Duffy, Patrick Emmet

Abstract Scientific foundation models learn high-dimensional representations from diverse data modalities, yet what they encode and how to extract that knowledge remain open questions. Here we show that probing the internal representations of Evo 2, a 7-billion-parameter genomic foundation model, enables accurate and interpretable genetic variant effect prediction. We introduce a covariance-based probe that captures second-order structure from Evo 2 sequence embeddings to predict variant pathogenicity across variant types and functional consequences, matching or exceeding specialized predictors within their domains. To ground these predictions in known biological mechanisms, we train a complementary panel of probes on existing annotations to detect which genomic properties are disrupted by a variant. This categorized evidence is then integrated with each variant’s genomic context through a language model to generate variant-specific mechanistic hypotheses. Our pathogenicity predictions correlate with experimental measures of variant function, clinical penetrance, and biobank disease associations while the mechanistic hypotheses are consistent with expert reviews, known mechanism classes, and downstream molecular readouts. We release pathogenicity scores, disruption profiles, and contextualized interpretations for 4.2 million variants from the ClinVar database as an open resource through the Evo Variant Effect Explorer (EVEE). More broadly, this structured probing approach offers a general framework for interrogating foundation models across scientific disciplines and grounding their outputs in existing domain concepts.

Abstract Introduction: Most rare disease patients (75 − 50%) undergoing genomic sequencing remain unsolved, often due to lack of information about variants identified. Data review over time can leverage novel information regarding disease-causing variants and genes, increasing this diagnostic yield. However, time and resource constraints have limited reanalysis of genetic data in clinical laboratories setting. We developed RENEW, (REannotation of NEgative WES/WGS) an automated reannotation procedure that uses relevant new information in on-line genetic databases to enable rapid review of genetic findings. Methods We tested RENEW in an unselected cohort of 1,066 undiagnosed cases from the Mayo Clinic Center for Individualized Medicine using new information in ClinVar, HGMD and OMIM between the date of previous analysis/testing and April of 2022. Results 5,741 variants prioritized by RENEW were rapidly reviewed by variant interpretation specialists. Mean analysis time was approximately 20 seconds per variant (32 hours total time). Reviewed cases were classified as: 879 (93.0%) undiagnosed, 63 (6.6%) putatively diagnosed, and 4 (0.4%) definitively diagnosed. Discussion New strategies are needed to enable efficient review of genomic findings in unsolved cases. We report on a fast and practical approach to address this need and improve overall diagnostic success in patient testing through a recurrent reannotation process.

Abstract Motivation: The Biological Reference Repository (BioR) is a toolkit for annotating variants. BioR stores public and user-specific annotation sources in indexed JSON-encoded flat files (catalogs). The BioR toolkit provides the functionality to combine and retrieve annotation from these catalogs via the command-line interface. Several catalogs from commonly used annotation sources and instructions for creating user-specific catalogs are provided. Commands from the toolkit can be combined with other UNIX commands for advanced annotation processing. We also provide instructions for the development of custom annotation pipelines. Availability and implementation: The package is implemented in Java and makes use of external tools written in Java and Perl. The toolkit can be executed on Mac OS X 10.5 and above or any Linux distribution. The BioR application, quickstart, and user guide documents and many biological examples are available at http://bioinformaticstools.mayo.edu . Contact: Kocher.JeanPierre@mayo.edu Supplementary information: Supplementary data are available at Bioinformatics online.