Classification Analysis of Surface-enhanced Laser Desorption/Ionization Mass Spectral Serum Profiles for Prostate Cancer

Classification analysis was performed on 322 SELDI-TOF-MS protein expression profiles for prostate cancer. Feature ranking was based on the F-test, information gain (entropy), and Gini diversity applied in a pairwise, one-against-all, and all-at-once modular form. Classifiers included 4NN, NBC, LDA, LVQ1, SVM, and ANN. 4-class bootstrap (0.632) accuracies were in the range 50-80%, with NBC resulting in the lowest average accuracy (50-66%) and SVM resulting in the greatest average accuracy (71-79%). A 12-peak model with 88% accuracy collapsed into 6 peaks with m/z values of 3460, 4172, 4581, 6890, 14281 and 14696. The peaks identified may be confirmed in the future to be markers of early detection and/or therapy.