Technical Highlight - May 2014
Short description: Combinatorial use of refinement algorithms leads to improved structure determination.
During structure refinement, crystallographers must choose a combination of the many available refinement algorithms that leads to the highest quality structure. To achieve this, it is common to use a decision tree in which single algorithms are evaluated for their ability to improve the model. The final refinement strategy is defined by the combination of those algorithms that effectively improved the model individually. This approach assumes, however, that each algorithm is independent, and that models will not worsen over the course of refinement.
Nettles and colleagues recently tested these assumptions by developing Extensive Combinatorial Refinement (ExCoR), a combinatorial refinement approach in which 256 distinct combinations of refinement parameters and algorithms are applied in parallel to determine an ideal refinement strategy. In applying ExCoR to 53 different protein structures, the authors showed that this method can reveal improved refinement protocols for each sample—even previously refined and published structures—and can allow for the automated correction of side chains, main chains and ligands.
Although the authors expected ExCoR might reveal optimal refinement strategies that could be generally applicable to most protein structures, no single strategy consistently produced the best model. Even more surprising was the observation that, in some cases, combining two algorithms that individually made the model worse could result in a better final model. These data suggest that refinement algorithms should not be evaluated in isolation, and that refinement strategies that temporarily worsen a model may be beneficial in forcing models out of local energy minima.
ExCoR was demonstrated to improve models at all stages of the refinement process, and should aid researchers in identifying which combination of refinement algorithms will yield the best final structure. Additionally, by eliminating the need to make decisions about which refinement algorithms to use, ExCoR may provide another step toward more automated structure solution.
J. C. Nwachukwu et al. Improved crystallographic structures using extensive combinatorial refinement.
Structure. 21, 1923-1930 (2013). doi:10.1016/j.str.2013.07.025