GCNCMI

GCNCMI predicts potential interactions between circular RNAs (circRNAs) and microRNAs (miRNAs) using a graph convolutional neural network to identify regulatory relationships relevant to gene expression and disease mechanisms.

Key Features:

• Graph convolutional neural network (GCNN): Employs a graph convolutional neural network to model circRNA–miRNA interaction networks.
• Recursive layer-wise propagation: Recursively propagates interaction information across multiple graph convolutional layers.
• Layer-wise embedding fusion: Unifies embeddings from each graph convolutional layer to form robust node representations for prediction.
• Adjacent-node interaction mining: Mines potential interactions between adjacent nodes in the interaction graph to capture local topological patterns.
• Cross-validation performance: Achieved an AUC of 0.9312 and an AUPR of 0.9412 in five-fold cross-validation.

Scientific Applications:

• CircRNA–miRNA interaction prediction: Predicts candidate circRNA–miRNA interactions to support studies of gene regulation and disease mechanisms.
• Case study validation: Demonstrated predictive capability in case studies involving hsa-miR-622 and hsa-miR-149-5p.

Methodology:

Uses a graph convolutional neural network that mines adjacent-node interactions, recursively propagates interaction information across multiple graph convolutional layers, and unifies embeddings from each layer to form representations for final prediction.