MULTIPLE COLLAPSE CLUSTERING
Tatiana V. Tatarinova, Alan Schumitzky · 2006
*SUMMARY In this manuscript, we present our Multiple Collapse Clustering (MCC) method for treatment of data-rich problems. Motivation: MCC is not limited to clustering of genes by similarity of their expression pattern: we suggest to compute parameters of piecewise continuous functions that approximates each gene. Our method is based on clustering of parameters of such curves. Results: We have developed a new method to analyze gene expression time series data. As a result of our clustering procedure for each cluster we obtain a smooth centroid curve and a set of curve mean parameters and standard deviations. On a test set MCC performed better compared to the K-means clustering.