Percolation clustering: a novel approach to the clustering of gene expression patterns in Dictyostelium development

We present a novel approach to the clustering of gene expression patterns based on the mutual connectivity of the patterns. Unlike certain widely used methods (e.g., self-organizing maps and K-means) which essentially force gene expression data into a fixed number of predetermined clustering structures, our approach aims to reveal the natural tendency of the data to cluster, in analogy to the physical phenomenon of percolation. The approach is probabilistic in nature, and as such accommodates the possibility that one gene participates in multiple clusters. The result is cast in terms of the connectivity of each gene to a certain number of (significant) clusters. A computationally efficient algorithm is developed to implement our approach. Performance of the method is illustrated by clustering both constructed data and gene expression data obtained from Dictyostelium development.