circ2DGNN: circRNA-Disease Association Prediction via Transformer-Based Graph Neural Network

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Understanding circular RNA (circRNA) and disease associations is crucial for disease mechanism elucidation and therapeutic development.
• Current computational methods often indirectly account for biomolecules' effects, limiting accuracy in predicting circRNA-disease interactions.
• A comprehensive approach integrating diverse biomolecular data is needed to improve circRNA-disease association prediction.

Purpose of the Study:

• To develop a novel computational model, circ2DGNN, for predicting circRNA-disease associations.
• To leverage heterogeneous graph neural networks and incorporate diverse biomolecule interaction data.
• To improve the accuracy and comprehensiveness of circRNA-disease association predictions.

Main Methods:

• Constructed a comprehensive heterogeneous network including human circRNAs, diseases, and other biomolecule interactions.
• Developed circ2DGNN, a heterogeneous graph neural network model utilizing graph representation learning.
• Employed a Transformer-like architecture with heterogeneous attention for message propagation and aggregation, incorporating residual connections.

Main Results:

• circ2DGNN effectively integrates heterogeneous network data for downstream link prediction.
• The model demonstrated superior performance compared to existing state-of-the-art methods on a test dataset.
• Fine-tuning via five-fold cross-validation optimized model hyperparameters.

Conclusions:

• circ2DGNN offers a powerful new approach for predicting circRNA-disease associations by directly utilizing heterogeneous network information.
• The model's ability to incorporate diverse biomolecular interactions enhances the understanding of disease mechanisms.
• This work provides a valuable tool for advancing research in circRNA-related diseases and potential therapies.