Gene clustering is based on a recursive, heuristic implementation of the Gene Recommender (GR) [26], automated for genome-wide application. GR is a rank-based biclustering algorithm with an existing, open source R implementation [28]. Here we outline relevant portions of the GR. First, gene expression data are normalized to a uniform distribution with mean zero and variance 1/3. A suggested gene list, based on prior knowledge, of at least two putative co-regulated genes is input to GR. For each putative cluster input, each sample's relative score, the "ZE(j)", is computed, as in Equation 1.
