On Friday 20 August 2010 08:53:23 am Eleanor Dodson wrote:
> That must be a bug..
>
> 2.00 is the lowest value permitted .
>
> Possble causes: You have very low resolution data, and the overall B
> value is badly estimated (not very common)
>
> common cause: you have done TLS refinement which reports b value
> differences from the TLS derived values. If you start from an
> artificially high initial B these will indicate that you should start
> with a lower initial b value. In fact after TLS the "iso B" isnt used
> for calculating structure factors so the reported value of 2 really
> doesnt matter for your map quality etc..
>
> If you run TLSANL then the B values should be reset to logical estimates.
> eleanor
I'm afraid that isn't the case in the commonly-used refinement protocol.
If one is refining Biso in addition to the TLS terms (which must be the
case here if it has bottomed out at 2.0) then the net B is the sum of the
TLS contribution and the Biso contribution. But since the Biso contribution
is erroneous (bottomed out at 2.0) TLSANL cannot reconstruct the true
net B. The information simply isn't available. There are a fair number
of depositions in the PDB that suffer from this problem. And yes, one
can improve the R factors for such structures by re-doing the refinement
taking extra care that the Biso contributions do not go out of range.
So it really is a problem that the Biso values are not allowed to
go negative in this refinement scenario. Operationally it might be more
practical for the program to deal with this by subtracting a delta-B from
the overall correction and boosting all the Biso values by that same
amount. But the code would have to check and correct the balance of
global and local B contributions after every cycle of applying B shifts.
> Ray Brown wrote:
> > Can anybody shed some light on why my atomic B factors all end up as 2.00
> > after
> > restrained refinement with REFMAC5? Is there a fix?
Reset all your protein Biso terms to something larger - e.g. the Wilson B.
Then reset the TLS parameters (keep only the RANGE terms; discard the
previous values for the T, L, and S terms) and start your refinement over again.
If you are working at resolution worse than about 2.8A, I suggest that you
try not refining Biso terms at all. The pure TLS model may be good enough.
In fact the pure TLS model may work better than a TLS+Biso model, due to
stability problems like the one you have encountered. Once you have reached
convergence with a pure TLS model, then you can try turning on Biso refinement
again. You might see further improvement in R_free, or you might not.
Ethan
> > Thanks.
> >
> > Ray
> >
> > email ray-br...@att.net
>
--
Ethan A Merritt
Biomolecular Structure Center, K-428 Health Sciences Bldg
University of Washington, Seattle 98195-7742
