A compact motif vocabulary is differentially interpreted across human cell types

CNN-CR learns a compact set of recurrent promoter motif families that retains most predictive information across 116 human biosamples, generalizing beyond discovery contexts.

• Why it matters: Understanding how one genome yields diverse human cell types is unclear if it depends on distinct regulatory vocabularies versus context-dependent reuse of shared motifs.
• What they did: The interpretable CNN-CR model predicts gene expression across 116 human biosamples while learning regulatory motifs; promoter regulation is tested for compressibility and motif conservation/enrichment at fine-mapped variants.
• The result: Chromatin state gates which motif instances are available, while reader identity shifts how accessible motifs are interpreted, supported by ChIP-seq reader occupancy changes and genome-scale perturbations.

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