Sequence-driven features for prediction of subcellular localization of proteins

Abstract

Prediction of the cellular location of a protein plays an important role in inferring the function of the protein. Feature extraction is a critical part in prediction systems, requiring raw sequence data to be transformed into appropriate numerical feature vectors while minimizing information loss. In this paper, we present a method for extracting useful features from protein sequence data. The method employs local and global pairwise sequence alignment scores as well as composition-based features. Five different features are used for training support vector machines (SVMs) separately and a weighted majority voting makes a final decision. The overall prediction accuracy evaluated by the 5-fold cross-validation reached 88.53% for the eukaryotic animal data set. Comparing the prediction accuracy of various feature extraction methods, provides a biological insight into the location of targeting information. Our experimental results confirm that our feature extraction methods are very useful for predicting subcellular localization of proteins.