Daniel McKinnon describes how the loss of his first son to an undiagnosed rare genetic disease motivated him to apply his AI background to genome interpretation. Years later, he says an agentic system reanalyzed the family's raw data and identified the diagnosis that an earlier neonatal whole-genome laboratory report had missed.
Daniel McKinnon explains that genome interpretation combines large sequencing datasets, variant filtering, scientific literature and clinical reasoning. He argues that this is well suited to AI agents because the search is complex but many intermediate claims can be checked against structured evidence.
Daniel McKinnon says his team's system uses an ensemble of frontier models, custom tools, scaffolding and evaluations to support physicians, hospitals and families with difficult cases. He reports examples in which the system surfaced an uncertain variant, accelerated contact with a relevant researcher and found a molecular cause after prior expert review, but this transcript-only review did not independently verify those medical outcomes.
Daniel McKinnon outlines a longer-term plan to pair computational diagnosis with laboratory experiments using CRISPR-edited cell lines. The aim is to test whether uncertain variants cause disease, feed verified findings back into the models and give patients clearer diagnostic and treatment paths.
Watch on YouTube
