Classification of lung cancer using ensemble-based feature selection and machine learning methods
Zhihua Cai1, Dong Xu, Qing Zhang
1Affiliated Cancer Hospital of Guangzhou Medical University, Guangzhou, Guangdong Province, China.
Molecular Biosystems
|December 17, 2014
Summary
Researchers identified 16 DNA methylation markers that can distinguish between three major lung cancer types: lung adenocarcinoma (LADC), squamous cell lung cancer (SQCLC), and small cell lung cancer (SCLC). This panel shows promise for improved lung cancer diagnosis.
Area of Science:
- Oncology
- Molecular Biology
- Biomarker Discovery
Background:
- Lung cancer is a leading global cause of mortality with distinct subtypes including non-small cell lung cancer (NSCLC), small cell lung cancer (SCLC), and carcinoid tumors.
- NSCLC is further categorized into lung adenocarcinoma (LADC), squamous cell lung cancer (SQCLC), and large cell lung cancer.
- DNA methylation markers have shown potential as lung cancer-specific biomarkers, but a panel capable of simultaneously differentiating the main lung cancer types remained undiscovered.
Purpose of the Study:
- To identify a compact panel of DNA methylation markers capable of distinguishing between LADC, SQCLC, and SCLC.
- To evaluate the efficacy of ensemble-based feature selection methods combined with machine learning for lung cancer classification.
- To assess the diagnostic power of the identified DNA methylation marker panel.
Main Methods:
- Utilized Receiver Operating Characteristic (ROC) curves, Random Forests (RFs), and Maximum Relevancy and Minimum Redundancy (mRMR) for feature selection.
- Employed machine learning algorithms for the classification of LADC, SQCLC, and SCLC based on DNA methylation profiles.
- Implemented leave-one-out cross-validation (LOOCV) and independent dataset testing to validate the classification performance.
Main Results:
- A panel of 16 DNA methylation markers demonstrated significant classification power.
- Achieved an accuracy of 86.54% in LOOCV and 84.6% in independent testing.
- Attained a recall of 84.37% in LOOCV and 85.5% in independent testing.
- Ensemble-based feature selection methods, when combined with incremental feature selection (IFS), proved superior in identifying informative and compact feature sets compared to individual methods.
Conclusions:
- The study highlights the effectiveness of ensemble-based feature selection approaches for identifying robust molecular signatures.
- A common panel of 16 DNA methylation markers shows potential for simultaneously distinguishing between LADC, SQCLC, and SCLC.
- These findings could significantly aid in the clinical diagnosis and treatment strategies for different lung cancer subtypes.