Predicting proteins interactions from protein sequence features using support vector machines

Abstract

Computational methods to predict protein-protein interactions are becoming increasingly important. This is due to the fact that most of the interactions data have been identified by high-throughput technologies like the yeast two-hybrid system which are known to yield many false positives. In this paper we investigate the use of two protein sequence features, namely, domain structure and hydrophobicity properties. The support vector machines (SVM) has been used as a learning system to predict protein interactions based only on protein sequence features. Protein domain structure and hydrophobicity properties are used separately as the sequence feature. Both features achieved accuracy of about 80%. But domains structure had receiver operating characteristic (ROC) score of 0.8480, while hydrophobicity had ROC score of 0.8159. These results indicate that protein-protein interaction can be predicted from domain structure with relatively better accuracy than hydrophobicity.