His dog was dying. The vets had given up. So an Australian tech entrepreneur turned to ChatGPT, Gemini, and Grok — and built a custom mRNA cancer vaccine from scratch.
June 2023. Paul Conyngham noticed something wrong with Rosie — swollen lumps on her head and leg. Over the next eleven months, three vet visits. Each time, the vet was dismissive. No biopsy. No diagnosis.
By May 2024, the swelling was severe and bleeding. Conyngham insisted the vet act. Surgery removed the masses. A biopsy was taken.
The result: mast cell cancer. The prognosis: fatal.
The fatal prognosis left Conyngham with a choice. He chose the second option. While running an AI consulting business full-time, he opened ChatGPT and began devouring everything about cancer biology.
Chemotherapy: very expensive. Standard immunotherapy: not effective. He needed a better path. One by one, vets failed him — until he found Dr. Mina Ghaly, the only veterinarian truly receptive to a "citizen scientist."
This was not a 23andMe cheek swab. This was full whole-genome sequencing — tumour DNA and matched healthy blood, plus later RNA sequencing of the cancer itself.
Conyngham drove the tissue samples himself from the vet to the Garvan Institute, where Pavel Bitter's team extracted the DNA. The samples were then couriered to UNSW's Ramaciotti Centre, where Professor Martin Smith ran the sequencing — a process that two decades ago would have cost over a billion dollars.
ChatGPT, Gemini, and Grok were all used to design the bioinformatics pipeline: BWA-MEM for alignment, GATK Mutect2 for variant calling, Ensembl VEP for annotation, pVACseq with NetMHCpan-4.1 — the same class of tools used in human precision oncology.
They found a mutation in the c-KIT gene — a known driver of mast cell cancer. AlphaFold 2 modeled the mutated protein. Now they needed a molecule to switch it off.
Two pathways opened. Pathway A: design a novel ligand using genetic algorithms. It worked in simulation — but would take years to validate in the real world. Rosie didn't have years.
Pathway B: dock a library of over one million existing compounds. After two weeks of computation — eureka. A match. Then the floor dropped out: the compound was patented. Conyngham sought compassionate use. The patent holder declined.
After two weeks of prioritizing quality time with Rosie, the idea came. Not a ligand to block the cancer — a vaccine to teach the immune system to hunt it.
He spent the night debating with ChatGPT. A peptide neoantigen vaccine looked promising. Then Dr. Deborah Burnett at UNSW suggested mRNA instead — faster to develop, more flexible.
The approach was a fundamental shift: rather than obstructing the proteins that fuel cancer growth, an mRNA vaccine would train Rosie's own T-cells to identify and kill the cancer cells themselves.
Gemini 2 Pro architected the multi-epitope construct. Grok 3 refined it for structural stability. Seven neoantigen targets, optimized linkers, codon optimization for canine cells.
UNSW could manufacture the vaccine but had no process for a trial of this kind. Creating one would take until mid-2026. Rosie did not have time.
Navigating ethics approval became a second full-time job. Conyngham turned to chatbots to decode the legalese and opaque regulatory language across Australian states.
Then a breakthrough. Dr. Mari Maeda of the Canine Cancer Alliance — arguably the world's foremost authority on canine cancer — connected him to Professor Rachel Allavena at the University of Queensland. Allavena had an existing compatible trial approval.
Six weeks of expert work by Professor Pall Thordarson's team at UNSW's RNA Institute. The vaccine was ready.
Rosie jumped into the car. Conyngham drove ten hours north to Queensland — only to discover the relevant campus was another four hours beyond Brisbane.
At the University of Queensland's School of Veterinary Science, they were finally ready to proceed: a precision-guided mRNA weapon designed by a pipeline that ChatGPT architected, Gemini implemented, and Grok validated.
But the vaccine couldn't work alone. Cancer doesn't just sit there waiting to be killed — it actively builds a suppressive microenvironment, corrupting immune cells, growing its own blood supply, hiding.
Conyngham's Star Trek analogy: the tumour microenvironment is the ship's shields. Fire all the mRNA-trained T-cells you want — if the shields are up, nothing gets through.
The protocol was trimodal:
The sequencing and timing mattered enormously — you can't give immunosuppressant treatments alongside an immune-activating vaccine. ChatGPT and Gemini helped design the phased rollout across weeks of iteration.
Three weeks in, the cancerous areas swelled. Pseudoprogression — a good sign. It meant the T-cells were swarming.
Six weeks in, two cancerous areas were shrinking. By February, the tumours around Rosie's legs were returning to what appeared normal. Residual flat bumps from presumably dead cancer cells remained — scar tissue.
One growing mass on Rosie's rear didn't respond. It was surgically removed, sent for genomic analysis. Early signs suggested differences in the non-responsive cancer — a different enemy hiding inside the same body.
This was not "upload DNA to ChatGPT, prompt: please make a vaccine with no mistakes." It took months. Three different AI systems, each for different things at different stages.
What they did not do: collect samples, isolate or sequence the DNA, physically manufacture the vaccine, or administer it. Many brilliant scientists were required at every step.
Rosie received a fully individualized, multimodal mRNA cancer protocol. One dog, one vaccine, designed from scratch. Three months in, she is showing strong signs of improvement.
That, Conyngham says, was the easy part.
He has spent the last week speaking to everyone involved to understand whether the process can be made more scalable. They believe it can.
It started with one dog. It will not end with one.
Source → @paul_conyngham on X