(360t) Elucidating Ligand Selectivity and Partial Agonism Towards Cannabinoid Receptors Using Machine Learning Approaches

The endocannabinoid system (ECS) consists of the two cannabinoid receptors (CBs), their endogenous ligands, and enzymes. ECS maintains homeostasis in neuron signaling and modulates a plethora of physiological processes, including pain, obesity, inflammation, and cardiovascular regulation. Therefore, cannabinoid receptors are important targets for neurological and physiological disorders. However, drug development targeting cannabinoid receptors faces overstimulation and selectivity issues. Full agonists such as Fubinaca of the cannabinoid receptors cause dangerous side effects,including impairment of fine motor skills and increased blood pressure, tachycardia. Conversely, partial agonists such as Δ⁹-tetrahydrocannabinol have been used as an appetite stimulant for AIDs patients and an antiemetic during chemotherapy with limited side effects. On the other hand, due to structural and sequence similarity, most of the drug molecules bind to both CB1 and CB2. Therefore, CB2 targeting molecules also bind to CB1, which is majorly expressed in the central nervous system (CNS) and causes psychological side effects. Thus, in this project, our objective is to develop a selective partial agonist using a data-centric machine learning approach. We build an extensive ligand activity dataset by curating information about ~6000 ligands from literature and databases. Using a graph neural network, we train our model to predict binding affinities of ligands for CB1 and CB2 separately. Our trained models show above 80% accuracy on the test dataset. These trained models were used to screen through molecules from the ZINC database to classify previously unknown molecules as selective CB-targeting partial agonists. Ultimately, our research will discover new scaffolds for developing selective partial agonists for therapeutic drug development.