Detection of pancreatic cancer biomarkers using mass spectrometry

Abstract

Mass spectrometry (MS) data analysis has been utilized to detect biomark- ers for early cancer detection. In this research, the aim of analysis was to search biomarkers for pancreatic cancer that is a highly lethal malignancy and the fourth leading cause of cancer-related deaths. With the biomarkers detected in this study, we could improve the prognosis and survivals of pancreatic cancer patients. MS data analysis typically consisted of three steps: preprocessing, clas- sification, and variable selection. The results of preprocessing were usually non-fully populated data structure, so it was essential to substitute missing values for further advanced statistical inference. We used three methods to address missing issues. We used four classification methods to discriminate the case and control group. Based on the prediction error, we compared the results of classification methods. We concluded that Peak Probability Contrast (PPC) method with bagging and random forest outperformed the others. Based on the variable importance measure in bagging and random forest, we selected the m/z 1,206, 1,465 as biomarkers for pancreatic cancer.