Non-Negative Matrix Tri-Factorization for Representation Learning in Multi-Omics Datasets with Applications to Drug...

Area of Science:

• Biomedical Informatics
• Computational Pharmacology
• Artificial Intelligence in Drug Discovery

Background:

• Heterogeneous biomedical data offers potential for drug design but requires integration.
• Current methods struggle to effectively fuse diverse data sources for AI analysis.
• Developing robust data representations is key for generating novel drug candidates.

Purpose of the Study:

• To propose Non-Negative Matrix Tri-Factorization (NMTF) for data integration and representation learning in drug design.
• To demonstrate the utility of NMTF-derived representations with traditional AI methods.
• To apply and validate NMTF in computational pharmacology, specifically for drug repurposing and predicting drug synergism.

Main Methods:

• Non-Negative Matrix Tri-Factorization (NMTF) for data integration, fusion, and representation learning.
• Application of NMTF to heterogeneous biomedical data sources (databases, ontologies, terminologies).
• Utilizing NMTF-learned representations with established artificial intelligence techniques.

Main Results:

• NMTF effectively integrates and fuses diverse biomedical data for AI analysis.
• NMTF-learned representations improve performance of traditional AI methods in drug design tasks.
• Validated novel drug-target predictions and demonstrated effectiveness in predicting drug synergism.

Conclusions:

• Non-Negative Matrix Tri-Factorization is a flexible and powerful tool for biomedical data integration and representation learning.
• The method significantly advances computational pharmacology, drug repurposing, and personalized medicine.
• NMTF shows broad applicability to various drug design and discovery challenges.