Granular decision fusion systems for effective protein methylation pPrediction

Protein methylation is one important type of post-translational modifications of proteins. Experimentally identifying methylation positions in protein sequences is time-consuming and costly. In order to provide insightful advice and reduce cost for further experiments, we propose a novel granular decision fusion framework based on granular computing, computational intelligence, and statistical learning. Algorithms are designed under this framework to predict methylation sites. Since methylation sites rarely appeared, the known data are imbalanced. Sampling and clustering is used to create different sub-sets and represent them with cluster centers. Support vector machine (SVM) classifiers are built for these sub datasets. Finally, granular decisions are fused to determine possible methylation sites. Simulation results show that the new granular decision fusion system has high prediction accuracy.