GPS-Net: Discovering prognostic pathway modules based on network regularized kernel learning.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Current prognostic biomarker discovery often relies on single-gene or global gene expression analysis.
• These gene-centric methods overlook crucial higher-order dependencies in co-regulated processes, pathways, and regulatory networks vital for complex diseases like cancer.
• Existing approaches struggle to capture the holistic biological context essential for accurate outcome prediction.

Purpose of the Study:

• To introduce GPS-Net, a novel computational framework for efficient identification of prognostic gene modules.
• To address the limitations of gene-centric approaches by incorporating pathway structures and gene interaction networks.
• To enable scalable and feasible genome-wide, pathway-level prognostic analysis.

Main Methods:

• Developed GPS-Net, a computational framework integrating multiple kernel learning and network-based regularization.
• Incorporated holistic pathway structures and gene interaction networks into the analytical model.
• Utilized extensive simulation studies to validate accuracy and computational efficiency.

Main Results:

• GPS-Net enhances the accuracy of biomarker and pathway identification compared to traditional methods.
• The framework significantly reduces computational complexity for genome-wide analyses.
• Identified key predictive pathways for patient outcomes in a cancer immunotherapy study using GPS-Net.

Conclusions:

• GPS-Net offers a scalable and feasible framework for pathway-level prognostic analysis in genomics.
• The approach effectively synergizes mechanism-driven and data-driven methodologies for precision genomics.
• This computational framework advances the discovery of prognostic biomarkers by considering biological network structures.