Tuesday, January 29, 2013
4:00 PM - 5:00 PM
Where: Technological Institute, M228, 2145 Sheridan Road, Evanston, IL 60208 map it
Audience: Faculty/Staff - Student - Public
Category: Lectures & Meetings
Virginia Commonwealth University
Abstract: This talk includes the development of a method for L1-norm principal component analysis (PCA), a description of its application to human microbiome data, and the results of an experiment to understand bias in microbiome experiments. Traditional principal component analysis (PCA) can be viewed as minimizing the sum of L2-norm distances of points to their projections on a subspace. In the interest of increasing resistance to outlier observations, several investigators have introduced the L1-norm in the optimization problem. A novel L1-norm PCA method is proposed that uses the global optimal solution of a nonlinear optimization problem obtained via a small number of linear programs. The human microbiome is the community of microbes that populate habitats on and around various body sites. Advances in high-throughput sequencing facilitate investigations into the impact of the microbiome on physiology and disease. The processing of microbiome samples introduces bias so that the relative quantities of bacteria that are observed are a distortion of the true population composition. We demonstrate that designed experiments and computational modeling identify sources of bias and quantify their impact. The models may be applied to clinical samples to recover the true proportions of bacteria, facilitating the testing of more specific hypotheses about bacterial community composition and its relationship to disease.
Biography: Paul Brooks is an Associate Professor of Operations Research and Fellow of the Center for the Study of Biological Complexity at Virginia Commonwealth University. He holds a Ph.D. in Operations Research from Georgia Tech. His interests are in applications of optimization to the development of novel data analysis algorithms and to medicine and bioengineering. He is currently Chair of the INFORMS Section on Data Mining. His research is published in Operations Research, Nature, Networks, Interfaces, BMC Systems Biology, BMC Genomics, and Computational Statistics and Data Analysis, and is supported by the NIH and NASA.