Predicting Drug-Disease Associations via Multi-Task Learning Based on Collective Matrix Factorization.

Area of Science:

• Pharmacogenomics
• Computational Biology
• Bioinformatics

Background:

• Drug-disease association identification is crucial for drug development and repurposing.
• Existing computational models often predict only one type of drug-disease association (e.g., indications or side effects) independently.
• These models overlook the inherent correlations between different types of drug-disease relationships.

Purpose of the Study:

• To develop a novel computational method for predicting multiple types of drug-disease associations simultaneously.
• To leverage the correlations between therapeutic effects and non-therapeutic effects (marker/mechanism) for improved prediction accuracy.
• To identify novel drug-disease associations and their types, aiding drug discovery and safety assessment.

Main Methods:

• A collective matrix factorization-based multi-task learning (CMFMTL) framework was proposed.
• Drug-disease associations were modeled as a bipartite network with two link types: therapeutic and non-therapeutic.
• Matrix tri-factorization was employed to approximate association matrices, sharing latent representations across tasks for collective learning.

Main Results:

• CMFMTL demonstrated superior performance compared to state-of-the-art methods in predicting both therapeutic and non-therapeutic drug-disease associations.
• The method successfully identified novel drug-disease associations not previously cataloged in the Comparative Toxicogenomics Database (CTD).
• Case studies validated the ability of CMFMTL to predict the specific types of these novel associations.

Conclusions:

• The proposed CMFMTL method effectively integrates information from different drug-disease association types.
• This multi-task learning approach enhances the accuracy and scope of computational drug-disease association prediction.
• CMFMTL offers a valuable tool for accelerating drug discovery and improving drug safety by identifying novel associations and their clinical relevance.