|Kernel Methods for Predicting and Filtering Protein-Protein Interaction Data
Department of Computer Science
Colorado State University
Monday, September 12, 2005
E203 Engineering Building
Most proteins perform their function by interacting with other proteins. Therefore, information about the network of interactions that occur in a cell can greatly increase our understanding of protein function. Experimental assays that probe interaction networks on a large scale are now available; and yet, interaction networks of even well studied organisms are still sketchy at best, and a need for computational methods for predicting novel interactions still exists. Moreover, the experimental data is highly noisy, so methods for assigning confidence levels to such data are also required.
We present kernel methods for both tasks that use a variety of data sources, including protein sequences, annotations of protein function, local properties of the network, and interactions in different species.
A classifier trained to predict interactions in yeast retrieves close to 80% of a set of trusted interactions at a false positive rate of 1%, demonstrating the ability of our method to make accurate predictions despite the sizable fraction of false positives that are known to exist in interaction databases.
[Refreshments will be served at 3:45 p.m. in Room 008 of the Statistics Building]