[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[ccp4bb]: Summary of responses, MR problems



Title: Summary of responses, MR problems

We recently posted a question about problems in solving MR using BEAST and AMORE, obtaining good rotation angles but translation solutions had significant steric clashes.  (original question below).  We’d like to conclude this tread and summarize the responses we received.

There were three overwhelming responses:
1.  The overlap is real and indicates a conformational change from the search model.  Although we considered that, the overlapping regions were not consistent from solution to solution and often we had overlaps into the core region of this protein.

2.  Try other MR programs -- EPMR in particular was mentioned.
3.  Possible wrong space group.

We are very grateful for all the replies, in particular Randy Read, Phil Jeffrey, and Eleanor Dodson.

Embarrassingly, what we discovered is that in fact we probably did have the wrong space group.  Instead of P212121, it behaves MUCH MORE NICELY as P21212 (we had to reindex to convert our original hkl to klh).  Although, I must say that this spacegroup is not real obvious in looking at expected extinctions from our SCALEPACK output.  After re-indexing, running auto-amore in CCP4i, we have correlation coefficients at ~70% and Rs at ~37%, and very nice packing.  After rigid body and simulated annealing, our R’s are ~27% at the moment.

We are attaching partial responses that we found particularly helpful:

From Phil Jeffrey:
Personally, I find the LLG values in Beast very hard to interpret since their magnitudes vary widely with several parameters.  I find the Correlation Coefficients in AMORE much more convenient.

For a molecule that is basically unchanged with respect to an existing structure, and which is relatively well-behaved (no weird anisotropies

etc) I'd expect to find final correlation coefficients for a complete model of the asymmetric unit to be >0.55.  For really good models perhaps something like 0.65.  For the first of two molecules in the asymmetric unit I'd expect a CC of >0.4.  (Typically using 8-4 Ang. resolution for Translation Function and Rigid Body).

If you're *not* getting those numbers, then that may be a clue that the solution isn't correct.

The second option is significant since I have sometimes been fooled by systematic absences, and ESPECIALLY in the case of really good models, a "wrong" space group can often produce quite high correlations (LLG, CC, whatever) because at least some of it's symmetry operators match that of the true space group.  So in primitive orthorhombic I always do the full

search:

P212121 (indexing: hkl)
P21212 (indexings: hkl, klh, lhk)
P2221 (hkl, klh, lhk)
P222 (hkl)

because it removes any possible ambiguity and because I like to see the CC of "wrong" space groups to compare to the best one.  (I use the CCP4 program REINDEX to get the new indexings for AMORE and BEAST - remember to permute the cell dimensions also).

From Randy Read
These results are a bit odd.  When there's such a large signal, I'd expect a much clearer separation between the best answer and everything else.  But for the first orientation you're getting several solutions with similar scores differing significantly with y coordinate, and for the second orientation several solutions with similar scores differing significantly in z.  In a case where the rotation function gives such a clear answer, I'd also expect a bigger jump in the LLG in the translation search, which makes me worry a bit about the space group.

…I think it's obvious that there is a good signal in there struggling to get out, but something has gone wrong.  I'd try first to do finer translation searches, and I'd do 6D refinements (rotate around, translate around) of the single molecule solutions.  Then I'd try several translation searches for the second molecule, making different choices for the first fixed molecule, especially if it is still ambiguous where it should be.

For a clear solution, one translation peak should stand out above the others. When you fix one molecule and look for the second, the peak you get should be related to something you got for the second molecule without fixing the first one.

If the solution isn't clear, then I think you have to re-examine your
assumptions.  Are you really sure about the space group?  Are there clear absences to support all 3 screw axes?  Is it possible there are 3 molecules in the asymmetric unit, 2 of them related by a translation?  (Thiswould show up as a big non-origin peak in the native Patterson map.)  Is there any possibility of twinning?

Original posting (partial) -------------------------------------------------
I am using MR to solve the structure of a protein that is a 4-site directed mutant of the wild type (known structure, 260 amino acids).  It crystallizes in a different space group (P21212121) from the native.  Its unit cell dimensions are 50.897, 94.758, 151.205 with an estimated 2 molecules in my asymmetric unit. I have used BEAST to find my rotation and translation results, but I am having some difficulty in interpreting some of the output.  I appear to get clear solutions, but the model exhibits considerable steric clashes between the molecules and their symmetry related mates.  I am using CCP4i, version 4.2.1 on a Compaq Alpha, Tru64.

My rotation search always gives me only 2 top rotation angles (with around 300 LLGs).
Partial output ----
Sorted list of best trials
       Mol      Euler angles        Fractional translation     LLG
BEST   mol1 116.85  73.67 178.69   282.679
BEST   mol1  49.66  41.01  38.95   257.279
BEST   mol1  52.30  45.09  36.00   240.956
BEST   mol1 114.08  77.75 181.32   179.224
BEST   mol1 114.34  69.59 178.61   171.144
After a fine search about these solutions (which increases the LLGs by about 20) I do a translation search with my 2 top rotation options.  I get a further increase in my LLGs.

Sorted list of best trials
       Mol      Euler angles        Fractional translation     LLG
BEST   mol1  52.89  42.72  37.04    0.1346  0.2227  0.4887   351.708
BEST   mol1  52.89  42.72  37.04    0.1444  0.1405  0.4887   338.858
BEST   mol1  52.89  42.72  37.04    0.1346  0.2776  0.4887   337.021
BEST   mol1  52.89  42.72  37.04    0.1346  0.3507  0.4887   332.362
BEST   mol1  52.89  42.72  37.04    0.1444  0.3050  0.4887   325.157
BEST   mol1 115.67  73.43 179.89    0.1226  0.0840  0.3668   325.893
BEST   mol1 115.67  73.43 179.89    0.1226  0.0840  0.0639   323.951
BEST   mol1 115.67  73.43 179.89    0.1226  0.0840  0.3252   322.643
BEST   mol1 115.67  73.43 179.89    0.1226  0.0840  0.1649   316.940
BEST   mol1 115.67  73.43 179.89    0.1281  0.0823  0.3639   312.382


I used PDBSET to rot/trans my search model to these new orientations and I merged them using MOLEMAN. But when I look at my new merged pdb file in "O" I have significant steric clashes.  The solution does not refine well using rigid body or simulated annealing refinement protocols.

Thanks,
Jean Dehdashti
University of Missouri-Kansas City
dehdashtis@umkc.edu