Presumably you only want to see if the mutation has been successful, and
check for other gross changes?
It would not be sensible to try to get a well refined model from such
low resolution, espec if there is already a structure M" with better data..
So the problem of model bias in refinement is not so serious.
I would merge the M2 data with the new mutants - get FCs for the M2
structure, perhaps after removing those residues you have mutated and
cutting back some side chains to see how well your new data produces the
missing density..
Then do Fmut - Fc maps, and maybe also Fo(mut) - fobs(M2) difference
maps with PHIC and FOM.
You should see large holes or peaks depending on the changes you have
done which should answer your questions.
Eleanor
[log in to unmask] wrote:
> Hi All:
>
> There are two published x-ray structures (very similar, RMSD<1A if ignore
> 2 loops) available for a wild-type protein; let's call these two
> structures M1 and M2. They belong to the same space group but with large
> differences in unit cell dimensions (over 10%, or 10A in each dimension,
> cross R>30%).
>
> Now I have some (deletion) mutant structures to solve and the unit cell
> dimensions of these mutant crystals are very similar to one of the crystal
> forms (say M2)mentioned above. All these crystals including the wild-type
> diffract to low resolution (~3.5-4A) and have no NCS, so the
> parameter/data ratio must be very high.
>
> My question is: would model bias be less if I use the structure model
> that is very different (i.e M1) in unit cell dimensions? In other words,
> if I only do molecular replacement and rigid body refinement, and look at
> the 2Fo-Fc and Fo-Fc maps right after this, will I see a less biased map
> by using M1 as the model instead of M2?
>
> In my naive thinking, since the M1 and M2 crystal forms are so different,
> and the parameters in the M1 model has essentially not been refined
> against the M2 data (which the mutant belong to). Using M1 as the model
> to calculate maps must be different from using M2, and presumably, with
> less model bias towards the wild type structure?
>
> In contrast, if the M2 model is used for these mutants. since all
> parameters of M2 has been fully refined against similar data in wild-type,
> all these parameters might presumably have been "contaminated" by the
> wild-type data? i.e. all the parameters (all parts of the wt model) have
> been adjusted "together" to fit the data. If I use this model on a very
> similar mutant data, will it be more likely to generate back the wt
> model(more biased)?
>
> Do you think this is a valid argument?
>
> Regards,
>
> Weikai
>
>
>
|