FDS-CAP: Modeling Fragmented Disease Subgraphs with Component-Level Attention for Comorbidity Prediction.

Area of Science:

• Computational Biology
• Network Medicine
• Bioinformatics

Background:

• Understanding disease comorbidity is crucial for identifying shared mechanisms and improving treatments.
• Existing methods often fail to account for the fragmented nature of disease modules within biological networks.

Purpose of the Study:

• To develop a novel graph-based deep learning framework, FDS-CAP, for predicting disease comorbidity.
• To address the challenge of fragmented disease modules in the human interactome.

Main Methods:

• Implemented Fragmented Disease Subgraphs with Component-Level Attention for Comorbidity Prediction (FDS-CAP), a graph-based deep learning framework.
• Utilized Subgraph Neural Networks (SUBGNN) with component-level attention to embed fragmented disease subgraphs.
• Employed a Variational Graph Auto-Encoder for comorbidity prediction within the Human Disease Network.

Main Results:

• FDS-CAP achieved state-of-the-art performance in comorbidity prediction, reaching an AUROC of 0.966 on a benchmark dataset.
• The framework demonstrated biological interpretability, with attention-weighted components capturing meaningful disease patterns in a glioma case study.

Conclusions:

• FDS-CAP provides a more accurate and expressive method for disease comorbidity prediction by effectively representing fragmented disease subgraphs.
• The attention mechanism enhances the understanding of complex disease associations and their underlying biological mechanisms.