Data-Driven Prediction and Design of bZIP Coiled-Coil Interactions
Fig 1
Schematic representation of the model-building and design protocol used to generate selective bZIP-interacting peptides.
(A) Experimentally determined dissociation constants for 4,549 bZIP interaction pairs were used as input to train a scoring model. (B) Weights corresponding to contributions from pairs and triplets of residues were fit to experimental binding data using a regression technique. The appropriate optimized weights can be summed to provide a predicted binding energy for two aligned bZIP coiled-coil sequences. (C) Binders were designed by using coiled-coil heptads as building blocks. Optimal combinations of heptads to construct tight-binding and selective designs were identified using integer linear programming in conjunction with the developed scoring function. (D) Designed sequences exhibit tight and selective binding to target bZIP coiled coils. Each square in the cartoon corresponds to a native human bZIP coiled coil, and cells are colored by the strength of interaction of each bZIP with the indicated designed peptide; darker shades correspond to stronger binding. Names of the bZIP proteins corresponding to each cell are given in S5 Table.