Fig. 11
From: Applying atomistic neural networks to bias conformer ensembles towards bioactive-like conformations
Pharmacophore searching hit rate selecting various fractions of conformers using different rankers, for the random (A) and scaffold (B) splits test sets. Each point represents a split. For early fractions (1% and 5%), TFD2SimRefMCS rankers shows a higher hit rate than bioactivity-unaware baselines, while for the 20% for the random split, ComENet shows a slightly higher hit rate