Predicting Protein-Protein Interactions for the Budding Yeast using Inductive Logic Programming and Multiple Sources

摘要

The interaction between proteins is fundamental to a broad spectrum of biological functions, including regulation of metabolic pathways, immunologic recognition, DNA replication, progression through the cell cycle, and protein synthesis. The function of a protein is also affected by the other biomolecules with which it interacts and reacts. For these reasons, an enormous amount of protein-protein interaction data have been obtained recently for yeast and other organisms using high-throughput experimental approaches such as two-hybrid, mass spectrometry, phase-display. However, a potential difficulty with these kinds of data is a prevalence of false positive (interactions that are seen in an experiment but never occur in the cell or are not physiologically relevant) and false negatives (interactions that are not detected but do occur in the cell). The objective of this work is to apply Inductive Logic Programming to predicting protein-protein interactions with high confidence for the budding yeast Saccharomyces cerevisiae. We used the Yeast Interacting Proteins Database provided by T.Ito, Tokyo University as training examples. Among 4,549 interactions detected using yeast-hybrid analysis, we used the "core" data including 841 interactions with more than two IST hits as positive examples, an the remainder as negative examples. Various kinds of background knowledge have been used by either extracting from protein databases or using computational approaches. Early results indicate that using Inductive Logic Programming is a promising approach for dealing with protein-protein network analysis, since it is able to obtain comprehensibly descriptive rules and the accuracy obtained using ten-fold cross-validation is rather high (nearly 80%).