Prediction Accuracy Evaluation of Domain and Domain Combination Based Prediction Methods for Protein-Protein Interaction

Since the proposal of domain based protein-protein interaction prediction method by [3], there are many attempts to improve domain based protein-protein interaction prediction methods [1]. Among them, domain combination based protein-protein interaction method by [2] is quite appealing in the sense that it achieves very impressive prediction accuracy in some situations. However there is no concrete evidence that domain combination based method generally achieves better prediction accuracy than domain based approaches and the reason is not clearly understood yet. This paper compares domain combination based protein-protein interaction prediction method with domain based protein-protein interaction method. The only difference of the methods is that domain based method treats a domain pair as a basic unit in protein interactions, whereas domain combination based method treats a domain combination pair as a basic unit in protein interactions. In that sense, the comparison can be considered as impartial and thus we can conclude which method is more competitive in achieving high prediction accuracy from the comparison. According to the comparison results, domain combination based prediction method achieved superior prediction accuracy to domain based prediction method in general. Several significant issues are also discussed from the in-depth analysis of the results. For example, the results back up the conjecture that a domain-domain interaction may be influenced by surrounding domains, is close to the truth. The description of such several significant facts revealed from the comparative studies is also included in this paper.