Could correlation-based methods be used to derive genetic association networks?

In recent years a number of methods have been proposed for reverse engineering of genetic networks from gene expression data. These methods work well on small genetic networks with very few connections between genes, but for larger networks and networks with higher connectivity, the computational cost increases dramatically and the performance of these methods is insufficient. In real systems, however, it is known that the networks are large and that genes typically have many interactions. In addition, the methods require abundant expression data for derivation of the networks. A method that can derive networks irrespective of these obstacles and have a low computational cost will be of importance. In this paper, three correlation-based methods are investigated as alternatives. Using correlation-based methods means that the computational cost-is reduced, since only N/2 correlations have to be computed-for a data set of N expression profiles. The presented methods are not limited by any maximum size of the network, nor by the connectivity of the network, or the amount of expression data. (C) 2002 Elsevier Science Inc. All rights reserved.