Hi, In such cases, I favor first trying to establish correctness of the heavy atom substructure. If you can have high confidence in the MSE sites, then there are various strategies you can try (phasing/density modification parameters) to squeeze out the best phase information from available data. On the other hand, if your sites are incorrect or mostly incorrect, you can be beating around the bush and not get anywhere. I would try to find the Se sites using HySS (phenix.hyss), SHELXD and SOLVE and then run EMMA (phenix.emma) to check how many of the sites are consistent, which would be indicative of correctness. Then take all the consistent sites and use them for generating phases and model tracing (partial models can be verified by performing MR on the native, or by running a DALI search to obtain better homologs for MR). If the latter works, you can then try to combine some MR (and also look into CNS DEN refinement and MR_Rosetta) and experimental phases to complete your heavy atom substructure and improve phases and model iteratively. Hope this helps. Thanks, Debanu. -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Francis E Reyes Sent: Thursday, February 16, 2012 3:10 PM To: PHENIX user mailing list Subject: Re: [phenixbb] AutoSol with 4.2A dataset for a protein/DNA complex Ahh low resolution structure solution, my favorite! Your SKEW isn't very promising... usually scores of 0.1 and above are better. I bet the maps before DM and after DM look bad. Getting the SeMet substructure at 4.2A is challenging, extending the phases from 4.2 to 3.2 without a buildable model is going to be even more challenging. [1] Are there any endogenous anomalous scatterers? I'm thinking Zn since this is a DNA:protein interaction. If you have crystals, run a fluorescence scan at the Zn edge. [2] Do you have any structural information on the protein ? (perhaps the apo state has been solved) If the answers to [1] and [2] are yes, then you can use the protein as an MR to find the Zn site and cross validate with the SeMet data. (use the Zn phases to find SeMet) If the answer is no... there's still other options. F On Feb 16, 2012, at 3:51 PM, Wei Shi wrote:

Hi everyone,

I have got a 4.2A SeMet MAD dataset for a protein-DNA complex. And, the native dataset for this protein-DNA complex is ~3.2A. The space group for the native and SeMet crystals seem to be different. I am not sure whether it's possible to solve the structure with the current data I have, and wondering whether any of you have experience with working with this low-resolution data and any suggestion for me.

I processed the data with XDS in space group P31, and used *.cns.hkl file from XDSCONV as an input for AutoSol and also include the sequence for protein only. The protein dimer has 418 residues and DNA is 32bp. I run Autosol with the default setting. Below is statistics I got. It seems that I didn't get anything promising. Any comment or suggestion about what to try next? Thank you so much!

Statistics:

Top solution: 2 Sites: 11. Space group: P32. FOM: 0.550.

BAYES-CC: 8.10. Residues: 465 Side-chains: 0. Chains: 60.

Model CC: 0.67 R-work: 0.4100 R-free: 0.4569.

Under Heavy-atom search and phasing:

Space group

# of refined sites

FOM

Overall score

R-factor

Map skew

Corr. of local RMS density

Solution1

P31

11

0.530

8.10+/-11.80

0.4184

-0.12

0.66

Solution2

P32

11

0.550

8.10+/-11.80

0.3881

-0.10

0.66

Best, Wei

_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb

--------------------------------------------- Francis E. Reyes M.Sc. 215 UCB University of Colorado at Boulder _______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb

I realized that I missed the fact that the P31 and P32 solutions yield very similar statistics - that's definitely not a good sign. I second Francis's suggestion for MR if there are any homologs, except I think you could use Phaser to go directly from the initial solution to Se phasing, and if you can actually find an MR solution it might work better to just start with the native data since it is potentially good enough to autobuild. -Nat On Thu, Feb 16, 2012 at 3:13 PM, Nathaniel Echols wrote:

On Thu, Feb 16, 2012 at 2:51 PM, Wei Shi wrote:

...

I processed the data with XDS in space group P31, and used *.cns.hkl file from XDSCONV as an input for AutoSol and also include the sequence for protein only.  The protein dimer has 418 residues and DNA is 32bp. I run Autosol with the default setting. Below is statistics I got. It seems that I didn’t get anything promising. Any comment or suggestion about what to try next? Thank you so much!

Statistics:

Top solution: 2           Sites: 11.                     Space group: P32.            FOM: 0.550.

BAYES-CC: 8.10.          Residues: 465           Side-chains: 0.             Chains: 60.

Model CC:  0.67           R-work: 0.4100          R-free: 0.4569.

At this resolution it's hard to know how much to trust those R-factors, and 60 chains is worrisome, but did you try running MR on the native dataset with the final model yet?  It may be too chopped up to be useful in a different crystal form, but if this works you're nearly there.

Other things to try: 1. Nucleic acids are much easier to see than protein at low resolution - see if you can find the double helix in the density-modified map.  (It will probably have protein residues built into it, but hopefully the shape is still distinctive.) 2. Generate an anomalous difference map in phenix.maps using the output model and either the peak or high remote (if you have it) wavelength (you might first need to combine the R-free flags generated by AutoSol with the original anomalous data) and look for clear Se peaks around the heavy atom sites.  Also check that the Se sites look like they're plausibly attached to Met residues. 3. Run phenix.find_helices_strands with the final map, and see how that works for MR on the native data.

-Nat