On Mining Biological Signals Using Correlation Networks

Correlation networks have been used in biological networks to analyze and model high-throughput biological data, such as gene expression from micro array or RNA-seq assays. Typically in biological network modeling, structures can be mined from these networks that represent biological functions, for example, a cluster of proteins in an interactome can represent a protein complex. In correlation networks built from high-throughput gene expression data, it has often been speculated or even assumed that clusters represent sets of genes that are co-regulated. This research aims to validate this concept using network systems biology and data mining by identification of correlation network clusters via multiple clustering approaches and cross-validation of regulatory elements in these clusters via motif finding software. The results show that the majority (81-100%) of genes in any given cluster will share at least one predicted transcription factor binding site. With this in mind, new regulatory relationships can be proposed using known transcription factors and their binding sites by integrating regulatory information and the network model itself.