MAQC-II Project: The Experience of the CDRH Data Analysis Team
*Samir Lababidi, U.S. Food and Drug Administration  *Francisco Martinez-Murillo, FDA / CDRH  *Gene Pennello, FDA / CDRH  *Reena Philip, FDA / CDRH  *Daya Ranamukhaarachchi, FDA / CDRH  *Rong Tang, U.S. Food and Drug Administration  *Zivana Tezak-Fragale, 2098 Gaither Road 

Keywords: MAQC, Microarray, Gene Expression, Genomic Classifiers, Biomarkers

MicroArray Quality Control (MAQC) Project is a large collaborative effort by FDA, Industry, and Academia toward consensus on “best practices” for the analysis and application of microarray data in the development and review of FDA-regulated products. The aim of MAQC-II is to reach consensus on “best practices” for developing and validating predictive models based on microarray data. Thirty data analysis teams generated 19,669 models (307 candidate models) on the 13 endpoints from six toxicogenomic / clinical data sets. Here, the CDRH data analysis team will describe its efforts toward developing genomic models for class prediction. For commercial purposes, we emphasized single array normalization, incorporating biological information as well as statistical analysis for feature selection, and inclusion of all model-building steps to ensure realistic estimates in confirmatory datasets.