Australian ML engineer used ChatGPT, AlphaFold, and genetic algorithms to design a personalized mRNA vaccine for his dog's cancer

Summary

Paul Conyngham, an Australian ML engineer, used ChatGPT, AlphaFold, and genetic algorithms to design a personalized mRNA vaccine for his dog Rosie's cancer after conventional treatments failed.

Rosie's cancer went undiagnosed for 11 months before a vet identified it during a third visit. Surgery removed the tumors, but the cancer had already spread. One tumor had grown so large it wrapped around her leg, leaving insufficient skin to close the surgical site. Chemotherapy did not slow the growth.

Conyngham started with ChatGPT in November 2024, asking how to design a drug to block the cancer. He had no prior cancer expertise. The LLM recommended he obtain Rosie's DNA sequence. Through connections, he reached Professor Martin at a Sydney institution, who provided the sequencing. The process returned 300 gigabytes of data on an external hard drive.

Conyngham built computational pipelines using ChatGPT, Gemini, and Grok to process the genome data, identify mutations, and narrow to a gene called c-KIT as a primary cancer driver. He compared healthy and cancer DNA sequences side by side, like a genetic diff, and modeled the mutant gene in AlphaFold.

He used two techniques to find blocking compounds. Genetic algorithms produced a novel chemical compound, but synthesis would require test-tube validation, mouse models, and further preclinical work—too slow. Molecular docking identified an existing drug that strongly inhibited c-KIT, but it was owned by a major U.S. company that declined his compassionate-use request.

In June 2025, after exhausting small-molecule approaches, Conyngham walked Rosie and realized he was already halfway to a vaccine strategy. ChatGPT confirmed the next steps. He designed the mRNA vaccine construct, summarized in half a page of text, and emailed it to the mRNA institute at UNSW.

Ethics approval became the critical bottleneck. Australia prohibits manufacturing mRNA vaccines outside licensed facilities. The approval process would have taken three months and required modifying the university's government license because the vaccine would be administered off-site. Through a connection in Seattle, he was introduced to a leading canine cancer researcher in Queensland who held an existing ethics approval for novel immunotherapies and offered to sponsor his work.

Conyngham drove Rosie to Queensland for the vaccine induction phase. Visual measurement shows roughly a 50 percent reduction in tumor size. Blood work is pending publication later in 2025.

Conyngham estimates he could now compress the pipeline to four to six weeks, critical because cancer mutates continuously. If the pipeline speed exceeds mutation speed, the approach can clamp down growth. The biggest opportunities to accelerate involve the computational pipeline itself, sequencing (which improves on a double exponential cadence), vaccine manufacturing, and most significantly regulatory approval pathways.

He frames the core constraint as one where technology advancement outpaces legal and institutional adaptation, and sees room to simplify the process more broadly.