HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Full Text (PDF) All Versions of this Article: jas.2007-0713v1 87/1/88    most recent Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Díaz, C. Articles by Carabaño, M. J. Search for Related Content PubMed PubMed Citation Articles by Díaz, C. Articles by Carabaño, M. J.

Model selection in a global analysis of a microarray experiment

C. Díaz^*, N. Moreno-Sánchez^*, Julia Rueda, A. Reverter, Y. H. Wang and M. J. Carabaño^*

^* Dpto. Mejora Genética Animal, INIA, Crta. de la Coruña km 7,5; 28040 Madrid, Spain , Dpto. Genética, Facultad de Biología, UCM, Ciudad Universitaria s/n, 28040 Madrid, Spain CSIRO Livestock Industries, Queensland Bioscience Precinct, 306 Carmody Rd., St Lucia 4067, QLD, Australia

cdiaz{at}inia.es

Abstract

Analysis of data from cDNA microarray experiments is an area of intense research. Options include models at a gene level or at a global level, the latter one joining information from all of the profiled genes. In general, a joint analysis is expected to be more powerful than gene specific analyses. Global analysis of microarray data requires fitting a model that jointly performs data normalization and analyses. The objective of this study was to assess the optimality of alternative models for data normalization and analysis in an experiment to identify differentially expressed genes between 2 muscles in Avileña Negra Ibérica calves. Three major groups of models were explored according to several aspects including spatial arrangement of spots, other technical sources of variation such as dye effects, assumptions related to effects included in the model and gene specific effects. In addition, 3 sources of heterogeneity of residual variance were investigated. All models were compared by Bayes Factors and cross-validation predictive densities. The model that included array-block, dye, muscle, and array-dye as systematic effects, and all gene related components as random effects was the best model for normalization and analysis of these data under heterogeneity of residual variances. Furthermore, level of intensity seemed to be the major source of heteroskadasticity for all models investigated. Such models rendered the best goodness of fit without compromising the predictive ability. The best model also provided the best performance to detect genes DE with the lowest FDR. The large differences found for the model comparison criteria across models, indicate the importance of defining the factors that more accurately account for experiment-wise variability in order to ensure a correct inference on differential expression of genes. Our results also illustrate the importance of the experimental setup in order to account for possible sources of bias in the detection of DE genes.

Key Words: Bayesian mixed linear model • differential gene expression • FDR • microarray • model selection • normalization,