What if the reason ~95% of cancer drugs fail in the clinic isn't the drug at all – but the patient they were given to? In this episode of Catalio Conversations, Martha Petrocheilos sits with Ron Alfa, Co-Founder & CEO of Noetik, who reveals a radically contrarian thesis: the industry is great at making drugs, but has never cracked how to match the right drug to the right patient, and that fixing this "translation gap" could rescue therapies once written off as failures. Ron's journey – physician-scientist with an MD/PhD from Stanford, a master's in the history of medicine, and six years at Recursion from seed stage to IPO as Head of Research and acting CSO – showcases turning a career at the frontier of tech-bio into a company built to solve oncology's hardest problem. From Noetik's conviction that the best model system for cancer is the human patient, to its industrial-scale data engine and Perturb-Map platform, to Octo, its "virtual cell" foundation model that integrates a tumor's DNA, proteins, gene activity, and tissue imaging into a single system, this episode pulls back the curtain on what genuinely AI-powered drug discovery looks like. And with GSK licensing Noetik's models for $50M upfront – a shift from selling services to licensing AI as infrastructure, an entirely new asset class – Noetik isn't just promising the future of oncology; they're proving pharma will pay for it. Perfect for healthcare leaders, AI enthusiasts, and anyone curious about the next frontier of cancer treatment, this episode reveals the paradigm shift redefining how cancer drugs are discovered – directly from human tumor biology, and matched to the patients most likely to benefit. Be here for the rare inside look at how AI could finally let oncology let go of the guesswork that's dominated it for far too long – because the next breakthrough cancer drug might already exist, waiting for the right patient to be found. Timestamped Summary: 00:04 - Introduction to Ron Alpha and the mission of Noetic 01:09 - The contrarian thesis: ~95% of cancer drugs fail on the wrong patients, not because they don't work 01:53 - Why enriching the right patient populations turns more therapeutics into successes 02:10 - How the definition of a tumor evolved: tissue, pathology, genomics, and now AI 02:31 - Six years at Recursion: what Ron carried into Noetic and what he left behind 03:42 - Making it concrete: what Noetic does with a tumor sample (the Genus/ASCO work) 04:27 - Measurement versus prediction, and selecting the patients most likely to benefit 04:55 - Why the best model system for cancer is the patient – not a mouse or a dish 06:13 - Building models from human tissue, with architectures that run "what if" simulations 06:49 - Perturb Map: manufacturing data at industrial scale by switching genes on and off 08:52 - Octo, the foundation model uniting protein, DNA, gene expression, and tissue imaging 09:06 - What it means for a model to truly understand a tumor, and how you know it's right 10:39 - The GSK deal: $50M to license Octo, and AI models as a new asset class 12:29 - Earning trust with a cautious, "show me the data" pharma industry 13:38 - Beyond the same dozen targets: surfacing hundreds more, from tumor to approved drug 15:17 - Myths and bets: more success over speed, and drugs discovered from human data 16:00 - Closing remarks
