Abstract
A common purpose of microarray experiments is to study the variation in gene expression across the categories of an experimental factor such as tissue types and drug treatments. However, it is not uncommon that the studied experimental factor is a quantitative variable rather than categorical variable. Loss of information would occur by comparing gene-expression levels between groups that are factitiously defined according to the quantitative threshold values of an experimental factor. Additionally, lack of control for some sensitive clinical factors may bring serious false positive or negative findings.
In the present study, we described a bootstrap-based regression method for analyzing gene-expression data from the non-categorical microarray experiments. To illustrate the utility of this method, we applied it to our recent gene-expression study of circulating monocytes in subjects with a wide range of variations in bone mineral density (BMD). This method allows a comprehensive discovery of gene expressions associated with osteoporosis-related traits while controlling other common confounding factors such as height, weight and age. Several genes identified in our study are involved in osteoblast and osteoclast functions and bone remodeling and/or menopause-associated estrogen-dependent pathways, which provide important clues to understand the etiology of osteoporosis.
SAS code is available from the authors upon request.
► Studied factors are often quantitative rather than categorical variables. ► Biased results may be obtained from microarray experiments of quantitative factors. ► A bootstrap-based regression method is proposed to control the bias. ► The utility of this method is illustrated on a microarray study of osteoporosis.