A Sydney tech entrepreneur with no medical training used artificial intelligence tools, including AlphaFold protein prediction software, to design a personalized cancer vaccine that shrank his rescue dog's aggressive tumors by roughly 75% in three months. The case represents a striking shift in how scientific discovery happens: AI is making it possible for people outside traditional research institutions to understand and act on complex biological science that was previously inaccessible to them. What Happened to Paul Conyngham's Dog Rosie? Paul Conyngham is a machine learning and data analysis professional with 17 years of experience in tech, but no background in medicine or biology. In 2024, his eight-year-old rescue dog Rosie, a Staffordshire Bull Terrier-Shar Pei cross, was diagnosed with aggressive mast cell cancer. After chemotherapy, surgery, and tens of thousands of dollars in treatment, the tumors kept returning. Veterinarians gave Rosie one to six months to live. Rather than accept that prognosis, Conyngham decided to treat the problem as a data challenge. He paid $3,000 to have Rosie's healthy DNA and tumor DNA sequenced at the University of New South Wales. He then used ChatGPT to understand how scientists design personalized cancer vaccines and map a treatment approach. Most critically, he used AlphaFold, a protein structure prediction tool from Google DeepMind, to identify which of Rosie's tumor mutations were most likely to trigger an immune response. Conyngham brought his half-page formula to UNSW's RNA Institute, where Professor Páll Thordarson synthesized a custom mRNA vaccine in under two months. An mRNA vaccine works by encoding specific proteins from a tumor and delivering those instructions to the immune system, which then learns to recognize and attack cells producing those proteins. This is the same underlying technology platform as the COVID-19 vaccines, applied to one dog's specific tumor profile. Rosie received her first injection in December 2025. By January, she was jumping fences at the dog park. By March, the tumor on her leg had shrunk roughly 75%. UNSW researchers described it as the first personalized cancer vaccine ever designed for a dog. How Does AlphaFold Enable This Kind of Scientific Work? AlphaFold is an AI system created by Google DeepMind that predicts the three-dimensional structure of proteins from their amino acid sequences. Before AlphaFold, scientists had to use slow and expensive laboratory methods like X-ray crystallography or cryo-electron microscopy to determine protein shapes. AlphaFold solved a 50-year-old problem in biology by making accurate predictions in minutes instead of months. AlphaFold 2, which won a major competition called CASP14, achieved accuracy within less than 1 Angstrom error, which is comparable to experimental methods. The latest version, AlphaFold 3 (released in 2024), expanded capabilities even further. It can now predict not just individual protein structures, but also how proteins interact with each other and with other molecules like DNA and RNA. In Conyngham's case, AlphaFold allowed him to predict which tumor mutations would be most immunogenic, or likely to trigger an immune response. This is precisely the kind of structural prediction that would have required years of laboratory work and specialized expertise just a decade ago. The AlphaFold Database now contains over 200 million precomputed protein structure predictions available free to the global public, making this information accessible to anyone with an internet connection. Steps to Access and Use AlphaFold for Protein Structure Prediction - Search the AlphaFold Database: Visit the official AlphaFold Database website and search for your protein of interest by name or sequence. The database contains over 200 million precomputed predictions, saving time and computational resources for most common proteins. - Understand Confidence Metrics: Learn to interpret results using confidence scores like pLDDT (predicted Local Distance Difference Test), where higher scores indicate more reliable predictions. Also review RMSD (Root Mean Square Deviation) and TM-score (Template Modeling score) to judge prediction accuracy. - Run New Predictions if Needed: If your protein isn't in the database, enter the amino acid sequence into AlphaFold's user-friendly interface or API. The system analyzes the sequence and predicts the three-dimensional structure, providing insights into protein function and potential drug targets. - Combine with Laboratory Validation: Use AlphaFold predictions as a starting point for further research, but validate findings through experimental methods. This hybrid approach combines computational speed with biological rigor. What Does This Mean for the Future of Scientific Discovery? For generations, scientific discovery has operated on a consistent model: credentialed researchers inside funded institutions, with access to equipment and peer networks, advance knowledge incrementally and publish it for others to build on. That model has produced extraordinary breakthroughs, but it has also confined the ability to act on cutting-edge science to a small population with the right training and affiliations. AI is beginning to change that model, not by replacing researchers, but by making it possible for people outside institutional structures to understand, navigate, and in some cases contribute to scientific work that was previously inaccessible to them. Conyngham's case is one documented instance of that gap closing, in a field that is difficult, high-stakes, and highly credentialed. Martin Smith, director of the UNSW Ramaciotti Centre for Genomics, posed a provocative question to The Australian: "If we can do this for a dog, why aren't we rolling this out to all humans with cancer?" An oncologist at Anglia Ruskin University offered appropriate caution, noting that this is one dog with no control group, and mast cell tumors can behave unpredictably. The broader implication is significant. Credentials have historically functioned as a proxy for fluency in expert domains. Institutions use them to identify who can engage meaningfully with a field. AI is creating an alternative path to that fluency, one that doesn't require years of training or institutional membership. Medicine is where that shift is becoming visible first. The same dynamic is already under way in law, finance, and policy. The long-term consequence of AI-assisted fluency across expert domains is one of the more significant open questions of this moment. It raises questions about how knowledge gets created, who benefits from it first, and how quickly it moves from research to application.
