ChAI-Net learns a regulatory code for chromatin co-accessibility

ChAI-Net is a deep learning model that has discovered a regulatory code governing the co-accessibility of chromatin in different regulatory elements of mouse embryonic stem cells. Chromatin accessibility is vital for cell function and identity. Single-molecule long-read footprinting technology has allowed for near-nucleotide-resolution mapping of chromatin fiber organization, revealing coordination of transcription factor binding and nucleosome positioning that was previously hidden in bulk methods.

ChAI-Net combines this footprinting data with machine learning to study how protein-DNA interactions are coordinated within different types of cis-regulatory elements. The model quantifies the frequency of simultaneous accessible positions on individual molecules and learns to predict locus-specific chromatin co-accessibility maps from DNA sequence.

By interpreting the model and running in silico experiments, the researchers found that ChAI-Net learns distinct combinations of sequence motifs associated with co-accessibility within enhancers, promoters, and CTCF-bound sites. Many of these motifs are not captured when modeling average accessibility, indicating their specific role in coordinated binding.

ChAI-Net's regulatory grammar also emphasizes the importance of transcription factor motif orientation, spacing, positioning, and co-occurrence in driving co-accessibility. This framework for extracting regulatory rules from footprinting data enhances understanding of transcription factor cooperativity and paves the way to uncover how these rules vary across cellular contexts as single-molecule genomics is applied at scale.