Fig. 5.

Scoring power of AEScore (with and without hydrogen atoms) as a function of the similarity threshold between the training and test sets, as defined by Su et al. [41]. The raw data for the RF and DT scoring functions was kindly provided by Su et al. [41] upon request. RF and DT are respectively the best and worst performing models (at the 95% similarity threshold) presented in Su et al. [41] and are consistently outperformed by AEScore