Case: Rosie, a non-biologist's AI-assisted cancer vaccine for his dog

The Sid Sijbrandij case followed one person investigating his own disease through data. This is the story of one dog and her owner. Paul Conyngham, a Sydney-based developer with 17 years of experience in data science and machine learning but no biology background, used AI tools to narrow down target candidates and worked with researchers at UNSW to design what was reported as the first personalized mRNA cancer vaccine for a pet dog. The patient was his dying dog, Rosie.

“In the truest sense of the word: Rose was a pioneer. It is my hope that her contribution will lead to many people never having to face what I had to go through: both humans and pets alike.” (Paul Conyngham)

This case follows the same broad path as Sid’s: data → target → treatment. What makes it different is that a non-biologist used AI to carry much of the analysis forward, and that the story has an honest ending that includes both a meaningful response and clear limitations.

1. What happened: Rosie’s diagnosis

In 2024, Rosie was diagnosed with a mast cell tumour¹, a common skin cancer in dogs. Chemotherapy and surgery did not shrink the tumours, and they kept returning. Standard treatment options had reached a dead end.

Paul was neither a physician nor a biologist. He did, however, know how to work with data and AI tools. He decided to approach the disease as a developer might approach a difficult bug.

2. What he did, in chronological order

Stage	What happened
Planning	Used ChatGPT to plan an approach, which led him toward immunotherapy and the UNSW genomics centre
Sequencing	Paid about $3,000 to sequence Rosie’s normal DNA and tumour DNA
Target discovery	Used AlphaFold2 to model the structures of mutated proteins and narrow down potential neoantigen3 targets that the immune system might recognize
Collaboration and design	Took the target list to the UNSW RNA Institute, which used the data to help design and manufacture a personalized mRNA vaccine4 in less than two months
Regulation	Spent two hours every night for three months preparing a 100-page animal ethics application
Treatment	Drove for ten hours for Rosie’s first injection in December 2025, followed by booster injections

A substantial part of the analysis was driven by one person using AI. Paul moved from sequencing data to mutation analysis and target exploration before the specialist institutions took on the work of review, vaccine production, and treatment.

3. Why his request for collaboration was taken seriously

Paul did not approach a university laboratory with only a request for help. He arrived with data and a large part of the preliminary work already done.

• He brought a completed preliminary analysis: rather than making a general appeal, he arrived with sequencing results and a target list narrowed down with AlphaFold. The researchers could begin by reviewing the work and designing the mRNA construct.
• He carried the cost and administrative burden: he paid for the sequencing and worked through the 100-page ethics application himself.
• He had relevant technical credibility: he did not know biology, but he understood data and AI well enough to work carefully with the results and communicate with specialists.
• The veterinary setting offered a possible path: this was not a human clinical trial. A research institution could explore a rapid personalized design through an animal ethics process with veterinary involvement.

4. The result, and the honest ending

By March 2026, the tennis ball-sized tumour on Rosie’s leg had shrunk by about 75%, and most of her tumours had become smaller. She entered remission⁵. Six weeks after treatment, she had recovered enough to jump a fence and chase a rabbit.

The story did not end there. After a later operation, the cancer came out of remission and tumours spread rapidly across her body. In mid-2026, as Rosie was suffering, Paul made what he described as the hardest decision of his life and had her put to sleep.

After remission

The substantial tumour shrinkage after vaccination and Rosie’s remission were meaningful results. The cancer later returned, and a single case cannot isolate how much of the response came from the vaccine. The significance of this case is not that it proved a cure, but that one person used AI to narrow down candidates and worked with researchers and veterinarians to carry the idea through to an actual treatment.

5. The regulatory barrier: Australia and South Korea

Paul described administration, or red tape, as the hardest part of the project. The exact barrier differs by country.

Stage	Australia in this case	In South Korea
Sequencing and target discovery	An individual paid for the work	Animal samples are outside the human Bioethics and Safety Act, so personal analysis with institutional services is comparatively accessible
Experimental treatment in an animal	Required university animal ethics approval and a 100-page application	The Animal Protection Act and Laboratory Animal Act generally require a registered facility and prior IACUC6 review
mRNA vaccine manufacture and administration	Manufactured by a registered research institution, UNSW	Veterinary biological products7 are subject to product and manufacturing authorization, so unapproved manufacture and administration are generally prohibited
Handling an mRNA construct	Managed through institutional biosafety procedures	May fall under the Living Modified Organisms framework and institutional biosafety review8
Diagnosis and administration	Involved veterinarians	Treatment is within the licensed scope of a veterinarian

The practical boundary is similar in both countries: once a project moves from analysis to manufacture and administration, qualified professionals, registered institutions, and regulatory review become essential. South Korea presents additional barriers around registered facilities and unapproved veterinary biological products, making it particularly difficult for an individual to carry the entire process alone.

6. Difficulties that apply beyond this case

Rosie’s story offers several practical lessons for anyone studying personal genomics.

• Regulation and cost become critical constraints: ethics applications and authorization can take longer than the analysis. Personalized cancer vaccines for humans have been estimated at around $100,000 per patient, although the cost depends heavily on the setting and process.
• A single case leaves attribution uncertain: even when tumours shrink, it can be difficult to separate a vaccine effect from spontaneous remission or the effects of other treatments. This is the same analytical problem that appears whenever there is no suitable comparison group.

