bulk solvent treatment inside protein cavities

From: Allister Crow
Date: 16 April 2012 11:37

Board members,

I have a couple of questions regarding how to improve the solvent model as applied to solvent-filled cavities inside proteins.

I am currently nearing the end of refinement of a protein structure at 2.8 A resolution.  I recently switched Refmac versions, upon doing this I noticed a modest improvement in R factors, but I also notice some new features in the difference maps.  These features don't show up in the sigma-weighted 2Fo-Fc maps and are unlikely to be 'ligands' of any form.  In fact, I suspect that the appearance of these features (which are all located in solvent channels within cavities inside the protein) are probably due to some difference in how the bulk solvent contribution has been applied.

I've attached a picture of one such feature showing the difference between Refmac 5.5 and 5.6.  (Both difference maps are contoured at 3 sigma- both using the same model and refinement parameters).

My questions are therefore:

1) has something substantial changed in the bulk solvent treatment between Refmac versions 5.5 and 5.6?

2) How can I go about changing the bulk solvent treatment to better account for solvent contribution inside the protein cavities?

Best wishes, and thanks in advance for all your help,

- Allister Crow



----------
From: Garib N Murshudov

Dear Allister

Could you please update refmac version. In the version you it seems that bulk solvent mask calculation has some problems. New version (at the moment) can be downloaded from this site:

http://www.ysbl.york.ac.uk/refmac/data/refmac_experimental/refmac5.7_linux.tar.gz

There is a mac version also.

regards

Garib

<bulk_solvent_inside_cavities.png>

Dr Garib N Murshudov

----------
From: Eleanor Dodson


Oh dear - this is the version of Refmac in the latest ccp4 release - can this be updated on the web site as soon as possible ?

Eleanor

----------
From: Keitaro Yamashita


Dear Garib,

Is there REFMAC option to output solvent mask information (e.g. Fmask
and PHImask in mtz to check with Coot)?

I tried to generate it by subtracting (FC, PHIC) from (FC_ALL,PHIC_ALL).
But I'm not sure that FC_ALL = FC + FMASK is correct or not.

Keitaro

----------
From: Garib N Murshudov <

Yes there is. If you use command line (it is not available on the ccp4i yet).

If you run with command lines

refmac5 <all others like hklin, xyzin etc> mskout <mask file name> <<eof

all options you want

eof

Then there will be a map and you can visualise it using coot.

regards

Garib

----------
From: Garib N Murshudov

A follow up:

In the new version there is FC_ALL_LS, PHIC_ALL_LS

That should be FC_ALL_LS = FC + FMASK. 

I have not tried but if you can use vector difference map then it should be: FMASK = FC_ALL_LS - FC

But it is after scaling. If you write out mask map then it is just 0 1 map (0 inside protein and 1 outside), except values are not 0 1 but 0 and some constant

Regards

Garib

 

----------
From: Keitaro Yamashita


Dear Garib,

Thank you very much for your quick reply.

I tried mskout option and the output looked almost the same as the map
generated by FC_ALL - FC.

By the way, when mskout option is specified, refmac stops before CGMAT cycles.
Is there any way to do refinement with mskout option?

> I have not tried but if you can use vector difference map then it should be:
> FMASK = FC_ALL_LS - FC

What is FC_ALL in the new version?


Thanks,

----------
From: Garib N Murshudov

Dear Ketaro

At the moment mskout option is a signal that the program should stop. Obviously I can add an option to continue. However if you have mskout option it is likely that you want to check what is going on with the mask. If you want to compare starting and final mask then you could run refmac with mskout in the beginning and after refinement.

If you need it urgently then I can add continuation of refinement with mskout option.

FC_ALL is ML scaled FC+FMASK

Sometime it may be different from least-squares scaled FC+FMASK

regards

Garib

----------
From: Keitaro Yamashita

Dear Garib,

I think it is better if refmac outputs final solvent mask when mskout specified.
If one just wanted to calculate the mask, NCYC 0 should be specified.

I hope my suggestion would be accepted, but I'm not in a hurry.

> FC_ALL is ML scaled FC+FMASK
>
> Sometime it may be different from least-squares scaled FC+FMASK

For what purpose is FC_ALL_LS written?
Can I check something by comparing FC_ALLto FC_ALL_LS?

I found somewhat large difference between FC_ALL_LS map and FC_ALL
map in Se position.
I used SAD function.
What does it mean?


Cheers,

Keitaro

----------
From: Garib N Murshudov

On 16 Apr 2012, at 17:14, Keitaro Yamashita wrote:

Dear Garib,

I think it is better if refmac outputs final solvent mask when mskout specified.
If one just wanted to calculate the mask, NCYC 0 should be specified.

I hope my suggestion would be accepted, but I'm not in a hurry.

ok

FC_ALL is ML scaled FC+FMASK

Sometime it may be different from least-squares scaled FC+FMASK

For what purpose is FC_ALL_LS written?
Can I check something by comparing FC_ALLto FC_ALL_LS?

There could be some differences, as I said FC_ALL maximum likelihood scaled (they are used for map calculation) and FC_ALL_LS least-squares scaled (it is used for Rfactor calculations)

I found somewhat large difference between FC_ALL_LS map and FC_ALL
map in Se position.

Without data I could not say what might be going on.

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

Dear Allister Crow,

in cases like these, I would recommend to apply the Babinet bulk solvent correction instead of the mask bulk solvent correction as a control (Refmac5 -> Scaling -> Use Babinet scaling; uncheck "Calculcate the contribution from the solvent region"). The Babinet bulk solvent correction only uses two overall scaling factors and is usually simple, robust and, in my experience, does not show any local difference density artefacts. The mask bulk solvent correction is more powerful, but, depending on the project and the various mask radii for generating and shrinking the mask, could produce false positive or negative difference density. To exclude these cases, you can always calculate the Babinet bulk solvent correction as a control.

Best regards,

Dirk.

Am 16.04.12 12:37, schrieb Allister Crow:

--   

----------
From: Ed Pozharski

On Tue, 2012-04-17 at 11:08 +0200, Dirk Kostrewa wrote:
> The mask bulk solvent correction is more powerful

Just to note that sometimes Babinet solvent correction returns lower
Rfree and thus may be preferred to mask (assuming that the Rfree is the
only thing that matters).

Beginning with 5.6.0078, you can also optimize the bulk solvent mask
parameters

http://www.ysbl.york.ac.uk/~garib/refmac/data/refmac_news.html#Bulk_solvent

This, imho, is not likely to remove the density artifacts, but is worth
a try.

--
"I'd jump in myself, if I weren't so good at whistling."
                              Julian, King of Lemurs