Improving Identification of Essential Proteins by a Novel Ensemble Method

Essential proteins are indispensable for cell survival, and the identification of essential proteins plays a critical role in biological and pharmaceutical design research. Recently, some machine learning methods have been proposed by introducing effective protein features or by employing powerful classifiers. Seldom of them focused on improving the prediction accuracy by designing efficient strategies to ensemble different classifiers. In this work, a novel ensemble learning framework called by Tri-ensemble was proposed to integrate different classifiers, which selected three weak classifiers and trained these classifiers by continually adding the samples that are predicted to have abnormally high or abnormally low properties by the other two classifiers. We applied Tri-ensemble on predicting the essential protein of Yeast and E.coli. The results show that our approach achieves better performance than both individual classifiers and the other ensemble learning methods.