MAGCNSE: predicting lncRNA-disease associations using multi-view attention graph convolutional network and stacking...

Area of Science:

• Genomics
• Computational Biology
• Biomedical Informatics

Background:

• Long non-coding RNAs (lncRNAs) play crucial roles in human biological processes and disease development.
• Accurate prediction of lncRNA-disease associations is vital for identifying disease biomarkers and advancing disease analysis and prevention.
• Developing effective computational methods for predicting these associations remains a critical challenge.

Purpose of the Study:

• To propose a novel computational model, MAGCNSE, for predicting underlying long non-coding RNA (lncRNA)-disease associations.
• To leverage multi-view similarity graphs and advanced deep learning techniques for enhanced prediction accuracy.
• To validate the model's efficacy through rigorous ablation studies and comparisons with existing state-of-the-art methods.

Main Methods:

• Utilized graph convolutional networks (GCNs) to extract features from multi-view similarity graphs of lncRNAs and diseases.
• Employed an attention mechanism to adaptively assign weights to different feature matrices.
• Integrated convolutional neural networks (CNNs) for feature extraction from multi-channel matrices and a stacking ensemble classifier for final prediction.

Main Results:

• Ablation studies confirmed the validity and contribution of each module within the MAGCNSE model.
• MAGCNSE demonstrated superior performance compared to six other state-of-the-art models in predicting lncRNA-disease associations.
• The effectiveness of utilizing multi-view data was verified, and case studies highlighted MAGCNSE's capability in identifying potential associations.

Conclusions:

• The MAGCNSE model provides a robust and effective computational approach for predicting lncRNA-disease associations.
• The findings suggest MAGCNSE's potential utility in disease biomarker discovery and understanding disease mechanisms.