Computational drug repositioning based on multi-similarities bilinear matrix factorization.

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• High-throughput technology generates vast biomedical data, enabling calculation of biological entity prior information from diverse aspects.
• Drug-drug similarities can be quantified using target profiles, interactions, and side effects.
• Multiple disease similarity measures arise from varied calculation methods and data sources, posing challenges for computational drug repositioning.

Purpose of the Study:

• To develop a dynamic method for optimizing the fusion of multiple similarities in computational drug repositioning.
• To propose the multi-similarities bilinear matrix factorization (MSBMF) method for predicting drug-associated indications.
• To enhance the accuracy and interpretability of drug repositioning predictions.

Main Methods:

• Concatenating drug and disease similarity matrices instead of fusing them into a single matrix.
• Applying matrix factorization to decompose the drug-disease association matrix into drug-feature and disease-feature matrices.
• Utilizing non-negative factorization to ensure biological interpretability and an efficient alternating direction method of multipliers algorithm for numerical solution.

Main Results:

• MSBMF achieves higher prediction accuracy compared to state-of-the-art drug repositioning methods in cross-validation experiments.
• The method demonstrates effectiveness in practical applications through case studies.
• Latent features extracted from feature matrices effectively infer missing drug-disease associations.

Conclusions:

• The proposed MSBMF method offers a robust and accurate approach for computational drug repositioning.
• MSBMF successfully integrates multiple similarity measures to improve the prediction of drug-associated indications.
• The method provides a valuable tool for identifying novel therapeutic uses for existing and new drugs.