Supporting information for Tudor et al. (2002) Proc. Natl. Acad. Sci. USA, 10.1073/pnas.242566899
Fig. 3.
Power (blue traces, left axes) and false-positive expectation (red traces, right axes) of different statistical methods used to analyze microarray data. The ability of the various tests to identify the manipulated genes (power) and sensitivity to Type I errors (false positives) was determined as a function of fold-change. The data represent the four MG-U74A experiments: P63 Mecp21/y vs. Mecp2+/y, cerebral cortex (n = 12, 2692 genes passed filters and were analyzed) and hippocampus experiments(n = 13, 2241 genes), and P135+ Mecp22/y; CamK-Cre93+/o vs. Mecp2+/y cortex (n = 15, 2407 genes) and hippocampus (n = 18, 4095 genes) experiments. These panels show the comparison of the parametric t test (P = 0.05) to various other methods of assessing statistical significance: VERA/SAM, maximum likelihood estimation of error model and underlying means, P = 0.05(7); PaGE, Patterns of Gene Expression error model with confidence >50% (3); BY, Benjamini-Yekutieli false-discovery rate correction of t test P values at q = 0.05 (6); WY, Westfall-Young family-wise error rate correction at P = 0.05 (4, 5); Neighborhood, signal-to-noise neighborhood analysis, P = 0.05 (2).