insitro, the AI-driven therapeutics company founded by AI pioneer Daphne Koller, has unveiled groundbreaking research demonstrating that artificial intelligence can unlock population-scale genetic analysis of brown adipose tissue (BAT), a tissue long difficult to study due to its diffuse nature and lack of scalable measurement methods. In a first-of-its-kind genome-wide association study (GWAS), insitro used machine learning to derive a BAT imaging phenotype from widely available MRI data in the UK Biobank, analyzing over 69,000 individuals. The approach leveraged Dixon MRI fat-signal fraction maps to calculate the difference between abdominal and supraclavicular fat signals—an AI-derived metric that correlates with BAT content. This innovation overcomes a major bottleneck in human genetics: the reliance on specialized PET scans, which are costly and impractical for large cohorts. By using AI to extract BAT measurements from standard MRIs, insitro enabled the first large-scale GWAS of BAT, identifying genetic loci linked to BAT biology. The phenotype showed expected seasonal variation, with peak activity in late winter—consistent with known BAT behavior—and correlated with metabolic health markers including body composition, lipid profiles, glucose homeostasis, and vascular function. A polygenic risk score for BAT also demonstrated causal links to multiple cardiometabolic traits, underscoring BAT’s role in metabolic regulation. Notably, the GWAS revealed several genes uniquely associated with BAT that had not been found in prior obesity studies, highlighting the discovery power of this AI-driven approach. Building on these genetic insights, insitro’s CellML™ platform was used to functionally validate targets in primary human adipocytes through high-content imaging, transcriptomics, and metabolic assays. This led to the prioritization of BAT-01, a gene whose knockdown via fat-targeted siRNA in diet-induced obese mice resulted in a 15% reduction in body weight and a 25% decrease in fat mass over four weeks—without altering food intake. Crucially, lean mass was preserved, and gene expression changes indicated a shift toward a beige-like, metabolically active adipose state, with increased Ucp1 and reduced Leptin in white adipose tissue. The results suggest BAT-01 modulation promotes fat loss through a peripheral mechanism, distinct from central appetite suppression—a key differentiator from current obesity drugs. “This is the difference between discovery driven by AI and human genetics, and discovery driven by trial and error,” said David Lloyd, Ph.D., Senior Vice President of Metabolic Disease at insitro. The findings point to a new class of obesity therapies that act directly on adipose tissue without affecting appetite. insitro’s approach integrates its ClinML™ and CellML™ platforms into a self-learning loop: AI generates scalable human phenotypes, genetic findings guide target prioritization, and functional validation confirms biological relevance. This industrialized pipeline accelerates drug discovery by reducing reliance on serendipity and increasing confidence in preclinical outcomes. Backed by over $800 million from top-tier investors including a16z, BlackRock, and GV, and with $150 million in revenue from partnerships with Bristol Myers Squibb, Eli Lilly, and Gilead, insitro is building a causal AI engine—Virtual Human™—to decode disease mechanisms and design precision medicines. The company is now expanding its pipeline of BAT-linked targets for obesity and other cardiometabolic diseases. Presented at the Keystone Symposia on Obesity Therapeutics, this research marks a pivotal step in translating AI into tangible therapeutic innovation, demonstrating how machine learning can transform the study of complex human biology and open new pathways for treating metabolic disorders.