Skip to main content
ARS Home » Research » Publications at this Location » Publication #204323
Title: A Comparative Study of Normalization Methods Used in Statistical Analysis of Oligonucleotide Microarray Data

Author
item SINGH, DHARMENDRA - MS VALLEY STATE UNIV
item Boykin, Deborah
item KELLEY, ROWENA - MSU DEPT PLT SCI
item NEWSOM, ABIGAIL - MS VALLEY STATE UNIV
item Williams, William

Submitted to: Mississippi Academy of Sciences Proceedings
Publication Type: Abstract Only
Publication Acceptance Date: 2/9/2007
Publication Date: 2/9/2007
Citation: Singh, D.K., Boykin, D.L., Kelley, R.Y., Newsom, A.S., Williams, W.P. 2007. A Comparative Study of Normalization Methods Used in Statistical Analysis of Oligonucleotide Microarray Data. Mississippi Academy of Sciences Proceedings. 52(1):Program Errata p. 4.

Interpretive Summary: Normalization methods used in the statistical analysis of oligonucleotide microarray data were evaluated. The oligonucleotide microarray is considered an efficient analytical tool for analyzing thousands of genes simultaneously in a single experiment. However, systematic variation in microarray, originating from various sources, affects the measurement of gene expression levels. The purpose of normalization methods is to identify and eliminate any systematic variance in the measurements. Several normalization methods, such as total intensity normalization, intensity-dependent normalization, and global normalization were studied. Our choice for the normalization method would depend on the nature of experiments, and the type of data set being used. Different normalization methods were compared to determine the most suitable method for achieving greatest precision and eliminate spatially dependent variability. Precision was evaluated using analysis of variance procedure and measuring the rate of erroneous decisions. A Type I error occurs when significance is falsely declared and a Type II error occurs when important differences are not detected. Decreasing variability using these normalization and statistical analysis methods will decrease Type I and Type II error rates. This is important because minimization in Type II error signifies greater precision or higher power of study. A lower Type I error, in the microarray case, resulted in a lower false discovery rate of responsible genes. Spatial patterns in variability was detected by plotting residuals from analysis of variance using the x, y coordinates from the microarray slide. Our method was focused on a research project to look at gene expression related to Aspergillus flavus resistance in maize.

Technical Abstract: Normalization methods used in the statistical analysis of oligonucleotide microarray data were evaluated. The oligonucleotide microarray is considered an efficient analytical tool for analyzing thousands of genes simultaneously in a single experiment. However, systematic variation in microarray, originating from various sources, affects the measurement of gene expression levels. The purpose of normalization methods is to identify and eliminate any systematic variance in the measurements. Several normalization methods, such as total intensity normalization, intensity-dependent normalization, and global normalization were studied. Our choice for the normalization method would depend on the nature of experiments, and the type of data set being used. Different normalization methods were compared to determine the most suitable method for achieving greatest precision and eliminate spatially dependent variability. Precision was evaluated using analysis of variance procedure and measuring the rate of erroneous decisions. A Type I error occurs when significance is falsely declared and a Type II error occurs when important differences are not detected. Decreasing variability using these normalization and statistical analysis methods will decrease Type I and Type II error rates. This is important because minimization in Type II error signifies greater precision or higher power of study. A lower Type I error, in the microarray case, resulted in a lower false discovery rate of responsible genes. Spatial patterns in variability was detected by plotting residuals from analysis of variance using the x, y coordinates from the microarray slide. Our method was focused on a research project to look at gene expression related to Aspergillus flavus resistance in maize.