GDCL-NcDA: identifying non-coding RNA-disease associations via contrastive learning between deep graph learning and...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Non-coding RNAs (ncRNAs) are crucial for biological processes.
• Computational methods complement wet experiments for ncRNA research but face challenges like false negatives and poor generalization.
• Existing methods often fail to fully leverage multi-source information and lack robustness across datasets.

Purpose of the Study:

• To develop an effective computational framework for identifying latent ncRNA-disease associations.
• To improve the accuracy, robustness, and generalization of ncRNA-disease association prediction.
• To integrate diverse multi-source heterogeneous networks (MHNs) for enhanced prediction.

Main Methods:

• Proposed GDCL-NcDA, an end-to-end framework combining deep graph learning and deep matrix factorization (DMF) with contrastive learning (CL).
• Utilized deep graph convolutional networks and attention mechanisms to integrate MHNs, including ncRNA, gene, and disease similarities and associations.
• Employed DMF to predict latent associations and CL to enhance model generalization and robustness on reconstructed and predicted graphs.

Main Results:

• GDCL-NcDA demonstrated superior performance compared to existing computational methods.
• Experimental results validated the framework's effectiveness in identifying diverse ncRNA-disease associations.
• Case studies confirmed the practical utility of GDCL-NcDA in biological discovery.

Conclusions:

• GDCL-NcDA offers an effective approach for predicting ncRNA-disease associations by leveraging multi-source data and advanced deep learning techniques.
• The framework addresses limitations of previous methods, including false negatives and poor generalization.
• GDCL-NcDA shows significant potential for advancing ncRNA research and understanding disease mechanisms.