Predicting abiotic stress-responsive miRNA in plants based on multi-source features fusion and graph neural network

Area of Science:

• Plant molecular biology
• Bioinformatics
• Computational biology

Background:

• MicroRNAs (miRNAs) are critical regulators of plant responses to abiotic stresses.
• Traditional experimental methods for identifying miRNA-abiotic stress associations are costly and time-consuming.
• Existing computational methods have not fully leveraged available miRNA and abiotic stress information.

Purpose of the Study:

• To develop an efficient and economical computational method for predicting miRNA-abiotic stress associations.
• To fully exploit multi-source feature information from miRNAs and abiotic stresses.
• To propose a novel approach based on graph neural networks.

Main Methods:

• Integrated multi-source feature information of miRNAs and abiotic stresses into a heterogeneous network.
• Employed Restart Random Walk (RWR) to extract global structural information and generate feature vectors.
• Utilized a graph autoencoder based on Graph Isomorphism Networks (GIN) for learning and reconstruction.

Main Results:

• The proposed model achieved superior performance in predicting potential miRNA-abiotic stress associations compared to existing methods.
• Achieved Area Under the Precision-Recall Curve (AUPR) of 98.24% and Area Under the Curve (AUC) of 97.43% under five-fold cross-validation.
• Demonstrated effective fusion of multi-source similarity networks and association information.

Conclusions:

• The developed graph neural network model is robust and effective for predicting miRNA-abiotic stress associations.
• This approach offers a valuable tool for advancing research in plant abiotic stress responses.
• The method highlights the potential of integrating diverse data sources for biological association prediction.