I check ha sites by 1) looking at the shelxe plots ( these are displayed in the CCP4 GUI task - prob also in phenix? ) If there is good contrast for one hand rather than the other the sites are likely to be correct - if not - hmmm; there might be a reason but they could be wrong..
Then I check the sites by giving the strongest ones to phaser_ep and seeing if it suggests more which agree with SHELX ones.. Bias is not usually a problem if the initial build is correct but if the Rfactors are not coming down that may not be the case. Can you see sidechains - espec MET near your Se sites? Eleanor On 12 July 2013 04:12, Yuan SHANG <shangyuan5...@gmail.com> wrote: > Dear All, > I got a Se-crystal diffracted to 3.3A. The protein is predicted to be a > coiled coil. After I got the heavy atom positions from a shelxd based > programs from AutoRickShaw, I run phaser and phenix.autobuild to get a > initial phase shown in the attachment. I fit a "standard helix" into the > density maps. After that, I did a two step iteration to improve my phases > similar to the methods mentioned in Roversi's paper "With phases: how two > wrongs can sometimes make a right". > Step 1: Run DM with current model, heavy atom positions, the initial > density maps to get a new DM map. > Step 2: In the new DM map, some new density appears. Fit new helixes in > this new map. > The iteration ends when I found no obvious helix densities. However, when > I run refinement, I still got very bad R/Rfree values. It seemed I got a > lot of model bias during the process above. > > Now, I have some questions that puzzled me a lot. > 1. Is there any programs that could help to compare and check the heavy > atom positions in shelxd solutions? Especially in cases when the solutions > are not obvious. SitCOM seemed very good, is there a 32-bit distribution of > this programs? Or any others have similar functions? > 2. How to reduce the model bias and phase errors in my situations? any > suggestions? > 3. Should I collect a MAD data instead of SAD data? How much would that > could help? > > Best regards & Many thanks! > > Yuan >