Technical Highlight - September 2012
Short description: RPF, an 'R-factor'-like protein structure validation tool, shows its mettle for the validation of NMR structures.
In the field of X-ray crystallography, the R factor, often referred to as the reliability factor, is a widely accepted measure of how well a solved structure satisfies raw experimental data. In the much younger field of protein NMR spectroscopy, a number of tools have been developed to validate various aspects of structure models, but thus far no single method has emerged to validate overall structure quality. To fill this niche, Montelione and colleagues (PSI NESG) have developed the RPF protein structure quality assessment program and compared its performance to that of other validation tools.
RPF requires atomic coordinates, chemical-shift data and nuclear Overhauser enhancement spectroscopy (NOESY) peak lists. An RPF analysis begins with a breakdown of the NOESY peaks. “Recall” measures how many of these peaks are explained by the 3D structure being analyzed, and “Precision” measures the relevance of these peaks 1H-1H distances ≤5Å from the query structure. These two scores are combined in an F-measure metric, which is then normalized to obtain a discrimination power (DP) score to report how the given structure explains the experimental NMR data relative to how a random-coiled structure would do so.
The RPF report features “Precision Violations” — short 1H-1H distances that are not supported by NOESY peaks. These false-positive interactions are mapped on the structural model based on a heat index from red (representing many or large violations) to blue. All Precision and Recall Violations are also tabulated to provide the details needed for local refinement of the 3D structure.
The authors also evaluated RPF performance relative to that of Verify3D, ProsaII, PROCHECK and MolProbity, each of which assesses various aspects of NMR structure. This was done for 63 proteins with NMR structure data obtained in a fully automated fashion that also had manually refined NMR structures and/or X-ray crystal structures, submitted for the Critical Assessment of Automated Protein Structure Determination by NMR 2010 session. For these 63 protein ensembles, the authors report a significant correlation between structure accuracy and DP score, but no strong parallel between structure accuracy and any of the other knowledge-based validation metrics tested.
Y.J. Huang et al. RPF: A quality assessment tool for protein NMR structures.
Nucl. Acids Res. 40, W542-W546 (2012). doi:10.1093/nar/gks373