question regarding secondary-structure restraints

From: Pete Meyer
Date: 22 September 2011 21:18


I've noticed that people seem to be using or recommending secondary structure restraints for low resolution refinement lately, but I'm somewhat confused about the logic underlying their use.

Using ballpark figures from a system I'm familiar with: 30000 atoms (90000 positional parameters), 4500 residues, 100000 reflections and 95000 geometric (bond and angle) restraints.
n_ref / n_param ~= 1.11
(n_ref + n_geom) / n_param ~= 2.16

Assuming all residues are localized, and each residue provides 2 secondary structure restraints (best-case scenario), this changes the effective observation to parameter ratio to:

(n_ref + n_geom + n_ss ) / n_param ~= 2.26

In short, the effective observation to parameter ratio improves by ~4%.  This seems like a relatively small improvement, especially if the trade-off is that Ramachandran statistics can't be used for validation anymore.  It also seems like the improvement would decrease with larger proteins (the number of additional parameters from adding a residues increase faster than the number of secondary structure restraints that residue could provide).

Does anyone have any suggestions that could help clear things up?

Thanks,

Pete

----------
From: Pavel Afonine

Hi Pete,

the rationale is: at that low resolution the density map and traditional set of restraints are not enough to preserve secondary structure elements during refinement. For example, if you start with a model having good secondary structure elements, they will be distorted after refinement against low resolution data.
See: around pages number 49 - 59 here:

http://www.phenix-online.org/presentations/latest/pavel_refinement_general.pdf

Pavel.

----------
From: Phoebe Rice

Its also a bit too simple to count secondary structure restraints as 2 restraints per residue, because if they're tight enough on, say, an alpha helix, in combination with other geometry restraints (good bond angles, no clashes, etc) you could probably turn the backbone of the entire helix into one nearly-rigid body.  Which is not a bad idea at 5A.

=====================================
Phoebe A. Rice
Dept. of Biochemistry & Molecular Biology
The University of Chicago
http://bmb.bsd.uchicago.edu/Faculty_and_Research/01_Faculty/01_Faculty_Alphabetically.php?faculty_id=123
http://www.rsc.org/shop/books/2008/9780854042722.asp

----------
From: Nat Echols

There is nothing preventing you from using Ramachandran statistics for validation if the secondary structure restraints only restrain the distances between hydrogen-bonded atoms.  (This is what the implementation of S.S. restraints in phenix.refine does.)  This actually contributes significantly fewer than two extra restraints per residue on average.  And it doesn't actually have much of an effect on Ramachandran statistics; the number of outliers is slightly reduced, as is the clashscore.  I suspect (although I haven't examined this in detail) that this is mostly keeping the ends of helices and sheets from unraveling, which is what they're most useful for.  (The change in R-factors is negligible in our tests, although they sometimes reduce overfitting a little bit.)

This is an entirely different issue from Ramachandran (i.e. phi/psi) restraints, which should only be used as a last resort at low resolution at the end of refinement, and after fixing all outliers manually in Coot.  (Restraining to a high-resolution model, if one is available, is a much better option.)

-Nat

----------
From: Tim Gruene



Dear Pete,

Coot offers using secondary structure restraints, i.e. this does not
refer to refinement but to model building where your calculations do not
apply.

It helps a great deal if you are looking at a patch of density at, say,
3.5A resolution which you recognise as alpha-helix or beta-strand and
simply ask your model building program of choice to "place a helix here"
which you can incorporate into your model or at least use as a guideline
for correcting yours.

Tim

- --
- --
Dr Tim Gruene
Institut fuer anorganische Chemie
Tammannstr. 4


----------
From: Dirk Kostrewa

Dear Nat and other interested colleagues,

when I played with H-bond restraints for secondary structures for the refinement of a 4.3 A structure (only a few weeks before they were introduced in phenix), I've made the following observation: at low resolution without H-bond restraints for secondary structures, the carbonyl groups of these secondary structures take the liberty within their globbish electron densities to deviate from their ideal H-bond conformation, resulting in a tight "belt" of outliers around the preferred Ramachandran regions, with typical deviations of only a few degrees. Introducing the additional H-bond restraints for maintaining secondary structures pulls these outlier carbonyl groups back into the preferred Ramachandran regions. In my case, the number of Ramachandran outliers was reduced to less than one half! Although, these H-bond restraints do not directly include information about allowed Ramachandran regions, the Ramachandran plot is actually affected by these restraints. Thus, at least in my opinion, the Ramachandran plot is then not a truly independent measure for model quality, anymore. The same holds true for all geometrical restraints, of course.

Best regards,

Dirk.

Am 22.09.11 22:49, schrieb Nat Echols:

--   ******************************************************* Dirk Kostrewa ******************************************************* 

----------
From: Nat Echols

It depends on how strictly you assess the "independence" of validation criteria.  The Ramachandran plot is considered valid in most cases because refinement programs traditionally do not restrain phi and psi angles, so we need to rely on the accuracy of the data (and our placement of atoms) and various complementary geometry restraints (especially nonbonded) to keep residues in the favorable regions of the plot.  There are a variety of ways to make the plot better by modification of the model and/or restraints (adding hydrogens, increasing the weight on the nonbonded restraints, secondary structure restraints, etc.), none of which are as drastic as directly restraining the model to the plot.  I don't really view this as biasing the plot, for two reasons: a) the quantity being measured is independent of the quantity restrained, and b) at least in my hands, these modifications never completely fix the problem of Ramachandran outliers.  (It's the loop regions that are really awful.)

Anyway, I don't think anyone should feel bad about using this kind of restraint at low resolution.  The caveat is that of all the specialized restraints that we (Jeff Headd and I) have been testing for low-resolution refinement (in Phenix), nothing works nearly as well in preserving good geometry, and usually improving the R-factors, as restraining model parameters to a related high-resolution structure, when one is available.  Fortunately, every modern refinement program has this ability in some form, and I expect that this is going to have the most impact in improving the overall quality of low-resolution structures.

-Nat

----------
From: Dirk Kostrewa

Dear Nat,

yes, I fully agree - all these restraints that improve the geometry either by restraining to high-resolution structures or by introducing H-bond restraints for secondary structures are very useful for low-resolution structures!

I see your argument with the Ramachandran plot. But imagine a set of very strong non-bonding/bonding restraints that would result in an absolutely clean Ramachandran plot for any structure, then the Ramachandran plot would become useless even in the absence of any phi/psi-restraints. So, I prefer to err a bit on the safe side here by saying "not a truly independent measure".

Personally, I think, that ALL refinement programs, including the real-space refinement in Coot, would benefit from inclusion of proper H-bonding terms (something, that for instance the very old X-Plor version did), since this would automatically restrain secondary structures and other hydrophilic interactions to some reasonable geometry, even at very low resolution.

Best regards,

Dirk.

Am 26.09.11 16:17, schrieb Nat Echols:

----------
From: Tommi Kajander


-Just to make a note, there has actually been some discussion in the published literature recently (ok  maybe past ten years) about what terms; simply steric (as originally) or hydrogen bonding etc
might be needed to explain observed backbone angular values.

Tommi

Tommi Kajander, Ph.D.