The Support Vector Machine has been introduced and shown to be well suited to a familiar machine learning application in pharmaceutical drug discovery. The theoretical advantages that the SVM brings to machine learning in drug discovery have been outlined and a specific area of the discovery process (SAR analysis) discussed. In a trial on data provided by GlaxoSmithKline Pharmaceuticals, the SVM outperformed four frequently used techniques, showing especially high accuracy in classifying the more important of two compound classes. Future work on this subject will examine methods of improving SVM performance on the negative class without affecting its high accuracy on the positive class.
展开▼
