembly, and mRNA optimization. A practical map of the tools — OptiType, pVACtools, ESM-2, LinearDesign — that run each one. · 2026-05-30
Foundation models have reached neoantigen discovery — but only some can actually be fine-tuned on your own data. A field guide to the ones that can, grouped by the problem they solve. · 2026-05-30
Search interest in “neoantigen” sat near zero for years, then broke out in 2026 to its highest point on record. What the curve means — and its limits. · 2026-05-30