Measuring similarities between gene expression profiles through new data transformations

Background: Clustering methods have been widely applied to gene expression data in order to group genes sharing common or similar expression profiles into discrete functional groups. In such analyses, designing an appropriate (dis)similarity measure is critical. In this study, we aim to develop a new distance measure for gene expression profiles. The new measure is expected to be especially efficient when the shape of expression profile is vital in determining the gene relationship, yet the expression magnitude should also be accounted for to some extent. Results: The new measure, named TransChisq, was developed by separately modeling the shape and magnitude information and then using the estimated shape and magnitude parameters to define a distance measure in a new feature space. The feature space was constructed based on the specific clustering purpose of grouping genes with similar shape of expression curves, while the magnitude information should also be considered when determining the shape similarity. The new measure was employed into a k-means clustering procedure for performing clustering analyses. Results from applications to a simulation dataset, a developing mouse retina SAGE dataset, a small yeast sporulation cDNA datase