Fast Calculation of Pairwise Mutual Information for Gene Regulatory Network Reconstruction Peng Qiu, Andrew J. Gentles and Sylvia K. Plevritis Department of Radiology, Stanford University, Stanford, CA Abstract We present a new software implementation to more efficiently com- pute the mutual information for all pairs of genes from gene expression microarrays. Computation of the mutual information is a necessary first step in various information theoretic approaches for reconstructing gene regulatory networks from microarray data. When the mutual in- formation is estimated by kernel methods, computing the pairwise mu- tual information is quite time-consuming. Our implementation signif- icantly reduces the computation time. For an example data set of 336 samples consisting of normal and malignant B-cells, with 9563 genes measured per sample, the current available software for ARACNE re- quires 142 hours to compute the mutual information for all gene pairs, whereas our algorithm requires 1.6 hours. The increased efficiency of our algorithm improves the feasibility of applying mutual information based approaches for reconstructing large regulatory networks. Keywords: mutual information, gene regulatory network, microarray. 1