Abstract
G protein-coupled receptors (GPCRs) are critical drug targets involved in numerous physiological processes, yet many of their structures remain unresolved due to inherent flexibility and diverse ligand interactions. This study systematically evaluates the accuracy of AlphaFold3-predicted GPCR structures compared to experimentally determined structures, with a primary focus on ligand-bound states. Our analysis reveals that while AlphaFold3 shows improved performance over AlphaFold2 in predicting overall GPCR backbone architecture, significant discrepancies persist in ligand-binding poses, particularly for ions, peptides, and proteins. Despite advancements, these limitations constrain the utility of AlphaFold3 models in functional studies and structure-based drug design, where high-resolution details of ligand interactions are crucial. We assess the accuracy of predicted structures across various ligand types, quantifying deviations in binding pocket geometries and ligand orientations. Our findings highlight specific challenges in the computational prediction of ligand-bound GPCR structures, emphasizing areas where further refinement is needed. This study provides valuable insights for researchers using AlphaFold3 in GPCR studies, underscores the ongoing necessity for experimental structure determination, and offers direction for improving protein-ligand interaction predictions in future computational models.