A statistical approach to selecting and confirming validation targets in -omics experiments

BMC Bioinformatics

Table 1 Simulation study to assess the properties of the validation probability

Scenario	Quantity	FDR	FDR	FDR
		5%	10%	50%
Errorless validation	Median Validation Probability	0.72	0.91	1.00
Prior = Uniform	Validation Probability IQR	(0.58, 0.87)	(0.72, 0.98)	(1.00,1.00)
	FDR 95% Credible Interval Coverage	0.98	0.68	0.00
	Median Posterior Expectation of FDR	0.04	0.07	0.35
Validation subject to error	Median Validation Probability	0.64	0.87	1.00
Prior = Uniform	Validation Probability IQR	(0.37, 0.83)	(0.71, 0.95)	(1.00,1.00)
	FDR 95% Credible Interval Coverage	0.98	0.68	0.00
	Median Posterior Expectation of FDR	0.05	0.08	0.24
Results should not validate	Median Validation Probability	0.00	0.00	0.25
Prior = Uniform	Validation Probability IQR	(0.00,0.00)	(0.00,0.00)	(0.15,0.51)
	FDR 95% Credible Interval Coverage	0.00	0.00	0.96
	Median Posterior Expectation of FDR	0.35	0.37	0.52
Errorless validation	Median Validation Probability	0.83	0.94	1.00
Prior = Adaptive	Validation Probability IQR	(0.70,0.94)	(0.79,0.99)	(1.00,1.00)
	FDR 95% Credible Interval Coverage	0.88	0.63	0.00
	Median Posterior Expectation of FDR	0.04	0.07	0.35
Validation subject to error	Median Validation Probability	0.76	0.91	1.00
Prior = Adaptive	Validation Probability IQR	(0.49, 0.91)	(0.78, 0.97)	(1.00,1.00)
	FDR 95% Credible Interval Coverage	0.91	0.77	0.00
	Median Posterior Expectation of FDR	0.04	0.07	0.23
Results should not validate	Median Validation Probability	0.00	0.00	0.25
Prior = Adaptive	Validation Probability IQR	(0.00,0.00)	(0.00,0.00)	(0.15,0.51)
	FDR 95% Credible Interval Coverage	0.00	0.00	0.95
	Median Posterior Expectation of FDR	0.35	0.37	0.52

For each of three scenarios and two choices for the prior distribution, 100 simulated gene expression studies were generated with 1,000 genes each. This table reports the median (25th percentile, 75th percentile) of the validation probability across the 100 studies, the coverage proportion of the 95% posterior credible interval for the estimated FDR in each scenario, and the median posterior expectation of the FDR.

ISSN: 1471-2105