Using Sequence Homology to Filter High-throughput Protein-protein Interaction Data

Protein-protein interaction data obtained from high-throughput experiments is thought to have a large number of false positives i.e. interactions that are spurious and do not occur in the cell. This fraction is estimated to be as high as 50% in yeast [3]. Hence it is important to quantitate the reliability of these interactions and identify those that are true positives i.e. those that actually occur in the cell. In this study, we show that an interaction can be judged as true if the interacting proteins have homologs that interact in one or more species. We used protein-protein interaction data for E. coli, S. cerevisiae, C. elegans, D. melanogaster, and H. sapiens from the Database of Interacting Proteins (DIP) [2], February 2004 version. PSIBLAST was used to detect the sequence homologs. Then, we proved our hypothesis first in yeast by estimating likelihood ratios for high-throughput interactions with and without homologs based on the number of known true positives and false positives using Bayesian approaches. Based on these results, we estimate the number of true positives in the high-throughput data sets of S. cerevisiae, C. elegans and D. melanogaster. We have also created a Database of Homologous Interactions (http://www.pdbj.org/dhi) to display our results and allow access to the homologous interactions studied.