Abstract
Discovering high-affinity ligands directly from protein structures remains a key challenge in drug discovery. BindCraft is a structure-guided generative modeling platform able to de novo design miniproteins with a high affinity for a large set of targets. While miniproteins are valuable research tools, short peptides offer substantially greater therapeutic potential. However, given their lack of stabilized tertiary structures, de novo generation of functional peptides is a remarkable challenge. Here, we show that BindCraft is able to generate high affinity peptides, solely based on target structure, with remarkable success rates. For the oncoprotein MDM2, BindCraft generated 70 unique peptides; 15 were synthesized, and 7 showed specific binding with nanomolar affinities. Competition assays confirmed site-specific binding for the intended target site. For another oncology target, WDR5, six out of nine candidates bound the MYC binding WBM site with submicromolar affinity. Bindcraft's high fidelity structure prediction enabled one shot peptide optimization via rational chemical modification, improving the potency of one WDR5 binder by 6-fold to a K(D) of 39 nM. BindCraft also generated candidate peptides for targeting PD-1 and PD-L1. However, none of the tested peptides showed detectable binding. Together, these results establish a first evaluation of BindCraft for peptide binder prediction. Despite remaining limitations, this tool shows the potential to rival display technologies in delivering high-affinity ligands for therapeutic development.