stanford.seal64.gif (1,768 bytes)

Broad Area Colloquium for Artificial Intelligence,
Geometry, Graphics, Robotics and Vision


Analyzing Gene Expression Patterns with Probabilistic Graphical Models

Nir Friedman
Hebrew University, Jerusalem, Israel

Monday, October 1, 2001, 4:15PM
Gates B01
http://robotics.stanford.edu/ba-colloquium/

Abstract

In recent years, biological and medical research is undergoing a major revolution. High-throughput methods allow to measure and record large amounts of data. In particular, microarray-based hybridization methods techniques allow to simultaneously measure the expression level of thousands of genes. It is clear that such measurements contain information about many different aspects of gene regulation and function.

The computational challenge is to extract new biological understanding from this wealth of data. In particular, to gain insight on the causal "structure" of the interactions between components in the systems, such as genes, proteins, and external signals. In this talk, I will describe an ongoing project that addresses this question using probabilistic graphical models. These models represent the measured expression level of each gene as a random variable and each regulatory interaction as a probabilistic dependency between these variables. In this project, we developed tools for automatic induction of networks from gene expression data, and more importantly, for analysis of the confidence in different features of these networks.

This analysis allows us to examine various aspects of gene interaction: In the interaction between genes direct, or mediated by other genes? Is there a causal direction in the interaction? Are there common factors that might determine co-expression? Although these questions cannot, in general, be completely determined by gene expression profiles alone, we show that we can construct partial answers. We demonstrate the utility of this approach in analysis of yeast expression data sets. In particular, our method manages to reconstruct fragments of known pathways from such expression profiles.

This is joint work with Dana Pe'er, Iftach Nachman, Michal Linial, Gal Elidan, and Aviv Regev.

About the Speaker

Nir Friedman received a Ph.D. in Computer Science from Stanford in 1997. He was a postdoctoral scholar in the Computer Science Division at the University of California, Berkeley till late 1998, and he is currently a faculty member in the School of Computer Science and Engineering at the Hebrew University, Jerusalem. His main research interest is learning and inference with probabilistic graphical models. In recent years, his work focuses on applications of these tools to computational biology.


Contact: bac-coordinators@cs.stanford.edu

Back to the Colloquium Page