Discussing with Fred Vellieux offlist I suggested the following: would it be
useful to have a sort of 'simulated annealing' of the B-factor values? - in
refinement I often try resetting them to higher values (Moleman has a function
for this I think) and then see which ones refine back down nicely - but it
would be good to have some sort of randomization method - that would be a test
of how much the values are inherited from the modelling early on - say from the
MR probe model for example.
We could do multiple B-factor refinements with different starting kicks to the
values and look for the best final Rfree or some measure of map quality?
Martyn
Martyn F. Symmons
Cambridge
'Chan fhiosrach mur feòraich.'
Gaelic proverb - Nothing asked, nothing learned.
________________________________
From: Pavel Afonine <pafon...@lbl.gov>
To: CCP4BB@JISCMAIL.AC.UK
Sent: Wednesday, 2 December, 2009 22:52:45
Subject: Re: [ccp4bb] Refining residues as rigid bodies
Hi,
you can do similar thing (that is resulting in similar outcome) in
phenix.refine by increasing the weight on ADP restraints term. Example:
increase "wu" or decrease "wxu_scale". Although I believe a regular
refinement of individual isotropic ADPs should normally work just fine
at 3A resolution in phenix.refine.
Pavel.
On 12/1/09 11:18 PM, Frederic VELLIEUX wrote:
If the problem is B-factor refinement, you can do that easily at low resolution
with CNS. You just give tight restraints.
>
>You modify the file bindividual.inp
>
>This section:
>{* target sigma values for restrained B-factor refinement *}
>
>{* mainchain bonds *}
>{===>} bsig_main=1.5;
>{* mainchain angles *}
>{===>} asig_main=2.0;
>
>{* sidechain bonds *}
>{===>} bsig_side=2.0;
>{* sidechain angles *}
>{===>} asig_side=2.5;
>
>I cannot remember in which direction the values have to go. I think up. I have
>done this with a very low resolution structures (4.5 A?) a few years ago, you
>get smoothly varying B values, very tightly restrained. I do not think we
>published that structure, we obtained a higher resolution structure later (I
>don't think the referees would have been very happy seeing B factor refinement
>at low resolution). But it worked.
>
>Fred.
>
>
>Message du 01/12/09 23:51
>>De : "Jason C Porta"
>>A : CCP4BB@JISCMAIL.AC.UK
>>Copie à :
>>Objet : Re: [ccp4bb] Refining residues as rigid bodies
>>
>>
>>Basically my reasoning for doing this is a low data-to-parameter ratio, which
>>makes B-factor refinement unfeasible. So far I have had nice results with
>>breaking the complex into rigid subdomains. So i was basically just thinking
>>of a way I could refine the structure best, without using too many parameters.
>>
>I see now how this can be done in Phenix. I will give it a try.
>
>Thanks for all of your suggestions!
>
