Discovering Effective DNA Methylation Biomarkers through Different Gene Mutation Groups for Lung Cancer Detection

Lung cancer is the leading causes of cancer-related deaths worldwide and the number of new lung cancer cases is continuously increasing for both sexes. Lung cancer is often diagnosed at its late stages when disease symptoms become apparent, and it would miss the optimal treatment options. Early detection of lung cancer could provide early treatment options and improve the survival rates. This study focused on analyzing genome-wide methylation differences to discover suitable DNA methylation biomarkers for lung cancer subjects mainly based on mutations occurred in EGFR, KRAS, or TP53 genes. In addition, the study applied neural network and deep SHAP to identify the common features for DNA methylation biomarker selection. The prediction models were constructed employing SVM, Logistic Regression, and XGBoost kernel techniques on cross-comparison and recommended 24 important methylation biomarkers. The goal is to identify key factors to pave a ground of a non-invasive detection toolkit for early lung cancer diagnosis with convenience and accuracy to improve survival rates. The results have shown that regardless of EGFR mutations, KRAS mutations, TP53 mutations, or normal subjects, the best selected methylation biomarker combinations performed very well across various machine learning models, with a median prediction accuracy of 93%, and it could be applied to liquid biopsy samples as well with an accuracy larger than 80%.