MDGF-MCEC

MDGF-MCEC predicts associations between circular RNAs (circRNAs) and diseases using a multi-view dual-attention graph convolution network (GCN) combined with cooperative ensemble learning to prioritize circRNA–disease links for biological investigation.

Key Features:

• Multi-view graph construction: Constructs two disease relation graphs and two circRNA relation graphs derived from different similarity measures to capture diverse relational information.
• Multi-view dual-attention GCN: Applies a multi-view graph convolution network with a dual-attention mechanism to learn representations from multiple relation graphs.
• Dual-attention mechanism: Adjusts contribution weights of different features at both channel and spatial levels to enhance discriminative feature learning.
• Embedding generation: Produces embedding features for diseases and circRNAs via the multi-view GCN.
• Multi-view feature combinations: Generates nine distinct feature combinations from the learned embeddings to create new multi-view data.
• Cooperative ensemble classifier: Uses a cooperative ensemble classifier to predict circRNA–disease associations from the multi-view feature combinations.

Scientific Applications:

• CircRNA–disease association prediction: Predicts potential associations between circRNAs and human diseases.
• Benchmark validation: Validated on the CircR2Disease database with an achieved AUC of 0.9744.
• Cross-dataset evaluation: Evaluated on additional databases including circ2Disease and circRNADisease.
• Literature corroboration: Generated predictions supported by literature, including associated circRNAs for hepatocellular carcinoma and gastric cancer.

Methodology:

Construct two disease relation graphs and two circRNA relation graphs based on different similarity metrics; process these graphs with a multi-view GCN employing a dual-attention mechanism for representation learning; generate nine distinct feature combinations from the embeddings and apply a cooperative ensemble classifier to predict circRNA–disease associations.