Predict lncRNA-drug associations based on graph neural network.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Long non-coding RNAs (lncRNAs) are implicated in diverse human diseases.
• Drug-mediated regulation of lncRNA expression presents therapeutic opportunities.
• Accurate prediction of lncRNA-drug associations (LDAs) is crucial for developing lncRNA-based therapies.

Purpose of the Study:

• To develop and validate an in-silico method for predicting LDAs.
• To leverage deep learning techniques for enhanced LDA prediction accuracy.
• To facilitate the identification of novel lncRNA-targeted drug candidates.

Main Methods:

• Utilized graph convolutional networks (GCN) and graph attention networks (GAT).
• Constructed lncRNA and drug similarity networks as input features.
• Evaluated model performance across five independent datasets.

Main Results:

• Achieved high prediction performance with average AUCs exceeding 0.92.
• Demonstrated model effectiveness through case studies and KEGG pathway analysis.
• Successfully identified novel potential LDAs.

Conclusions:

• The proposed deep learning framework provides a robust approach for predicting LDAs.
• This computational tool can significantly accelerate the discovery and development of lncRNA-targeted drugs.
• The study offers a valuable resource for advancing precision medicine in lncRNA-related pathologies.