Novel Ensemble Predictor for Gram-positive Bacterial Protein Sequences

in the fields of bioinformatics and drug discovery, prediction of bacterial proteins is an important research. We proposed a novel classifier ensemble scheme for the prediction of Gram-positive bacterial protein. The proposed methodology exploits diversity in decision space through different machine learning techniques. However, diversity in feature spaces is exploited using physicochemical properties of amino acids compounds. Further, three different classification models are employed to further explore diversity in decision space. First individual ensemble (IE) predictors are developed using a single feature extraction technique. Their results are then combined to develop improved performance composite ensemble (CE) for Gram-positive bacterial protein sequences. The predictive performance of IE and CE predictors are evaluated for two standard datasets. We investigated the prediction results of our novel scheme for this problem is better than any previous approach so far to the best of our knowledge.