One-Class Classifier to Predict Protein-Protein Interactions based on Hydrophobibity Properties

Abstract

Protein-protein interactions are important in a wide range of biological processes. The development of drugs that target such interactions is a very active research field. Hence predicting protein-protein interactions represent an important challenge in bioinformatics research. Machine learning techniques have been applied to predict protein-protein interactions. Most of these techniques address this problem as a binary classification problem. While it is easy to get a dataset of interacting protein as positive example, there are no experimentally confirmed noninteracting proteins to be considered as a negative set. Therefore, in this paper we solve this problem as a one-class classification problem using One-Class SVM (OCSVM). The hydrophobicity properties have been used in this research as the protein sequence feature. Using only positive examples (interacting protein pairs) for training, the OCSVM achieves accuracy of 72% using RBF kernel. These results imply that protein-protein interaction can be predicted using oneclass classifier with reliable accuracy.