Researchers have developed a new diagnostic method to distinguish between sequence predictability and actual regulatory function in genomic foundation models. This approach was applied to three distinct models, Caduceus-Ph, HyenaDNA, and Enformer, analyzing over 30,000 dark genome elements. The findings indicate a consistent 10kb proximal-regulatory horizon but reveal that language model-derived hierarchies do not reliably predict regulatory impact, with a simpler linear model achieving comparable results. AI
IMPACT Provides a methodological tool to better interpret genomic foundation models, potentially improving their application in biological research.
RANK_REASON The cluster contains an academic paper detailing a new methodology and findings in the field of genomic foundation models.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →