Analysis of Paired miRNA-mRNA Microarray Expression Data Using a Stepwise Multiple Linear Regression Model

MicroRNAs are small endogenous RNAs that play important roles in gene regulation. With the accumulation of expression data, numerous approaches have been proposed to infer miRNA-mRNA regulation from paired miRNA-mRNA expression profiles. These mainly focus on discovering and validating the structure of regulatory networks, but do not address the prediction and simulation tasks. Furthermore, functional annotation of miRNAs relies on miRNA target prediction, which is problematic since miRNA-gene interactions are highly tissue-specific. Thus a different approach to functional annotation of miRNA-mRNA regulation that can generate context-specific expression levels is needed. In this study, we analyzed paired miRNA-mRNA expressions from breast cancer studies. The expression of mRNAs is modeled as a multiple linear function of the expression of miRNAs and the parameters are estimated using stepwise multiple linear regression (SMLR). We demonstrate that the SMLR model can predict mRNA expression patterns from miRNA expressions alone and that the predicted gene expression levels preserve differentially regulated gene sets, as well as the functional categories of these genes. We show that our quantitative approach can determine affected biological activities better than the traditional target-prediction based methods.