. 2019 Sep 13;59(10):4438–4449. doi: 10.1021/acs.jcim.9b00236

Table 5. Multilabel Classification of the 8336 Molecules with an Average of 1.2 Classes Assigned to Each Molecule^a.

problem group	metric	IMG + CNN
12-class multilabel	thresh. accuracy^b	0.954 ± 0.00133
	F_β^b	0.635 ± 0.0168
	ROC AUC	0.938 ± 0.00353
	ave. prec. score	0.837 ± 0.00953

Results are from fivefold cross-validation with folds determined by iterative stratification. Accuracy, ROC AUC, and average precision scores are not directly comparable to the 12-class single-class formulation (Table 4) because the number of molecules differs.

Class score thresholds set to 0.5.

Table 5. Multilabel Classification of the 8336 Molecules with an Average of 1.2 Classes Assigned to Each Moleculea.

Table 5. Multilabel Classification of the 8336 Molecules with an Average of 1.2 Classes Assigned to Each Molecule^a.