Manuela Hummel1, Reinhard Meister, Ulrich Mansmann. 1. IBE, University of Munich, Technische Fachhochschule Berlin and Department of Statistics, University of Munich, Germany. hummel@ibe.med.uni-muenchen.de
Abstract
MOTIVATION: Several authors have studied expression in gene sets with specific goals: overrepresentation of interesting genes in functional groups, predictive power for class membership and searches for groups where the constituent genes show coordinated changes in expression under the experimental conditions. The purpose of this article is to follow the third direction. One important aspect is that the gene sets under analysis are known a priori and are not determined from the experimental data at hand. Our goal is to provide a methodology that helps to identify the relevant structural constituents (phenotypical, experimental design, biological component) that determine gene expression in a group. RESULTS: Gene-wise linear models are used to formalize the structural aspects of a study. The full model is contrasted with a reduced model that lacks the relevant design component. A comparison with respect to goodness of fit is made and quantified. An asymptotic test and a permutation test are derived to test the null hypothesis that the reduced model sufficiently explains the observed expression within the gene group of interest. Graphical tools are available to illustrate and interpret the results of the analysis. Examples demonstrate the wide range of application. AVAILABILITY: The R-package GlobalAncova (http://www.bioconductor.org) offers data and functions as well as a vignette to guide the user through specific analysis steps.
MOTIVATION: Several authors have studied expression in gene sets with specific goals: overrepresentation of interesting genes in functional groups, predictive power for class membership and searches for groups where the constituent genes show coordinated changes in expression under the experimental conditions. The purpose of this article is to follow the third direction. One important aspect is that the gene sets under analysis are known a priori and are not determined from the experimental data at hand. Our goal is to provide a methodology that helps to identify the relevant structural constituents (phenotypical, experimental design, biological component) that determine gene expression in a group. RESULTS: Gene-wise linear models are used to formalize the structural aspects of a study. The full model is contrasted with a reduced model that lacks the relevant design component. A comparison with respect to goodness of fit is made and quantified. An asymptotic test and a permutation test are derived to test the null hypothesis that the reduced model sufficiently explains the observed expression within the gene group of interest. Graphical tools are available to illustrate and interpret the results of the analysis. Examples demonstrate the wide range of application. AVAILABILITY: The R-package GlobalAncova (http://www.bioconductor.org) offers data and functions as well as a vignette to guide the user through specific analysis steps.
Authors: Fabrice Berger; Bertrand De Meulder; Anthoula Gaigneaux; Sophie Depiereux; Eric Bareke; Michael Pierre; Benoît De Hertogh; Mauro Delorenzi; Eric Depiereux Journal: BMC Bioinformatics Date: 2010-10-13 Impact factor: 3.169
Authors: Uwe Thissen; Suzan Wopereis; Sjoerd A A van den Berg; Ivana Bobeldijk; Robert Kleemann; Teake Kooistra; Ko Willems van Dijk; Ben van Ommen; Age K Smilde Journal: BMC Bioinformatics Date: 2009-02-07 Impact factor: 3.169