Pancreatic cancer biomarker detection by two support vector strategies for recursive feature elimination.

Aim: Pancreatic cancer is one of the worst malignant tumors in prognosis. Therefore, to reduce the mortality rate of pancreatic cancer, early diagnosis and prompt treatment are particularly important.
Results: We put forward a new feature-selection method that was used to find clinical markers for pancreatic cancer by combination of Support Vector Machine Recursive Feature Elimination (SVM-RFE) and Large Margin Distribution Machine Recursive Feature Elimination (LDM-RFE) algorithms. As a result, seven differentially expressed genes were predicted as specific biomarkers for pancreatic cancer because of their highest accuracy of classification on cancer and normal samples.
Conclusion: Three (MMP7, FOS and A2M) out of the seven predicted gene markers were found to encode proteins secreted into urine, providing potential diagnostic evidences for pancreatic cancer.