Machine and Deep Learning Reveal Sequence Determinants Encoding Bivalent Histone Modifications

Bivalent histone modifications, marked by the coexistence of activating and repressive histone marks, define a distinctive chromatin state with key roles in developmental gene regulation. However, the specific sequence features that distinguish bivalent chromatin regions remain unclear. Here we show that genome-wide profiling of H3K4me3, H3K27me3, and H3K9me3 in mouse embryonic stem cells revealed that bivalent domains have higher GC content and stronger evolutionary conservation than monovalent regions. Genes marked by bivalency were enriched in developmental signaling pathways, including Hippo, MAPK, and TGF-β. Using machine learning models trained on k-mer sequence features, we accurately distinguished bivalent from monovalent regions. Feature analysis identified informative motifs such as TCTGAA and TCACAG, associated with pluripotency transcription factors including OCT4, SOX2, ESRRB, and TCFCP2l1. Deep learning models further improved predictive accuracy and uncovered motifs enriched at the boundaries of bivalent peaks, suggesting positional specificity. These findings reveal that bivalent chromatin states are encoded by distinct sequence features. Machine learning and deep learning reveal that bivalent histone modifications in mouse embryonic stem cells is encoded evolutionarily conserved DNA sequences and motifs associated with transcription factors that regulate developmental genes