Integrated Reliability-Assessment for Protein-Protein Interaction Data

Interaction among genes and gene products has a major role in many cellular processes. Obtaining an accurate picture of protein-protein interactions is one of the major challenges to the systematic understanding of cellular activity. While there are reliable but time-consuming experimental techniques to determine interactions, there have been developed a variety of high-throughput methods to obtain vast amounts of protein-protein interaction data at a time. With accumulation of high throughput protein-protein interaction data, evaluation of these data becomes of importance. Previous studies indicate that local connectivity pattern of interaction data can be used for measuring the reliability of individual interaction [3]. However, these approaches have one major drawback: they are excessively dependent upon connectivity patterns of interaction data which contain false negatives as well as false positives. Here we propose an integrated approach for reliability assessment of protein-protein interaction data from yeast two-hybrid system. Our approach uses both local connectivity pattern of interaction data and other information of each protein as attributes of individual interaction. Information of proteins includes cellular localization and function. Using these attributes, we employ decision tree as machine learning architecture. For evaluating this method, the yeast two-hybrid interaction data of Saccharomyces cerevisiae by Ito et al. (full data: 4549 interactions) [2] is used as an evaluation target and DIP-core is used as golden standard.