Correspondence analysis of microarray time-course data in case-control design

Although different statistical approaches have been proposed for analyzing microarray time-course data, method for analyzing such data collected using the popular case-control design in clinical investigations has not been proposed perhaps due to the increased complexity for the existing parametric or non-parametric approaches. In this paper, we introduce a new multivariate data analyzing technique, the correspondence analysis, to analyze the high dimensional microarray time-course data in case-control design. We show, through an example on type 2 diabetes, how the nice features of the correspondence analysis can be use to explore the various time-course gene expression profiles that exist in the data. By coordinating and examining the projections on the reduced dimensions by both the genes and the time-course experiments, we are able to identify important genes and time-course patterns and make inferences on their biological relevance. Using the sample replicates, we propose a bootstrap procedure for inferring the significance of contributions on the leading dimensions by both the time-course experiments and the genes. Striking differences in the time-course patterns in the normal controls and diabetes patients have been revealed. In addition, the method also identifies genes that display similar or comparable time-course expression patterns shared by both the cases and the controls. We conclude that our correspondence analysis based approach can be a useful tool for analyzing high dimensional microarray data collected in clinical investigations.