[ccp4bb] inflated BOND_RMSD with external restraints (refmac)

2010-10-14 Thread Ed Pozharski
It appears that external restraints are included in bond_rmsd
calculation.  When they are used to restrain the hydrogen bonds to
maintain the Watson-Crick pairing in a 3A resolution structure of a
protein-DNA complex, the bond_rmsd is inflated about 5 times.  To verify
this, the refmac run was done with external restraints removed and zero
refinement cycles.  With external restraints I get 0.028A, without -
0.006A.

This is with v.5.5.0109.  I assume this may be classified as a bug -
there is no limitation to using external restraints only for the
covalent bonds, thus they should not be included in bond_rmsd
calculation.  

On a practical side, I now have a misleading bond_rmsd value.  The
correct one can be calculated as described, but this may make geometry
weight optimization cumbersome.  Do I understand correctly that an
alternative is to monitor Zbonds, with a rule that it should be around
1.0?  And more generally, shouldn't we not look at rmsd_bonds at all and
only use Zbonds instead (which is, I assume, an average bond length
deviation to the target value ratio?)  I suspect that "acceptable"
bond_rmsd value is slightly affected by sequence.

Ed.

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


Re: [ccp4bb] inflated BOND_RMSD with external restraints (refmac)

2010-10-14 Thread Garib N Murshudov
Hi Ed

refmac 5.6 should not have this problem. Yes, you are right. It should be 
considered as a bug. 
I think I have fixed it. Could you please try 5.6version from:


www.ysbl.york.ac.uk/refmac/latest_refmac.html

You need to take experimental version (it should be stable enough, although I 
update it more often than older versions). In this version to make external 
restraints as covalent bonds you need to specify type 1 (type 0 means 
dictionary values will be overwritten and type 2 is external restraints for 
non-covalent bonds)

I hope it helps

regards
Garib


On 14 Oct 2010, at 21:51, Ed Pozharski wrote:

> It appears that external restraints are included in bond_rmsd
> calculation.  When they are used to restrain the hydrogen bonds to
> maintain the Watson-Crick pairing in a 3A resolution structure of a
> protein-DNA complex, the bond_rmsd is inflated about 5 times.  To verify
> this, the refmac run was done with external restraints removed and zero
> refinement cycles.  With external restraints I get 0.028A, without -
> 0.006A.
> 
> This is with v.5.5.0109.  I assume this may be classified as a bug -
> there is no limitation to using external restraints only for the
> covalent bonds, thus they should not be included in bond_rmsd
> calculation.  
> 
> On a practical side, I now have a misleading bond_rmsd value.  The
> correct one can be calculated as described, but this may make geometry
> weight optimization cumbersome.  Do I understand correctly that an
> alternative is to monitor Zbonds, with a rule that it should be around
> 1.0?  And more generally, shouldn't we not look at rmsd_bonds at all and
> only use Zbonds instead (which is, I assume, an average bond length
> deviation to the target value ratio?)  I suspect that "acceptable"
> bond_rmsd value is slightly affected by sequence.
> 
> Ed.
> 
> -- 
> "I'd jump in myself, if I weren't so good at whistling."
>   Julian, King of Lemurs


Re: [ccp4bb] inflated BOND_RMSD with external restraints (refmac)

2010-10-15 Thread Ian Tickle
Ed,

> On a practical side, I now have a misleading bond_rmsd value.  The
> correct one can be calculated as described, but this may make geometry
> weight optimization cumbersome.  Do I understand correctly that an
> alternative is to monitor Zbonds, with a rule that it should be around
> 1.0?

No, the average value of Zbonds over the whole PDB is almost exactly
0.5 (there's no clear resolution-dependence), so that would be a good
starting point.  Re-refinement of ~98% of the PDB (PDB-REDO project)
using weight optimisation showed a linear dependence of Zbonds on
resolution in the resolution range 1 to 3 Ang.:

Zbonds = 0.85 - 0.146*dmin  (SD = 0.2)

so e.g.  = 0.56 @ 2 Ang.

Cheers

-- Ian

> And more generally, shouldn't we not look at rmsd_bonds at all and
> only use Zbonds instead (which is, I assume, an average bond length
> deviation to the target value ratio?)

Correct.

> I suspect that "acceptable"
> bond_rmsd value is slightly affected by sequence.

Not really, the effect of sequence on the bond lengths is
insignificant in comparison with the uncertainty.

Cheers

-- Ian


Re: [ccp4bb] inflated BOND_RMSD with external restraints (refmac)

2010-10-15 Thread Ian Tickle
>> And more generally, shouldn't we not look at rmsd_bonds at all and
>> only use Zbonds instead (which is, I assume, an average bond length
>> deviation to the target value ratio?)

Sorry, I just realised that I misread what you wrote (that's what
comes of speed-reading!).  What you said above is NOT correct: yes you
should monitor Zbonds, but Zbonds is defined in (almost) the usual way
for a statistical RMS(Z-score), i.e. it's the RMS value of:

Z' = (deviation of observed from target value) / (standard uncertainty
of the target value).

