Fig. 2: Deep learning accelerated virtual screening finds KLHDC2 binders.

a 7 out of 29 initial synthesized and assayed compounds from the initial virtual screening. b 6 out of 21 synthesized and assayed compounds from the focused screening. Seven compounds in total show low micromolar binding affinity (indicated by the boxes). The substructure highlighted in red is used for focused library generation. c AlphaLISA assay and the IC₅₀ values of the seven compounds from the initial screening. The statistics were calculated based on three technical replicates (n = 3) for each concentration of the compounds in the 12-point titration curves. Data are presented as mean values +/− SD. d AlphaLISA assay and the IC₅₀ values of the six compounds from the focused screening. The statistics were calculated based on three technical replicates (n = 3) for each concentration of the compounds in the 12-point titration curves. Data are presented as mean values +/− SD. e Close-up view of C29 bound to KLHDC2 together with its mF_o-DF_c map calculated before the compound was built into the model and contoured at 1.5 σ. f Comparison of experimentally resolved and the computationally predicted binding pose of C29. The high-resolution X-ray crystal structure in yellow is superimposed on predicted docked pose in magenta.