Gene module identification from microarray data using nonnegative independent component analysis

Genes mostly interact with each other to form transcriptional modules for performing single or multiple functions. It is important to unravel such transcriptional modules and to determine how disturbances in them may lead to disease. Here, we propose a non-negative independent component analysis (nICA) approach for transcriptional module discovery. nICA method utilizes the non-negativity constraint to enforce the independence of biological processes within the participated genes. In such, nICA decomposes the observed gene expression into positive independent components, which fits better to the reality of corresponding putative biological processes. In conjunction with nICA modeling, visual statistical data analyzer (VISDA) is applied to group genes into modules in latent variable space. We demonstrate the usefulness of the approach through the identification of composite modules from yeast data and the discovery of pathway modules in muscle regeneration.