University of Connecticut

Events Calendar

CCAM/ISG Joint Seminar - Brown

Thursday, October 2, 2014
4:00pm – 5:00pm

UConn Health
CGSB, CCAM Conference Room, R1673, 400 Farmington Ave

Center for Cell Analysis & Modeling/Institute for Systems Genomics Joint Seminar

Speaker: Kevin Brown, Assistant Professor, Biomedical Engineering/Physics/Chemical and Biomolecular Engineering/Marine Sciences Institute for Systems Genomics University of Connecticut

Title: "The Statistics of Statistical Mechanics of Sloppy Models"

Abstract: Scientists are often forced to construct models of natural systems from grossly incomplete information, as in cellular signal transduction, oceanic biogeochemistry, and terrestrial nutrient cycling. Dynamical modeling of such complex systems presents four key challenges: (i) the presence of poorly known parameters, (ii) the necessity of using simplified dynamics, (iii) uncertain model connectivity, and (iv) limited data availability. Can such models even be predictive, and if so, how do we use them to properly infer the results of perturbations to the system under study? I designate problems with these four challenges sloppy models. While sloppiness is ubiquitous in physical, chemical, and biological systems, not all models are sloppy. Sloppy models form a well-defined class of problem which tends to arise in nonlinear systems with weak prior parameter constraints, and the resulting model space geometry has important implications for understanding such systems. By treating models as ensemble objects, in which all sets of parameters consistent with the available experimental data are considered when making predictions, I can generate confidence intervals on model predictions and yield robust, falsifiable models. These kind of Monte Carlo calculations can be directly extended to both model selection, in which one wants to find the “best” model from among a set of candidate models, and optimal experimental design, in which the goal is to acquire additional experimental data in order to constrain a model prediction to have minimum variance for fixed experimental effort. I will discuss all these issues with repeated reference to two specific models: one from cancer biology and another from estuarine biogeochemistry.

Hosts: Dr. Ion Moraru & Dr. Marc Lalande

Contact:

Tiffany Gough. tgough@uchc.edu

Center for Cell Analysis and Modeling (primary)

Control Panel