Hi,
Though I do not claim to be a data guru (like some of the true wizards of crystallographic theory who post here) - I would suggest that the 'improvement' in the Rfree does not in any way represent a better fit of the model to the data but rather is a consequence of the fact that our new Rfree is not 'free' since it is now selected mostly from reflections that were previously used to refine the model. Including your previous test data into the new refinement may have made an insiginficantly small impact on the R, but the fact that your R didn't even budge suggests that the only effect was the bias introduced into the Rfree (hence the apparently lower Rfree). In my limited experience straight runs of Refmac, regardless of the number of cycles, do not have the capacity to 'liberate' your Rfree from previously introduced bias. I believe that a while ago I've read a theoretical treatise on the Rfree contamination and methods suitable for reduction of bias - but to my shame I cannot recall whose work it was. What I do recall is that even simulated annealing may not be sufficient to completely dis-associate the Rfree set from the model if the reflections in the Rfree were refined against in prior runs. Conversion from P222 to P212121 implies removal of systematically absent reflections from scaling. Number of such reflections is quite small as compared to the overall number of reflections, especially at relatively high resolution of 2A (since the number of reflections overall grows much more rapidly with resolution than that of systematic absences). Therefore, it is not very likely at all that you've significantly affected the outcome of your scaling. I am sure that a much more eloquent answer is forthcoming shortly :-) Artem _____ From: CCP4 bulletin board [mailto:[EMAIL PROTECTED] On Behalf Of Young-Tae Lee Sent: Wednesday, November 19, 2008 7:27 PM To: [email protected] Subject: [ccp4bb] Order of scale/merge and reindexing Hi, I have been working on the 2A data. Though almost done, R/Rfree was around 0.20/0.27. Then I noticed that structural factors were reindexed from P222 to P212121 after scaling/merging in the early process (I am finishing up a little bit old data). To escape any problems from wrong merging of structural factors, I reindexed the raw mtz first and scaled/merged again. As new scale/merge assigned new test set of structural factors, R/Rfree after 10 new cycles of refmac5 refinement was quite low around 0.20/0.23. With a hope to make the new test set truly free, I ran almost 300 cycles of refmac5 refinement. Then R/Rfree converged to 0.20/0.25. However, I am not sure whether this improvement was really from correcting order of scale/merge and reindex, or still from including previous test data in the new refinement. Thanks, Young-Tae Young-Tae Lee, Ph. D. Research Associate The Scripps Research Institute Dept. of Molecular Biology
