Skip to Main Content
Protein--protein interaction (PPI) is a major way for proteins to perform their biological functions. Due to the lack of the PPI experimental data, theoretical prediction seems important to understand protein functions. Up to now, there have been various computational methods as for how to predict PPI. This report, however, will present a novel approach to the prediction of PPI by analyzing protein secondary structures. In the model, a support vector machine (SVM) was trained through a positive data set and a negative data set, each of which contains 7,714 protein pairs involving 1,730 proteins. To select PPI pairs in the training negative data sets, a new parameter that describes protein interaction relative bias (PIRB) was introduced as a measure of PPI propensity. The prediction accuracy was 88.01% when the model was employed to predict PPI in Saccharomyces cerevisiae.