Table 3. Assessment of performance across compound embedding methods.

	Activatory DTIs		Inhibitory DTIs
	AUROC	AUPR	AUROC	AUPR
MACCS	0.883±0.023	0.852±0.028	0.941±0.005	0.923±0.006
Morgan	0.861±0.026	0.837±0.033	0.935±0.004	0.923±0.006
Mol2vec	0.885±0.024	0.863±0.027	0.945±0.004	0.934±0.005

Boldface indicates the highest value for each performance metric. Logit, logistic regression; RF, random forest; ERT, extremely randomized trees; MLP, multilayer perceptron; CDF, cascade deep forest.

^# CDF model with 2 estimators in each cascade layer and 100 trees in each forest.