Empirical Bayes method for reducing false discovery rates of correlation matrices with block diagonal structure

BMC Bioinformatics

Table 3 Simulation results with known groups (first three columns) and using a known threshold level to estimate groups or blocks from the data (threshold) and the EB AIC method for when neither the threshold or groups are known

		EB	Corpcor	Pearson	EB threshold	CorpCor threshold	Pearson threshold	EB AIC
n=10	FDR mean	0.00	0.00	0.00	0.04	0.26	0.16	0.04
	FDR sd	0.00	0.00	0.00	0.09	0.12	0.09	0.09
	TPR mean	0.70	0.54	0.66	0.39	0.56	0.66	0.36
	TPR sd	0.21	0.12	0.15	0.21	0.13	0.15	0.21
n=15	FDR mean	0.00	0.00	0.00	0.09	0.16	0.13	0.06
	FDR sd	0.00	0.00	0.00	0.11	0.12	0.11	0.11
	TPR mean	0.90	0.32	0.83	0.75	0.59	0.81	0.61
	TPR sd	0.10	0.08	0.10	0.19	0.19	0.12	0.21
n=20	FDR mean	0.00	0.00	0.00	0.11	0.11	0.10	0.10
	FDR sd	0.00	0.00	0.00	0.10	0.10	0.09	0.10
	TPR mean	0.97	0.44	0.90	0.91	0.55	0.88	0.79
	TPR sd	0.04	0.09	0.06	0.09	0.14	0.08	0.11

We calculate the FDR and TPR for all the variables by comparing them to the true matrix, significance of correlations was determined using a t-test. We see improved or comparable FDR’s for the EB methods across all sample sizes. There is a particular improvement for n=10, however, there is a trade-off in terms of lower TPR. However, together we would still expect the EB method to find high value interactions as significant, which is important in designing downstream validation experiments

ISSN: 1471-2105