It strongly depends on the resolution of your structure and number of residues how many outliers are acceptable. At low resolution there have to be very good reasons for disallowed conformations. Here you can consider stubs to just model the main chain if there is big trouble. In contrast, at 1 A resolution you will clearly see the atoms and therefore an outlier can be easily accepted if present.

In general keep the number as low as possible and fine-tune the refinement parameters for optimization of the Ramachandran. If that doesn't help optimize by hand.

It depends on your resolution and quality of model.

@Edward - very nicely explained.

I would only add two things. Glycines can lie outside the allowed regions and often do. And at very high resolution, if you find real outliers, they are often active sie residues, interestingly enough.

Thanks for all your comments, this is really interesting. I have read before about active site residues being in disallowed regions. Why these in particular? And is this only in the apo forms that the outliers are seen?

I'm not sure why. One possibility is a conformational star that is relieved during catalysis, as a driving force, but I know of no evidence for that. Relief of strain could also be used to tune substrate binding or aid product release. There is no study I am aware of comparing apo enzymes to liganded ones. I might expect, perhaps naively, that substrates and TS analogs would be more relevant than other kinds of inhibitors.

Dear Claire-Louise,

All of the above remarks are correct, as they address different points of the final but unique, single model that we customarily derive form protein crystallography. However, the data supports the view of proteins as dynamical ensembles, what is equivalent that proteins are mobile.

Proteins are partially solid and partially liquid what means that a part of every protein will have well defined local architecture and these regions should not have any violations, whatsoever, except for a very good chemical reasons mentioned above (i.e. prolyl isomerases). Every violation in these fragments cost free energy, which means, three to five such violations would provide sufficient energy to unfold the fragment. (For protein energetics read some external literature. Energetics is the reason for changing the standards for Ramachandran).

The other part (liquid) is seen in crystal structures in three different states. Regions are: (i) ordered but different in different crystal structure determinations, (ii) in some crystal structures are visible in some not, (iii) there are not clearly visible in any crystal structure. If these regions are refined and included in the model, they should always be represented as ensembles which they are not. In a single model representation these regions have usually distorted geometry, as the refinement programs attempt to force a single model to imitate an ensemble. If a single model is selected to represent such a fragment it must have as ideal geometry as possible, because, in absence of constraints the fragment would attain a minimal local energy and avoid violations. (By the way every violation can be translated to partial interpenetration of VdW surfaces - rigid sphere model - that initially led Ramachandran to propose his plot.)

The percentage of both regions changes and is unique to every protein (read PNAS 106(2009)10505), but every protein must have some disorder (usually deeply under-reported). The newest trend advocated by developers of Phoenix is to include the ensemble, (but abundant prior literature exist on this point).

In essence avoid violations, if you cannot minimize them, if included they must have a very good reason to exist because they cost energy so there must be a compensating factor. By the way, paradoxically, glycines in unusual conformations are unusually well conserved among the homologs.Your original question concerning what is the percentage of usually reported is a sociology of the crowd question, and as such needs to be avoided and replaced with with the science question. What our general knowledge tell us about the the possibility of existence. For instance, researchers were quite reluctant to report cis conformations of peptide bonds besides cis-Pro. In my hands the frequency of other residues in cis conformers is higher than reported.

Typically, the fraction of residues in the (residue-specific) disallowed regions should be (very) low (

Hi this question isn't quite the right thing to ask. If you look around you will find rama chandran plots shown as density plots. I think Roland Dunbrak's site has these at high resolution. Once you plot these as probability density then your question sort of vanishes. TO create a plot that dichotomizes this to "allowed" or disallowed you first need to choose a cut off probability density. then you can color it as allowed (greater than) or "disallowed" (less than) the probability you chose. But this begs the question. that is, by choosing the cutoff to generate your allowed/disallowed plot you have already answered your question!

Furthermore, this too is ill posed. If you draw the axes in "degrees" you will get one answer. But if you draw them as cosine(angle) then the jacobian for this gives you a slightly different answer.

Finally, there is the situation that poor resolution x-ray structures may have unphysical backbone angles reported. This does not mean the protein has those backbone angles. it means the structure was solved poorly and mis-reported. It might mean that the atomic positions reported are an average of many polycrystaline structures so that computing an bond angle between the average positions does not make physical sense.

Thus a second way of asking this question is how many reported structures have improbable bond angles versus how many real protiens have these.

Thank you for all your responses.

As a modeller that has never had any experience with X-ray it is great to get feedback. When performing quality assessment of models we consider the percentage of residues in "disallowed" regions and I've always wondered how much of the "disallowed" data was "real" and a sign of something far more interesting. This is why finding out about the relationship with active site positions intrigued me.

So, Charles, I guess that was what I was really trying to get at... How often do experimentalists come across situations where they have improbable phi/psi angles? And, based on these observations how many "real" proteins are estimated to contain the same?

As for missing electron density; I know this is mostly attributable to intrinsic disorder. However, what other reasons are there for missing density? If not disorder, does it come down to the quality of the crystal?

Thanks again.

You please mention the resolution of your protein structure solved and corresponding R-factors etc. Your refinement is complete?? Did you added properly the water (or other ions, ligands if present) near or around the active site??