[buster-discuss] Temp factors for solvent refining to very low values?
Ashley Pike
ashley.pike at sgc.ox.ac.uk
Fri Apr 21 15:27:23 CEST 2017
Hi Clemens,
Thanks for taking time to respond (the first response to this enquiry!). For info - got to the bottom of the problem. Only appeared to occur when data processed in a certain way - odd behaviour observed with DIALS processed data (DIAMOND auto pipeline) and/or data run through STARANISO truncation process. Dataset is mildly anisotropic (3.1 x 3.7A) so suspect something going wrong with scaling/compensation carried out by these processes. Solution was to simply re-process (XDS then aimless) and resulting dataset gives OK results (no atoms with B's at 3 although some are or the lower than expected side).
Thanks again,
Ash
PS. Waters are 'high confidence' - manually placed based +fofc peaks and environment although there is no higher resolution structure to fall back on to confirm existence for sure.
-----Original Message-----
From: Clemens Vonrhein [mailto:vonrhein at globalphasing.com]
Sent: 21 April 2017 10:58
To: Ashley Pike
Cc: buster-discuss at globalphasing.com
Subject: Re: [buster-discuss] Temp factors for solvent refining to very low values?
Dear Ashley,
(sorry for the late reply).
On Wed, Apr 05, 2017 at 08:13:34PM +0000, Ashley Pike wrote:
> Have run into an issue with a structure I am refining (3.2A). The
> temperature factors for the refined solvent molecules show a very
> weird distribution with half having B's below 10 and the remainder
> around 30-50 (I have modelled around 30 solvent in total) - see
> excerpt below.
How many waters are we talking about: hundreds or a few dozen?
What is the source of that water model: are these waters from a high-resolution native/apo structure or does it come fro automatic or manual water update starting with an empty model?
> Average B for protein chains is around 45 and other "solvent"
> heterogroups (lipid) have B's around 30-70. The solvent is not part of
> a TLS group (single TLS group per protein chain). Refining with
> autoncs / TLSalternate
Sounds all sensible, yes.
> HETATM10043 O HOH C 1 37.384 85.915 48.227 1.00 3.00 O
> HETATM10044 O HOH C 2 41.165 85.554 55.195 1.00 33.13 O
> HETATM10045 O HOH C 3 38.207 79.024 102.082 1.00 40.10 O
> HETATM10046 O HOH C 4 31.928 83.353 106.446 1.00 3.00 O
> HETATM10047 O HOH C 5 59.895 83.126 112.069 1.00 53.13 O
> HETATM10048 O HOH C 6 71.980 110.118 85.303 1.00 41.10 O
> HETATM10049 O HOH C 7 35.067 77.191 94.667 1.00 23.21 O
> HETATM10050 O HOH C 8 58.834 101.563 100.861 1.00 7.26 O
> HETATM10051 O HOH C 9 62.444 104.317 96.908 1.00 3.00 O
> HETATM10052 O HOH C 10 31.634 95.839 49.139 1.00 6.41 O
> .......
>
> Solvent peaks return to +fofc maps if I leave them out. If I reset B's
> to 30.0 and re-refine, they converge to similar values. I am assuming
> 3.00 is a minimum Bvalue allowed (similar to Bmax of 300)
Correct.
> oddly it is always the same solvent molecules that hit 3.00.
If it were different waters hitting that, it would be even weirder I
think: then it would show some instability in parameter refinement. Whereas the fact that it is always the same waters seem to suggest there is something "special" about those.
> I do not think there are likely to be many additional ions about - I
> do have anomalous peaks for calcium sites (whose B's are in the range
> 25-40).
What other ions (apart from Ca) have you already fitted? Could it be Mg or Na - which wouldn't show up in anomalous maps. These abnormal B-factors could also arise because peaks corresponding to something like disordered phosphate groups on the surface of the lipid membrane have been assigned to waters, that then need a very low B-factor in order to pretend to be a disordered phosphate.
Anything interesting or special about the environment of those waters (H-bonding partners, geometry)?
> I am at the end of rebuilding with Rwork/Rfree of 25.0/27.0, fom 0.8.
>
> Any reason for this behaviour (maybe I am trying to push my luck
> refining B's but BUSTER has produced/refined sensible solvent
> molecules in the past with other datasets at similar resolutions).
3.2A doesn't seem to be too unusual when refining B's in BUSTER. Unless of course that 3.2A label hides other data issues like anisotropy, incompleteness, ice-rings etc.
> Hints for what to look for to diagnose the problem?
Cheers
Clemens & Gerard (for buster-develop)
More information about the buster-discuss
mailing list