Graph transformer with disease subgraph positional encoding for improved comorbidity prediction.

Area of Science:

• Computational biology
• Medical informatics
• Graph-based machine learning

Background:

• Comorbidity significantly impacts disease management and patient outcomes.
• Understanding complex disease interconnections is crucial for effective healthcare.
• Existing methods may not fully capture the nuances of disease associations.

Purpose of the Study:

• To develop an advanced method for predicting disease comorbidities.
• To leverage human interactome data and graph methodologies for improved prediction.
• To introduce Transformer with Subgraph Positional Encoding (TSPE) for enhanced comorbidity prediction.

Main Methods:

• Utilized transformer's attention mechanisms and subgraph positional encoding (SPE).
• Developed a novel SPE inspired by biologically supervised embedding.
• Compared TSPE against Laplacian positional encoding in graph transformers.

Main Results:

• TSPE demonstrated superior performance in predicting disease comorbidities.
• Achieved up to 28.24% higher ROC AUC and 4.93% higher accuracy on benchmark datasets.
• The proposed SPE method proved more effective than Laplacian positional encoding.

Conclusions:

• TSPE offers a promising approach for disease comorbidity prediction.
• The method shows potential for adaptation to other complex graph-based tasks.
• Integrating clustering and disease-specific information enhances predictive accuracy.