Fig. 6From: Applying atomistic neural networks to bias conformer ensembles towards bioactive-like conformationsMedian BEDROC of bioactive-like conformations depending on the size (number of heavy atoms) of the MCS to the closest training molecule, for the random (A) and scaffold (B) split test sets. Each point represents a split. For MCS sizes between 20 and 40 heavy atoms, AtNNs and TFD2RefSim rankers outperforms the bioactivity-unaware baselinesBack to article page