We meet Daniel McKinnon (@danielmckinn0n), CEO of Gamow Labs, who is building AI systems to reanalyze unresolved rare-disease genomes—and using a rescued laboratory to generate new biological evidence. We spend much of the conversation at the VUS boundary: a biology study can add ACMG points to a patient's case while also teaching the machines what that region of the genome does. We trace the personal origin of the work through his son Owen: McKinnon's own reanalysis found what a prenatal sequencing lab had missed. We then follow Owen's missed diagnosis to the workflow level: a disease-causing deletion fell outside a clinical lab's search cutoff, while an AI agent kept looking. The conversation is worth our time because it gets concrete about the AI layer: a harness, tools, and trace analysis are used to catch failures as basic as changing a gene's name. We also visit the lab to examine the proposed hill climb from $50,000 per functional study toward a 50-cent target, with a robot arm moving plates through high-throughput experiments. That target is not an achieved cost; it is the engineering question on the table. We learned a ton. Enjoy! TIMESTAMPS 0:00 Can a rescued lab teach AI? 3:09 Why AI genetics needs biological experiments 6:10 A failed drug program unlocked Gamow's lab 7:51 Owen's missed diagnosis became a life's mission 10:21 AI agents can search beyond human cutoffs 13:14 Sequencing got cheap; interpretation stayed the bottleneck 15:20 Why Gamow chose diagnostics over screening 18:41 Reanalysis in a week for ten dollars a genome 20:28 What Gamow builds on top of frontier models 23:10 Stronger models can break existing tools 25:45 Can robots make genetic experiments affordable? 27:08 Can biological experiments teach genetic AI?

Rare-disease diagnostics tend to be slow and expensive. They don't need to be. Within an hour of seeing @cgeorgiaw's @huggingface challenge to diagnose a real child living with Mosaic Variegated Aneuploidy (MVA), @Gamowlabs submitted a correct molecular diagnosis identified by George, our clinical genomics agent. Every child should have access to SOTA clinical genomics. This is just one demonstration of how this is now possible with AI. Now onto phase 2 to try to find a treatment!

Rare-disease diagnostics tend to be slow and expensive. They don't need to be. Within an hour of seeing @cgeorgiaw's @huggingface challenge to diagnose a real child living with Mosaic Variegated Aneuploidy (MVA), @Gamowlabs submitted a correct molecular diagnosis identified by George, our clinical genomics agent. Every child should have access to SOTA clinical genomics. This is just one demonstration of how this is now possible with AI. Now onto phase 2 to try to find a treatment!

Grok