Integrative Deep Learning for Identifying Differentially Expressed (DE) Biomarkers

Table 3

Simulation results of metalogistic regression, meta-SVM, and the integrative deep learning (inclusion of 2 random studies).

Methods	σ	Sensitivity (s.e)		Specificity (s.e)		Youden J index
Meta-SVM	0.1	0.8284	(0.008)	0.9815	(0.001)	0.8099
	0.3	0.9090	(0.006)	0.1453	(0.005)	0.0543
	0.5	0.9990	(0.001)	0.0010	(0.0003)	0
	0.7	0.8518	(0.013)	0.2736	(0.012)	0.1254
	0.9	0.9944	(0.0017)	0.0056	(0.001)	0
Metalogistic regression	0.1	0.1423	(0.008)	0.9062	(0.003)	0.0485
	0.3	0.1861	(0.01)	0.9145	(0.003)	0.1006
	0.5	0.2319	(0.01)	0.8678	(0.004)	0.0977
	0.7	0.2527	(0.01)	0.8170	(0.006)	0.0697
	0.9	0.2583	(0.01)	0.8359	(0.005)	0.0942
Integrative deep learning	0.1	0.7666	(0.018)	0.9135	(0.012)	0.6802
	0.3	0.7562	(0.023)	0.7338	(0.02)	0.4901
	0.5	0.7208	(0.022)	0.6968	(0.018)	0.4177
	0.7	0.7187	(0.03)	0.6192	(0.029)	0.3380
	0.9	0.7770	(0.02)	0.5229	(0.03)	0.3000