CIRCE: Web-Based Platform for the Prediction of Cannabinoid Receptor Ligands Using Explainable Machine Learning

Area of Science:

• Pharmacology and Medicinal Chemistry
• Computational Chemistry
• Artificial Intelligence in Drug Discovery

Background:

• The endocannabinoid system regulates various physiological processes and diseases.
• Cannabinoid receptor 1 (CB1R) and 2 (CB2R) are key targets, but their structural similarity hinders selective ligand development.
• Existing challenges in developing subtype-selective cannabinoid receptor ligands.

Purpose of the Study:

• To develop an explainable machine learning platform, CIRCE, for predicting selective CB1R and CB2R ligands.
• To support the rational design of novel cannabinoid receptor modulators.
• To overcome the challenge of subtype selectivity in cannabinoid drug discovery.

Main Methods:

• Implementation of multilayer classifiers combined with Shapley value analysis for explainable predictions.
• Development of the Cannabinoid Iterative Revaluation for Classification and Explanation (CIRCE) compound prediction platform.
• Utilizing explainable machine learning for ligand design and prediction.

Main Results:

• CIRCE achieved approximately 80% accuracy in predicting selective ligands.
• Identified and rationalized structural features crucial for ligand-target interactions.
• Demonstrated the utility of explainable AI in guiding the design of selective CB1R and CB2R ligands.

Conclusions:

• The CIRCE platform offers a powerful, explainable approach to designing selective cannabinoid receptor ligands.
• Explainable AI can effectively address challenges in developing subtype-selective drugs.
• CIRCE is available as a free web-based tool to advance cannabinoid research.