Multiple Imputation for Missing Values in Microarray Data

• Activity Number: 673
• Type: Contributed
• Date/Time: Thursday, August 5, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Richard Kennedy*+ and Hemant K. Tiwari
• Companies: The University of Alabama at Birmingham and The University of Alabama at Birmingham
• Keywords: microarray ; missing data ; imputation

Abstract

Imputation algorithms have been proposed to estimate the missing values encountered in analyzing microarray gene expression data. These methods utilize single imputation, which provides estimates of missing gene expression values but not measures of uncertainty associated with the estimates. We present an application of multiple imputation (MI) as an alternative to impute probable expression values and associated measures of uncertainty. We validate the MI process with a missing at random (MAR) simulation using other covariate information in the linear model context, as well as missing completely at random (MCAR) and not missing at random (NMAR) deletion of entire genes used in previous studies, across a range of percentages for missingness. We investigate bias and root mean square error (RMSE) of the estimates, as well as effects of MI on declaration of differential gene expression.