home · mobile · calendar · colloquia · 2004-2005 · 

Colloquium - Goldberg

Network Topology Reveals Biological Hypotheses
Harvard Medical School

Network analysis has become an indispensable tool for understanding complex phenomena as diverse as social relationships, the internet, and chemical reactions in a living cell. Studies have shown that diverse real networks share large-scale topological properties. In particular, networks of genes or gene products (proteins) that are systematically determined using high-throughput experimental methods have some well-studied properties. These inaccurate biomolecular networks are increasingly being used to predict protein function, a task that is critical for finding new drug targets and for gaining a systems-level understanding of any organism. My recent work has involved exploiting the network topology to make specific inferences about the objects and relationships represented in the network.

In this talk, I will show how to predict protein interactions and compute accurate confidence assessments of observed interactions by adapting a measure of small-world-topology. I will present several methods that integrate distinct types of data with topological information to predict protein function. It is generally believed that these networks have distinct types of edges (e.g., true and false positive edges), and I will show evidence for such multiple edge-types and characterize their distributions using a multinomial perspective and model-selection techniques. This is joint work with Fritz Roth.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)