ACLNDA: an asymmetric graph contrastive learning framework for predicting noncoding RNA-disease associations in...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Noncoding RNAs (ncRNAs), including long noncoding RNAs (lncRNAs) and microRNAs (miRNAs), are vital regulators of gene expression with significant implications in disease.
• Accurate prediction of ncRNA-disease associations is crucial for understanding disease pathogenesis and developing novel therapeutic strategies.
• Current prediction methods often address individual association types (e.g., lncRNA-disease, miRNA-disease) and do not fully leverage the rich information within heterogeneous biological networks.

Purpose of the Study:

• To develop an advanced computational framework, ACLNDA, for predicting associations between ncRNAs and diseases.
• To effectively utilize heterogeneous graph characteristics for more accurate and comprehensive association predictions.
• To provide a versatile tool applicable to lncRNA-disease associations (LDAs), miRNA-disease associations (MDAs), and lncRNA-miRNA interactions (LMIs).

Main Methods:

• ACLNDA employs an asymmetric graph contrastive learning framework to analyze heterophilic ncRNA-disease associations.
• It constructs a triple-layer heterogeneous graph by creating inter-layer adjacency matrices and using a Top-K intra-layer similarity edges approach.
• The method uniquely integrates node attribute and preference features, maximizing neighborhood context and similarity without relying on graph augmentations or homophily assumptions.

Main Results:

• ACLNDA demonstrates superior performance compared to existing state-of-the-art methods in predicting ncRNA-disease associations.
• The framework effectively extracts ncRNA-disease features, maintaining data integrity and reducing computational costs.
• Experimental results confirm the broad applicability of ACLNDA for LDA, MDA, and LMI predictions.

Conclusions:

• ACLNDA offers a powerful and efficient approach for predicting ncRNA-disease associations, advancing our understanding of disease mechanisms.
• The framework's ability to integrate diverse association types and leverage heterogeneous graph structures holds significant potential for disease diagnosis and therapeutic target discovery.
• The public availability of ACLNDA's source code and data facilitates further research and application in the field.