The Statistics Seminar Speaker for Wednesday, May 9, 2018, will be George Michailidis, professor and director of the Informatics Institute at the University of Florida. George Michailidis received his Ph.D. in mathematics from UCLA and then did a postdoc in operations research at Stanford University. Subsequently, he spent 17 years in the faculty of the University of Michigan before joining the University of Florida in 2015 as the director of the Informatics Institute. He served as editor of the Electronic Journal of Statistics and has also served on the editorial boards of JASA, JCGS, Technometrics, JSPI and J Nonparametrics Statistics. He is a fellow of IMS, ASA and ISI. His research interests are in the areas of modeling, inference and analysis of high-dimensional data with network structure, change point problems, stochastic control, optimization and their applications to biology, smart grid and Internet traffic, and economic/finance problems. Further, he has mentored to date 43 doctoral students and 10 postdocs.
Talk: Graphical Modeling based Integrative Analysis of Metabolomics Data
Abstract: Recent technological advances in mass spectrometry, development of richer mass spectral libraries and data processing tools have enabled large scale metabolic profiling that provides a direct functional readout of cellular activity and physiological status. However, biological interpretation of metabolomics studies heavily relies on knowledge-based tools that contain information about metabolic pathways. Incomplete coverage of different areas of metabolism and lack of information about non-canonical connections between metabolites limits the scope of applications of such tools. Data driven approaches for building interactions networks from metabolomics data have proved useful.
In this talk, we propose an integrative framework based on Gaussian graphical models for joint estimation of such interaction networks from metabolomics profiles across multiple experimental conditions or disease subtypes. We discuss in detail modeling, estimation and statistical inference aspects, and provide extensions of the framework for integrating metabolomics data with other Omics modalities (e.g. transcriptomics data, etc.). We further show how these interaction networks can be used in combination with topology based pathway enrichment methods to identify differentially enriched subnetworks across disease subtypes. The methodology is illustrated on three different independent cohorts of patients with early and late stage chronic kidney disease.