Integrative Deep Learning for Identifying Differentially Expressed (DE) Biomarkers

Table 2

Simulation results of metalogistic regression, meta-SVM, and the integrative deep learning (inclusion of 1 random study).

Methods	σ	Sensitivity (s.e)		Specificity (s.e)		Youden J index
Meta-SVM	0.1	0.8514	(0.008)	0.9531	(0.003)	0.8045
	0.3	0.7804	(0.009)	0.8502	(0.005)	0.6345
	0.5	0.8868	(0.007)	0.4567	(0.008)	0.3435
	0.7	0.9930	(0.002)	0.0208	(0.002)	0.0138
	0.9	0.8465	(0.012)	0.4899	(0.013)	0.3364
Metalogistic regression	0.1	0.1347	(0.007)	0.9392	(0.002)	0.0739
	0.3	0.2131	(0.007)	0.9548	(0)	0.1680
	0.5	0.2638	(0.01)	0.9338	(0.003)	0.1977
	0.7	0.2555	(0.01)	0.8965	(0.004)	0.1520
	0.9	0.2652	(0.01)	0.8706	(0.004)	0.1359
Integrative deep learning	0.1	0.7708	(0.024)	0.8345	(0.02)	0.6052
	0.3	0.6812	(0.026)	0.7302	(0.023)	0.4114
	0.5	0.6708	(0.022)	0.6989	(0.022)	0.3697
	0.7	0.6979	(0.027)	0.6395	(0.029)	0.3375
	0.9	0.7583	(0.028)	0.5828	(0.028)	0.3411