Dear Richard,
at first, I would like to call your and others attention that in a recent issue of Prot. Sci. we published a detailed methodological paper about incorporating an ab initio loop modeling method into MODELLER. The paper touches on many methodological aspects of MODELLER, discussing new developments.
Fiser et al. Prot Sci 2000,9,1753-1773 http://www3.ncbi.nlm.nih.gov/htbin-post/Entrez/query?db=0&form=1&ter...
For instance, regarding your question: the original dihedral angle preferences were derived (year ~1993) on a small set of proteins (~100), using only 3 different definition of Ramachandran map for the 20 residues regarding the 5 main conformational classes. The observed frequencies were approximated by normal gaussian distributions. In the present paper we describe that the set of proteins was updated and increased by 10 fold (~1000 proteins), now each residue type has its own individually designed conformational preference regions on the Ramachandran map, which is defined by 5x5 degrees, and the distributions are approximated by a more realistic binormal gaussian fitting. The details of this library, including the calculated deviation values that you asked for, can be found in $MODELLER/modlib/mnch.lib. You can edit this file and probably influence the shape of the distribution.
However, in case of modeling, if a highly similar template is present, the fi-psi preferences will be dominantly inherited from the available template and the role of the statistical observations are suppressed. The physically meaningful angles within each conformational preference region are ensured by the terms of the CHARMM potential function. The final conformation will depend on all the other restraints as well.
I looked at a few models, and i also saw an obvious high density around the most typical -60,-40 peak, and can agree that it is more regular than certain x-ray structures although in a well refined structure >90 % of the helical angles will also massively fall into this region.
As I mentioned if you find the distributions too tight you might try to modify the weights in mnch.lib, but it will not necessarily effect the distribution of fi-psi's in the calculated models.
You can see in the file e.g. that Pro sigma values are around 8-13 degrees, which looks usual. Accidently a 3 dimensional plot about the Pro fi psi distribution i sincluded in the article, regarding both the observed frequencies and the fitted functions.
best regards,
Andras
> > Dear modeller users, > I have a question concerning the phi/psi distribution in the alpha > region. The homology models which I generate seem to have a curious > phi/psi distribution within the allowed alpha region. Almost all the > residues within procheck's red region are sharply clustered into two > peaks (one at the classical alpha helical angles of > around -60, -40) and the other at the gamma region ( -90, 0). I often > get a couple of minor lateral peaks as well. This is very different > from crystal structures (including my template), where there is a more > homogenous spread within the allowed region, although obviously showing > a greater density of points towards -60, -40. I wonder why this happens > and if I am doing something obviously stupid. It would seem that the > sigmas on binormal restrainsts for phi/psis are very small. Can I > easily adjust these values and what is the easiest way? > This question is related to a similar one concerning proline phis > angles which I note have a tendancy to drift away from -65. I have in > the past used special restraints on all prolines in order to reduce > sigma(phi) for which the default value seems to be rather generous. > I would be most grateful for any comments/suggestions/help. > Thanks, > Richard Garratt > University of Sao Paulo