AQuaRef: Machine learning accelerated quantum refinement of protein structures.

Area of Science:

• Structural biology
• Computational chemistry
• Biophysics

Background:

• Cryo-electron microscopy (Cryo-EM) and X-ray crystallography are vital for atomic-level biomacromolecular models.
• Current refinement methods use library-based restraints, limiting accuracy for unknown chemical entities and noncovalent interactions.
• Quantum mechanical (QM) calculations offer higher accuracy but are computationally prohibitive for large biomolecules.

Purpose of the Study:

• To develop a computationally efficient method for refining biomacromolecular models.
• To improve the geometric quality and accuracy of atomic models derived from experimental data.
• To integrate the accuracy of quantum mechanics into macromolecular structure refinement.

Main Methods:

• Development of AI-enabled Quantum Refinement (AQuaRef) using the AIMNet2 neural network potential.
• Mimicking quantum mechanical calculations at reduced computational expense.
• Application of AQuaRef to refine 41 cryo-electron microscopy (cryo-EM) and 30 X-ray crystallography structures.

Main Results:

• AQuaRef achieved superior geometric quality in refined atomic models compared to standard techniques.
• The method maintained an equal or better fit to experimental cryo-EM and X-ray data.
• Substantially reduced computational costs were observed compared to traditional QM methods.

Conclusions:

• AI-enabled Quantum Refinement (AQuaRef) offers a powerful and efficient approach for improving biomacromolecular models.
• This method enhances structural accuracy by incorporating QM-level accuracy at a lower computational cost.
• AQuaRef represents a significant advancement in the refinement of structures obtained from cryo-EM and X-ray crystallography.