Predicting single-molecule chromatin distribution from DNA sequence

Mola, a sequence-to-distribution model, predicts single-molecule chromatin accessibility states from DNA sequence and identifies sequence determinants for nucleosome positioning, including YY1 as the strongest +1 factor.

• Why it matters: Existing deep learning approaches predict population averages but miss single-molecule heterogeneity and dynamics that reflect genome regulation mechanisms.
• What they did: The authors built an S2d framework to model full probabilistic distributions of chromatin states from DNA sequence alone, then trained Mola for single-molecule accessibility and analyzed sequence-driven nucleosome regulation and long-context co-accessibility/co-footprinting.
• The result: Mola reveals distinct transcription factor engagement modes, including a nucleosome-trapping factor ZNF146, generates pseudo-dynamic chromatin trajectories, and predicts variant effects at the single-molecule level, including a methylation-mediated bimodal state.

AI summary of the abstract · may contain errors · not medical advice · read the paper.