Improving gene expression data interpretation by finding latent factors that co-regulate gene modules with clinical factors

In the analysis of high-throughput data with a clinical outcome, researchers mostly focus on genes/proteins that show first-order relations with the clinical outcome. While this approach yields biomarkers and biological mechanisms that are easily interpretable, it may miss information that is important to the understanding of disease mechanism and/or treatment response. Here we test the hypothesis that unobserved factors can be mobilized by the living system to coordinate the response to the clinical factors.

We developed a computational method named Guided Latent Factor Discovery (GLFD) to identify hidden factors that act in combination with the observed clinical factors to control gene modules. In simulation studies, the method recovered masked factors effectively. Using real microarray data, we demonstrate that the method identifies latent factors that are biologically relevant, and extracts more information than analyzing only the first-order response to the clinical outcome.

Finding latent factors using GLFD brings extra insight into the mechanisms of the disease/drug response. The R code of the method is available at http://userwww.service.emory.edu/~tyu8/GLFD.