Fig. 3: An interface-focused ZF design model.
From: A universal deep-learning model for zinc finger design enables transcription factor reprogramming
a, The model comprises two modules trained on single-helix B1H selections to predict residues in partially masked helices that bind 4-mer nucleotide sequences. b, The residue embeddings generated from these modules are fed into a third module that learns interhelix compatibility. The full model is trained on two-helix B1H selection data to predict residues in partially masked helix pairs that bind 7-mer nucleotide sequences. In the model architecture schematic, layer normalization is abbreviated to "layer norm." and concatenation is abbreviated to “concat”.
