<p>Spearman’s rank correlation coefficient of the 50 best predicted features, for multiple combinations of network hyperparameters and input-output combinations, for a deep neural network model.</p>

Performance was computed on a validation set, separate from the test sets described in the Methods section. An augmentation factor equal to 1 indicates that no augmentation was applied, while an augmentation factor equal to n > 1 indicates that the final number of data points is equal to the initial size of the dataset multiplied by n. The best result for each input-output combination is highlighted. (XLSX)