rfree problem

Gerard 'CD' Kleywegt GERARD at XRAY.BMC.UU.SE
Fri May 10 03:50:24 EST 1996


large differences between r and rfree can have many causes,
for example:

- the model may still be very incomplete; the refinement program
  doesn't know that not all scattering matter is yet accounted
  for and will try to explain Fobs by manipulating those atoms
  it does know, resulting in low R but not necessarily a low
  Rfree; this is normal and shouldgo away as the model gets
  more complete

- overfitting the data; e.g., if you have non-crystallographic
  symmetry and low resolution data, and you don't use ncs
  constraints or restraints, R will go down at the expense
  of the model quality; since the changes do not improve the
  model Rfree will stay high

- problem with the data, e.g. a dataset for which in the highest
  resolution ranges really only HKL have been measured and the
  Fobs are more or less random; this should show up as a sharp
  increase in Rfree for those shells (of the order of 45-55%)

- the model may be largely correct, but there are a couple of
  bits in the structure which are very wrong (this once almost
  caused me to abort a mol replacement case; after initial 
  refinement Rfree was 45% so I assumed the solution was wrong,
  until I looked at the maps which were great for 80% of the
  structure and crap for the rest)

- finally, there may be one or more serious errors, e.g. register
  errors, parts mistraced, etc.; check the Ramachandran to see
  if there are continuous stretches of suspicious residues, use
  SA omit maps etc.

it's difficult to give advice without knowing more specifics
(resolution, ncs, model completeness and quality, etc.)

--Mr Gerard ;-)



More information about the X-plor mailing list