Not medical advice. The Journal of Biohacking reports published research and teaching models. Nothing on this page is medical advice, a diagnosis, or a recommendation to use any compound, product, or therapy. Where a dose appears, it is a figure reported in a cited study or by its sponsor, not an instruction. Editorial position on dosing.

Personalised vaccines · Cancer and immune tolerance · November 2026

The Dog, the Chatbot, and the Vaccine That Points Both Ways

A research question in two directions. If a vaccine can be built for one animal’s tumour, what do the human trials actually show, and does the mirror image, a vaccine that teaches the immune system to stand down, have any pivotal win?

Grades used below. A, randomised controlled human data. B, early or single-arm human data. C, animal data only. D, a single case, a press release, or a report without a peer-reviewed paper.

Rosie

Rosie was a rescue dog, a Staffordshire bull terrier crossed with a Shar-Pei. In May 2024 she was diagnosed with mast cell cancer. Surgery, chemotherapy and immunotherapy did not stop it. Her vets gave her months.

Her owner, Paul Conyngham, is a Sydney machine-learning engineer. By his own account he had no biology background. He paid about three thousand dollars to have Rosie’s tumour DNA and her healthy DNA sequenced at the Ramaciotti Centre for Genomics at the University of New South Wales. The currency is reported inconsistently across outlets, so the figure stays “about three thousand dollars.” ChatGPT helped him plan the analysis and navigate the literature. DeepMind’s AlphaFold modelled mutated proteins. Conyngham later corrected the viral version of the story: Grok, not ChatGPT, produced the final mRNA construct design. The design targeted seven neoantigens, mutations found in the tumour and not in her healthy tissue.

He did not make the vaccine in a garage. Scientists at UNSW reviewed the data. Professor Páll Thordarson, director of the UNSW RNA Institute, manufactured it in that lab in under two months. Veterinary researchers at the University of Queensland, who held the ethics approvals for experimental canine immunotherapy, gave the first injection at their Gatton campus in December 2025. The University of Queensland’s veterinary newsletter (May 2026) confirms the people, the institutions, and a “substantial but not complete remission.” UNSW’s Ramaciotti Centre published its own feature on 17 March 2026.

The large tumour on Rosie’s leg shrank by roughly half within weeks and by about 75 percent over the following months, in the accounts that rest on the owner and the treating institutions. A tumour on her hindquarters never responded. Rosie entered remission. Later, after a subsequent surgery, the cancer came back. Gamgee’s launch release (17 August 2026) says the cancer returned and does not state a final outcome. Conyngham has said he is under no illusion that this is a cure.

Grade for the Rosie case: D. One dog. No comparison group. Other drugs in her system. A tumour type that can behave unpredictably. No peer-reviewed paper, and details such as the tumour grade were not published.

What the case shows, at that grade and no higher, is a logistics result. A personalised vaccine pipeline, from sequence to syringe, ran in months for one animal, outside a pharmaceutical company, with university manufacturing and an ethics approval. The institutions were the bottleneck. That is not evidence the vaccine caused the shrinkage.

The human version is already an industry

In October 2026 Moderna and Merck reported detailed Phase 3 results for intismeran autogene, a personalised mRNA vaccine encoding up to 34 neoantigens from a patient’s own tumour, given with pembrolizumab in resected melanoma. The sponsor-reported schedule in that programme is 1 mg of the vaccine every 3 weeks for up to 9 doses, together with the checkpoint drug. That schedule is the trial regimen as the companies described it. It is not a dose for any other use.

Grade: A. Randomised, double-blind, 1,137 patients in INTerpath-001.

Recurrence-free survival hazard ratio 0.58. Two-year recurrence-free survival 82.2 percent against 70.0 percent. Overall survival was not significant (hazard ratio 0.77). The curves barely separated during the first six months, then separated after. The vaccine took six to eight weeks to manufacture. The earlier randomised Phase 2, KEYNOTE-942, Weber et al., Lancet 2024, had pointed the same way in 157 patients. A five-year update held the recurrence signal; overall survival was still immature. The product is not approved. No filing had been confirmed in primary sources as of 6 November 2026.

BioNTech’s autogene cevumeran produced the starkest single-arm result in pancreatic cancer. Of 16 vaccinated patients, half mounted strong T-cell responses, and at more than three years those responders had not reached median recurrence, against 13.4 months for non-responders (Rojas et al., Nature 2023; Sethna et al., Nature 2025).

Grade: B. Tiny, uncontrolled, and “responder” is defined after treatment, so the fitter immune systems may be the ones that both respond and stay well. A randomised Phase 2 is running.

The same product missed a randomised trial in colorectal cancer and was stopped in 2026. A shared KRAS vaccine from Elicio was immunogenic in 98.9 percent of patients and still missed its randomised endpoint. Grade A, negative. Gritstone went bankrupt in 2024 before its data matured. Being able to manufacture a bespoke vaccine is not the same as changing survival. Indication, partner drugs and disease setting decide the result.

