Mimosa: Mixture model of co-expression to detect modulators of regulatory interaction

Here we present a novel mixture modeling approach where a TF-Gene pair is presumed to be significantly correlated (with unknown coefficient) in an (unknown) subset of expression samples. The parameters of the model are estimated using a Maximum Likelihood approach. The estimated mixture of expression samples is then mined to identify genes potentially modulating the TF-Gene interaction. We have validated our approach using synthetic data and on four biological cases in cow, yeast, and humans.While limited in some ways, as discussed, the work represents a novel approach to mine expression data and detect potential modulators of regulatory interactions.Eukaryotic gene regulation is carried out, to a significant extent, at the level of transcription. Many functionally related genes, e.g., members of a pathway, involved in the same biological process, or whose products physically interact, tend to have similar expression patterns [1,2]. Indeed, co-expression has been used extensively to infer functional relatedness [3-6]. Various metrics have been proposed to quantify the correlated expression, such as Pearson and Spearman correlation [2], and mutual information [5]. However, these metrics are symmetric and they neither provide the causality relationships nor do they discriminate between indirect relations. For instance, two co-expressed genes may be co-regulated, or one may regulate the other, directly or indirectly.A critical component of transcription regulation relies on sequence-specific binding of transcription factor (TF) proteins to short DNA sites in the relative vicinity of the target gene [7]. If one of the genes in a pairwise analysis of co-expression is a TF then the causality is generally assumed to be directed from the TF to the other gene. In the absence of such information, an additional post-processing step [5] can be used to infer directionality between the pair of genes with correlated expression. Moreover, a first order conditional independence metr