Biomedical data and computational models for drug repositioning: a comprehensive review

Area of Science:

• Pharmacology and Bioinformatics
• Computational Drug Discovery

Background:

• Drug repositioning offers a cost-effective and time-efficient alternative to traditional drug development.
• Advancements in high-throughput technologies generate vast biological and medical data, fueling computational approaches.
• Computational drug repositioning systematically identifies potential drug-target and drug-disease interactions.

Purpose of the Study:

• To review available biomedical data and public databases relevant to drugs, diseases, and targets.
• To discuss and categorize existing computational drug repositioning approaches.
• To analyze standard datasets, evaluation metrics, and compare prediction methods.

Main Methods:

• Summarization of biomedical data sources and public databases.
• Categorization of computational drug repositioning methods: classical machine learning, network propagation, matrix factorization/completion, and deep learning.
• Analysis of common datasets, evaluation metrics, and comparative assessment of prediction methods.

Main Results:

• Comprehensive overview of data resources for drug repositioning.
• Detailed discussion of various computational models employed in drug repositioning.
• Comparative analysis of prediction method performance on benchmark datasets.

Conclusions:

• Identified key challenges in computational drug repositioning, including data noise, incompleteness, and sparseness.
• Highlighted the need for ensemble methods, reliable negative sample selection, and robust benchmark datasets.
• Emphasized the importance of analyzing and explaining the mechanisms behind predicted drug-disease interactions.