Technical Highlight - January 2014
Short description: Combining the sampling methodology and energy function of Rosetta and the X-ray refinement methodology of Phenix improves refinement of low-resolution X-ray crystallographic data.
Improved approaches and refinement strategies have allowed for enormous progress in the determination of medium- to high-resolution macromolecular structures. However, refinement of structures obtained from low-resolution data has lagged behind, due to the poor ability of existing methods to converge on a model structure with reasonable geometry.
To address this issue, DiMaio, Echols and colleagues have developed a refinement method (phenix.rosetta_refine) that combines the all-atom force field energy and conformation sampling functions of Rosetta with the maximum-likelihood refinement function of Phenix to optimize poor starting models against low-resolution diffraction data. Phenix is used to calculate electron density maps, perform bulk solvent correction and calculate B factors, while Rosetta optimizes model geometry via force field, minimizer and sampling functions. The method switches between real- and reciprocal-space refinement; the Rosetta force field limits reciprocal-space refinement to physically possible conformations while the electron density maps guide real-space backbone and side-chain sampling. These two features are what distinguish Rosetta-Phenix from existing low-resolution refinement methodologies.
The authors assess their methodology using a test set comprising 15 low-resolution (3–4.5 Å) data sets. Refinement using Rosetta-Phenix was performed following molecular replacement of a suitable starting model in Phaser. Rosetta-Phenix refinement was compared against three other low-resolution refinement methodologies: phenix.refine (as a control), CNS-DEN and REFMAC5. The final refined models were validated in MolProbity and compared against PDB-deposited protein structures. In nearly all cases, Rosetta-Phenix yielded better model quality. In particular, while CNS-DEN and REFMAC5 were able to extend the radius of convergence, they could not improve model geometry.
Rosetta-Phenix was also able to refine a high-resolution test set, achieving improved MolProbity scores, but not changing reported Rfree values. Rosetta-Phenix requires installation of Rosetta (version 3.6 or newer) and Phenix (version 1.8.3 or newer)—both free to academics—and can be automated using the phenix.rosetta_refine program in Phenix.
F. DiMaio et al. Improved low-resolution crystallographic refinement with Phenix and Rosetta.
Nat. Methods 10, 1102-1104 (2013). doi:10.1038/nmeth.2648