A Novel Drug-Disease Association Prediction Method Based on Deep Non-Negative Matrix Factorization with Local Graph...

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Traditional drug screening is costly and inefficient.
• Existing computational models struggle with deep feature extraction for drug repurposing.
• Accurate prediction of drug-disease associations is crucial for efficient drug development.

Purpose of the Study:

• To develop a novel computational model, DNMF-DDA, for enhanced drug repurposing.
• To improve the accuracy of predicting drug-disease associations, especially for novel drugs.
• To leverage deep matrix factorization with graph Laplacian and regularization for complex relationship modeling.

Main Methods:

• Developed a DNMF-DDA model integrating drug/disease similarity and association data.
• Applied k-nearest neighbors (KNN) for preprocessing to enhance matrix density.
• Incorporated graph Laplacian and relaxed regularization for feature optimization.
• Used non-negativity constraints for biologically meaningful predictions.

Main Results:

• DNMF-DDA demonstrated superior performance in predicting drug-disease associations.
• The model significantly outperformed five state-of-the-art methods in cold-start tests and cross-validation.
• Achieved high accuracy in handling high-dimensional data and mitigating cold-start issues.

Conclusions:

• DNMF-DDA offers a powerful and accurate approach for computational drug repurposing.
• The model provides valuable insights for drug development and efficiently handles complex data.
• Case studies confirmed the practical applicability and significant value of the DNMF-DDA model.