Predicting circRNA-disease associations with shared units and multi-channel attention mechanisms

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Circular RNAs (circRNAs) are implicated in disease but their associations are costly to study experimentally.
• Existing computational models often fail to fully leverage multi-view data or account for varying view significance.

Purpose of the Study:

• To develop an efficient computational model for predicting circRNA-disease associations.
• To overcome the limitations of experimental methods and existing computational approaches.

Main Methods:

• Developed MSMCDA, integrating multi-view shared units and multichannel attention mechanisms.
• Constructed similarity and meta-path networks with shared units for interactive learning.
• Optimized network weights using multichannel attention and strengthened features with contrastive learning.

Main Results:

• MSMCDA significantly outperformed baseline methods across five public datasets.
• Validated effectiveness through case studies on colorectal, gastric, and nonsmall cell lung cancers.
• Successfully uncovered novel circRNA-disease associations.

Conclusions:

• MSMCDA provides a powerful and cost-effective computational approach for predicting circRNA-disease associations.
• The model's ability to identify new associations holds promise for disease research and biomarker discovery.