Predicting lncRNA-disease associations based on heterogeneous graph convolutional generative adversarial network.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Long non-coding RNAs (lncRNAs) are implicated in various diseases, but predicting their associations with diseases is challenging.
• Existing methods struggle with heterogeneous data and data imbalance in lncRNA-disease association prediction.

Purpose of the Study:

• To develop a novel computational method, HGC-GAN, for accurate prediction of lncRNA-disease associations.
• To address limitations in handling heterogeneous information and sparse, unbalanced data.

Main Methods:

• Constructed a lncRNA-miRNA-disease heterogeneous network integrating multiple association data and sequence information.
• Employed a heterogeneous graph convolutional neural network (GCN) as a generator to obtain node embeddings.
• Utilized a generative adversarial network (GAN) discriminator to refine the generator's prediction accuracy.

Main Results:

• HGC-GAN achieved high performance in predicting lncRNA-disease associations, with AUC of 0.9591 and AUPR of 0.9606.
• Demonstrated effectiveness in predicting associations for novel lncRNAs.
• Experimental results confirmed the method's superior predictive capability.

Conclusions:

• HGC-GAN offers a promising computational approach for predicting lncRNA-disease associations.
• The method has potential implications for disease diagnosis, treatment strategies, and drug development.
• Highlights the utility of integrating GCN and GAN for complex biological network analysis.