Inferring Protein Interaction Network by Boosting Algorithm

One of major goals of functional genomics is to elucidate protein interaction networks for whole organisms. Determining protein interactions provides not only detailed functional insights on characterized proteins, but also an information base for identifying biological complexes and metabolic or signal transduction pathways [1]. The recent emergence of high-throughput proteomics techniques has opened new prospects to systematically characterize physical interactions between proteins. Based on experimental dataset, many computational algorithms have been developed to infer the protein-protein or domain-domain interactions. For instance, for inferring protein interactions, there are the gene fusion (Rosetta Stone) method, the phylogenetic profile method, the interaction domain pair profile method, the probabilistic method, the SVM-based method, and the LP-based approach, whereas for inferring domain interactions, there are the association method, the EM algorithm. Despite the relative success, there is much room for improvement of protein interaction inference in terms of prediction quality and computational efficiency. Based on the association method, we propose an association probabilistic method (APM) to infer protein interactions directly from the experimental data, and then further improve the accuracy of APM [1] by adopting boosting algorithm. By the numerical simulation, we show that the proposed method achieves the highest accuracy among the existing approaches for the measures of root mean square error and the Pearson correlation coefficient with the efficiency.