Pathway-based subnetworks enable cross-disease biomarker discovery.

Area of Science:

• Computational biology
• Genomics
• Biomarker discovery

Background:

• Precision medicine relies heavily on biomarkers, but current methods for biomarker development are often bespoke and lack generalizability.
• There is a critical need for systematic approaches to generate biologically interpretable molecular models for robust phenotype prediction.

Purpose of the Study:

• To introduce SIMMS (Subnetwork Integration for Multi-Modal Signatures), a novel algorithm for systematic biomarker discovery.
• To demonstrate the capability of SIMMS in integrating multi-modal data for robust phenotype prediction across various diseases.

Main Methods:

• SIMMS fragments biological pathways into functional subnetworks.
• These subnetworks are then used to build predictive models for phenotypes.
• The algorithm was applied to multiple data types across five diseases, including profiling the PI3K pathway in breast tumors.

Main Results:

• SIMMS reproducibly identified known and novel disease subtypes across five diseases.
• The algorithm demonstrated superior predictive performance compared to existing bespoke methods.
• A SIMMS-derived breast cancer model outperformed a clinically-validated test in an independent patient cohort.

Conclusions:

• SIMMS provides a generic and systematic methodology for data integration and biomarker discovery.
• This approach enables the development of robust, biologically interpretable molecular models for precision medicine.
• SIMMS has the potential to significantly advance biomarker development and clinical application.