ASSESSING TOXICOGENOMIC DATA: ANALYSIS OF MICROARRAY QUALITY CONTROL MEASURES

Burgoon, L.D.*^1,3, Boverhof, D.R.^2,3, Burt, J.^2,3, Fong, C.J.^2,3, Ramer, M.², Zacharewski, T.R.^2,3

¹Dept of Pharmacology & Toxicology, ²Dept of Biochemistry & Molecular Biology, ³Institute for Environmental Toxicology, National Food Safety & Toxicology Center, Michigan State University.

Adoption of the microarray technology brings with it many benefits, such as the ability to measure the expression of thousands of genes simultaneously. However, in addition to substantial costs, it also harbors significant technical difficulties. Although the technology continues to mature, quality assurance and quality control measures for assay performance and data quality are lacking. Average signal intensity, background intensity and dynamic range were compared across five independent microarray data sets to assess their predictive value in determining data quality using spread vs. level (SVL), principal components biplots, shewhart, feature not detected (FND), boxplots, and saturated features (SF) plots. The anomalies identified included 1) high global background, 2) low feature signal intensity, and 3) compressed dynamic range which skewed or abbreviated normalized distributions, resulting in disrupted comparisons and confounding the results of statistical tests as well as pattern recognition algorithms (i.e., clustering). Average signal intensity, background intensity and dynamic range proved to be valuable measures to monitor assay and operator performance within and between studies. Moreover, these analyses suggest background intensity values should be sacrificed in favor of a higher PMT gain setting in order to maximize dynamic range. This work was supported by NIH grants: *T32 ES07255 and ES011271.