Identification and Analysis of Cell Cycle Phase Genes by Clustering in Correspondence Subspaces

Correspondence analysis (CA) is a statistical method that is widely used in multiple disciplines to reveal relationships amongst variables. Among others, CA has been successfully applied for microarray data analysis. One of CA’s strengths is its ability to help visualize the complex relationships that may be present in the data. In this sense, CA is a powerful exploratory tool that takes advantage of human pattern analysis abilities. The power of CA can, however, be diluted, if the patterns are embedded in data clutter. This is because CA is a dimensionality reduction approach and not a data reduction method; thus, is powerless to remove clutter. Unfortunately, our visual analysis abilities can be overwhelmed in such conditions causing failures in identifying relationships. In this paper, we propose a solution to this problem by combining CA with oneway analysis of variance (ANOVA) and subsequently by clustering in the lowdimensional space obtained from CA. We investigate the proposed approach using microarray data from 6200 S. cerevisiae genes and demonstrate how visual analysis is facilitated by removal of unnecessary clutter as well as facilitating the discernment of complex relationships that may be missed through application of CA alone.