Bayesian Approaches to Improve Metabolic Networks
Mindy Xinghua Shi
Computer Science Department
University of Chicago

April 10 , 2007 11:00 AM
Van Zoeren 142

With the rapid availability of hundreds of sequenced genomes, the
reconstruction of genome-scale metabolic networks for these organisms
has attracted much attention. A metabolic network can be viewed as a
directed graph with nodes representing metabolites and edges
representing biochemical reactions catalyzed by enzymes that are
encoded by corresponding genes. Although current genome/pathway
databases provide a large proportion of metabolic information that can
be used directly to reconstruct metabolic networks, networks
reconstructed this way turn out to be incomplete. The incomplete
metabolic networks contain network holes when there are certain
metabolites that cannot be produced or consumed. Efforts have been
carried out to fill these network holes by introducing reasonable
reactions into networks. However, most of these efforts require a big
amount of manual work. With the availability of hundreds of complete
genomes, it is desirable to automate this process using computational
techniques.

Towards this direction, we propose a set of Bayesian
approaches that integrate multiple topological and biological
evidences from different databases to fill network holes and improve
metabolic model reconstructions. Seven individual Bayesian-based
predictors are built by combining various network topological and
biological evidences. These evidences come from features extracted out
of the published model repository, the BIGG, and two genome/pathway
databases, the SEED and KEGG. After individual predictors are trained
on observed data, three mechanisms including Majority Vote, Naïve
Bayes Classifiers and Bayesian Networks are utilized to integrate
individual predictors and produce unified predictions based on results
of the individual predictors.

Colloquium Schedule