GSLCDA: An Unsupervised Deep Graph Structure Learning Method for Predicting CircRNA-Disease Association.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Molecular Medicine

Background:

• Circular RNAs (circRNAs) are increasingly recognized for their roles in cellular processes and disease pathogenesis.
• Accurate prediction of circRNA-disease associations (CDAs) is crucial for developing novel therapeutic strategies.
• Existing CDA prediction methods struggle with noisy graph structures, limiting their performance.

Purpose of the Study:

• To develop an unsupervised deep graph structure learning method, GSLCDA, for predicting potential circRNA-disease associations (CDAs).
• To address the limitations of existing methods that rely heavily on graph networks and are susceptible to noisy connections.

Main Methods:

• Integration of multi-source circRNA and disease data to construct a heterogeneous network.
• Unsupervised deep graph structure learning to enhance network topology and uncover essential features.
• Application of a graph space sensitive k-nearest neighbor (KNN) algorithm for latent CDA identification.

Main Results:

• GSLCDA achieved 92.67% accuracy and 0.9279 AUC on a benchmark dataset.
• The method demonstrated exceptional performance on independent datasets.
• Case studies in breast, colorectal, and lung cancers showed high validation rates for predicted CDAs.

Conclusions:

• GSLCDA effectively predicts underlying circRNA-disease associations.
• The method offers new perspectives for the diagnosis and therapy of complex human diseases.
• GSLCDA provides a robust approach for CDA prediction, overcoming limitations of previous graph-based methods.