Classifying Protein-Protein Interaction Type based on Association Pattern with Adjusted Support

Abstract

Proteins carry out their functions by means of interaction. There are two major types of protein-protein interaction (PPI): obligate interaction and transient interaction. In this paper, residues with geographical information on the binding sites are used to discover association patterns for classifying protein interaction type. We use the support of a frequent pattern as its inference power. However, due to the number of transient examples are much less than the number of obligate examples, therefore there needs adjustment on the imbalance. Three methods of applying association pattern to classify PPI type are designed. In the experiment, there are almost same results for three methods. And we reduce effect which is correct rate decreased by data type imbalance.