Detecting Perturbed Biological Pathways Through Latent Network Modeling of Gene Expression

• Activity Number: 218
• Type: Invited
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: IMS
• Author(s): Eric Kolaczyk*+ and Lisa Pham and Luis E. Carvalho and Scott E. Schaus
• Companies: Boston University and Boston University and Boston University and Boston University
• Keywords: Conditional auto-regressive model ; Latent factor model ; Biological networks

Abstract

Proteins interact with each other to form biological pathways that perform numerous different tasks in the cell. These pathways, in turn, work together to achieve cell homeostasis, through a network of functional relationships. Here our goal is the identification of pathways that are perturbed, due to disease or drug perturbation, given only high-throughput gene expression measurements and information from biological databases. Approaching this task as one of statistical modeling and inference, we develop a two-level model, consisting of (i) a confirmatory latent factor model that captures the relationship between gene expression and biological pathways, and (ii) a simultaneous equation model of the behavior within an underlying network of pathways induced by an unknown perturbation. Detection of a perturbed pathway(s) is accomplished through statistical inference on latent variables representing perturbation targets, using principles of classical Bayesian analysis. We describe both the modeling framework and certain mathematical and computational challenges it poses. We illustrate using simulation studies and perturbation data from the DREAM 7 challenge.