An ensemble learning approach for prediction of phosphorylation sites

Protein phosphorylation plays a fundamental role in most of the cellular regulatory pathways. Experimental identification of phosphorylation sites is labour-intensive and often limited by the availability and optimisation of enzymatic reaction. An ensemble learning approach that combines different encodings using a meta-learner was developed which was catalyzed by four protein kinase families and three residues. A predictor is constructed to predict the true and false phosphorylation sites based on Support Vector Machines (SVM), and knowledge based encoding method is used for amino sequences. Different encoding methods catch different aspects of amino sequences feature. The stacking SVM approach was applied to combine these aspects and improved both sensitivity and specificity.