On Fri, Jun 25, 2010 at 12:41 AM, Francois Berenger <[email protected]>wrote:
> When you say "After molecular replacement you should check the result > with the model building program of your choice and correct as many > errors as possible before running a refinement program." > > Can this step be automated in some way (without any > human intervention)? > > Are there programs to do this? > I am interested even if some program does only one part of the job. > Both ARP/wARP and phenix.autobuild will do this (again, given data of sufficiently high quality). But I disagree with the original statement. Molecular replacement is typically done at reduced resolution, and particularly where the search model is different in conformation or sequence from the protein in the crystal, there will be many errors, even if the solution is generally excellent. Attempting to correct each of these manually is a waste of time, since the refinement or rebuilding programs will do it faster and better. If there are severe problems like overlapping molecules, this will become obvious very quickly whether or not automation is used. Is it even possible to solve, rebuild, and refine a structure from MR in the wrong space group? I would expect the R-factors to be unreasonably high and the maps uninterpretable. I do agree that inspecting the lattice visually after MR is a good idea (I like PyMOL for this). I don't know whether there are programs that will do this automatically - it would be a useful tool to have. As always, there will still be corner cases, where the search model is too remote for the automated steps to work and manual fixing is required, or you get packing clashes before all copies are found and need to prune the search model, or the crystal resolution is too poor, and so on. It is also possible to waste time trying to improve a nonsensical solution - until your R-free is well below 0.5, you can't be certain that you've solved the structure, and if the automated methods don't push it below 0.4, it's probably worth checking your MR solution to make sure it's already as good as possible. The more general point, that no one should publish or deposit a structure without ever thoroughly checking the maps, validation, and lattice contacts, I of course agree with - but I haven't heard anyone propose doing away with human intervention altogether, nor are any of the automated pipelines anywhere close to making this feasible. -Nat
