Table 4:

Comparison of BNN with SIS-SCAD, GAM, RF, BART, and BRNN in gene selection for 24 drugs: “Training error” denotes the average mean squared fitting error, “Prediction error” denotes the average mean squared prediction error, and #gene denotes the average number of selected genes, where the averages are over 24 drugs; the numbers in parentheses denote the standard deviations of the corresponding averages;

Method	Training error	Prediction error	#gene
SIS-SCAD	0.0248(0.0020)	0.1757(0.0132)	62.92(0.97)
BRNN	0.0587(0.0059)	0.1420(0.0151)	84.06*(7.64)
GAM	0.0719(0.0063)	0.1253(0.0091)	41.75(2.76)
RF	0.0137(0.0010)	0.1176(0.0087)	30.54(1.75)
BART	0.0366(0.0031)	0.1249(0.0092)	7.92(0.47)
BNN	0.0855(0.0061)	0.1201(0.0089)	2.38(0.40)

^*BRNN reports the effective number of parameters of the neural network instead of the number of selected genes.