Cancer classification and biomarker selection via a penalized logsum network-based logistic regression model

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Identifying molecular biomarkers and signaling pathways is crucial in genome research.
• Logistic regression models are effective for classification but lack robust biomarker selection capabilities.
• High-dimensional biological data presents challenges for traditional analysis methods.

Purpose of the Study:

• To enhance logistic regression models with efficient gene selection capabilities.
• To develop a method for accurate cancer classification using genomic data.
• To address limitations in biomarker discovery within high-dimensional datasets.

Main Methods:

• A novel penalized logsum network-based regularization logistic regression model was proposed.
• The model integrates network-based regularization for improved gene selection.
• The approach is designed for analyzing high-dimensional genomic data.

Main Results:

• The proposed method demonstrated effectiveness in analyzing simulated high-dimensional datasets.
• Achieved high Area Under the Curve (AUC) performances: 89.66% (training) and 90.02% (testing) on a large dataset.
• Outperformed mainstream methods by an average of 5.17% (training) and 4.49% (testing) in AUC.

Conclusions:

• The developed penalized logsum network logistic regression model is a promising tool.
• It offers effective gene selection and cancer classification for high-dimensional biological data.
• This method advances the analysis of complex genomic information.