Prediction of Protein Essentiality by the Support Vector Machine with Statistical Tests

Essential proteins affect the cellular life deeply, but it is extreme time-consuming and labor-intensive to discriminate them experimentally. The goal of this paper is to identify the features which are crucial for discriminating protein essentiality and build learning machines for prediction. We first collect features from a variety of sources. Then we adopt a backward feature selection method and use the selected features to build SVM predictors. The cross validations are conducted on the originally imbalanced data set as well as the down-sampling balanced data set. The performance of these feature subsets are then subject to the statistical test to confirm their significance. For the imbalanced data set, our best values of F-measure and MCC are 0.549 and 0.495, respectively. For balanced data set, our best values of F-measure and MCC of our models are 0.770 and 0.545, respectively. The results are superior to all previous results under various performance measures.