Fig. 9: Exploring performance in molecular property prediction.
From: A systematic study of key elements underlying molecular property prediction
a Distribution of metric variability of different models in opioids-related datasets. b Relationship difference between metric mean and metric variability. c Relationship between label value and molecular descriptors in ESOL and Lipop. d Pearson_R between label value and molecular descriptors in MoleculeNet datasets. e Pearson_R between label value and molecular descriptors inactivity datasets by Cortés-Ciriano et al. and Tilborg et al. f Prediction performance in MolWt datasets of varying dataset sizes. g Prediction performance in NumAtoms datasets of varying dataset sizes. Metrics include root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R2), and Pearson correlation coefficient (Pearson_R). Whiskers in the box plots denote the range within 1.5 times interquartile from the median. ESOL and Lipop are two datasets from MoleculeNet. Error bar denotes standard deviation over 30 splits. Data are in the Source Data.
