Support vector machines (SVMs) are becoming a standard tool in bioinformatics, where SVM is a kind of machine learning algorithm developed in 1990's. Indeed, SVMs have been applied to many important problems in bioinformatics, which include motif detection, protein secondary structure prediction, fold recognition, protein subcellular localization prediction, prediction of protein-protein interaction, and microarray data analysis. There are two phases for using an SVM: training and prediction. In thetraining phase, training data are given as points in the Euclidian space where each point corresponds to an object (e.g., a protein sequence) and is labeled as either positive or negative (see Fig.1). Then, a hyperplane (e.g., line in 2-diemnsional caseand plane in 3-dimensional case) separating positive training data and negative training data in some optimal way is computed. In the prediction phase, each object is also given as a point without label. If the point lies above the hyperplane, this object is predicted as positive. Otherwise, it is predicted as negative. Though an SVM is basically capable of binary classification only, multiple classes can be handled by combining multiple SVMs. One of the important points is that several efficient SVM programs are available free-of-charge at least for academic use and these are very easy to use.
展开▼
