Resolution-standardized evaluation of ligand atomic coordinates in crystallographic structures using machine learning

Accurate evaluation of ligand coordinate-density consistency across varying resolutions has long been a challenge in macromolecular crystallography. To address this issue, the authors have introduced the atomic Box Correlation Coefficient (aBCC), an atom-level metric designed to assess the consistency between ligand atomic coordinates and electron density within a resolution-standardized framework.

The development of QAEmap, a machine learning model based on three-dimensional convolutional neural networks (3D-CNNs), enabled the prediction of aBCC values from electron-density maps. Trained using Fourier-truncated electron-density maps and ligand coordinates generated from high-resolution structures in the Protein Data Bank, QAEmap was evaluated using both truncated electron-density maps and experimentally determined PDB structures.

The results showed that while the prediction accuracy of aBCC gradually decreased with decreasing resolution, it remained reliable up to a resolution of approximately 3.5 angstrom. These findings demonstrate that aBCC provides a valuable tool for resolution-standardized atom-wise evaluation of coordinate-density consistency across various resolutions, laying the groundwork for further advancements and refinements in machine learning-based coordinate validation techniques.