Iterative improvement of deep learning models using synthetic regulatory genomics.

ABSTRACT: Deep learning models can accurately reconstruct genome-wide epigenetic tracks from the reference genome sequence alone. But it is unclear what predictive power they have on sequence diverging from the reference, such as disease- and trait-associated variants or engineered sequences. Recent work has applied synthetic regulatory genomics to characterized dozens of deletions, inversions, and rearrangements of DNase I hypersensitive sites (DHSs). Here, we use the state-of-the-art model Enformer to predict DNA accessibility and RNA transcription across these engineered sequences when delivered at their endogenous loci. At a high level, we observe a good correlation between accessibility predicted by Enformer and experimental data. But model performance is best for sequences that more resembled

SUBMITTER: Ribeiro-Dos-Santos AM 

PROVIDER: S-EPMC12581950 | biostudies-literature | 2025 Nov

REPOSITORIES: biostudies-literature