Fig. 1: Design of β-strand pairing binders.
From: Improved protein binder design using β-pairing targeted RFdiffusion
a Representation of β-strand interface conditioning information provided as a tensor to RFdiffusion to generate β-strand pairing binders. b Example RFdiffusion denoising trajectories without (top) and with (bottom) β-strand interface conditioning. Conditioning indicates that part of the binder scaffold should be a β-strand (gold) that contacts the indicated target edge strand (cyan). This information influences the denoising in very early trajectory timesteps (t), with the tertiary fold determined within 5 timesteps and the final output at t = 50. c Binder design success rates using β-strand interface conditioning or RFdiffusion default settings with hotspots indicating the target edge strand of interest. d Structural clustering of RFdiffusion output binder scaffolds using hotspot conditioning (left) or β-strand interface conditioning (right) t-distributed Stochastic Neighbor Embedding (t-SNE) transforms of all-by-all pairwise template modeling (TM) scores60 among binder scaffolds across all targets are plotted, with close proximity of points representing structural similarity. Output fold secondary structures are classified by color as indicated in the legend. Bold bordered data points indicate in silico successes (red circles, pAE interaction <10, pLDDT >85, ΔΔG < −30) and experimentally validated binders (black stars). Source data are provided as a Source Data file.
