GoodfireAI Releases Open Source Database Of 4.2M Disease-Causing Genetic Variants
Reactions from ranked influencers
6 posts
3 months ago, we used interpretability to predict which of 4.2 million genetic variants cause disease. Now, we've validated several of those predictions with real-world datasets, including a national biobank, clinical data, and RNA sequencing data. 4 examples: (1/6)
We achieved state-of-the-art performance in predicting which of 4.2 million genetic variants cause diseases by interpreting a genomics model, in a new preprint with @MayoClinic. We're now releasing an open source database for all variants in the NIH's clinvar database. 🧵(1/8)
Do EVEE’s predicted pathogenic variants actually show up in human disease? @7uomoki checked them against the FinnGen biobank, and found that they were strongly enriched for disease associations across the Finnish population—an independent validation of EVEE’s predictions. (2/6)
We also found evidence that EVEE’s pathogenicity scores track how often a variant causes disease (clinical penetrance) as part of our collaboration with @MayoClinic. EVEE scores appear to predict the severity of familial hypercholesterolemia (FH) better than other computational predictors. (3/6)
We’ve also further validated EVEE’s mechanistic hypotheses, which predict *how* a variant affects downstream function. RNA-seq data confirms many predicted effects from EVEE's disruption profiles, including specific impacts on splicing at nucleotide-level resolution. (4/6)
Lastly, by combining EVEE’s mechanistic hypotheses with evidence from the literature, we identified 6 variants that may warrant reclassification in ClinVar (the NIH database of clinically interpreted variants). We’ve submitted supporting evidence which is now under review. (5/6)
We’re now deploying EVEE with collaborators for clinical use cases, as well as continuing to validate more of its predictions. Read more on how we built EVEE: https://www.goodfire.ai/research/evee-explaining-genetic-variants#