Biophysical Society Thematic Meeting | Bucharest 2026

Biophysics of Membrane Reactions in Brian

Tuesday Speaker Abstracts

ML MAPPING OF GPCR INTRACELLULAR POCKET CONFORMATIONS TO MULTIDIMENSIONAL PHARMACOLOGICAL EFFICACY David Minh Illinois Institute of Technology, Chemistry, Chicago, IL, USA We have developed machine learning methods to train models that map from the probability of intracellular pocket conformations observed in molecular simulations of ligand-bound G protein coupled receptors to their efficacy in multiple pharmacological assays. Simulated structures are clustered into conformations that are input into multiple linear regression models. Using fewer than 15 complexes per receptor, we have trained accurate models (mean absolute error of Emax <20%) for CNS-relevant opioid (MOR, DOR, and KOR), 5-HT2A serotonin, and CB2 cannabinoid receptors in transducer recruitment (both G protein subtype and arrestin) and secondary messenger (cAMP) assays. These empirical results suggests a linear proportionality between intracellular pocket conformation probability and signaling efficacy. In addition to outputting multidimensional efficacy, the ML models are highly interpretable as three dimensional structures with features characteristic of GPCR activation including outward motion of transmembrane helices 5 and 6 and rotation of microswitches.

12