DTi2Vec: Drug–target interaction prediction using network embedding and ensemble learning

Table 3 Performance of DTi2Vec in terms of AUPR using AdaBoost and XGBoost classifiers on each dataset with multiple FVs generated by applying different edge representation functions for the random CV setting, and averageAUPR for each fusion function across all datasets for each classifier

Model	Fusion function (FF)	Yamanishi_08 datasets				FDA_DrugBank	AVG AUPR per FF
Model	Fusion function (FF)	NR	GPCR	IC	E	FDA_DrugBank	AVG AUPR per FF
DTi2Vec AdaBoost	Concatenate	0.74 (0.145)	0.83 (0.039)	0.97 (0.010)	0.97 (0.01)	0.77 (0.014)	0.856
	Hadamard	0.85 (0.127)	0.89 (0.037)	0.93 (0.020)	0.96 (0.007)	0.82 (0.009)	0.89
	Average	0.64 (0.132)	0.78 (0.054)	0.90 (0.029)	0.92 (0.007)	0.75 (0.022)	0.798
	Weighted L1	0.92 (0.082)	0.84 (0.052)	0.93 (0.017)	0.96 (0.008)	0.82 (0.009)	0.894
	Weighted L2	0.92 (0.082)	0.84 (0.054)	0.94 (0.018)	0.96 (0.008)	0.82 (0.008)	0.896
DTi2Vec XGBoost	Concatenate	0.72 (0.118)	0.87 (0.039)	0.98 (0.0096)	0.98 (0.007)	0.82 (0.011)	0.87
	Hadamard	0.81(0.115)	0.90 (0.036)	0.93 (0.017)	0.97 (0.006)	0.88(0.0087)	0.898
	Average	0.68 (0.142)	0.81 (0.042)	0.91 (0.024)	0.95 (0.009)	0.78 (0.020)	0.826
	Weighted L1	0.88 (0.107)	0.84 (0.055)	0.94 (0.014)	0.97 (0.006)	0.87 (0.009)	0.9
	Weighted L2	0.89 (0.103)	0.84 (0.055)	0.94 (0.014)	0.97 (0.006)	0.87 (0.009)	0.902

AUPR in bold font with underline indicate the best result in each dataset, and the italic values between parentheses are the standard deviations

ISSN: 1758-2946