7. What followed: Gamgee

Paul turned this experience into a company. Gamgee⁹ is developing personalized mRNA cancer vaccines for dogs. Its first product, ze-ets-001, is presented as an end-to-end personalized cancer treatment protocol for dogs. The company manages the process from application review and tissue collection through genomic analysis, vaccine design, treatment, and monitoring. Gamgee says it is running a trial in Australia, working with UNSW and the University of Queensland, and is backed by investors including Y Combinator and Founders Fund.

Gamgee describes Rosie’s story as one treatment adding two years to her life. That account should be read alongside the full outcome: Rosie later came out of remission after surgery and was put to sleep when the cancer spread. In a single case involving multiple interventions, it is not possible to calculate precisely how much survival time one component added.

The company is still an important part of the story. The same general pipeline taught in this handbook, a non-biologist using AI to move from tumour data toward potential targets, became the starting point for a funded company attempting to make the process available to other dogs.

8. What we can learn from Paul

The way Paul worked is at least as instructive as the vaccine itself.

• Ownership: he did not stop when he was told there were no more standard options. He made the problem his responsibility, much like Sid’s statement that saving his life had become his job.
• Translate the problem into the skills you have: he did not let the absence of a biology degree end the attempt. He reframed the problem in terms of data, AI, and debugging.
• Bring work, not only a request: sequencing results and a preliminary target list lowered the cost for a research group to engage with the project.
• Do not avoid the tedious critical path: he spent months on ethics paperwork rather than focusing only on the interesting AI analysis.
• Divide the work clearly: Paul led the analysis, while UNSW and veterinarians handled review, manufacture, and administration.
• Share the full outcome: he disclosed not only the initial response but also Rosie’s death, calling her a pioneer whose contribution might help both people and pets.

In one line: take ownership → translate the problem into skills you have → do enough work to make collaboration possible → complete the tedious parts → share what happened honestly.

9. Strategy: why target more than one thing?

Both Sid and Paul avoided relying on a single target, but they did so at different levels.

	Sid	Paul and Rosie
Approach	Pursued several different treatment strategies in parallel, including radioligand therapy, a neoantigen vaccine, TCR therapy, and MDM2	Included several neoantigens in one mRNA vaccine, then prepared to iterate with a second vaccine
Structure	A portfolio of relatively independent bets	A broad first construct followed by sequential iteration if necessary
Main constraint	A founder’s capital, team, and network of collaborators	One dog, one owner, and a limited budget

Paul’s approach was closer to sequential iteration than to a fully parallel portfolio. The difference largely reflected available resources. Even so, both cases point toward the value of avoiding dependence on a single target.

• Avoid a single point of failure: cancer is heterogeneous and evolves, so resistance can defeat a single target. One of Rosie’s tumours did not respond, while a B7-H3 target considered in Sid’s case was rejected because it was also expressed in normal liver tissue.
• The best target is uncertain in an individual case: several independent targets increase the chance that at least one will be useful and create opportunities to learn sooner.
• Time is a critical resource: in advanced disease, trying one option at a time may take longer than the patient has.

Parallel strategies are not free

Adding targets increases cost and expands the surface for safety risks. It can also make attribution harder: when several things are tried together, it becomes difficult to determine which component produced a response. The principle is not to parallelize everything. When the cost of a single failure is high and resources permit, independent targets can be hedged in parallel, provided each target is checked carefully and the limits of causal attribution are acknowledged.

10. How this connects to HomeGenomics

Sid was a person and Rosie was a dog, but both stories follow the same broad path:

Data (sequencing) → targets (variants and neoantigens) → treatment (personalized design).

Rosie’s case adds another point: a non-biologist can now use AI to explore part of the path from data to potential targets. In the Rosie proof-of-concept exercise, you filter variants using synthetic tumour and normal DNA plus RNA evidence, then generate mutant peptide candidates.

---

Sources: UNSW News · Fortune, March 2026 · The Conversation: an oncologist’s caution · Gamgee · public updates from Paul Conyngham on LinkedIn.

Footnotes

1. A mast cell tumour is a common skin cancer in dogs. Mast cells are immune cells involved in allergic and inflammatory responses. The disease can behave very differently between dogs, and some tumours can regress spontaneously, which makes treatment effects difficult to interpret. ↩

2. AlphaFold is an AI system developed by Google DeepMind that predicts a protein’s three-dimensional structure from its amino acid sequence. Paul used protein modelling while exploring how Rosie’s tumour mutations might alter potential targets. ↩

3. A neoantigen is a protein fragment created by a mutation found in cancer cells but not normal cells. If the immune system can recognize it, the fragment can serve as a target for a personalized cancer vaccine. ↩

4. An mRNA vaccine delivers temporary genetic instructions that cause cells to produce selected protein fragments so the immune system can learn to recognize them. The platform became widely known through COVID-19 vaccines and can also be customized around mutations found in an individual tumour. ↩

5. Remission means that the signs and symptoms of cancer have decreased or disappeared. It is not the same as a cure: cancer may remain and later return. Rosie entered remission before the disease progressed again. ↩

6. An Institutional Animal Care and Use Committee reviews proposed animal research before it begins. In South Korea, registered animal research facilities generally require prior review under the Animal Protection Act and the Laboratory Animal Act. ↩

7. A biological product is a medicine such as a vaccine or antibody that comes from biological material or biotechnology. Veterinary biological products in South Korea are regulated through product and manufacturing authorization requirements. ↩

8. LMO stands for living modified organism. Research involving genetically engineered organisms or related constructs may require review under biosafety rules and by an institutional biosafety committee, depending on the material and procedure. ↩

9. Gamgee is a startup developing personalized cancer treatment protocols for dogs, based on the process that grew out of Rosie’s case. ↩