A Comparison of two Methods for Finding Groups using Heat Maps and Model Based Clustering

Abstract

We are concerned with methods to investigate homogeneous patterns among clinical heterogeneous complex diseases. This methodology involves (1) a cluster analysis to group individuals by similar disease patterns, (2) a visualization step to characterize the cluster patterns and (3) an evaluation step to ascertain the reliability of discovered patterns. It will be applied to individuals affected by osteo arthritis (OA) at multiple joint sites. Here, we present and compare two methods that are used to find groups of individuals sharing similar OA patterns. The first approach uses hierarchical clustering to derive the groups, model based clustering to assess their reliability and heat maps to characterize them. The second approach uses model based clustering to derive the groups, BIC to select the optimal model and heat maps to characterize each group. Our experimental results show that for this data set the second approach, which uses model based clustering and heat maps, works much better.