Swarthmore College Department of Computer Science

Talk by Ameet Soni, University of Wisconsin Department of Computer Science

Probabilistic Techniques for Protein-Structure Determination
Tuesday, February 15, 2011
SCI 240, 4:00 pm (refreshments at 3:45)

Abstract

The field of artificial intelligence (AI) aims to develop agents (or machines) that receive observations from the world in order to make decisions and perform actions. Unfortunately, it is rare for agents to decide with complete certainty; the observations may be noisy or incomplete, or the consequences of decisions may be non-deterministic. One example is the task of determining protein structures given low-quality images produced via X-ray crystallography. Proteins are essential to almost all cellular functions in an organism, and understanding their structure is key to problems such as disease diagnosis and drug design. The noisy/low feature images and the dynamic process by which proteins form their structure make this a computationally demanding task that requires techniques for handling uncertainty. In my work, I utilize a probabilistic graphical model known as a Markov Random Field (MRF) to compactly represent a protein structure. MRFs allow us to combine various pieces of (uncertain) information - such as evidence from the image as well as biochemical constraints - to provide a probabilistic representation of a protein structure.

The first part of my talk will concentrate on our technique, Automated Crystallographic Map Interpretation (ACMI), which outperforms competing methods in determining protein structures from low-quality images. In the second part of the talk, I will concentrate on my contributions to the problem of probabilistic inference - the process by which we draw conclusions from probabilistic models. Inference is a major area of interest in the artificial intelligence community as we deal with increasingly complex modeling problems similar to ACMI. The development of improved inference methods has implications in many areas of AI including natural language processing, computer vision, and statistical relational learning.