Authors

Roman Zubatyuk, Malgorzata Biczysko, Kavindri Ranasinghe, Nigel W. Moriarty, Hatice Gokcan, Holger Kruse, Billy K. Poon, Paul D. Adams, Mark P. Waller, Adrian E. Roitberg, Olexandr Isayev, Pavel V. Afonine

Purpose

The purpose of AQuaRef is to refine atomic model coordinates using geometric restraints derived from AI-accelerated quantum-mechanical calculations. Real- and reciprocal-space refinements are supported using crystallographic or cryo-EM experimental data.

Expectations

Atomic models resulting from AQuaRef refinement are expected to have improved geometry, as evaluated by standard validation tools such as MolProbity, Rama-Z, and hydrogen bond skew-kurtosis plots. They should also maintain a similar or improved fit to the experimental data compared to standard classic refinement.

Requirements and Current Limitations

Protein-only (plus water) models are supported at the moment. Static disorder (i.e., alternative conformations) is not supported.

The atomic model needs to be nearly final (in the later stages of standard classic refinement) and atom-complete (e.g., all hydrogen atoms are present explicitly). The model is also expected to have meaningful geometry, such as being devoid of overlapping (clashing) atoms or severely distorted covalent bonds.

Hardware: While this can possibly work on both CPU and GPU computers, currently available builds are specific to Linux machines with CUDA 11 or 12. Additionally, runtimes on CPU-based computers are expected to be much slower compared to GPUs.

No GUI: Command line only.

Availability

AQuaRef is available as part of Phenix or separately from Phenix as part of Q|R refinement project (www.qrefine.com or https://github.com/qrefine/qrefine).

Examples

Reciprocal space refinenemt:

qr.refine model.pdb data.mtz engine=aimnet2

Real space refinenemt:

qr.refine model.pdb map.mrc engine=aimnet2

Optimization of model geometry (no experiemntal data used):

qr.refine model.pdb engine=aimnet2

Literature