Table 1.

Type I error rates of gene set tests when genes in the set are correlated

Test method	Nominal P-value
	0.01	0.02	0.05	0.10
geneSetTest (modt)	0.2779	0.3275	0.4157	0.4950
geneSetTest (ranks of modt)	0.2826	0.3319	0.4144	0.4955
sigPathway (t)	0.2524	0.3025	0.3880	0.4704
PAGE (logFC)	0.2441	0.2900	0.3709	0.4503
CAMERA (modt)	0.0087	0.0187	0.0477	0.0990
CAMERA (ranks of modt)	0.0086	0.0173	0.0473	0.1003

CAMERA holds its size correctly whereas existing methods are highly liberal.

Entries are probabilities of rejecting the null hypothesis when conducting a gene set test to compare two groups of four arrays. Set size is 100 with inter-gene correlation 0.05. The remainder of 10 000 genes are uncorrelated. Results are based on 10 000 simulated data sets, so the standard error with which the error rate is estimated ranges from slightly < 0.001 (for rates near 0.01) to slightly < 0.005 (for rates near 0.5).