[ccp4bb]: answer to forwarded question on r factors in refmac5 and sfcheck

["M.D.Winn" <M.D.Winn@dl.ac.uk>]

***  For details on how to be removed from this list visit the  ***
***          CCP4 home page http://www.ccp4.ac.uk         ***


And my reply ....


> Dear Tom,
> 
> I'm afraid we're having problems with the BB - everyone got
> unsubscribed! It's being sorted now, and your message will
> be sent on soon. Meanwhile, I can answer some of your
> points:
> 
> First off, your obsevrations are quite correct. I tried for
> an example here and got R factors:
> 
> refmac + TLS                        0.17
> sfcheck                             0.24
> refmac with TLS-derived B's only    0.21
> as previous, with bulk solvent off  0.24
>    (SOLV NO - can't do this in GUI!)  
> 
> I'm pretty sure that sfcheck doesn't use ANISOU lines, so that 
> explains a lot of the difference. Also, sfcheck uses a very
> different scaling function. Refmac5 uses a mask generated
> solvent correction in addition to the Babinet-style correction.
> sfcheck certainly doesn't have the former, and removing that
> (4th number above) account for the rest of the difference.
> 
> Note that the 3rd and 4th numbers are without any refinement.
> Some refinement would lower the difference, as the model
> without TLS and bulk solvent correction will adapt to their
> absence.
> 
> Re: your 3rd point. Certainly quote the R factor from refmac5.
> That is the only program which uses your full model. R factors
> calculated from only a subset of your model parameters are not
> an accurate reflections of your model. For submission to PDB,
> you should submit TLS parameters (refmac5 includes these in PDB
> header) not ANISOU lines, since this is your model. But you
> can discuss aniso U's provided there are clearly flagged as
> derived values rather than refined values.
> 
> People on BB will undoubtedly give you some different opinions ...
> 
> Martyn
> 
> > Hi,
> > 
> > I have a question regarding Refmac5 and Sfcheck.  I used Refmac5 with =
> > TLS refinement.  After Refmac5, I run TLSANL to add TLS contributions to =
> > the isotropic B factors as well as to add "ANISOU" lines for anisotropic =
> > B factors.  Please correct me if I'm wrong.  My questions are:
> > 
> > 1. When I run SFCHECK on the output pdb of TLSANL, it calculates R =
> > factors ~2% higher than those from Refmac5 output.  I think part of this =
> > is because SFCHECK does not use ANISOU records to calculate Rs.  Is =
> > there a way to include anisotropic B factors when calculating R factors?
> > 
> > 2.  Are there other reasons that result in the difference in the R =
> > factors from Remac5 and Sfcheck?  If I delete the ANISOU records from =
> > the output of TLSANL and use the resulting pdb as input pdb for another =
> > round of Refmac5 (thus no ansiotropic TLS contributions,  just for =
> > checking R factors), I notice the starting R factors is only ~1% higher =
> > than those from previous Refmac (with TLS), but still ~1% lower than =
> > those from Sfcheck.  One reason I can think of is that maybe Sfcheck =
> > uses a different way in scaling and bulk solvent correction.  But it =
> > seems to me the difference shouldn't be
> > so big.
> > 
> > 3. What criteria do people use to judge whether to report anisotropic B =
> > factors?  Only with very high resolution?  I only have 1.9 A data, if I =
> > leave them out, what R factors should I report, those from Remac5 or =
> > those from Sfcheck?
> > 
> > Tom Lee