ASCO 2026: where personalized cancer vaccines stand — and the landscape around them

The ASCO Annual Meeting is the largest clinical-oncology gathering of the year, and its 2026 edition (May 29 – June 2, Chicago) ran under the theme "The Science and Practice of Translation." For a field like personalized cancer vaccines — which has spent a decade proving it can make a bespoke product per patient, and the last few years trying to prove that product changes outcomes — translation is exactly the right word. This is a builder's reading of the meeting: which abstracts move our field, which map the landscape around it, and which expected readouts simply did not show up.

A note on stance before the data. This site is written by an engineer, not an oncologist — there is no lab and no position in any of these companies here. The aim is to be positive about the field as a whole while staying neutral about any single datapoint, and to name the disappointments as plainly as the wins. Every number below is taken from the presented abstract, its simultaneous publication, or the presenting institution's own release; where a figure is early or single-sourced, it is flagged as such.

The headline belongs to intismeran autogene — the mRNA individualized neoantigen therapy formerly known as mRNA-4157 / V940, from Moderna and Merck. The five-year update of the randomized phase 2b KEYNOTE-942 trial (Abstract 9500, published simultaneously in the Journal of Clinical Oncology) showed that adding the vaccine to pembrolizumab in resected high-risk stage III/IV melanoma sustained its benefit out to a median follow-up of about 60 months: a 49% reduction in the risk of recurrence (hazard ratio 0.51) and a 59% reduction in the risk of distant metastasis or death (hazard ratio 0.41), across 157 patients. Each vaccine encoded up to 34 patient-specific neoantigens. That durability — the benefit holding, not eroding, at five years — is the single most important fact for the field this year, because the central worry about adjuvant vaccines has always been whether an early recurrence-free-survival signal survives long follow-up. Here it did.

The second readout extends the approach into one of oncology's hardest settings. A Dana-Farber team (David Reardon and Catherine Wu) presented a phase 1 trial of NeoVax — a personalized neoantigen peptide vaccine — plus pembrolizumab in newly diagnosed glioblastoma (Abstract 2006, oral session). Median overall survival reached 36.9 months in MGMT-methylated patients versus 25.3 months in propensity-matched standard-of-care controls, and 19.0 versus 16.7 months in the harder-to-treat unmethylated group. The trial deliberately avoided dexamethasone, an immune suppressant, and the investigators confirmed that vaccine-specific T cells trafficked into the brain and tumor — mechanistic evidence, not just a survival curve, that a personalized vaccine can mount a response inside the central nervous system.

The third points the modality at prevention rather than treatment. NOUS-209, an off-the-shelf vaccine encoding shared frameshift neoantigens common to mismatch-repair-deficient tumors (Nouscom; presented by Eduardo Vilar-Sanchez of MD Anderson), induced a neoantigen-specific T-cell response in all 37 evaluable Lynch-syndrome carriers, durable at one year in 85%, with no serious adverse events. The program received FDA Fast Track designation on June 1, 2026, and the underlying data were published in Nature Medicine in January 2026. Lynch syndrome — an inherited high risk of colorectal and other cancers — is a clean test bed for cancer interception, and a 100% immune-response rate in people who are not yet sick is a genuinely distinct thread from the adjuvant-treatment story.

The most interesting structural shift at this meeting was not a new vaccine — it was that neoantigens showed up as targets for cell therapy. NT-175, from Neogene Therapeutics (an AstraZeneca company), is an engineered TCR-T product whose receptor recognizes the TP53 R175H neoantigen presented on HLA-A*02:01 — and it reported an objective response rate of 47.6% in a first phase 1 cohort of 21 infused patients with that specific mutation (Abstract 2506). TP53 is the most commonly mutated gene in cancer, and R175H is its single most frequent hotspot, so a therapy that drugs it directly addresses a target that small molecules have spent forty years failing to hit. A second program directed at the KRAS G12V neoantigen on HLA-A*11:01 was flagged among the meeting's TCR-T abstracts to watch.

Why does a vaccine site cover a cell therapy? Because the hard part is shared. A neoantigen vaccine and a neoantigen-reactive TCR-T solve the same upstream problem — read the tumor's mutations, work out which mutated peptides are actually presented on the patient's HLA, and pick the ones a T cell can see — and then diverge only in the delivery: teach the patient's own immune system to find the target (vaccine) versus manufacture T cells that already recognize it (cell therapy). ASCO 2026 is the clearest sign yet that "personalized = the tumor's own mutations" is a platform thesis, not a single-modality bet. The target-identification engine we track every day feeds both.

