Finding Sparse Gene Networks

Abstract

DNA microarray technology enabled us to produce time series of gene expression patterns. Our research group launched a project whose purpose is to reveal the gene regulatory networks among the 6,200 genes of Saccharomyces cerevisiae. We have introduced a weighted network model as an edge-weighted graph, where each weight reflects the strength of the interaction, and analyzed its computational complexity [4]. We also proposed an algorithm to adjust the weights incrementally. Based on the algorithm, we have been developing a system to find genetic networks and visualize them [3]. One of the most serious problem in our model is that the network produced by our system might be dense, since we have not put any restrictions on in-degree nor out-degree of the network, although many biologists claim that the network should be sparse. We need some methods to reduce the edges so that the resulting network is sparse enough. In this paper, we propose a method to reduce the edges inspired by the covariance selection models [2].