An Efficient and Explainable Ensemble-learning Framework for Early Lung Cancer Biomarkers Detection.

Area of Science:

• Computational Biology and Bioinformatics
• Artificial Intelligence in Medicine
• Oncology and Cancer Research

Background:

• Early detection of non-small cell lung cancer (NSCLC) is critical for patient outcomes.
• Interpretability of machine learning models in biomarker discovery remains a significant challenge.
• Gene expression data offers potential for identifying NSCLC biomarkers but requires sophisticated analysis.

Purpose of the Study:

• To propose an Efficient and Explainable Ensemble-learning framework (EEE-framework) for early NSCLC biomarker detection.
• To address the trade-off between accuracy and interpretability in NSCLC detection models.
• To identify and understand the importance and interrelationships of NSCLC biomarkers using explainable AI.

Main Methods:

• Developed an ensemble-learning model using five interpretable base learners (Decision Tree, Random Forest, XGBoost, AdaBoost, Gaussian Naive Bayes) with voting ensemble integration.
• Employed a multi-perspective approach utilizing six explainable artificial intelligence (XAI) methods for biomarker identification (Local, Global, Intrinsic, Post-hoc Interpretability).
• Validated the EEE-framework on the TCGA RNA_seq dataset and a self-constructed methylation dataset.

Main Results:

• The ensemble-learning model achieved high classification performance with F1 scores of 0.9983 (AUC: 0.9993) and 0.8462 (AUC: 0.9249) on the respective datasets.
• Global and Local Interpretable Visualization provided significant insights into NSCLC biomarker importance, interrelationships, and causality.
• The EEE-framework demonstrated strong generalization ability and enhanced interpretability in NSCLC detection outcomes.

Conclusions:

• The EEE-framework effectively balances accuracy and interpretability for early NSCLC detection and biomarker discovery.
• Explainable AI methods significantly enhance the understanding of complex biomarker data in NSCLC.
• This framework offers a promising tool for improving NSCLC diagnosis and guiding further research.