Predicting miRNA-disease associations based on multi-view information fusion

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial in biological processes; their dysregulation is linked to various diseases.
• Identifying miRNA-disease associations is vital for diagnosing and treating complex diseases.
• Existing databases offer opportunities for computational discovery, but integrating multi-source data remains challenging.

Purpose of the Study:

• To propose a novel computational method, MVIFMDA, for predicting miRNA-disease associations (MDA).
• To effectively learn and fuse information from multi-source data for improved MDA prediction.
• To enhance the diagnosis and treatment of complex diseases through accurate MDA identification.

Main Methods:

• Constructed multiple heterogeneous networks using known MDAs and miRNA/disease similarities.
• Employed graph convolutional networks (GCNs) to extract topology features from each network view.
• Utilized an attention strategy for adaptive fusion of topology representations and CNNs for attribute representations, followed by a bilinear decoder.

Main Results:

• The proposed MVIFMDA model demonstrated superior performance compared to baseline methods on a public dataset.
• Experimental results confirmed the model's effectiveness in predicting underlying miRNA-disease associations.
• Case studies further validated the model's capability in inferring novel associations.

Conclusions:

• MVIFMDA offers an effective approach for miRNA-disease association prediction by leveraging multi-view information fusion.
• The method successfully integrates topology and attribute information for robust association inference.
• This work contributes to advancing computational methods for understanding disease mechanisms and potential therapeutic targets.