I thank Dr Zwart and Dr. Matias for their replies. This made me review
many notes and computer files.
Peter Zwart wrote:
Detwinning data and refining against detwinned data is not the best
thing to do, especially when you realise that detwinning data with
twin fraction larger than 45% is done using model information in CNS:
the detwinned data you get out is not experimental anymore.
Subsequently, you cannot trust your R and Rfree statistics as you
would normally and any 'improvement' you see in your free R or your
maps, might be due to the 'detwinning'
Have you tried using TLS refinement in combination with twin refinement?
Peter
Thank you for this good point. By the time I performed the detwinning,
the structure had already been refined against the original (not
de-twinned) MAD data. After this was finished, I used molecular
replacement before refining the structure against a better resolution,
native data set (2.1 A as opposed to 2.6 A). It was only after this, and
a little more refinement, that I detwinned the native data. Therefore I
was already working with model-biased maps at this stage, and the
protein structure was essentially finished. No nucleotides were present
in the model used to the detwin the data, so that should not have biased
the interpretation of the DNA density.
Nonetheless, your comment about the artificial lowering of R and
R-free is well taken. The de-twinning process could explain this. I need
to keep this in mind for the future--although I hope not to need to work
with twinned data again! Regarding this structure, I think that we had
an essentially correct strucutre (with some degree of coordinate error)
but with 'statistics problems' due to the twinning, and very little DNA
density to model (maybe due to twinning, maybe to flexibility). I didn't
try TLS.
--
*******************************************
Gregg Crichlow
Dept. of Pharmacology
Yale University
P.O. Box 208066
New Haven, CT 06520-8066
*******************************************
