An Interpretable Approach for Lung Cancer Prediction and Subtype Classification using Gene Expression.

Area of Science:

• Computational biology
• Genomics
• Machine learning in oncology

Background:

• Lung cancer is a leading cause of cancer death, characterized by significant heterogeneity even within subtypes.
• Accurate histological subtype diagnosis is crucial for identifying target genes and tailoring therapy.
• Current deep learning models for cancer gene expression analysis lack interpretability, hindering biological insight.

Purpose of the Study:

• To develop interpretable machine learning algorithms for lung cancer prediction and subtype classification.
• To address limitations in feature weight extraction from deep learning models.
• To identify key genes associated with cancer development and specific lung cancer subtypes.

Main Methods:

• Utilized tree-based learning algorithms for analyzing gene expression data.
• Developed methods for extracting interpretable feature weights from models.
• Compared performance against existing deep learning approaches.

Main Results:

• Proposed tree-based methods demonstrated superior performance compared to related research.
• The developed models provide interpretable insights into gene significance.
• Successfully identified gene sets for differentiating normal vs. cancerous tissue and LUAD vs. LUSC subtypes.

Conclusions:

• Interpretable machine learning offers valuable biological insights for lung cancer research.
• The identified gene sets can aid in cancer prediction and subtype classification.
• This approach enhances the understanding of lung cancer drivers for improved treatment strategies.