Abstract:Cryo-EM and X-ray crystallography provide crucial experimental data for obtaining atomic-detail models of biomacromolecules. Refining these models relies on library-based stereochemical restraints, which, in addition to being limited to known chemical entities, do not include meaningful noncovalent interactions relying solely on nonbonded repulsions. Quantum mechanical (QM) calculations could alleviate these issues but are too expensive for large molecules. We present a novel AI-enabled Quantum Refinement (AQuaRef) based on AIMNet2 neural network potential mimicking QM at substantially lower computational costs. By refining 41 cryo-EM and 30 X-ray structures, we show that this approach yields atomic models with superior geometric quality compared to standard techniques, while maintaining an equal or better fit to experimental data.

What problem does this paper attempt to address?

The main goal of this paper is to propose a new method—Artificial Intelligence Accelerated Quantum Refinement (AQua Ref)—for the refinement of protein structures. Traditionally, protein structure refinement relies on library-based stereochemical constraints, which are limited to known chemical entities and do not include meaningful non-covalent interactions. Additionally, while quantum mechanics (QM) calculations can improve these issues, they are computationally too expensive for large molecules. The AQua Ref method is based on the AIMNet2 neural network potential, which can simulate QM behavior at a cost far lower than QM calculations. By refining 41 cryo-electron microscopy (cryo-EM) and 30 X-ray crystallography structures, researchers demonstrated that the atomic models produced by this method have better geometric quality than standard techniques while maintaining or improving the match with experimental data. Specifically, the study addresses the following issues: 1. **Improving structural accuracy**: Enhancing the geometric quality of protein structure models using quantum mechanics calculations, especially in low-resolution cases. 2. **Reducing overfitting**: The AQua Ref method allows researchers to reduce the overfitting between the model and experimental data. 3. **Enhancing hydrogen bond quality**: Considering hydrogen bond parameters during refinement to improve the quality of hydrogen bonds in the model. 4. **Increasing consistency with the true structure**: After comparison with high-resolution homologous models, it was found that models refined using AQua Ref are closer to the true structure. 5. **Reducing computational complexity**: By using the AIMNet2 neural network potential, linear computational scaling is achieved, making quantum refinement of large protein systems possible. In summary, this study aims to leverage machine learning to accelerate quantum mechanics calculations, enabling more efficient and accurate refinement of protein structures to obtain higher quality structural models.

AQuaRef: Machine learning accelerated quantum refinement of protein structures

Quantum refinement in real and reciprocal space using the Phenix and ORCA software

Atomic protein structure refinement using all-atom graph representations and SE(3)-equivariant graph neural networks

Accelerating reliable multiscale quantum refinement of protein-drug systems enabled by machine learning

Atomic protein structure refinement using all-atom graph representations and SE(3)-equivariant graph transformer

A new method on reconstructing protein structure from NOESY distances

A physics-aware neural network for protein–ligand interactions with quantum chemical accuracy

MD-assisted Refinement of X-Ray Coordinates

A Physics-Aware Neural Network for Protein-Ligand Interactions with Quantum Chemical Accuracy

Accurate flexible refinement for atomic-level protein structure using cryo-EM density maps and deep learning

A New Protocol for Atomic-Level Protein Structure Modeling and Refinement Using Low-to-Medium Resolution Cryo-EM Density Maps

High-accuracy protein structures by combining machine-learning with physics-based refinement

Protein model refinement for cryo-EM maps using AlphaFold2 and the DAQ score

Automated multiconformer model building for X-ray crystallography and cryo-EM

Uncovering Protein Ensembles: Automated Multiconformer Model Building for X-ray Crystallography and Cryo-EM

Enhancing cryo-EM structure prediction with DeepTracer and AlphaFold2 integration

Machine Learning Builds Full-Qm Precision Protein Force Fields in Seconds.

Machine Learning-Enhanced Quantum Chemistry-Assisted Refinement of the Active Site Structure of Metalloproteins

Ensemble Refinement of mismodeled cryo-EM RNA Structures Using All-Atom Simulations

Automated Real-Space Refinement of Protein Crystal Structures Using a Realistic Backbone Move Set

Novel Artificial Intelligence-Based Approaches for Ab Initio Structure Determination and Atomic Model Building for Cryo-Electron Microscopy