Table 4.

Baseline methods features.

Characteristics	Methods
	1. KronRLS	2. SimBoost	3. DeepDTA	4. WideDTA	5. PADME
Datasets	Davis, Metz	Davis, Metz, Kiba	Davis, Kiba	Davis, Kiba	Davis, Metz, Kiba, ToxCast
ML/DL	AI/ML	AI/ML	DL	DL	DL
Similarity (OR) Feature based method	Similarity-based	Similarity and feature based	Feature-based	Feature-based	Feature-based
Drug representation (or features)	PubChem Sim Chemical kernels	PubChem Sim + statistical and network features	SMILES	SMILES + LMCS	SMILES / ECFP
Protein representation (or features)	SW sim score, Normalized SW sim score	SW sim score	aaseq	aaseq + PDM	PSC
NN type for features learning			CNN	two 1D-CNN	GCNN
NN type for prediction			3 FC layers	FC layer	Feedforward NN
Regressor/OR/activation function	KronRLS model	Gradient boosting model	ReLU	ReLU	ReLU
Validation setting	S1, S2, S3	S1	S1	S1	S1, S2, S3
Cross Validation	Repeated 10-folds CV, Nested CV, LDO-CV, LTO-CV	10 times 5 folds CV, LDO-CV, LTO-CV	5 folds CV	6 folds CV	5 folds CV, LDO-CV, LTO-CV
Performance metrics	CI, MSE	CI, RMSE	CI, MSE, PCC	CI, MSE, PCC	CI, RMSE, R2
Classification/Regression	Both	Both	Regression	Regression	Both
Year	2014	2017	2018	2019	2018

ML, Machine Learning; DL, Deep Learning; Sim, Similarity; aaseq, amino-acid sequence; SPS, structural property sequence; PSC, protein sequence composition; PDM, protein domain and motif; ECFP, extended-connectivity fingerprint; LMCS, ligand maximum common substructure; KronRLS, Kronecker Regularized Least Square; CNN, convolutional neural network; GCNN, graph convolution neural network; RNN, recurrent neural network; FC, fully connected; ReLU, rectified linear unit; CV, cross validation; LDO, leave one drug out; LTO, leave one target out; MSE, Mean Square Error; RMSE, root square of mean square error; CI, concordance index; PCC, Pearson correlation coefficient.