ResProx

ResProx estimates atomic resolution of protein structures using machine learning to quantify structural accuracy and detect refinement defects.

Key Features:

• Machine Learning Estimation: Predicts resolution from 25 structural features derived from atomic coordinates, including ensemble precision, torsion angle normality, and distance restraints per residue.
• Comprehensive Evaluation: Integrates multiple structural features to assess resolution with a mean absolute error of 0.28 Å.
• Structural Defect Detection: Identifies under-restrained, poorly refined, or inaccurate NMR structures using coordinate data as a proxy for X-ray diffraction data.
• Performance: Correlation coefficient between observed and calculated resolutions reported as 0.92.

Scientific Applications:

• Structural Comparison and Validation: Provides a common resolution metric for evaluating NMR and X-ray crystallography-derived protein structures.

Methodology:

Machine learning model trained on 25 structural features derived from atomic coordinates, including ensemble precision, torsion angle normality, and distance restraints per residue, to predict resolution and compare calculated versus observed resolutions (correlation 0.92, MAE 0.28 Å).