Gene Regulatory Networks: the Prior Can Matter More Than the Downstream Model

bravo_abad · x · 2026-09-15

Gibbs et al. fine-tune a 250M-parameter genomic language model (pretrained on 850 species) to predict plausible transcription factor–gene regulation from DNA sequence, using it as a prior for reconstructing cell-specific regulatory networks from single-cell expression data. Strikingly, changing this prior map can matter more than swapping the downstream model.

Original post →

More from Research

Research channel →