The Dark Regulome: Disentangling Predictability from Regulation in Genomic Foundation Models
arXiv:2606.06834v1 Announce Type: new Abstract: High-grade gliomas integrate into neural circuits through functional synapses with neurons, raising the question of which noncoding elements shape synaptogenic gene expression in tumor cells. The regulatory program written across the dark genome, what we call the $\textit{dark regulome}$, is the natural substrate to probe, and sequence foundation models offer a zero-shot route through in-silico mutagenesis (ISM); yet likelihood-based scoring is tautologically coupled to local sequence predictability, leaving the regulatory interpretation underdet
The proliferation of genomic foundation models enables zero-shot routes for biological discovery, increasing the urgency to understand their limitations and underlying mechanisms.
This research addresses a fundamental limitation in interpreting genomic foundation models, directly impacting the discovery and understanding of disease mechanisms, particularly in cancer.
A clearer distinction between predictability and genuine regulatory insight in genomic AI models will lead to more robust and biologically meaningful discoveries, improving drug target identification and therapeutic strategies.
- · Biotech companies
- · Oncologists
- · Pharmaceutical R&D
- · AI in healthcare
- · Researchers relying solely on likelihood-based scoring in ISM
- · Inefficient drug discovery pipelines
- · Undifferentiated AI genomic platforms
Improved understanding of noncoding elements influencing gene expression in diseases like high-grade gliomas.
Accelerated development of targeted therapies for complex diseases by identifying critical regulatory components.
Enhanced ability to engineer genetic code for synthetic biology applications, moving beyond correlation to causation in genomic manipulation.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL