Evaluation of microarray normalization strategies to detect cyclic circadian genes.

Abstract

Microarrays are a widely used research tool in gene expression analysis. A large variety of preprocessing methods for raw intensity measures is available to establish gene expression values. Normalization is the key stage in preprocessing methods, since it removes systematic variations in microarray data. Then, the subsequent analyses may be highly dependent on normalization strategy employed. Our research focuses on detecting rhythmic signals in measured circadian gene expressions. We have observed that rhythmicity detection depends not only upon the rhythmicity detection algorithm but also upon the normalization strategy employed. We analyze the effects of well-known normalization strategies in literature within three different rhythmicity detection algorithms; JTK, RAIN and our recently proposal ORI, a novel statistical methodology based on Order Restricted Inference. The results obtained are compared using artificial microarray data and publicly available circadian data bases.