Crytallographic resolution versus R factor

Ethan A Merritt merritt at
Thu Feb 25 14:06:48 EST 1993

In article <93056.112831SML108 at>, <SML108 at> writes:
|> Hi, in the hopes of starting a heated debate I would like to ask the
|> following questions:
|> Which is a better measure of the accuracy of a protein's crystal structure:
|> Its resolution in angstroms or its overall R factor?  And what would you
|> consider reasonable cutoffs for each of these?
|> Scott

		(followups redirected to bionet.xtallography)

	Resolution in Angstroms is a statement of the data collected, not
a measure of accuracy in refinement.  I.e., you know the resolution when
you collect the data, long before you have solved the structure!  It is
true, however, that the eventual precision of the refined structure is
limited by the resolution of the data used in refinement.  Generally the
resolution is limited by (1) how well the crystals diffract and (2) the
amount of time needed to collect higher resolution data.

	The standard crystallographic R factor is a measure of how well
the refined structure predicts the observed data.  As such it is not a
perfect measure by any means, and in fact it is theoretically possible 
to have a very low R factor for a totally incorrect model, although
in general the "model" in question will not look much like a real protein.
More common (though fortunately not _too_ common :-) is a mistake in
connectivity or other specific features in what would otherwise be a
correct model.  This can produce an R factor which looks reasonable, though
not as good as the correct model would have produced. 

	Resolution and R factor do interact.  One possible rule of thumb
is that a reasonable model for the protein in the presence of a minimal
solvent model (I'm not going to get into that here) should give an R
no greater than  (resolution in A)/10.  E.g. a 2.4A model with an R factor
greater than 0.24 is suspect.

	In any event the plausibility of the model itself needs to be
considered, not just the R factor or resolution.  A standard crystallographic
paper will report how far the stereochemistry of the model departs from the
"ideal" (i.e. most commonly observed) bond lengths, angles, etc for proteins.
A related measure is the reported distribution of backbone torsion 
angles (PHI/PSI), often called Ramachandran plot.

	The question of evaluating the goodness of a protein structure is
one that many people have thought about.  I suggest that you check, for
instance, the paper: "Stereochemical Quality of Protein Structure Coordinates"
by Morris, MacArthyr, Hutchinson & Thornton (1992) Proteins 12:345-364.

					hope that helps,

					Ethan A Merritt

Dept of Biological Structure                H510 Health Sciences
University of Washington SM-20              (206)543-1421
Seattle, WA 98195                           merritt at

More information about the Xtal-log mailing list