Abstract
BACKGROUND: Lymph node (LN) metastasis was an independent risk factor for stomach cancer recurrence, and the presence of LN metastasis has great influence on the overall survival of stomach cancer patients. Thus, accurate prediction of the presence of lymph node metastasis can provide guarantee of credible prognosis evaluation of stomach cancer patients. Recently, increasing evidence demonstrated that the aberrant DNA methylation first appears before symptoms of the disease become clinically apparent. OBJECTIVE: Selecting key biomarkers for LN metastasis presence prediction for stomach cancer using clinical DNA methylation based on a machine learning method. METHODS: To reduce the overfitting risk of prediction task, we applied a three-step feature selection method according to the property of DNA methylation data. RESULTS: The feature selection procedure extracted several cancer-related and lymph node metastasis-related genes, such as TP73, PDX1, FUT8, HOXD1, NMT1, and SEMA3E. The prediction performance was evaluated on the public DNA methylation dataset. The results showed that the three-step feature procedure can largely improve the prediction performance and implied the reliability of the biomarkers selected. CONCLUSIONS: With the selected biomarkers, the prediction method can achieve higher accuracy in detecting LN metastasis and the results also proved the reliability of the selected biomarkers indirectly.