Making Cancer Vaccines
The Default For Dogs

On August 19th, Moderna and Merck announced that a personalized mRNA cancer vaccine slowed melanoma recurrence in the first randomized Phase 3 trial of its kind. 1,137 patients. Moderna's stock roughly doubled in a day. After a decade of "promising early data," the neoantigen thesis finally cleared the bar that matters.

Five months earlier, an Australian data scientist named Paul Conyngham had used ChatGPT and AlphaFold to design a personalized mRNA vaccine for his own dog, Rosie, who had aggressive mast cell cancer and months to live. It shrank her tumours 75%. That experiment became Gamgee, a Y Combinator Summer 2026 company with a $4M seed led by Founders Fund, now running canine trials in Australia with UNSW and the Garvan Institute.

Read together, those two events settle the two questions that usually kill a company like this before it starts. Does the science work? Yes — a Phase 3 says so. Will anyone fund the dog version? Yes — Founders Fund already did.

I built an open-source pipeline that does the computational half of this — tumour variant file in, ranked canine neoantigens and a synthesis-ready mRNA construct out. That part is solved and nearly free. What follows is my attempt to write down honestly what it would take to do the other half, structured as a retrospective from a future where it worked.

The instinct is to assume genome sequencing is the expensive part. It was, a decade ago. Clinical-grade tumour-normal whole genome sequencing has fallen under $1,500 a pair, with turnaround inside two weeks. Running my pipeline on that data costs somewhere around $15 in cloud compute.

Then you try to turn the output into something you can put in a syringe.

Human personalized neoantigen vaccines currently cost north of $100,000 per patient to manufacture — BioNTech has said publicly it is trying to get below $100K a dose, which tells you where they are now. A single clinical-trial drug product batch generally runs $300K to $1M regardless of how many doses come out of it. And a personalized vaccine is, by construction, a batch of one. You pay the full batch-release and QC overhead to produce a single injection for a single animal.

Which raises an obvious question, and it is the one that reframed this entire thing for me.

Conyngham has said his out-of-pocket cost was a few thousand dollars. Against a six-figure commercial manufacturing bill, that looks impossible. The breakdown explains it:

He never paid a contract manufacturer, because he never used one. The vaccine was produced at a government-funded academic RNA facility, for a compelling one-off research case, at roughly the cost of consumables — and the expertise around it was donated by researchers who found the problem interesting.

For a Toronto-based attempt, the structural analogue is CCRM's Centre for Cell and Vector Production at MaRS — a 20,000 sq ft GMP facility built with UHN, designed precisely to carry academic work toward the clinic, and sitting in the same building cluster as the Princess Margaret Genomics Centre and OICR. One relationship, three stages of the pipeline.

The honest caveat: CCRM's publicly documented GMP capability is cell and gene therapy and viral vectors, not mRNA-LNP specifically. Whether that extends — or whether they would stand it up for a landmark case the way UNSW did — is a phone call, not a search result.

The third one is the one I have to actively resist. Writing more pipeline code feels like progress and requires no one's permission. Recruiting a board-certified oncologist co-founder feels like nothing is happening for months. The second is worth more.

Roughly one in four dogs develops cancer, with risk climbing sharply after age ten. Owners routinely authorise five figures for protocols that buy months. A vaccine at $8–15K sits inside that envelope rather than above it — which is exactly why the strategic problem is manufacturing cost and not willingness to pay.

A working end-to-end pipeline exists today: variant file in, ranked canine neoantigens and a synthesis-ready mRNA construct out, validated on real canine coordinates, deployed publicly, open source. That is the part most people in this space are still pitching on a slide.

What it needs now is two people. A board-certified veterinary oncologist who wants to run the trial that makes this the default treatment. And a manufacturing lead who believes the per-dose cost curve can be broken and has opinions about how.

I am neither of those people. I am a builder who got a pipeline to work and then read enough to understand which half of the problem is still open. If the backcast above is wrong somewhere, I would genuinely rather find out now — the whole point of writing it down is to make the errors visible while they are still cheap.