Effect of missing values in detecting significant genes in a cDNA microarray experiment
by
Byung-Soo Kim
Dept. of Applied Statistics, Yonsei University
Coauthors: Sun Young Rha (Cancer Metastasis Research Center, College of Medicine, Yonsei Unviersity, Seoul, Korea)

The aim of this paper is to discuss the effect of missing values in detecting significant genes in a cDNA microarray experiment in a one sample problem. We conducted a cDNA microarray experiment to detect significant genes for the metastasis of colon cancer based on twenty patients who underwent liver resection due to liver metastasis from colon cancer. Total RNAs' from metastatic liver tumor and adjacent normal liver tissue were labeled with Cy5 and Cy3, respectively, and competitively hybridized to a cDNA microarray with 8045 genes.

We use M=log2(R/G) for the signal evaluation, where R and G denote the fluorescent intensities of Cy5 and Cy3 dyes, respectively. The statistical problem comprises a one sample test of testing E(M)=0 and involves multiple tests. The twenty cDNA microarray data would comprise a 8045 by 20 matrix if there were no missing values. However, missing values occur for various reasons. For each gene, the missing proportion (MP) is defined to be the proportion of missing values out of twenty with the from 0 to 1. We consider the genes whose MP is less than or equal to 0.4 and decrease MP by 0.1 for the further analyses. For each fixed MP, we impute the missing values with K-nearest neighbor method (K=10) and apply one sample t-test by slightly modifying Dudoit et al¡¯s procedure (Dudoit et al., 2000), Tusher et al.¡¯s SAM (Tusher et al., 2001) and other procedures to find out the effect of missing values in the final outcome.

Date received: August 29, 2002