Table 3

EASE themes are consistent despite the poor overlap of gene lists derived from the same experiment by various analytical methods

Method

A

B

C

D

E

F

G

H

Normalization

MAS 4

MAS 4

dChip

dChip

Rank remap

Rank remap

NP

NP

Intensity calculation

MAS 4

MAS 4

dChip

dChip

dChip

dChip

dChip

dChip

Method

Normalization

Intensity calculation

Gene selection

t-test

SAM

t-test

SAM

t-test

SAM

t-test

SAM


A

MAS 4

MAS 4

t-test

[72]

72 (60%)

20 (7%)

17 (12%)

18 (8%)

19 (12%)

24 (8%)

22 (10%)

B

MAS 4

MAS 4

SAM

[120]

27 (9%)

22 (12%)

25 (9%)

25 (13%)

34 (10%)

31 (12%)

C

dChip

dChip

t-test

[220]

81 (36%)

130 (48%)

70 (27%)

105 (29%)

71 (22%)

D

dChip

dChip

SAM

[86]

56 (27%)

55 (40%)

49 (18%)

47 (27%)

E

Rank remap

dChip

t-test

[180]

95 (50 %)

109 (35%)

76 (32%)

F

Rank remap

dChip

SAM

[105]

68 (24%)

67 (59%)

G

NP

dChip

t-test

[242]

154 (59%)

H

NP

dChip

SAM

[173]


Gene lists resulting from the same experiment can differ greatly as a result of selection criteria. Eight different methods were used to select genes upregulated in peripheral blood mononuclear cells (PBMCs) of HIV patients after discontinuation of antiretroviral drug therapy. The various lists resulted from four different array-to-array normalizations, two different methods of intensity calculation and two methods of statistical selection of genes (see text). The values in square brackets on the diagonal show the total number of genes yielded by each method. The unbracketed numbers give the absolute number of genes shared by any two methods, with the percentage of genes in both lists from the combined gene lists in parentheses. MAS 4, Affymetrix analysis software version 4.0; dChip, dChip software [8]; Rank remap, unpublished method of D.A.H; NP, nonparametric method of Sidirov et al. [9]; t-test, Student T statistic; SAM, statistical analysis of microarrays software [10].

Hosack et al. Genome Biology 2003 4:R70   doi:10.1186/gb-2003-4-10-r70

Open Data