Fig. 10
From: Applying atomistic neural networks to bias conformer ensembles towards bioactive-like conformations
Successful docking rate of the highest score pose (A, B) or lowest ARMSD pose (C, D) for GOLD rigid-ligand redocking of PDBbind selecting various fractions of conformers per docked ligand using different rankers, for the random (A, C) and scaffold (B, D) splits test sets. Each point represents a split. For early fractions, ComENet and TFD2SimRefMCS rankers retrieve a higher rate of successful docking than bioactivity-unaware baselines