Predicting Residence Time of GPCR Ligands with Machine Learning

Area of Science:

• Pharmacology
• Computational Chemistry
• Biophysics

Background:

• Drug-target residence time, the duration a drug binds to its target, is increasingly recognized as critical for therapeutic efficacy, sometimes surpassing binding affinity.
• Predicting and optimizing residence time is essential for efficient drug discovery, but is hindered by limited available data.
• G protein-coupled receptors (GPCRs) are a major class of drug targets, making kinetic data for GPCR-ligand interactions particularly valuable.

Purpose of the Study:

• To establish the largest publicly available repository of GPCR-ligand kinetic data.
• To summarize experimental evidence on factors influencing drug-target residence time.
• To outline machine learning workflows for predicting residence time.

Main Methods:

• Compilation of all currently available ligand kinetic data, focusing on GPCRs.
• Systematic review of experimental findings on properties influencing residence time.
• Development and description of two machine learning prediction workflows: a single-target model using ligand features and a multi-target model using molecular dynamics features.

Main Results:

• Creation of the most extensive public dataset of GPCR-ligand kinetic data to date.
• Identification of key experimental factors influencing residence time for inclusion in predictive models.
• Demonstration of two distinct machine learning approaches for residence time prediction.

Conclusions:

• The curated kinetic dataset and summarized experimental evidence provide a foundation for developing robust residence time prediction models.
• Machine learning offers a viable strategy for optimizing drug-target residence time in drug discovery.
• The presented workflows offer practical approaches for computational modeling of drug kinetics.