Fine-Tuned Deep Transfer Learning Models for Large Screenings of Safer Drugs Targeting Class A GPCRs

Area of Science:

• Pharmacology
• Computational Biology
• Drug Discovery

Background:

• G protein-coupled receptors (GPCRs) are crucial for cell signaling and are major drug targets.
• Linking drug efficacy to specific signaling pathways and predicting therapeutic windows remains difficult.
• Understanding low intrinsic efficacy and ligand bias is key to developing safer drugs.

Purpose of the Study:

• To develop predictive AI models for identifying GPCR ligands with low intrinsic efficacy or biased agonism.
• To overcome data limitations in deep learning for GPCR drug discovery.
• To facilitate the discovery of safer drug candidates by predicting their bioactivity profiles.

Main Methods:

• Pretrained a deep learning model on class A GPCR sequences and ligand data.
• Utilized transfer learning and a neural network with natural language processing.
• Incorporated receptor mutation effects on signaling for model refinement.

Main Results:

• Developed two fine-tuned models: one for low-efficacy agonists and one for biased agonists.
• Models are available on demand for individual class A GPCRs.
• Enabled virtual screening of large chemical libraries for potential drug candidates.

Conclusions:

• The developed AI models can predict GPCR ligands with improved safety profiles.
• These models significantly advance drug discovery by facilitating the identification of safer compounds.
• The approach addresses the challenge of limited high-quality data in GPCR drug development.