Regularized logistic regression with network-based pairwise interaction for biomarker identification in breast cancer

Area of Science:

• Genomics
• Bioinformatics
• Biomarker Discovery

Background:

• Personalized medicine requires predictive, stable, and interpretable biomarkers.
• Traditional methods often overlook gene interactions and suffer from low reproducibility.
• Robust biomarker identification is crucial for accurate disease interpretation and treatment.

Purpose of the Study:

• To develop a robust method for identifying disease-associated gene interactions.
• To improve the stability and interpretability of biomarkers.
• To enhance diagnostic and prognostic accuracy in complex diseases.

Main Methods:

• Proposed a regularized logistic regression model incorporating gene pairs from protein-protein interaction networks.
• Constructed a line graph to represent pairwise interaction adjacencies.
• Utilized an adaptive elastic net incorporating network degree information for model robustness.

Main Results:

• The model achieved competitive classification performance on six breast cancer datasets.
• Demonstrated high stability in variable selection, identifying reproducible biomarkers.
• Discovered biomarkers largely verified in existing biochemical and biomedical research.

Conclusions:

• The method offers a promising approach for disease diagnosis and understanding pathogenesis.
• Enables more accurate and stable biomarker discovery for monitoring disease-related functional changes.
• Facilitates potential suggestions for novel therapeutic strategies based on biomarker predictions.