Analysis of Proteomic Pattern Data for Cancer Detection

Abstract. In this paper we analyze two proteomic pattern datasets containing measurements from ovarian and prostate cancer samples. In particular, a linear and a quadratic support vector machine (SVM) are applied to the data for distinguishing between cancer and benign status. On the ovarian dataset SVM gives excellent results, while the prostate dataset seems to be a harder classification problem for SVM. The prostate dataset is futher analyzed by means of an evolutionary algorithm for feature selection (EAFS) that searches for small subsets of features in order to optimize the SVM performance. In general, the subsets of features generated by EAFS vary over different runs and over different data splitting in training and hold-out sets. Nevertheless, particular features occur more frequently over all the runs. The role of these “core” features as potential tumor biomarkers deserves further study.

• Type: Conference
• Year: 2004
• Where: EVOW
• Authors: Kees Jong, Elena Marchiori, Aad van der Vaart
• Keywords: Artificial Intelligence, EVOW 2004, Ovarian Dataset Svm, Prostate Cancer Samples, Prostate Dataset