Biochemically Constrained Multi-Omics Integration Reveals Protein-Metabolite Dependencies Across Diseases

Area of Science:

• Biochemistry
• Computational Biology
• Systems Biology

Background:

• Integrating proteomic and metabolomic data is crucial for understanding complex diseases.
• Current methods often miss biochemical relationships and lack stability in small cohorts.

Purpose of the Study:

• To introduce ProMetNet, a novel framework for biochemically constrained multi-omics integration.
• To improve the robustness and interpretability of cross-omics analyses in disease research.

Main Methods:

• Developed ProMetNet, a neural network framework encoding protein-metabolite relationships using Reactome pathway topology.
• Applied the framework to Alzheimer's disease, type 2 diabetes, COVID-19, and glioblastoma cohorts.
• Validated findings using UK Biobank population data.

Main Results:

• ProMetNet outperformed existing multi-omics integration methods across diverse disease cohorts.
• Demonstrated high discriminative performance even with significant data downsampling.
• Identified biologically plausible protein-metabolite dependencies missed by traditional analyses.

Conclusions:

• ProMetNet offers a robust, interpretable, and biologically grounded approach for multi-omics integration.
• The framework enhances the identification of structured molecular dependencies for disease understanding.
• ProMetNet shows generalizability and robustness in large-scale population data.