Graph in Graph (GiG): A novel graph AI framework for integrating and interpreting medical and omics data

Area of Science:

• Computational biology
• Artificial intelligence in healthcare
• Precision medicine

Background:

• Medical records and omics data offer comprehensive patient insights but are challenging to integrate.
• Systematic interpretation of combined datasets is crucial for disease diagnosis and target discovery.

Purpose of the Study:

• To propose and evaluate a novel graph AI model, Graph in Graph (GiG), for integrating and interpreting whole-person medical and omics data.
• To characterize individual patients and prioritize omics biomarkers and phenotypes.

Main Methods:

• Modeled medical records as a person-phenotype graph.
• Developed omics signaling graphs for individual patients.
• Integrated phenotype features and omics data using the GiG framework.

Main Results:

• GiG achieved high prediction accuracy in distinguishing type 2 diabetes (T2D) and pre-T2D from healthy individuals.
• The model successfully interpreted predictions by ranking essential clinical and omic biomarkers.
• Applied to the Long Life Family Study (LLFS) cohort, revealing insights into healthy aging.

Conclusions:

• The GiG framework provides an effective method for integrating and interpreting complex medical and omics datasets.
• This approach holds potential for advancing disease diagnosis and uncovering pathogenesis.
• GiG can be broadly applied to diverse studies requiring multi-modal data integration.