Abstract
While computational systems biology provides a rich array of methods for network clustering, most of them are not suitable to capture cellular network dynamics. In the most common setting, computational algorithms seek to integrate the static information embedded in near-global interaction networks with the temporal information provided by time series experiments. We present a novel technique for temporally informed network module detection, named TD-WGcluster (Time Delay Weighted Graph CLUSTERing). TD-WGcluster utilizes four steps: (i) time-lagged correlations are calculated between any couple of interacting nodes in the network; (ii) an unsupervised version of k-means algorithm detects sub-graphs with similar time-lagged correlation; (iii) a fast-greedy optimization algorithm identify connected components by sub-graph; (iv) a geometric entropy is computed for each connected component as a measure of its complexity. TD-WGcluster notable feature is the attempt to account for temporal delays in the formation of regulatory modules during signal propagation in a network.