Inferring LncRNA-disease associations based on graph autoencoder matrix completion.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Long non-coding RNAs (lncRNAs) are integral to numerous biological processes.
• Understanding lncRNA-disease associations is vital for deciphering disease mechanisms and advancing diagnostics and therapeutics.

Purpose of the Study:

• To develop a novel computational framework, GAMCLDA, for accurate prediction of lncRNA-disease associations.
• To enhance the understanding of molecular mechanisms underlying human diseases through lncRNA-disease association identification.

Main Methods:

• Employed a graph autoencoder matrix completion (GAMCLDA) approach.
• Utilized graph convolutional networks for encoding node features and local graph structures.
• Integrated a cost-sensitive neural network to address class imbalance in sample data.

Main Results:

• GAMCLDA demonstrated superior prediction performance compared to existing state-of-the-art methods.
• Evaluated performance using AUC, AUPR, PPV, and F1-score metrics.
• Case studies confirmed GAMCLDA's efficacy in predicting potential lncRNA-disease relationships.

Conclusions:

• GAMCLDA provides an effective computational tool for identifying lncRNA-disease associations.
• The framework aids in understanding disease pathogenesis and offers potential for clinical applications.
• Accurate prediction of these associations is crucial for future biomedical research and development.