Chemical toxicity prediction based on semi-supervised learning and graph convolutional neural network

Table 3 The average test performance of conventional ML models on the 12 prediction tasks in 5 repeated experiments

Tasks	KNN		NN		RF		SVM		XGBoost
Tasks	AUC	Std.	AUC	Std.	AUC	Std.	AUC	Std.	AUC	Std.
NR-AR-LBD	0.6955	–	0.6671	0.0244	0.7323	0.0267	0.6795	–	0.6784	–
NR-AR	0.6527	–	0.6806	0.0088	0.6836	0.0266	0.7193	–	0.6818	–
NR-AhR	0.7639	–	0.7628	0.0177	0.8243	0.0074	0.7794	–	0.8287	–
NR-Aromatase	0.5576	–	0.5127	0.0772	0.6900	0.0092	0.6873	–	0.7106	–
NR-ER-LBD	0.6191	–	0.5387	0.1171	0.6169	0.0300	0.6078	–	0.6250	–
NR-ER	0.6597	–	0.6549	0.0162	0.6316	0.0080	0.6126	–	0.6745	–
NR-PPAR-gamma	0.6182	–	0.5558	0.0736	0.7135	0.0258	0.6454	–	0.6414	–
SR-ARE	0.6366	–	0.5656	0.0251	0.6603	0.0018	0.6843	–	0.6640	–
SR-ATAD5	0.5866	–	0.6240	0.0537	0.6928	0.0189	0.6546	–	0.6841	–
SR-HSE	0.6574	–	0.6143	0.0222	0.6852	0.0131	0.6858	–	0.6647	–
SR-MMP	0.7057	–	0.6551	0.0612	0.7818	0.0065	0.7794	–	0.7656	–
SR-p53	0.6778	–	0.5963	0.0075	0.7263	0.0130	0.7051	–	0.6942	–

The bold number denotes the best result among all conventional ML models in the corresponding task

ISSN: 1758-2946