Relevance networks was the simplest method used in our study, but it performed quite well given its simplicity. An implementation of the algorithm is available on the RELNET website. We used matlab however for our implementation. In matlab (or any other statistical software package), you need only generated the gene-gene pairwise matrix of dependencies (using any statistical metric of dependency, but we found correlation to perform best). Then to build a network, you just decide what is the minimum dependency value to keep, and you remove all edges below your threshold.