Feature cohesion and coupling based object-object matching prediction method

Abstract

Object-object matching is one of the most frequently observed events in nature. A marriage is a matching between a man and a woman to form a family, and a protein-protein interaction is a matching between two proteins to achieve a specific function. Although matching problems in distinctive domains look different, they usually have several points in common. Since we should undergo many trials of unsuccessful matchmakings before a matching is made, it will be profitable if we can predict if a matching would be successfully made or not before the matchmaking is tried. The recent accumulation of object-object matching and its associated feature data has allowed us to develop methods and techniques to predict object-object matching based on the accumulated data. In this paper, we propose a general computational framework to predict object-object matching based both on intra-object feature cohesion and inter-object feature coupling information. Unlike conventional methods, the framework incorporates intra-object feature cohesion information additionally in the prediction. In addition, we introduced application of object matching prediction like protein interaction, match-making, recommend attending doctor and wine-food marriage. For the validation of the framework, we developed a protein-protein interaction prediction and primary interacting feature combination pair prediction methods within the framework. According to our evaluation, the framework was effective in developing a protein-protein interaction prediction method, which showed comparable or superior prediction accuracy to the conventional protein-protein interaction prediction methods. In primary interacting domain or domain combination pair prediction, among 253 primary interacting domain combination pairs reported in PDB, 163 domain pairs and 47 domain combination pairs were correctly predicted by the method.