First Phase 3 Win for a Personalized mRNA Cancer Vaccine — Built on a Neoantigen-Picking Algorithm

For more than a decade, personalized cancer vaccines have lived in the gap between elegant theory and unproven medicine. On August 19, 2026, that gap narrowed decisively: Merck and Moderna announced that their individualized neoantigen therapy intismeran autogene (V940/mRNA-4157), combined with the checkpoint inhibitor Keytruda (pembrolizumab), met both its primary and key secondary endpoints in the Phase 3 INTerpath-001 trial in resected high-risk melanoma. It is the first time a personalized mRNA cancer vaccine has succeeded in a late-stage trial — and the first Phase 3 win for a treatment whose active ingredient is, in a very real sense, selected by a machine-learning algorithm.

What the trial found

INTerpath-001 enrolled 1,137 patients with completely resected stage IIB–IV cutaneous melanoma — people whose tumors had been surgically removed but who faced a high risk of the disease coming back. Patients were randomized 2:1 to receive intismeran (1 mg every three weeks, up to nine doses) plus pembrolizumab 400 mg, or pembrolizumab alone.

The combination met the primary endpoint of recurrence-free survival (RFS) and the key secondary endpoint of distant metastasis-free survival (DMFS), meaning patients on the vaccine arm went significantly longer without their cancer returning, and longer without it spreading to distant organs, than those on Keytruda alone. Merck and Moderna said the results were statistically significant and clinically meaningful; full hazard-ratio numbers will be presented at an upcoming medical meeting.

Context for what “meaningful” looks like in this program: in the earlier Phase 2b KEYNOTE-942 study, the combination cut the risk of recurrence or death by 49% (HR 0.51) at five years of follow-up — a bar analysts considered remarkably strong for adjuvant melanoma. Moderna shares more than doubled on the Phase 3 news, and the companies said they intend to seek regulatory approval on the INTerpath-001 results.

The algorithm in the manufacturing loop

What makes intismeran unusual isn’t just that it works — it’s how each dose is made. This is not an off-the-shelf drug. For every single patient:

1. Tumor and blood samples are collected and subjected to next-generation sequencing.
2. A proprietary AI/machine-learning algorithm compares the tumor’s mutations against the patient’s normal genome and predicts which mutated protein fragments — neoantigens — the patient’s immune system is most likely to recognize.
3. The algorithm selects up to 34 neoantigens, ranked by predicted immunogenicity.
4. That selection is encoded into a single mRNA construct, manufactured per-patient, and delivered as a vaccine course alongside Keytruda.

Moderna has described this pipeline publicly since 2023: the AI takes sequencing data from tumor and blood, reviews the genetic mutations, and predicts up to 34 neoantigens per patient. The machine-learning layer is what makes per-patient turnaround feasible at all — a human team manually triaging hundreds of candidate mutations for every one of 1,137 trial patients would be logistically impossible. In other words, the Phase 3 result is simultaneously a validation of the immunology and of the computational screen upstream of it.

Why “first Phase 3” matters

Cancer vaccines are one of the oldest dreams in oncology, and one of the most snakebit. Decades of attempts — peptide vaccines, dendritic-cell therapies like Provenge, personalized neoantigen programs from BioNTech and others — produced tantalizing Phase 2 signals that repeatedly failed to replicate at scale. Moderna itself endured years of post-COVID skepticism as its mRNA pipeline moved slowly. INTerpath-001 breaking that pattern changes three things at once:

The platform is validated, not just the product. If algorithm-selected neoantigens plus a checkpoint inhibitor work in melanoma, the same design logic — sequence, predict, manufacture, dose — transfers to any tumor type with enough mutations. Reuters noted the breakthrough “could usher in a wave of cancer vaccines.”

The pipeline behind it is enormous. INTerpath-001 is the first of nine ongoing Phase 2 and Phase 3 trials in the INTerpath program, spanning multiple tumor types and disease stages — including INTerpath-002 in resected non-small-cell lung cancer and INTerpath-009 (680 patients) in earlier-stage lung cancer, plus studies in renal cell carcinoma and beyond. Melanoma is the beachhead; lung cancer is the far bigger prize.

It reframes pharma’s AI race. While the industry’s AI narrative has centered on drug discovery timelines, intismeran demonstrates a different deployment: AI as a per-dose manufacturing decision engine. Every patient who eventually receives this vaccine will have their treatment’s composition chosen by a prediction model. That is a template — and a regulatory precedent — that extends well beyond oncology.

The caveats worth keeping

The announcement is topline only. Specific hazard ratios, confidence intervals, and subgroup analyses haven’t been disclosed yet, and “met endpoints” leaves room for the effect size to be smaller than the Phase 2 49% risk reduction that excited the field. Timeline to approval and the practical challenge of per-patient manufacturing at commercial scale — cost, turnaround time, cold-chain logistics — remain open questions. And melanoma is a immunologically “hot,” mutation-rich tumor; the harder test is whether the same approach works in colder tumors like pancreatic cancer, where a Moderna trial is also underway.

None of which dims the headline. For the first time, a vaccine designed from scratch for one person — with an algorithm choosing the targets — has cleared the Phase 3 bar. As Nature put it, the results suggest cancer vaccines simply work, and could be aimed at other tumors. After thirty years of false starts, that sentence has never been writable before.