I say 'almost' because it's not a true Z-score which would be defined as:

Z = (deviation of observed from target value) / (standard uncertainty
of the deviation)

The expected value of Z^2 is 1 as you say, the expected value of Z'^2
is not (it's always < 1).

Unfortunately we can't easily calculate the true Z-score: the
deviation in question is (dobs - dtarg); the standard uncertainty of
this cannot easily be calculated.  We know SU(dtarg) but we don't know
SU(dobs) and most importantly we don't know COV(dobs,dtarg) (dobs and
dtarg are obviously not independent so we cannot even assume that the
covariance is zero).

-- Ian


Re: [ccp4bb] inflated BOND_RMSD with external restraints (refmac)

2010-10-15 Thread Ed Pozharski
Garib,

indeed - I tested 5.6.077 and the bond_rmsd is down to 0.006 as it
should.  Thanks,

Ed.


On Thu, 2010-10-14 at 22:04 +0100, Garib N Murshudov wrote:
> Hi Ed
> 
> refmac 5.6 should not have this problem. Yes, you are right. It should be 
> considered as a bug. 
> I think I have fixed it. Could you please try 5.6version from:
> 
> 
> www.ysbl.york.ac.uk/refmac/latest_refmac.html
> 
> You need to take experimental version (it should be stable enough, although I 
> update it more often than older versions). In this version to make external 
> restraints as covalent bonds you need to specify type 1 (type 0 means 
> dictionary values will be overwritten and type 2 is external restraints for 
> non-covalent bonds)
> 
> I hope it helps
> 
> regards
> Garib
> 
> 
> On 14 Oct 2010, at 21:51, Ed Pozharski wrote:
> 
> > It appears that external restraints are included in bond_rmsd
> > calculation.  When they are used to restrain the hydrogen bonds to
> > maintain the Watson-Crick pairing in a 3A resolution structure of a
> > protein-DNA complex, the bond_rmsd is inflated about 5 times.  To verify
> > this, the refmac run was done with external restraints removed and zero
> > refinement cycles.  With external restraints I get 0.028A, without -
> > 0.006A.
> > 
> > This is with v.5.5.0109.  I assume this may be classified as a bug -
> > there is no limitation to using external restraints only for the
> > covalent bonds, thus they should not be included in bond_rmsd
> > calculation.  
> > 
> > On a practical side, I now have a misleading bond_rmsd value.  The
> > correct one can be calculated as described, but this may make geometry
> > weight optimization cumbersome.  Do I understand correctly that an
> > alternative is to monitor Zbonds, with a rule that it should be around
> > 1.0?  And more generally, shouldn't we not look at rmsd_bonds at all and
> > only use Zbonds instead (which is, I assume, an average bond length
> > deviation to the target value ratio?)  I suspect that "acceptable"
> > bond_rmsd value is slightly affected by sequence.
> > 
> > Ed.
> > 
> > -- 
> > "I'd jump in myself, if I weren't so good at whistling."
> >   Julian, King of Lemurs

-- 
Edwin Pozharski, PhD, Assistant Professor
University of Maryland, Baltimore
--
When the Way is forgotten duty and justice appear;
Then knowledge and wisdom are born along with hypocrisy.
When harmonious relationships dissolve then respect and devotion arise;
When a nation falls to chaos then loyalty and patriotism are born.
--   / Lao Tse /


Re: [ccp4bb] inflated BOND_RMSD with external restraints (refmac)

2010-10-15 Thread Ed Pozharski
Thanks, Ian, this is excellent.  It appears that depending on the
sequence the "ideal target rmsd" may vary from 0.018 for a poly-H to
0.024 for a poly-P.  Except for some really short sequences, in PDB the
variation is generally between 0.021-0.022, indeed undetectable.

On Fri, 2010-10-15 at 10:26 +0100, Ian Tickle wrote:
> Ed,
> 
> > On a practical side, I now have a misleading bond_rmsd value.  The
> > correct one can be calculated as described, but this may make geometry
> > weight optimization cumbersome.  Do I understand correctly that an
> > alternative is to monitor Zbonds, with a rule that it should be around
> > 1.0?
> 
> No, the average value of Zbonds over the whole PDB is almost exactly
> 0.5 (there's no clear resolution-dependence), so that would be a good
> starting point.  Re-refinement of ~98% of the PDB (PDB-REDO project)
> using weight optimisation showed a linear dependence of Zbonds on
> resolution in the resolution range 1 to 3 Ang.:
> 
> Zbonds = 0.85 - 0.146*dmin  (SD = 0.2)
> 
> so e.g.  = 0.56 @ 2 Ang.
> 
> Cheers
> 
> -- Ian
> 
> > And more generally, shouldn't we not look at rmsd_bonds at all and
> > only use Zbonds instead (which is, I assume, an average bond length
> > deviation to the target value ratio?)
> 
> Correct.
> 
> > I suspect that "acceptable"
> > bond_rmsd value is slightly affected by sequence.
> 
> Not really, the effect of sequence on the bond lengths is
> insignificant in comparison with the uncertainty.
> 
> Cheers
> 
> -- Ian

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