A strategy for evaluating pathway analysis methods

BMC Bioinformatics

Table 2 Average recall values for the six pathway analysis (PA) methods, using the 10 gene expression datasets

Sample Size	ORA	GSA	GSEA	GSEAs	AFC	AFCs
Significantly up-regulated pathways
20 MSP^a
6	0.19	0.13 ^b	0.25		0.44 ^c
12	0.28	0.31	0.30	0.23	0.54	0.31
24	0.38	0.45	0.34	0.35	0.64	0.45
48	0.51	0.58	0.40	0.48	0.74	0.60
50 MSP
6	0.23	0.27	0.39		0.50
12	0.33	0.43	0.46	0.35	0.59	0.42
24	0.42	0.55	0.52	0.46	0.68	0.55
48	0.55	0.68	0.60	0.59	0.78	0.68
Significantly down-regulated pathways
20 MSP
6	0.37	0.14	0.35		0.56
12	0.49	0.31	0.40	0.27	0.64	0.34
24	0.58	0.43	0.45	0.38	0.72	0.46
48	0.68	0.55	0.49	0.51	0.80	0.60
50 MSP
6	0.39	0.29	0.48		0.59
12	0.51	0.45	0.54	0.41	0.66	0.47
24	0.60	0.56	0.60	0.51	0.73	0.59
48	0.70	0.68	0.65	0.63	0.81	0.70

^aMSP most significant pathway
^bUnderlined text indicates the minimal value in a row
^cBold text indicates the maximal value in a row
The computations are based on each method’s identification of the top 20 or 50 most significantly up- or down-regulated pathways for all datasets (and 500 randomly resampled sub-datasets for each original dataset). The sample size corresponds to the number of samples contained in the sub-datasets. A higher average recall reflects a more consistent PA method

ISSN: 1471-2105