earticle

영어

Predicting subcellular localizations of human proteins become crucial, when new unknown proteins sequences do not have significant homology to proteins of known subcellular locations. In this paper, we present a novel approach to develop CE-Hum-PLoc system. Individual classifiers are created by selecting a fixed learning algorithm from a pool of base learners and then trained by varying feature dimensions of Amphiphilic Pseudo Amino Acid Composition. The output of combined ensemble is obtained by fusing the predictions of individual classifiers. Our approach is based on the utilization of diversity in feature and decision spaces. As a demonstration, the predictive performance was evaluated for a benchmark dataset of 12 human proteins subcellular locations. The overall accuracies reach upto 80.83% and 86.69% in jackknife and independent dataset tests, respectively. Our method has given an improved prediction as compared to existing methods for this dataset. Our CEHum-PLoc system can also be a used as a useful tool for prediction of other subcellular locations.