Assessment of statistical performances through a simulation study.

* Multiple Testing Correction: performed via Benjamini-Yekutieli procedure with a 5% threshold.In each simulation, 100 gene sets are simulated and significance level α = 5% is applied. This table displays the Type I error and the statistical power means over a hundred simulation runs for 3 different situations (0%, 27% and 85% of simulated gene sets are simulated under H1). Whenever the percentage of gene sets simulated under H1 is not null, 25% of the gene sets simulated under H1 are heterogeneous, the remaining 75% being homogeneous. Type I error is the probability of rejecting H0 given that H0 is true, i.e. for declaring a gene set significant when it actually is not. Statistical power is the probability of rejecting H0 given that H1 is true, i.e. for declaring a gene set significant when it actually is. Four methods are evaluated: i) TcGSA, the proposed approach, fitted either with a linear or with a cubic function of time; ii) the GlobalANCOVA procedure [29] in which p-values are either computed by permutation (10,000) or approximated; iii) the GSEA for time series [14], iv) CAMERA [31]. Default values are used for the various methods (see S1 Text and S1 Software).Assessment of statistical performances through a simulation study.