Multicategory Classification using Extended SVM-RFE and Markov Blanket on SELDI-TOF Mass Spectrometry Data

Surface-enhanced laser desorption/ionization time-of-flight (SELDI-TOF) mass spectrometry data has been increasingly analyzed for identifying biomarkers for disease to help early detection of the disease. Recently, support vector machine (SVM) algorithm based on recursive feature elimination (RFE) was proposed to find a set of genes for cancer classification. In our study, we extend the SVM-RFE such that it can be used in the multicategory classification work using SELDI-TOF mass spectrometry data and propose a new feature selection algorithm (SVM-MB/RFE : SVM-Markov Blanket/Recursive Feature Elimination). In the preprocessing task of SVM-MB/RFE, ANOVA (Analysis of Variance) and binning methods are used for feature filtering. We demonstrate that the performance is improved through the preprocessing work. Compared with other methods such as not only SVM-RFE and Markov blanket but also PCA (Principle Components Analysis)+LDA (Linear Discriminant Analysis) and other feature selection algorithms, SVM-MB/RFE performs better than them.