Statistical tools for synthesizing lists of differentially expressed features in related experiments

Table 3 Performance of Hwang et al.'s method on simulated data for scenario I

	DE	nonDE	FP (%)	TP (%)	FN (%)	TN (%)	Global error	Global error R(q₂)
Independent case: n = 3000, common = 0, DE1 = 1000, DE2 = 800	320	2,680	320 (10.7)	0	0	2,680 (89.3)	320	0
A: n = 3000, common = 700, DE1 = 1000, DE2 = 800
Case A1	1,121	1,879	440 (19.1)	681 (97.3)	19 (2.7)	1,860 (80.9)	459	82
Case A2	409	2,591	188 (8.2)	221 (31.6)	479 (68.4)	2,112 (91.8)	667	544
B: n = 3000, common = 200, DE1 = 700, DE2 = 500
Case B1	999	2,001	805 (28.8)	194 (97.0)	6 (3.0)	1,996 (71.2)	811	31*
Case B2	427	2,573	333 (11.9)	94 (47.0)	106 (53.0)	2,467 (88.1)	439	165
C: n = 3000, common = 100, DE1 = 500, DE2 = 400
Case C1	816	2,185	718 (24.8)	97 (97.1)	3 (2.9)	2,182 (75.2)	721	19*
Case C2	346	2,654	299 (10.3)	47 (47.0)	53 (53.0)	2,601 (89.7)	352	84

Average simulation results: we present the results from Hwang et al.'s method on the simulated data under scenario I. DE1 and DE2 are the differentially expressed genes in the first and the second experiment respectively. We used the Fisher's weighted F defined as $F_{g} = - 2 \sum_{k = 1}^{2} w_{k} l n (p_{g k})$ , where w_kis the weight for the k^thexperiment and p_gkis the p value for the gene g in the experiment k. We present the non-parametric rule to select the differentially expressed (DE) genes, as suggested by the authors. The method is implemented in Matlab. In the last column we report the Global error (FP + FN) of our procedure for q₂ (see Table 2) for ease of comparison. *There is no ratio larger than 2 so the maximum rule has been used in this case.

ISSN: 1474-760X