The Open-Source Toolkit That's Putting Personalized Cancer Vaccines Within Reach
In June 2026, a 62-year-old melanoma patient who had already failed two lines of therapy received an experimental vaccine. It wasn’t an off-the-shelf drug. It was built from the DNA of their own tumor, encoding 34 custom neoantigens selected by an algorithm trained on thousands of other patients’ cancers. Six months later, scans showed no new lesions.
That vaccine, Moderna’s mRNA-4157 (now called V940), is the most visible example of a quiet revolution in the open-source code repositories where computational biologists build the tools that make personalized cancer immunotherapy possible. One of those tools, pVACtools, just got its biggest update yet.
Background & Context
The immune system patrols the body looking for anything that doesn’t belong. Cancer cells, born from our own tissue, are hard to spot. But every tumor carries mutations: spelling mistakes in DNA that produce abnormal proteins. Some of those proteins get chopped into fragments called peptides and displayed on the cell surface like a distress flag. These are neoantigens, molecular signatures unique to the tumor.
The idea behind neoantigen vaccines is elegant: sequence a patient’s tumor, find the mutations most likely to produce visible flags, and manufacture a custom vaccine that trains the immune system to attack those targets. Between the concept and the injection lies a computational nightmare. A single tumor biopsy can yield tens of thousands of mutations, each generating dozens of candidate peptides. Each peptide must be checked against the patient’s specific HLA type, the molecular lock that determines which peptides can be displayed. Which ones will actually provoke a T-cell response? Which might accidentally trigger autoimmunity? And once you have your shortlist, how do you arrange them into a vaccine that can be manufactured at all?
That last constraint is brutal. Peptide synthesis is expensive and time-consuming. DNA and mRNA vaccines have a hard physical limit: Moderna’s platform maxes out at 34 epitopes, BioNTech’s at 20. Picking the right 20 targets out of thousands is not a luxury. It is the entire problem.
What the Researchers Did
pVACtools was first released in 2016 by the Griffith Lab at Washington University in St. Louis, and it became one of the most widely used neoantigen analysis platforms in the world: 500 citations, 225,000 downloads, deployed in more than 12 clinical trials. Version 6, described in a preprint posted June 25, is a substantial overhaul.
The suite has grown from three tools to seven. pVACseq handles neoantigens from single-letter DNA mutations. pVACfuse catches neoantigens from gene fusions. The new pVACsplice targets a completely different source, errors in RNA splicing. When a cell misreads its own genetic instructions and stitches exons together in the wrong order, it can produce a novel protein sequence the immune system has never seen, a potential target even in tumors with very few DNA mutations.
The update also brings pVACbind for non-canonical neoantigens, pVACcompare for cross-run validation, and a completely rewritten pVACvector algorithm that arranges selected peptides into an optimal multi-epitope vaccine sequence.
| pVACvector Metric | Version 4 | Version 5 | Improvement |
|---|---|---|---|
| Average runtime (IC1500 setting) | 334.6 min | 131.2 min | 61% faster |
| Median runtime (IC1500 setting) | 201.9 min | 97.2 min | 52% faster |
| Average runtime (IC1500_Clip5) | 326.0 min | 134.1 min | 59% faster |
| Core sequence preservation | Missed in some cases | Always preserved | Safer designs |
The new algorithm also preserves the binding core of each peptide, the region that actually interacts with the T-cell receptor. The old version clipped through these in edge cases. The new one navigates around them.
What They Found
The team validated their tools on real patient data. On a cohort of glioblastoma patients, a brain cancer with very few mutations that makes neoantigens hard to find by standard methods, pVACsplice identified 1,264 tumor-specific splice junctions absent from normal tissues. Seven out of 30 patients had at least one predicted strong-binding neoantigen from splicing alone. For a cancer with a median survival of 15 months and few effective therapies, that is a meaningful new target source.
In a small cell lung cancer cohort, the results were even stronger: 42 out of 57 samples had splicing neoantigens, and 34 had strong-binding epitopes, including in key driver genes TP53 and RB1. These are not fringe events. They are common, targetable, and until now invisible to standard pipelines.
The most convincing validation came from a mouse model. The team sequenced tumors, ran pVACseq predictions, then performed mass spectrometry on the actual peptides displayed on the tumor cell surface: a direct biochemical check. Four predicted neoantigens were experimentally confirmed. One, a peptide from the Lama4 gene, is a known immunodominant antigen that has already been shown to induce antitumor immunity as a vaccine. The other three, Vkorc1, Mphosph8, and Zc3h13, are entirely novel. All four were ranked in the top tier by pVACseq’s prioritization system.
Why It Matters
Personalized cancer vaccines have been a tantalizing idea for over a decade, but the path from “we sequenced your tumor” to “here is your vaccine” is choked with computational bottlenecks. Most research groups build ad hoc pipelines that die when the grant ends or the postdoc leaves. pVACtools solves this by putting everything in one open-source, documented, version-controlled suite.
The clinical momentum is real. The KEYNOTE-942 phase 2b trial showed that mRNA-4157 combined with the checkpoint inhibitor pembrolizumab significantly improved recurrence-free survival in high-risk melanoma compared to pembrolizumab alone. Three phase 3 trials, INTerpath-001, INTerpath-002, and INTerpath-009, are now enrolling. Without automated, reliable neoantigen prediction, you cannot manufacture 30 vaccines for 30 different patients in parallel. You cannot scale.
How It Could Change Our Lives
For a patient, this changes the conversation. Instead of “you have Stage III melanoma, here are the approved drugs,” it becomes “we sequenced your tumor. We found targetable mutations. We are building a vaccine for exactly your cancer, ready in about 6 weeks.” That shift, from population-average medicine to individual biology, is the real story.
And pVACtools is free. Any hospital with sequencing capability and a bioinformatics team can use it. The technology is not locked inside a single company’s patent portfolio; it can be adopted, modified, and improved by anyone.
The Bigger Picture
pVACtools v6 arrives as the field transitions from proof-of-concept to production. The individual components (sequencing, prediction, manufacturing) have matured separately. What the field now needs is integration: tools that guide the entire pipeline from biopsy to injection. pVACtools is the closest thing to that in the open-source world. The broader lesson: some of the most important advances in cancer treatment are not drugs or devices. They are software. Open-source, peer-reviewed, reproducible software creates infrastructure no single company could build alone.
Limitations & What’s Next
Neoantigen vaccines are not a cure-all. Tumors evolve: a vaccine that works today may lose its targets as the cancer mutates. Manufacturing timelines of 4-8 weeks are too slow for some patients with aggressive disease. The pVACtools team also points to areas for improvement: better support for MHC class II neoantigens, integration with single-cell data, and better immunogenicity prediction, which remains the field’s hardest problem.
Still, the direction is clear. We are moving toward a future where cancer treatment begins with a computation: a read of your tumor’s genome, a set of predicted targets, and a vaccine built from code.
📄 Source: pVACtools v6: A comprehensive suite for neoantigen prediction, visualization, and therapy design — Hoang, Kiwala, et al. (2026). Preprint on arXiv.