Reply To: protein structure prediction using relax — high-powered jobs

[Anonymous]

If you’re attempting to use models for molecular replacement, you’ll probably be interested in the following paper:

DiMaio et al. “Improved molecular replacement by density- and energy-guided protein structure optimization.” (2011) Nature, doi:10.1038/nature09964

My understanding is that, while lower energy structures in the Rosetta energy function were more likely to give good results, having a large diversity of structures and a thoroughly sampled conformational space was also key to finding a good search model. So instead of taking a single structure and really optimizing it with relax, its likely better to create a range of structures, and consider each of them for molecular replacement.

For the ordering, you’re probably best served by loop modeling, then backrub, then relax. That is, going from more course, drastic operations to finer, more refinement operations. Loop modeling makes big changes to backbone structure, whereas backrub makes finer, more local changes, and relax is to some extent a final minimization step. Another technique to try is to apply them in cycles to walk from one minima to the other. (Loop modeling moves you to another region of search space, and then backrub and relax reorganizes you around that new conformation, which sets you up to find a different set of conformations in the next round of loop modeling.)