Prediction of protein subcellular localization using machine learning with novel use of generic feature set

The method of identifying the location of protein within a cell is called subcellular localization of proteins. This area of research in Bioinformatics is pivotal for protein synthesis and drug discovery of several medical conditions and diseases. This paper introduces a new machine learning approach for subcellular localization of proteins, which used 18 basic and physicochemical features novel for such methods. A model with support vector machine (SVM) was developed at first to learn these properties of proteins from 6 locations inside a cell, and then test the model on another independent set of protein sequences. The proposed multi-class classification algorithm achieved an accuracy of about 94%. The results show superior performance with minimal computations when compared to similar algorithms in the literature.