Predicting Drug-Disease Associations Through Similarity Network Fusion and Multi-View Feature Projection...

Area of Science:

• Computational biology
• Pharmacogenomics
• Bioinformatics

Background:

• Drug-disease association (DDA) prediction is crucial for efficient drug development.
• Existing computational methods often overlook potential associations in unconfirmed pairs.
• Integrating multiple data resources into heterogeneous networks is a common approach.

Purpose of the Study:

• To propose a novel computational model for predicting new drug-disease associations (DDAs).
• To address the limitation of existing methods in considering unconfirmed drug-related or disease-related pairs.
• To enhance the accuracy and efficiency of DDA prediction in drug development.

Main Methods:

• Construction of a heterogeneous network with drugs, targets, cell lines, and diseases.
• Application of an updating and merging-based similarity network fusion (UM-SF) method.
• Utilizing an intermediate layer-mediated multi-view feature projection representation (IM-FP) method for DDA scoring.

Main Results:

• The proposed model demonstrates effectiveness through comparative experiments.
• 10-fold cross-validation shows superior performance on AUROC and AUPR metrics compared to state-of-the-art models.
• The model successfully predicted 107 novel high-ranked drug-disease associations.

Conclusions:

• The novel computational model significantly improves the prediction of drug-disease associations.
• The UM-SF and IM-FP methods offer innovative solutions for integrating diverse biological data.
• This approach holds promise for accelerating drug discovery and development.