A comparative analysis of computational drug repurposing approaches: proposing a novel tensor-matrix-tensor...

Area of Science:

• Computational drug discovery
• Pharmacology
• Bioinformatics

Background:

• Drug repurposing accelerates the identification of new therapeutic uses for existing drugs.
• Developing efficient computational methods is crucial for successful drug repurposing.
• Various computational strategies exist, each with unique strengths and weaknesses.

Purpose of the Study:

• To introduce a novel tensor-matrix-tensor (TMT) formulation for drug repurposing.
• To compare the performance of factorization-based, machine learning, deep learning, and graph neural network methods.
• To evaluate the efficacy and reliability of different computational drug repurposing approaches.

Main Methods:

• Developed a novel tensor-matrix-tensor (TMT) data array method.
• Employed a gradient-based factorization procedure within the TMT framework.
• Compared four distinct computational drug repurposing strategies: factorization-based, machine learning, deep learning, and graph neural networks.
• Tested methods on two distinct datasets.

Main Results:

• Deep learning methods exhibited superior performance and reliability in drug repurposing tasks.
• Factorization-based methods and traditional machine learning showed moderate effectiveness.
• Graph neural networks require inductive capabilities for reliable predictions.
• The TMT formulation provides a novel approach to data representation in drug repurposing.

Conclusions:

• Deep learning represents a promising avenue for enhancing drug discovery through repurposing.
• The TMT method offers a new computational tool for analyzing drug-gene-disease relationships.
• Further development of graph neural networks is needed to optimize their application in drug repurposing.