ChromBPNet: bias factorized, base-resolution deep learning models of chromatin accessibility reveal cis-regulatory sequence syntax, transcription factor footprints and regulatory variants

ChromBPNet accurately predicts chromatin accessibility and regulatory variants at base resolution, outperforming larger models with a lightweight design across diverse assays and contexts.

• Why it matters: Understanding the sequence rules and genetic variants that control transcription factor binding and chromatin accessibility is crucial for decoding gene regulation and disease mechanisms, yet remains largely unknown.
• What they did: The authors developed ChromBPNet, a deep learning model that learns and deconvolves enzyme biases from regulatory sequences, enabling precise identification of TF motifs, cooperative syntax, and footprints from chromatin accessibility data.
• The result: ChromBPNet effectively predicts variant impacts on chromatin accessibility and TF binding, aiding in prioritizing regulatory variants linked to complex traits and rare diseases, and advancing our understanding of regulatory DNA.