Different normalization strategies for microarray gene expression traits affect the heritability estimation

Abstract

Abstract Several studies have been conducted to assess the influence of genetic variation on genome-wide gene expression profiles measured by the microarray technologies. Due to substantial noise in microarray-based experiments, it has long been recognized that proper normalization is a crucial step to ensure sensitive and reliable downstream analyses. This is especially true when large number of samples were collected and analyzed. In this study, we investigated the impact of different normalization strategies on genome wide linkage analyses, in particular, the estimation of heritability of gene expression traits. We used the Genetics Analysis Workshop 15 Problem 1 data. We found that there are significant differences in the estimated number of genes showing heritability when different normalization strategies were used. RMA (robust multiarray average) and dChip identify 45% and 13% more genes showing heritability than MAS 5.0, respectively. Our study also reveals that a large number of genes show strong "family effect" in their expression levels but no significant heritability. Analysis of their annotation indicates different types of genes were enriched in this group compared to genes showing strong heritability.