Thank you Pavel for the clarification!
What I was really trying to point out is that a "missing atom", occ=0.00
and occ=0.01 are not as similar as one might naiively think. Also, if
you put a ligand into a pocket and the occupancy refines to > 0, that
does not necessarily mean the ligand is "partially occupied". If the
pocket is actually filled with flat bulk solvent, then you expect the
ligand occupancy to be non-zero, simply because something is better than
nothing. However, if the bulk solvent mask were somehow "smarter" and
filled the pocket of a, say, 60% occupied ligand with flat bulk density
at 40% the level of bulk density used far away from any atoms, then one
might actually see the occupancy of a bogus ligand refine to zero. That
is, a ligand built into a pocket that is truly "empty" (filled with flat
bulk solvent) and then occupancy refined would actually be a
"competition" between two alternative hypotheses: 1) ligand in the
pocket, 2) nothing but solvent in the pocket. If the occupancy of the
ligand refines to zero in this context, then you can be quite confident
that it didn't bind, at least not in the given orientation.
I fully realize that the implementation of this is easier said than
done, but perhaps it would be worth the effort?
-James Holton
MAD Scientist
On 6/16/2014 3:04 PM, Pavel Afonine wrote:
Hi James,
a remark: different programs may treat occ=0 differently. In
phenix.refine (phenix.maps, etc) atoms with zero occupancy will be
ignored for bulk-solvent mask calculation, unless you ask to do
otherwise. For example, this means that if you want to calculate a
ligand OMIT map both options
- removing the ligand from PDB file;
- setting its occupancy to zero and making sure mask does not ignore
occ=0 atoms)
are a) not equivalent and b) both not good.
In first case (removing atoms from file) bulk-solvent will flatten
residual map (as you pointed out). In second case bulk-solvent will be
excluded in a very specific area, so that residual ("green") density
you see there may be either just bulk-solvent or ligand in question or
a mixture; obviously not a very useful information! This highlights
the fundamental problem of flat bulk-solvent model the way it's
currently used.
Pavel
On Sun, Jun 15, 2014 at 3:01 PM, James Holton <jmhol...@lbl.gov
<mailto:jmhol...@lbl.gov>> wrote:
The principle difference between occ=0 and omitting the atom
entirely is that occ=0 atoms exclude bulk solvent. Or at least
they do for typical operation of contemporary refinement
programs. So, by defining occ=0 you are forcing all map voxels
within ~0.6A or so of your "invisible" atom to be vacuum. If you
omit it, then the bulk solvent may "flood in", perhaps enough to
pull the fo-fc peak down below 3x rms. How much the bulk solvent
floods in depends on how nearby atoms exclude the bulk solvent,
and this, in turn, depends on which refinement program you are
using. Different bulk solvent implementations use different
radii, "shrink" parameters, etc. In addition, bulk solvent always
"bleeds" a bit into surrounding areas because the solvent B factor
is never zero.
The real problem, I think, is that for any voxel of the map there
is ALWAYS "something there". The only question is: what is it?
Is there a 100% occupied ligand? 100% occupied solvent? Two
conformers of the ligand? Or is it some mixture of all these? If
you are asking these questions I think it is most likely a
mixture, and mixtures are hard to model. What is worse, mixtures
of a partially-occupied ligand with bulk solvent taking up the
slack is currently impossible to model. We will have to wait for
partial-occupancy-bulk-solvent to be implemented before we can
build representations of these alternative hypotheses and and test
them with competitive occupancy refinement.
The bulk solvent is actually a very good example of something for
which we see "no evidence" in our electron density maps, yet we
model it in because 1) we know it must be there, and 2) it makes
our R factors lower. What more could you want?
-James Holton
MAD Scientist
On 6/13/2014 7:45 PM, Frank von Delft wrote:
Hi all - talking about ligands, a quick question on that old
conundrum, of what to do about invisible atoms -- build them with
occ=0, or omit them?
For bits of protein, I know all the arguments; personally I
prefer omitting atoms because:
* for amino acid sidechains, their presence is implied in the
residue name.
* for whole residues, their presence is implied in the sequence
numbering
However: what about ligands? Nowhere else in the PDB file is
their presence implied - or have I missed something?
(Certainly disorder in a ligand is important information that
needs to be captured!)
Cheers
phx