A hybrid feature extraction framework combining PCA and mutual information for gene expression based lung cancer...

Area of Science:

• Bioinformatics and Computational Biology
• Genomics and Cancer Research
• Machine Learning in Healthcare

Background:

• Lung cancer is a major cause of cancer mortality, necessitating improved diagnostic accuracy.
• High-dimensional gene expression data offer insights but pose computational and overfitting challenges.
• Effective feature extraction is crucial for accurate classification of complex diseases like lung cancer.

Purpose of the Study:

• To develop a hybrid feature extraction framework combining Principal Component Analysis (PCA) and Mutual Information (MI) for lung cancer classification.
• To integrate gene expression data from TCGA and ICGC for robust analysis.
• To enhance biological interpretability through protein-protein interaction (PPI) network analysis.

Main Methods:

• Implemented a hybrid PCA-MI framework for dimensionality reduction and feature selection on integrated gene expression datasets.
• Trained a Convolutional Neural Network (CNN) classifier using selected features for lung cancer classification.
• Utilized PPI networks to identify biologically significant hub genes from the ranked feature set.

Main Results:

• The hybrid PCA-MI framework achieved superior performance compared to ten other feature extraction methods.
• The CNN classifier attained 98% accuracy and 98% precision for lung cancer classification using PCA-MI features.
• PPI analysis confirmed the biological relevance of the identified hub genes.

Conclusions:

• The proposed hybrid PCA-MI framework effectively addresses high-dimensional gene expression data challenges for lung cancer classification.
• Integration of deep learning and network analysis provides a robust and interpretable diagnostic strategy.
• This framework lays the groundwork for future multi-omics data integration and advanced diagnostic approaches in cancer research.