Leveraging machine learning for selective cannabinoid ligand discovery: methods, challenges, and opportunities.

INTRODUCTION: Selective modulation of cannabinoid receptors, particularly achieving CB2 selectivity over CB1, represents a promising strategy for developing safer therapeutics with reduced psychotropic effects. This review examines how machine learning (ML) approaches can address persistent challenges in cannabinoid receptors selectivity and accelerate drug discovery. AREAS COVERED: The authors summarize current ML-based methodologies applied to cannabinoid ligand discovery, focusing on strategies for predicting receptor affinity and selectivity. The literature covered was identified through a PubMed search followed by manual screening to retain studies directly relevant to cannabinoid-focused AI-driven ligand discovery. The review discusses feature engineering approaches, including molecular fingerprints, physicochemical descriptors, and SMILES-based representations, as well as classification and regression algorithms for selectivity prediction. The authors evaluate model performance metrics, dataset limitations, and interpretability challenges. Recent advances in deep learning and generative models for de novo molecular design are also highlighted, with emphasis on their potential to expand chemical space and improve selective ligand identification. EXPERT OPINION: ML has significantly advanced the prediction of cannabinoid receptor selectivity, yet progress remains constrained by data quality, endpoint inconsistency, and limited interpretability. Future efforts integrating curated datasets, mechanistically informed modeling, and generative AI frameworks are expected to substantially enhance the discovery of selective cannabinoid therapeutics.