Dear Kevin Please re-read my earlier email paying particular attention to the words "explicitly" and "expressly". To my mind, the notion of moving a Ramachandran outlier by moving a dot across a graph (which you feel is a danger) is identical to expressly forcing that residue into a favoured Ramachandran region. What you suggest doing is pretty much what I do do. I pay a great deal of attention to the Ramachandran plot at every stage of refinement, noting outliers and looking for the sorts of problems you mention, but I do not attempt to correct register errors (for example) by restraining phi/psi angles! Gert and Robbie have explained clearly why phi/psi angles do not lend themselves easily to fitting, and it seems best (to me at least) to leave the Ramachandran plot as an unfudged indicator of the interesting/problematical residues, not only for subsequent PDB users but for the crystallographer him/herself during refinement.
I think we are in perfect agreement up to this point. Where we differ is in your apparent view that an outlier residue must be "fixed". The first structure I ever solved had one residue with unusual backbone geometry, and I was given substantial grief from a modeller in the group over it - but it turned out to be a crucial functional residue at a hinge which flips between open and closed forms. Of course, it can be a pain explaining to a reviewer/editor/supervisor/PDB-bod why your model is flagged by validation software, but sometimes this is where the interesting bits are. If we have such a strongly defined preconception of what proteins look like then isn't the experimental data becoming undervalued? There are many parts of proteins that are not modelled well by single conformations, and the expectation that every atom will sit nicely in one place and behave is sometimes a little unrealistic. Dear Judit I am sure there are cases at AstraZeneca where repeated refinement of a well-known protein with umpteen different test drugs is expedited by the tools you suggest, but the aim of such experiments is rather different from an initial structure determination. I suspect that if Michael Murphy had wanted to know how to use a prior high-resolution model as a set of restraints for later lower resolution refinements then he would have framed his original question rather differently. When you say "many crystallographers now prefer better quality structures over pedantic validation tools..." I have to wonder what is meant by "quality", but that is perhaps opening too large a can of worms. Basically though, beauty and truth are not synonymous, whether or not we regard protein crystallography as a video game where the person with the lowest R-factor wins. best wishes to all! Jeremy the Pedant Begin forwarded message: > From: Kevin Jude <kj...@stanford.edu> > Date: February 27, 2015 3:25:24 AM GMT+09:00 > To: Jeremy Tame <jt...@tsurumi.yokohama-cu.ac.jp> > Cc: CCP4BB@jiscmail.ac.uk > Subject: Re: [ccp4bb] adjusting bad Ramachandran angles > > I think the Ramachandranplot should be used in the refinement and rebuilding > process - a Ramachandran outlier is a flag that that region of the model > needs a closer look, and the fix may be more complicated than simply rotating > a peptide. Maybe a C-beta is pointing the wrong way, maybe there is a > register error. The danger is that people will treat a Ramachandran outlier > by moving a dot across a graph without addressing the underlying structural > problem. > > kmj > > On Wed, Feb 25, 2015 at 8:37 PM, Jeremy Tame > <jt...@tsurumi.yokohama-cu.ac.jp> wrote: > I think Goodhart's Law applies here (see the Wikipedia page): > When a measure becomes a target, it ceases to be a good measure. > > From memory I believe Randy Read and George Sheldrick have commented that > Ramachandran plots are a good measure of structure quality, and therefore > should > not be used explicitly at the modelling stage. Some residues may be difficult > if they have more than one backbone conformation or are just mobile, but > expressly holding them in favoured regions of the Ramachandran plot is not a > good > idea. The most interesting proteins are of course enzymes, and the > Ramachandran > outliers are often among the most interesting active site residues. So you > may be trying > to eliminate something which is actually more important than a low Rfactor. > > The idea of limiting data use may seem counter-intuitive, but to take another > example from economics, John Cowperthwaite was in charge of Hong Kong's > financial affairs in the 1960s. He attributed the success of the economy > under his > tenure to his adamant refusal to collect any economic data whatsoever! > > > On Feb 26, 2015, at 2:08 AM, Michael Murphy wrote: > > > Does anyone know of a way to adjust Ramachandran angles so that they fall > > within the preferred range? Either in Coot or possibly some online server? > > I have been trying to do it manually without much success, I was wondering > > whether there might another way to do it. -Thanks >
