Comparing models and experimental structures of the GPR101 receptor: Artificial intelligence yields highly accurate...

Area of Science:

• Structural biology
• Computational modeling
• G protein-coupled receptors (GPCRs)

Background:

• Experimental cryo-electron microscopy structures for GPR101, a GPCR linked to X-linked acrogigantism (X-LAG), are now available.
• Previous computational models, including homology and AI-generated models, were developed for GPR101.

Purpose of the Study:

• To compare the accuracy of previously published computational models (homology, AlphaFold2, AlphaFold-Multistate) against new experimental GPR101 structures.
• To evaluate the relative strengths of different modeling approaches for GPCRs, particularly GPR101.

Main Methods:

• Comparative analysis of experimental cryo-EM structures with in-house homology models and third-party AI models (AlphaFold2, AlphaFold-Multistate).
• Assessment of model accuracy across different regions of the GPR101 receptor.
• Evaluation of the impact of molecular dynamics simulations on model accuracy.

Main Results:

• Both homology and AI models demonstrated considerable accuracy, with AI methods generally showing superiority.
• AlphaFold2 models exhibited high fidelity in capturing structural features, including the challenging second extracellular loop.
• A homology model accurately predicted G protein binding and showed superior accuracy for the sixth transmembrane domain (TM6) compared to AI models.

Conclusions:

• AI methods, especially AlphaFold2, are highly effective for GPCR modeling, but homology modeling may be superior for specific domains when suitable templates exist.
• Molecular dynamics simulations had inconsistent effects on model accuracy.
• This study provides valuable insights for modeling GPCRs lacking experimental structures, guiding future computational efforts.