Skip to main content
ARS Home » Research » Publications at this Location » Publication #197722
Title: A COMPARATIVE STUDY ON METHODS FOR ANALYSIS OF TWO COLOR MICROARRAY DATA CONTAINIG BIOLOGICAL AND TECHNICAL REPLICATION

Author
item ZHANG, WENSHENG - UNIVERSITY OF GEORGIA
item ROBBINS, KELLY - UNIVERSITY OF GEORGIA
item REKAYA, ROMDHANE - UNIVERSITY OF GEORGIA
item BERTRAND, KEITH - UNIVERSITY OF GEORGIA
item Barb, Claude
item Hausman, Gary

Submitted to: World Congress of Genetics Applied in Livestock Production
Publication Type: Abstract Only
Publication Acceptance Date: 6/30/2006
Publication Date: 8/1/2006
Citation: Zhang, W., Robbins, K., Rekaya, R., Bertrand, K., Barb, C.R., Hausman, G.J. 2006. A comparative study on methods for analysis of two color microarray data containig biological and technical replication [abstract]. World Congress of Genetics Applied in Livestock Production. p. 207.

Interpretive Summary:

Technical Abstract: It was proposed that a two-step procedure could be used to analyze cDNA microarray gene expression data containing biological and technical replications. In the first step, a normalization procedure is conducted in order to alleviate the bias introduced by systematic effects. In the second step, the adjusted data is analyzed, for each gene separately, via a mixed linear model with the biological replication as a random effect. In this study, an ANOVA normalization based method (M1) and a LOWESS normalization-based method (M2) were compared and a compromise implementation (M3) of the two methods was proposed. The results from the analysis of ovarian growth and development in prepuberal pigs at different age classes showed large discrepancies between the lists of differentially expressed genes identified by M1 and M2. Further, it seems that LOWESS normalization in M2 was more efficient in alleviating the gene-specific dye biases. The proposed compromise method (M3) had similar results to M2. The repeatability of gene expression calculated using the three methods showed different distribution profiles. A real data-based simulation proved that M2 was superior to M1 in sensitivity for identifying differentially expressed genes, and the proposed method (M3) improved the sensitivity compared to M2 when the simulated inter-class difference was relatively small.