Paul Conyngham's pipeline for designing his dog Rosie's mRNA cancer vaccine started with roughly 300 gigabytes of raw sequencing data - whole genome sequencing of Rosie's tumor and a matched normal blood sample, plus RNA sequencing of the tumor. The final output was half a page: an mRNA vaccine construct encoding seven neoantigen targets. Everything in between was built with AI chatbots.

Conyngham, an AI consultant based in Australia, published a step-by-step account of the process on X, revealing which tools did what and where human expertise remained indispensable, as "Hvylya" reports.

The core bioinformatics pipeline used BWA-MEM for alignment, GATK Mutect2 for variant calling, Ensembl VEP for annotation, and pVACseq with NetMHCpan-4.1 - a neural network predicting whether mutant peptides bind to Rosie's specific immune molecules. "These are the same class of tools used in human precision oncology pipelines," Conyngham wrote. He relied on chatbots to design and implement the pipeline, troubleshooting tool failures, dependency conflicts, and reference genome incompatibilities as they arose. Annotation format mismatches proved especially stubborn - the dog genome is far less documented than the human one.

Each chatbot played a distinct role. ChatGPT handled planning, pipeline design, and the overall bioinformatics workflow. Gemini 2 Pro architected the multi-epitope vaccine construct, incorporating optimized linkers and adjuvants. Grok 3 Thinking provided secondary heuristic refinement of the final design, ensuring structural stability. Conyngham used AlphaFold 2 to model Rosie's mutated c-KIT protein - a step that less than a decade ago would have required expensive lab techniques like X-ray crystallography.

The vaccine was not given alone. The chatbots helped design a multimodal treatment protocol: the mRNA neoantigen vaccine training Rosie's T-cells against seven specific targets, a tyrosine kinase inhibitor targeting the c-KIT mutation, and a PD-1 checkpoint inhibitor removing the brakes from the immune system. Conyngham compared the combined approach to a Star Trek battle: the TKI and PD-1 inhibitor take the cancer's shields down from different angles, letting the mRNA-trained T-cell "torpedoes" get through.

"I do not think everyone should have to know what I know to be able to use these tools," Conyngham wrote. "I think this was far more complex than it needed to be." The entire process also required 120 hours of ethics approval paperwork - a second full-time job on top of his consulting career and the scientific work itself.

Also read how OpenAI set a deadline for its first autonomous AI agent capable of conducting independent research.