Clustering the non-uniformly sampled time series of gene expression data

This paper introduces a new principled method for clustering the time series of non-uniformly sampled gene expressions. A first stage is fitting a minimum description length model to each gene, which will make possible the interpolation of the data at a finer regular grid. The optimum parameters can be used for clustering the genes according to the underlying dynamics in the data. The correlation coefficient is shown to generalize to a dynamic correlation coefficient which can be used with arbitrary clustering methods. Experiments with gene expression time series collected from the development of the cerebellum and dentate gyrus demonstrate the ability of the new clustering method to group the genes according to functional classes, which was not always possible using classical clustering methods.