Fig. 3: Evaluations on protein-ligand docking task.
From: Interformer: an interaction-aware model for protein-ligand docking and affinity prediction
a The bar plot of the successful docking rate on PoseBusters version 2 benchmark, * denotes using the provided start conformation as initial ligand sampling conformation. b The docking performance bar plot on three levels of homology sequence similarity subsets. c The histogram depicts the number of specific interactions recovered by various methods within the five recovery rate ranges. d The predicted five MDN distributions, the x-axis is the VdW radius distance of the atoms pair d, and the y-axis is the probability of the MDN. Yellow is a hydrogen bond pair, orange is hydrophobic, and brown are other non-interaction pairs. e The upper diagram illustrates the predicted fusion coefficients α for the hydrogen bond interaction between the 20th ligand atom (a hydrogen bond acceptor) and all other protein atoms that are hydrogen bond donors. The model predicts the top-2 highest α value for the true hydrogen bonds on 20-143 and 20-280. A similar analysis applies to the lower diagram but for another hydrophobic pair set. f Two docking poses were generated by Interformer. The white represents the crystal structures, the green indicates the docking poses, and the yellow dashed lines signify the hydrogen bond interactions. As can be observed, the poses can directly form hydrogen bond interactions.
