Table 5. Summary of Model Performances of the ChemBioSim Models and Existing Methods.

endpoint	model	mean sensitivity	mean specificity	evaluation	modeling approach	comments
MNT	Yoo et al.	0.54–0.74	0.77–0.93	5% leave-many-out	Leadscope Enterprise and CASE Ultra software	variations related to different modeling approaches
	our method	0.78	0.76	5-fold CV	CP built on RF models	CHEMBIO model with feature selection
DILI	Ancuceanu et al.	0.83	0.66	nested CV	meta-model with a naïve Bayes model trained on output probabilities of 50 ML models
	our method	0.78	0.78	5-fold CV	CP built on RF models	CHEMBIO model with feature selection
DICC	Cai et al.	0.69–0.75	0.72–0.81	5-fold CV	combined classifier using neural networks based on four single classifiers	results refer to five cardiological complications endpoints evaluated independently
	our method	0.83	0.86	5-fold CV	CP built on RF models	CHEMBIO model with feature selection