Hierarchical signature clustering for time series microarray data.

Existing clustering techniques provide clusters from time series microarray data, but the distance metrics used lack interpretability for these types of data. While some previous methods are concerned with matching levels, of interest are genes that behave in the same manner but with varying levels. These are not clustered together using an Euclidean metric, and are indiscernible using a correlation metric, so we propose a more appropriate metric and modified hierarchical clustering method to highlight those genes of interest. Use of hashing and bucket sort allows for fast clustering and the hierarchical dendrogram allows for direct comparison with easily understood meaning of the distance. The method also extends well to use k-means clustering when a desired number of clusters are known.