An update for AlphaFold3 versus experimental structures: assessing the precision of small molecule binding in GPCRs.

Abstract

G protein-coupled receptors (GPCRs) are key drug discovery targets with many of them modulated by small molecules via diverse binding mechanisms. AlphaFold3, a leading structure prediction tool, models GPCR-small molecule complexes, but its accuracy remains insufficiently evaluated. In this study we compared 74 AlphaFold3-predicted structures to experimental counterparts. We showed that while AlphaFold3 accurately captured global receptor architecture and orthosteric binding pockets, which was consistent with our previous research, its ligand positioning was highly variable and often inaccurate, rendering predictions unreliable, particularly for allosteric modulators. The significant divergence from experimental structures, particularly for complex ligand interactions, highlighted AlphaFold3's limitations and underscored that experimental structures remained essential for validating ligand-binding accuracy in GPCR complexes. These findings suggest that while AlphaFold3 offers potential for structure-based drug design, its current inaccuracies necessitate substantial refinement and integration with experimental data. This study highlights the limitation of AlphaFold3 in predicting small molecule binding and reinforces the critical role of high-resolution experimental validation for reliable GPCR-ligand interactions.