deepDR: a network-based deep learning approach to in silico drug repositioning

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Traditional drug discovery is slow and expensive.
• Drug repurposing offers a faster, cheaper alternative.
• Existing computational methods struggle with complex biological network data.

Purpose of the Study:

• To develop an advanced in silico drug repurposing method.
• To improve the identification of novel drug-disease associations.
• To leverage heterogeneous biological networks for drug repositioning.

Main Methods:

• Developed deepDR, a network-based deep learning approach.
• Integrated 10 diverse biological networks (drug-disease, drug-side-effect, drug-target, drug-drug).
• Utilized multi-modal deep autoencoder and variational autoencoder for feature learning and prediction.

Main Results:

• Achieved high performance in predicting drug-disease associations (AUROC = 0.908).
• Outperformed existing network-based and machine learning methods.
• Validated predictions using ClinicalTrials.gov (AUROC = 0.826) and identified potential treatments for Alzheimer's and Parkinson's diseases.

Conclusions:

• deepDR provides a powerful tool for in silico drug repurposing.
• The approach effectively captures complex network structures for accurate predictions.
• Identified promising drug candidates for neurodegenerative diseases, warranting further investigation.