Skip to Main Content
In order to deal with the high-dimensional and small sample size problem of proteomic peptide profile, a feature selection method, weighted maximum margin criterion (WMMC), was proposed in this paper. The profile was firstly preprocessed through iterative minimum in adaptive setting window (IMASW) method for baseline correction, discrete wavelet transform (DWT) for filtering the spectral background and noise, and univariate significance analysis (SA) for ordering the variables according to their ability to discriminate between classes. Then, WMMC and support vector machines (SVM) technique were used to build prediction model. With an optimization of the parameters involved in the modeling, a satisfactory model was achieved for ovarian cancer diagnosis in two proteomic peptide profile datasets. The results show that WMMC method is more efficient than principal component analysis (PCA), independent component analysis (ICA) and locally linear embedding (LLE) for classification.