Multi-Omic Graph Diagnosis (MOGDx): a data integration tool to perform classification tasks for heterogeneous...

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in medicine

Background:

• Human diseases exhibit significant heterogeneity, complicating diagnosis and treatment.
• Integrative machine learning with multi-omic data offers potential for granular disease classification.
• Existing methods face challenges in scalability, data simplification, and handling missing data.

Purpose of the Study:

• To introduce Multi-Omic Graph Diagnosis (MOGDx), a tool for integrating multi-omic data for disease classification.
• To address limitations of current integrative methods in scalability and data handling.
• To improve the classification of heterogeneous diseases using a network-based approach.

Main Methods:

• MOGDx integrates multi-omic data using a network taxonomy.
• It fuses patient similarity networks and augments them with reduced genomic data vectors.
• Classification is performed using a graph convolutional network.

Main Results:

• MOGDx achieved state-of-the-art performance on three cancer datasets (breast, kidney, glioma).
• The tool successfully identified relevant multi-omic markers for each disease.
• MOGDx demonstrated superior integration of genomic measures and patient coverage compared to existing methods.

Conclusions:

• MOGDx is a flexible and effective tool for multi-omic data integration and disease classification.
• It aids in the interpretation of genomic marker data for heterogeneous diseases.
• The approach shows promise for advancing precision medicine through improved disease subtyping.