Biologically Supervised Graph Embedding for Disease Comorbidity Prediction: A Versatile Framework for Various...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Comorbidity, the co-occurrence of diseases, is critical for disease management and understanding.
• Genetic mutations and protein-protein interactions (PPI) within the human interactome are believed to underlie comorbidity.
• The complexity and incompleteness of the human interactome hinder effective feature extraction for comorbidity prediction.

Purpose of the Study:

• To introduce a novel framework, Biologically Supervised Graph Embedding (BSE), for selecting relevant features in graph embeddings.
• To improve the accuracy of disease comorbidity prediction by leveraging biologically relevant features.
• To evaluate the performance of BSE against state-of-the-art techniques.

Main Methods:

• Developed the Biologically Supervised Graph Embedding (BSE) framework.
• Applied BSE to graph embeddings capturing disease subgraph relationships.
• Evaluated BSE's performance on centered and uncentered embedding methods using various metrics and classifiers.

Main Results:

• BSE demonstrated consistent superiority over existing state-of-the-art techniques.
• BSE achieved significant improvements in prediction performance, up to 50% by ROC AUC score.
• BSE effectively selects features with a higher ratio of disease associations to gene connectivity, indicating biological relevance.

Conclusions:

• BSE offers a versatile and broadly applicable approach for precise disease comorbidity prediction.
• The framework successfully extracts biologically insightful features, uncovering latent factors affecting comorbidity.
• BSE presents novel avenues for advancing comorbidity research and related applications.