Run the film backwards

The question on the other side of the same technology: if a vaccine can teach the immune system what to attack, can one teach it what to leave alone?

In 2021 researchers at BioNTech and TRON in Mainz, including Uğur Şahin, Özlem Türeci and Katalin Karikó, published a mouse study in Science (Krienke et al.). The mRNA used a modified nucleoside that removes the inflammatory alarm, and it was given without an adjuvant. In mice with the MS-like disease EAE, it stopped progression and expanded regulatory T cells specific to a myelin antigen. Those cells also suppressed responses to other myelin antigens nearby, while responses to unrelated targets stayed intact.

Grade: C. Animal data only. As of late 2026, no reported trial had given this vaccine to a person. A claim that mRNA has cured MS in people is not supported by this paper.

The nanoparticle version is further along in humans, and it has not won a pivotal trial either. Cour Pharmaceuticals’ gliadin nanoparticle suppressed the immune response to gluten by about 90 percent on a validated biomarker in a randomised gluten-challenge study (Kelly et al., Gastroenterology 2021). Grade A for that biomarker. In July 2026 Takeda discontinued the programme without publishing the larger trial’s data. Cour’s primary biliary cholangitis product, CNP-104, showed controlled Phase 2 signals after two doses, in a small trial whose full paper was not yet out. Anokion reported positive Phase 2 symptom data for a liver-targeted coeliac product in 2025, still largely a press release. Selecta’s rapamycin nanoparticle tolerised people to a therapeutic enzyme at Phase 3 scale in gout; the FDA’s 2026 rejection was about manufacturing paperwork, not about the patients. That product was never tested against a self-antigen.

The failures belong in the same paragraph. Diamyd’s single-antigen vaccine for type 1 diabetes missed a Phase 3 interim analysis and was stopped for futility in April 2026. A large randomised prevention trial of oral insulin did not prevent disease. The largest tolerance trial in MS missed its endpoint in 2008. The pattern, as far as these trials go: antigen-specific tolerance has been safe wherever it has been tried, and it has not been proven effective in a pivotal trial.

What a tolerance vaccine for rheumatoid arthritis would still have to be

Rheumatoid arthritis is the disease this journal keeps returning to, so the contrast is concrete. Cancer personalisation starts from private mutations. Rheumatoid arthritis targets are mostly public: citrullinated proteins, presented by HLA molecules that many antibody-positive patients share. An Australian product, DEN-181, already tested a fixed citrullinated peptide in genetically selected patients. It was safe and it moved immune markers in a randomised Phase 1 of 17 people. It has no Phase 2.

Grade: B, and stalled.

A personalised version would not be a search for secret antigens. It would be a ranking of known ones: HLA type, antibody fine specificities, and which citrullinated epitopes that person’s T cells are responding to, chosen by an algorithm and encoded in non-inflammatory mRNA or loaded into a tolerance nanoparticle. Researchers discuss doing this early, before joint damage and epitope spreading. That is a hypothesis about trial design. It is not a product, and it is not a regimen.

What exists today is every component, somewhere, in someone else’s trial. What does not exist is the assembled product, a biomarker that can certify tolerance, or a regulatory template that has approved one. The oncology path that has actually worked is unglamorous: one protocol, many bespoke batches, biomarker endpoints, years of follow-up, and a control group. A single dog cannot measure that.

The point

Rosie’s story, graded honestly, is about the cost and time of going from one individual’s sequence to one manufactured batch falling far enough that a determined non-biologist could reach two universities. The human melanoma trial says a related collapse is real in oncology, slower and regulated, and still short of an approved drug or a significant survival result. The autoimmune mirror is safe in the trials that measured safety, mechanistically interesting, backed at its mRNA frontier by one mouse paper, and still waiting for a pivotal win.

Sources

University of Queensland School of Veterinary Science newsletter, May 2026, and UNSW Ramaciotti Centre feature, 17 March 2026. Institutional accounts of who treated Rosie and what remission was claimed.
Gamgee launch release via GlobeNewswire, 17 August 2026, on recurrence after a later surgery. No peer-reviewed case report existed as of November 2026.
Conyngham’s own public correction on which model produced the construct design, as reported by Reason and other outlets in 2026. The case details beyond the university newsletters rest on the owner’s account.
INTerpath-001, Moderna and Merck, ESMO 2026. KEYNOTE-942: Weber et al., Lancet 2024. The 1 mg schedule is the sponsor-reported trial regimen.
Rojas et al., Nature 2023, and Sethna et al., Nature 2025, on autogene cevumeran in pancreatic cancer. The colorectal randomised trial was stopped in 2026.
Krienke et al., Science 2021, on non-inflammatory mRNA in the mouse MS model. Kelly et al., Gastroenterology 2021, on the gliadin nanoparticle. Diamyd Medical, DIAGNODE-3 discontinuation, 2026.

Research summary only. Nothing on this page is medical advice or a treatment recommendation.