Personalized vaccines do not compete in a vacuum; they compete for the same patients, trials and investor attention as the rest of personalized oncology. Three competing modalities had loud meetings. Bispecific antibodies kept their momentum: ivonescimab, a PD-1×VEGF bispecific, beat an active PD-1 (tislelizumab)–based regimen in first-line squamous lung cancer in the HARMONi-6 trial (late-breaking abstract LBA4, overall-survival hazard ratio 0.66) — the first time a regimen has topped an active PD-1 control head-to-head in that setting, reinforcing the bispecific as the off-the-shelf challenger to checkpoint immunotherapy. Antibody-drug conjugates continued their expansion — sacituzumab tirumotecan, a TROP2 ADC, posted a large progression-free-survival benefit in first-line lung cancer (Abstract 8506). And targeted small molecules produced the meeting's pancreatic-cancer headline: daraxonrasib, a pan-RAS inhibitor, roughly doubled median overall survival versus chemotherapy in metastatic pancreatic cancer in the plenary RASolute302 trial (LBA5; 13.2 vs 6.7 months).

That pancreatic result matters to us in an oblique way. KRAS is exactly the target that shared-neoantigen vaccines such as Elicio's ELI-002 are built around, so a drug that validates hitting mutant RAS directly both proves the target is worth chasing and raises the bar a vaccine must clear. The same double-edged logic runs through the ctDNA story below.

The dominant paradigm of "personalized medicine for cancer" remains genomic, not immunological — matching a drug to a tumor's mutations, and increasingly using circulating tumor DNA to decide who needs treatment at all. The CIRCULATE trial (LBA3500) was the cleanest example and the most instructive caution: among stage II colon-cancer patients who were ctDNA-positive after surgery, chemotherapy sharply cut recurrence (disease-free-survival hazard ratio 0.31) — but the trial's overall primary endpoint was statistically negative and it closed early with only 41 ctDNA-positive patients. Tumor-informed minimal-residual-disease testing (the Natera/Signatera and Guardant programs together accounted for dozens of abstracts) is conceptually close kin to neoantigen work: both start by sequencing a patient's tumor to build a bespoke set of mutations to track. MRD is, plausibly, the enrollment gate and early-readout tool that adjuvant vaccine trials will increasingly lean on.

Artificial intelligence was a cross-cutting theme — ASCO launched a dedicated AI-in-oncology platform earlier in 2026 — though none of the standout models were neoantigen-prediction-specific. The pattern worth noting is adjacent to ours: AI as the tool that decides who responds. Noetik's TARIO-2 foundation model, run on routine H&E pathology slides, identified a checkpoint-inhibitor-responsive subgroup in microsatellite-stable colorectal cancer (Abstract 2535) — an AI-selected group with roughly a 64% response rate versus 9% in everyone else. A separate foundation model, GigaTIME, read bone-marrow slides to estimate immune fitness and help guide therapy in newly diagnosed multiple myeloma (Abstract 7520). These are the same problem we cover from the antigen side — turning messy tumor data into a confident prediction about a single patient — approached from the image side. It is a reasonable bet that the immunopeptidome and the histology slide eventually feed the same model.

Honesty about a field means covering its disappointments. The most conspicuous was an absence: BioNTech and Genentech's autogene cevumeran (BNT122), the personalized mRNA vaccine that produced the celebrated pancreatic-cancer phase 1 data, did not have a marquee ASCO 2026 readout — its pivotal pancreatic data have slipped toward 2027, and its adjuvant colorectal program crossed a futility boundary. Two other closely watched personalized programs, Evaxion's EVX-01 and Transgene's TG4050, presented their most recent data at other 2026 venues rather than ASCO, so anyone reading an "ASCO vaccine" headline should check the dateline. And as above, the marquee ctDNA-guided trial, CIRCULATE, missed its overall primary endpoint even as its biomarker-positive subgroup looked striking. None of this dents the three positive readouts; it simply means the field's progress in 2026 is uneven, which is the normal texture of translation.

Three things to carry out of ASCO 2026. First, the durability question that has shadowed adjuvant neoantigen vaccines now has a five-year answer in melanoma, and the modality reached into glioblastoma and into cancer prevention — three different bets, all of them landing. Second, neoantigens graduated from a vaccine ingredient to a cross-modality target class, with the first cell-therapy responses against shared hotspot neoantigens like TP53 R175H. Third, the competitive landscape sharpened rather than receded: bispecifics, ADCs, pan-RAS inhibitors and ctDNA-guided therapy all advanced, which is good for patients and a reminder that a personalized vaccine has to earn its place against fast-moving alternatives. The throughline, as ever on this site, is the prediction engine underneath — the part that reads a single tumor and decides what to aim at. ASCO 2026 made the case that getting that part right is now the bottleneck for more than one kind of medicine.