Interpreting regulatory mechanisms of Hippo signaling through a deep learning sequence model

How specific cells respond to signaling pathways is largely encoded in the DNA sequence. However, the sequence rules result from complex interactions between signaling and cell type-specific transcription factors and are considered intractable by traditional methods. Here, we leverage interpretable deep learning on high-resolution data and extensive validation experiments to identify the sequence rules for the Hippo pathway in mouse trophoblast stem cells. We show that TEAD4 and YAP1 engage in two types of cooperativity. First, their binding is enhanced by cell type-specific transcription factors, including TFAP2C, in a distance-dependent manner. Second, a strictly-spaced Tead double motif is a canonical Hippo pathway element that mediates strong Tead4 cooperativity through transient protein-protein interactions on DNA. These mechanisms occur genome-wide and allow us to predict how small sequence changes alter the activity of enhancers in vivo. This illustrates the power of interpretable deep learning to decode canonical and cell type-specific sequence rules of signaling pathways. These super series include ChIP-nexus data for TF binding (TEAD4, YAP1, CDX2, TFAP2C, GATA3), H3K27ac Chip-seq, ATAC-seq, RNA-seq, and Nascent RNA captured through the TT-seq method in mouse trophoblast stem cells.