Artificial Intelligence/Machine Learning-Driven Small Molecule Repurposing via Off-Target Prediction and...

Area of Science:

• Computational chemistry
• Drug discovery and development
• Pharmacology

Background:

• Small molecule drug discovery faces high attrition rates (~90%) due to toxicity or efficacy issues.
• Approved drugs interact with multiple targets (6-11 on average), suggesting repurposing potential.
• Leveraging off-target interactions can uncover new therapeutic applications for existing compounds.

Purpose of the Study:

• To develop and validate a computational framework for small molecule drug repurposing.
• To identify novel off-target interactions for FDA-approved drugs.
• To explore potential new therapeutic applications based on predicted drug-target interactions.

Main Methods:

• Integrated artificial intelligence/machine learning (AI/ML) and chemical similarity approaches.
• Employed eight distinct target prediction methods, including three machine learning models.
• Analyzed a dataset of 2766 FDA-approved drugs and cross-species transcriptomics data.

Main Results:

• Identified 27,371 off-target interactions across 2013 protein targets for 2766 drugs.
• Found 150,620 structurally similar compounds to the drugs in the dataset.
• Confirmed 63% (17,283) of predicted off-target interactions in vitro, with many showing high affinity (IC50 < 100 nM or < 10 nM).
• GPCRs, enzymes, and kinases were the most frequent target classes for predicted interactions.

Conclusions:

• The computational framework effectively predicts numerous off-target interactions for drug repurposing.
• Validated interactions and tissue-specific expression patterns provide a basis for exploring new therapeutic uses of approved drugs.
• This approach offers a promising strategy to overcome drug discovery attrition and accelerate the development of new treatments.