Machine learning to estimate the local quality of protein crystal structures

Area of Science:

• Structural biology
• Computational biology
• X-ray crystallography

Background:

• Low-resolution electron density maps present significant challenges in protein structure determination and application.
• Accurate protein structures are crucial for understanding biological function and developing therapeutics.

Purpose of the Study:

• To introduce QAEmap, a novel method for evaluating local protein structure quality using low-resolution electron density maps.
• To demonstrate QAEmap's potential for correcting structural errors and assessing ligand binding in X-ray crystallography.

Main Methods:

• Developed QAEmap, a quality assessment tool based on electron density maps.
• Employed a three-dimensional deep convolutional neural network that takes electron density maps and coordinates as input.
• Predicted the correlation between local structure and a putative high-resolution experimental electron density map.

Main Results:

• QAEmap provides a metric for assessing local protein structure quality from low-resolution data.
• The predicted correlation can be utilized to refine and correct structural errors.
• The method shows potential for evaluating ligand binding, even at low resolutions.

Conclusions:

• QAEmap offers a promising approach to overcome limitations posed by low-resolution electron density maps in structural biology.
• This method can enhance the accuracy of protein structure determination and analysis.
• QAEmap may significantly aid in the study of ligand-protein interactions.