Statistical analysis of gene expression profiles

Abstract

Function divergence after gene duplication has been considered to be an important mechanism for the evolution of new functions. Although gene expression profiles have been treated as an important indicator of gene function, large scale gene expression analysis has mostly focused on current relationships among genes, instead of their evolutionary relationships. By putting expression analysis into the framework of evolution, we make inferences about expression divergence after gene duplication. Based on the Brownian-based model (Gu 2004), the posterior distribution of the ancestral expression profiles are shown to follow a multivariate-normal distribution. This approach provides not only the estimates of the ancestral expression profiles, but also provides a measure of the precision of the estimation/predict!on, thereby, filtering significant information from the background noise of the data. Introduction Ancestral state reconstruction within an evolutionary tree is at the center of comparative studies in evolutionary biology. At the morphological level, comparative analysis looking for evidence of correlated change in two characters may need the help of inferred ancestral states (Harvey and Pagel, 1991). At the molecular level, the properties of ancient molecules (Malcolm et al., 1990; Stackhouse et al., 1990; Adey et al., 1994; Jermann et al., 1995) can be examined through the inferred ancestral amino acid sequences, which can be further tested in vivo or in vitro. Massive microarray expression profiles make it possible to reconstruct the ancestral expression pattern. In the course of such endeavor, appropriate methodology for ancestral state inference is essential.