GPCR_LigandClassify.py; a rigorous machine learning classifier for GPCR targeting compounds.

Area of Science:

• Computational chemistry
• Pharmacology
• Machine learning

Background:

• G-Protein Coupled Receptors (GPCRs) are a major drug target family.
• Drug repurposing offers a faster route to new therapeutics.
• Developing accurate predictive models for GPCR-ligand interactions is crucial.

Purpose of the Study:

• To construct and evaluate ligand-based machine learning models for drug repurposing against GPCRs.
• To compare the performance of various machine learning algorithms and featurization methods.
• To identify novel drug-GPCR associations.

Main Methods:

• Collected over 500,000 data points from the GPCR-Ligand Association (GLASS) database.
• Employed diverse molecular featurization techniques.
• Developed and compared supervised machine learning algorithms, including ensemble decision trees, gradient boosted trees, and deep neural networks (DNNs).

Main Results:

• Machine learning models, particularly ensemble and gradient boosted trees with molecular fingerprinting, demonstrated performance comparable to DNNs.
• Models achieved approximately 90% classification accuracy on a test dataset.
• Identified potential drug-GPCR associations, aligning with existing literature.

Conclusions:

• Ligand-based machine learning models are effective tools for GPCR drug repurposing.
• Ensemble and gradient boosted trees offer a competitive alternative to DNNs for this task.
• The developed models can accelerate computational drug discovery and identify novel therapeutic applications.