SGFCCDA: Scale Graph Convolutional Networks and Feature Convolution for circRNA-Disease Association Prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Circular RNAs (circRNAs) are non-coding RNAs with significant roles in disease development.
• Computational models for circRNA-disease associations offer insights into disease mechanisms and potential diagnostics/therapeutics.
• Existing methods may not fully capture complex topological and attribute information in circRNA-disease networks.

Purpose of the Study:

• To propose SGFCCDA, a novel computational model for predicting circRNA-disease associations.
• To leverage scale graph convolutional networks and feature convolution for enhanced prediction accuracy.
• To reduce the need for extensive and costly laboratory experiments in identifying these associations.

Main Methods:

• Constructed a heterogeneous network integrating circRNA/disease similarity and known associations.
• Employed scale graph convolutional networks to capture network topology and node attributes.
• Utilized convolutional neural networks for feature learning and a multilayer perceptron for final association prediction.
• Integrated circRNA and disease features using the Hadamard product.

Main Results:

• SGFCCDA demonstrated accurate prediction of potential circRNA-disease associations.
• Five-fold cross-validation on the CircR2Disease dataset validated the model's performance.
• Case studies confirmed SGFCCDA's effectiveness in identifying disease-associated circRNAs.

Conclusions:

• SGFCCDA is an effective computational tool for predicting circRNA-disease associations.
• The model's approach enhances understanding of disease pathogenesis.
• SGFCCDA facilitates the development of novel diagnostic and therapeutic strategies.