Citations

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 Background Approximately 12 000 adolescents and young adults (ages 15-39, AYAs) are diagnosed with a primary central nervous system (CNS) tumor each year. DNA methylation profiling has transformed CNS tumor classification by refining diagnostic accuracy and identifying biologically distinct subtypes, but its application has not been systematically evaluated in the AYA population. In this study, we examined the spectrum of CNS tumor types across age groups, with a focus on the AYA population. Methods We assembled a large dataset of CNS tumor samples with age annotations and methylation profiles matching with high confidence to a CNS tumor type using the NCI/Bethesda classifier. Prevalence of tumor type, methylation class and DNA copy number aberrations were compared across age strata (pediatric, AYA, and adult). Results The cohort of 21 712 CNS tumors included 5351 tumors from AYAs (25%), which showed a mixed pattern, with tumors typical of childhood (eg, medulloblastoma) as well as those common in older adults (eg, glioblastoma). Several tumor types were specifically enriched in the AYA, including IDH-mutant astrocytoma, pleomorphic xanthoastrocytoma, posterior fossa group B ependymoma, and diffuse hemispheric glioma, H3 G34-mutant, among others. Distinct methylation subclasses of multiple tumor types were observed in the AYA, and patterns of genomic aberrations showed age-specific distributions. Conclusion Large-scale methylation profiling revealed unique classification patterns of CNS tumors in AYAs, with specific tumor types, subclasses, and genomic alterations enriched in this population. These data may serve as a valuable reference resource for better understanding the spectrum of CNS tumors affecting AYA patients.