Discovery of Perturbations in Multi-Attribute Biological Networks

• Activity Number: 569
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Paula J. Griffin*+ and Eric Kolaczyk
• Companies: Boston University School of Public Health and Boston University
• Keywords: biological networks ; Gaussian graphical models ; penalized regression ; sparse estimation

Abstract

As a greater variety of high-throughput biological data becomes available, demand is increasing for statistical methods that integrate multiple data types. To infer mechanism-of-action in biological systems, there is the added complication that small perturbations can have wide-ranging downstream effects. Given a snapshot of cellular activity, it can be difficult to tell where a disturbance originated. We approach this problem by extending network filtering to multi-attribute data. We first estimate a joint Gaussian graphical model across multiple data types using penalized regression, then filter for network effects to identify the site of the original perturbation. We discuss applications to disease studies and drug-targeting experiments using mRNA microarrays and reverse-phase protein arrays.