Heterogeneous graph neural network for lncRNA-disease association prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Experimental validation of long noncoding RNA (lncRNA)-disease associations is costly and time-consuming.
• Computational prediction methods are crucial for efficient discovery of lncRNA-disease links.
• Existing computational models struggle to fully leverage multi-source network topology for accurate predictions.

Purpose of the Study:

• To propose a novel computational method, HGNNLDA, for predicting lncRNA-disease associations.
• To enhance the utilization of network topology information from multi-source data.
• To improve the accuracy and efficiency of identifying potential lncRNA-disease relationships.

Main Methods:

• Constructed a heterogeneous network integrating lncRNA similarity, lncRNA-disease, and lncRNA-miRNA association networks.
• Employed restart random walk for sampling correlated neighbors within the heterogeneous network.
• Utilized a heterogeneous graph neural network with an attention mechanism for type-based neighbor aggregation and feature embedding.

Main Results:

• Achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.9786 and an Area Under the Precision-Recall Curve (AUPR) of 0.8891 via fivefold cross-validation.
• Demonstrated superior prediction performance compared to five state-of-the-art methods.
• Case studies confirmed the method's efficacy in predicting potential lncRNA-disease associations and identifying novel disease associations.

Conclusions:

• HGNNLDA offers a powerful and accurate computational approach for lncRNA-disease association prediction.
• The method effectively integrates diverse network information, overcoming limitations of existing models.
• HGNNLDA shows promise for advancing disease diagnosis, treatment, and prevention through enhanced understanding of lncRNA functions.