L.A.Gilhuijs-Pederson and A.H.C. van Kampen
Bioinformatica Laboratorium, Academisch Medisch Centrum / Universiteit van Amsterdam
Analysis of microarray data relies heavily on determining the statistical significance of a given gene specific variance between 2 (or more) variety samples of interest. If there are multiple sources of noise (print tip variance, irregular spot sizes/shapes, uneven sticking of excess dye on the background, and nonlinearities of laser intensities, etc.) which can show up inconsistently for any given array within the dataset, often times, the noise obscures true statistically significant gene-variety variances. Traditional ANOVA analysis of microarray datasets by Kerr. et. al., although giving highly reliable estimates for statistical significance on a dataset including the noise, fails to tease out all sources of noise. Extensions of traditional ANOVA analysis are presented which aim toward detecting and subtracting various sources of noise before determining statistical significance. A variation of the traditional reference design ANOVA analysis is also presented which allows one to visualize and characterize any remaining sources of noise on an individual array within the dataset.