Just to add a pedantic note to this conversation — if you look at the B-factor term in the structure factor equation, that term will drop to zero for all but the 000 reflection, for which the B-factor term is one. So, in the electron density equation, the electrons from that atom will be spread evenly over the unit cell, i.e. not over all space. This actually makes sense, when you consider that the structure factor equation is effectively considering our crystal to be infinite. The contribution of an infinite number (over all unit cells) of infinitely disordered atoms corresponds to the multiplication of an infinite number by an infinitesimal, which cancels out (in the limit) to give the finite average electron density.
Best wishes,
Randy Read
-----
Randy J. Read
Department of Haematology, University of Cambridge
Cambridge Institute for Medical Research Tel: +44 1223 336500
Wellcome Trust/MRC Building Fax: +44 1223 336827
Hills Road E-mail: [log in to unmask]
Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk
> On 15 Nov 2015, at 05:31, Ed Pozharski <[log in to unmask]> wrote:
>
> The B-factor approaching infinity will allow resulting electron density to become increasingly uniform. And yet, exactly because it cannot reach infinity (because, as you correctly point out, average density will then go down to zero, unless we allow occupancy to go to infinity as well), in practice bunch of modeled waters would not resemble flat solvent. Not sure what are you disagreeing with.
>
> I neither suggested placing waters on a grid nor randomly filling crystal space with it. I am just saying that knowing that a particular chemical entity is present within the crystal is in most cases clearly not enough to justify adding corresponding atoms to a structural model. Waters are one such case - we know they are there, yet unless electron density clearly indicates a discrete spatial position, individual water molecules are not added because B-factors will not "take care of it".
>
> Water molecules will not organize themselves into evenly spaced peaks because protein is present and the portion of data that these waters will try to fit is basically residual Fo-Fc, which is not flat. There isn't much useful signal in it either (well, that's whole another debate), but the outcome is that such waters will contribute to fitting noise and errors, contributing to model bias (if you want to discuss that, we'll need to define terminology first because I certainly tend to interpret model bias beyond just map issues).
>
> I renew my plea for an explanation as to *how placing atoms to where no discernible density peak exists is improving a crystallographic model*. It certainly does not help Rfree, quite the opposite.
>
> Cheers,
>
> Ed.
>
> On 11/14/2015 10:56 PM, James Holton wrote:
>>
>> No, if the B factor of an atom is infinity then its electron density is zero everywhere. That won't change the R factor.
>>
>> However, a "grid" of waters spaced every 1.3 A with occupancy 0.1 and B=25 is a pretty good approximation to flat bulk water with scale=1 and B=33. As long as the B factor and occupancy are constrained to all be the same, the "bulk" is flat to within 2%. I don't think that introduces any more "model bias" than doing the bulk solvent as electron density. Do you?
>>
>> True, if you let all the individual positions, occupancies and B factors float then macromolecular refinement programs will get very cross with you. Cross enough to make your Rwork and Rfree blow through the roof for some reason. However, if you fit a train of Gaussians to a flat boxcar function in gnuplot it works just fine. The Gaussians organize themselves into evenly-spaced peaks with the same occupancy and B factor. No "model bias" there. Maybe the difference is full-matrix vs sparse-matrix optimization?
>>
>> -James Holton
>> MAD Scientist
>>
>> On 11/14/2015 6:33 PM, Ed Pozharski wrote:
>>> No - a set of discrete water molecules would approach a uniform density as you elevate B-factors to infinity. But it would never be exactly identical to it, still having hills and valleys. Naturally, in practical terms such approach would be problematic because of potential model bias.
>>>
>>>
>>>
>>> Happy Connecting. Sent from my Sprint Samsung Galaxy S® 5
>>>
>>>
>>> -------- Original message --------
>>> From: James Holton <[log in to unmask]>
>>> Date: 11/14/2015 7:27 PM (GMT-05:00)
>>> To: Ed Pozharski <[log in to unmask]>, [log in to unmask]
>>> Subject: Re: [ccp4bb] proper modeling of residues into patchy electron density
>>>
>>>
>>> Isn't that exactly what the bulk solvent correction is? Except that the B factor and occupancy of all the disordered water "atoms" are constrained to be the same?
>>>
>>> -James Holton
>>> MAD Scientist
>>>
>>> On 11/14/2015 4:23 PM, Ed Pozharski wrote:
>>>> This is a bit confusing. Don't you already know the "location" from backbone coordinates?
>>>>
>>>> I still would like to hear exactly how placing atoms to where no discernible density peak exists is improving a crystallographic model. I also know waters are there in large numbers, should I flood the bulk solvent area with discrete waters and state that B-factors will take care of "it"?
>>>>
>>>> Cheers,
>>>>
>>>> Ed.
>>>>
>>>>
>>>>
>>>> Happy Connecting. Sent from my Sprint Samsung Galaxy S® 5
>>>>
>>>>
>>>> -------- Original message --------
>>>> From: Quyen Hoang <[log in to unmask]>
>>>> Date: 11/14/2015 6:23 PM (GMT-05:00)
>>>> To: [log in to unmask]
>>>> Subject: Re: [ccp4bb] proper modeling of residues into patchy electron density
>>>>
>>>> Let's see if we can settle this. Ed might remember where I stood, but my view had changed a bit.
>>>> 1. Delete atoms of a side-chain for which no density is visible if one is using density to find out what the residue looks like.
>>>> 2. Model a complete residue even when density is missing for a side-chain atom if one is using the density to find the location of the residue.
>>>>
>>>> Can we settle with this?
>>>>
>>>> Cheers,
>>>> Quyen
>>>>
>>>>
>>>> On Nov 14, 2015, at 5:40 PM, Ed Pozharski <[log in to unmask]> wrote:
>>>>
>>>> > On 11/14/2015 04:36 PM, Artem Evdokimov wrote:
>>>> >>
>>>> >> I would agree with both sides, since absence of evidence is not evidence of ansence.
>>>> >>
>>>> > Well, that's agnostic :)
>>>> >
>>>> > Just a comment - omitting side chain atoms from the model does not assert that they are somehow missing from the chemical structure one is modeling. It means that their spatial distribution cannot be adequately approximated from experimental data via simple 3D gaussian. So when I am excluding atoms from disordered side chains, I am not saying that an X-ray fairy has cut off covalent bonds with a tiny magic chainsaw. I am just saying I don't have sufficient experimental evidence to locate these atoms.
>>>> >
>>>> > Cheers,
>>>> >
>>>> > Ed
>>>
>>
>
|