Plant lncRNA-miRNA Interaction Prediction Based on Counterfactual Heterogeneous Graph Attention Network.

Area of Science:

• Computational Biology
• Plant Molecular Biology
• Bioinformatics

Background:

• Long non-coding RNAs (lncRNAs) and microRNAs (miRNAs) play crucial roles in regulating plant life processes.
• Interactions between lncRNAs and miRNAs (LMIs) are vital for understanding these regulatory networks.
• Computational methods, particularly graph neural networks (GNNs), are increasingly used to predict LMIs, but existing approaches suffer from low-semantic graph information and noise.

Purpose of the Study:

• To develop a robust and accurate computational method for predicting plant lncRNA-miRNA interactions (LMIs).
• To address the limitations of existing GNN-based methods, specifically their susceptibility to noise and low-semantic graph data.
• To enhance the understanding of regulatory relationships in plant molecular biology.

Main Methods:

• Construction of a real-world based lncRNA-miRNA (L-M) heterogeneous network.
• Development of a novel Counterfactual Heterogeneous Graph Attention Network (CFHAN) incorporating node-level attention, semantic-level attention, and counterfactual links.
• Utilizing enhanced node embeddings as input for a Multilayer Perceptron (MLP) to predict LMIs.

Main Results:

• CFHAN demonstrated superior performance compared to five state-of-the-art methods on a benchmark plant LMI dataset.
• Achieved high prediction accuracy with an average AUC of 0.9953 and an average ACC of 0.9733.
• Showcased promising cross-species prediction capabilities.

Conclusions:

• CFHAN effectively predicts plant LMIs with improved robustness and accuracy.
• The method offers valuable insights for experimental LMI research in plants.
• CFHAN represents a significant advancement in computational prediction of regulatory interactions in plants.