Dog owner uses ChatGPT and AlphaFold to design mRNA cancer vaccine
TL;DR: Paul Conyngham, a Sydney data scientist, used ChatGPT as a research assistant and AlphaFold for protein modelling to design a personalised mRNA vaccine for his rescue dog’s incurable cancer. The tumour shrank by 75% within a month. Scientists at UNSW and the University of Queensland produced and administered the vaccine, calling it the first personalised cancer vaccine designed for a dog. The case is a single anecdotal result, not clinical evidence, but the speed of the AI-assisted pipeline is what makes it noteworthy.
When conventional veterinary treatment failed to stop the mast cell tumours growing on his rescue dog Rosie’s leg, Conyngham turned to the tools from his professional life. A 17-year veteran of machine learning and data science, he used ChatGPT to plan a pipeline: sequence the tumour’s DNA, compare it to healthy cells, identify the mutations driving the cancer, and design a vaccine targeting those specific neoantigens.
AlphaFold, Google DeepMind’s protein structure prediction tool, modelled the 3D structures of proteins from Rosie’s tumour mutations. Conyngham’s own machine learning algorithms then selected which neoantigens would most likely trigger an immune response. The output was a half-page formula describing the mRNA sequence.
Real institutions, real results
This was not a DIY experiment. UNSW’s RNA Institute designed and produced the bespoke vaccine. The University of Queensland, which held the ethical approvals for experimental veterinary treatments, administered the injections in December 2025. Within roughly a month, the targeted tumour had shrunk by 75%.
UNSW computational biologist Martin Smith reacted bluntly: “It was like holy crap, it worked!” Several UNSW academics noted the significance of a non-academic executing a pipeline that would typically require a specialised research team.
The important caveats
One tumour on one dog responded to one vaccine. A second tumour on Rosie did not shrink, and Conyngham is already designing a follow-up treatment for it. There are no controlled trials, no sample sizes beyond one, and no long-term data. The lighter regulatory requirements for veterinary experimental treatments made the rapid timeline possible, but scaling this to standardised treatment for either animals or humans would require years of clinical work.
The headline “AI creates cancer vaccine” also overstates what happened. AI accelerated the research and modelling stages, but human expertise from immunologists, RNA engineers, and veterinary oncologists was essential at every decision point.
Looking forward
What this case demonstrates is not that AI can cure cancer, but that AI-assisted pipelines can compress months of literature review and computational modelling into weeks. For UK biotech firms and veterinary researchers, the speed and accessibility of the approach is the element worth watching. The tools Conyngham used, ChatGPT and AlphaFold, are publicly available. The bottleneck is no longer the computation; it is the institutional infrastructure to validate and scale the results.