Deep learning design of programmed enhancer pairs [Fiber-Seq]

We extrapolate principles of activation, identifying GC content and nucleosome occupancy as key determinants. Finally, we use the model to create programmed enhancer pairs (PEPs) by editing endogenous loci in embryonic stem cells. We engineer ~3,000-fold induction of sonic hedgehog (Shh) expression, surpassing levels seen in endogenous tissues, by creating a de novo, highly cooperative programmed enhancer pair.

Model guided edits raise GC content and remove nucleosomes, activating expression at an otherwise unexpressed gene. Together, these results provide insight into how regulatory elements function in cis to yield complex patterns and provide a method for engineering de novo gene expression in mammalian cells.