"Better model building tool" and "twin data" are two separate problems.

I like the program Phenix for refinement, and have used it for twinned structures (although mention of LLG indicates you are already using Phaser or similar?). What are your R factors?

Perhaps this link will be of use to you:

http://www.phenix-online.org/documentation/tutorial_twin.htm

Twinned structures are always limited in interpretability, particularly as the twin fraction approaches 0.5, due to overlapping model bias inherent in the mathematics of dealing with intensities of twins.

Good luck!

You can try to improve your MR search model without model building, namely by leaving out wrong or unreliable parts like side chains (at least in regions of random sequence similarity), surface loops or insertions/deletions. A brute force approach would be to convert the whole model to a polyalanine/glycine chain, i.e. deleting all side chains (e.g. with MOLEMAN). A more elegant solution would be to define the atoms to be included/excluded from the search explicitly (certainly possible with CNS, probably as well with PHENIX).

Try Balbes: http://www.ysbl.york.ac.uk/YSBLPrograms/index.jsp

is an automated molecular replacement program and it may help with a model building. But as Christina mentioned you have two separate problem. You may include twinning option in Refmac while refining. OR do you have other crystals?

For model rebuilding I suggest you try more than one tool (e.g. ARP/wARP and BUCCANEER). They are all fully automated so you can just run them and inspect the results. It's impossible to predict which one will work best for your specific case.

AFAIK all major refinement packages can deal with twinned data. So here too, I suggest you try more than one. I personally like Refmac, because it is fast and I have a lot of experience using it, but I've seen pretty amazing results with Buster too. Phenix and CNS have a large radius of convergence which may be good at the start of the building and refinement process.

Just keep in mind that you probably spent a lot of time getting the crystals and the data. So, it's perfectly acceptable to also spent a reasonable amount of time getting a decent structure model.

As Christine and Andrei have mentioned, you have two separate problems. One is getting the best map you can with your working model and your data from a twinned crystal; the other is building the best model you can with this map. Most popular refinement programs (e.g., refmac, phenix.refine, CNS) can give you a reasonably good map if you provide a twin law, though if the crystal is highly twinned you may find these maps somewhat biased towards your model.

A composite omit map such as you have made, or the related omit procedure mentioned by Karsten are both good ideas a to try.

If you find your composite omit map shows features that are believable but are not in your model (i.e., good side chain density for side chains that are not in your model), then it may be a very good starting point for model-building either by hand (as you have done) or with automated model-building.

You should be wary of the LLG gain obtained after model-building and refinement. You would expect pretty much any model to have an improved LLG in this case, so the higher LLG does not really show that the model is improved. A better measure would be the free R factor after refinement.

For automated model-building, you have many choices; some popular ones are ARP/wARP, buccaneer, phenix.autobuild. All of these can take a data file, a starting map, and a starting model and improve it. You will want to specify the twin law when using most of these procedures. In phenix.autobuild you will also want to specify use_two_fofc_in_rebuild=True so that it does not try to carry out density modification which may not work so well with twinned data.

I would second Robbie's point that all of the major packages should be able to handle twinned refinement. In most cases, an important point is to accurately estimate the twin factor, and the xtriage tool in PHENIX is very powerful for this.

One approach that I used successfully in a similar situation was to run density modification (DM from CCP4, in my case) on the MR solution, and inspect the map. It didn't show anything much new, but it did highlight regions of the model that were completely wrong. Once I had removed these and done a little manual rebuilding, Arp/Warp did the rest.

Good luck!

I wouldn't look too much on the LLG in phaser, they are not always a good indicator and I already had quite some false positives (many false solutions with LLG>750).

The correct MR solution is the one that refines.This remains true for twinning. It just takes a little longer.

As far as I can tell, you have a poor MR solution problem and twinning is just on top of that.

Just keep model building/refining if you trust you solution. Are you sure you have the twin law right? I am pretty sure MR-Rosetta doesn't work with twinned data but you might want to check that yourself. It is a great tool for poor MR

solutions. You can also emulate MR-Rosetta by running Rosetta in stand alone without inputting your data and try to refine a bunch of models at the position(s) you found in MR.

In any case. It is just twinning, so don't panic and stick to the basics of post-MR refinement. I'd recommend to initially truncate your solution even further down to the most reliable regions of the structure (generate an OMIT) and run rigid body refinement with very small domain definitions to account for slight but crucial differences in core packing. Do you have beta sheets in the core region? Variation from as little as 10 degrees between secondary structure element orientations can make your refinement/map calculation fail if you do not separate them well enough in the rigid domain definitions. If your refinement picks up you expect to see a steep drop in the residuals. If you have mainly helices in the core you might want to cut back the resolution for rigid body refinement and truncate all sidechains at c-beta. Helices are well defined to low(est) resolution, let's say 5A, and what you are looking for is the correct mainchain position. If you already had that one right, your map would be interpretable at 2.5A. Even if the structure is twinned, as long as you have the twin law right and the data processed correctly.

You also might want to try simulated annealing with high initial energies and low cooling rates but i have no positive experience with that.

The first thing I would do is going back to the raw data to make sure everything was processed properly. I always invest quite some time in data analysis/processing. The last thing you want to do is fitting poor models into inconsistent data. Is it twinning by merohedry or twinning by pseudo-merohedry or else? Do you have a rhombohedral or any other 'problematic' space group? What does the Patterson map look like? Do you have additional NCS or even translational NCS? Do you see reflections splitting at high resolution? You also might want to use a different program for integration and scaling since 'one of the more prominent ones' quite often does a poor job....

good luck.

First you go for a scattering experiment with x-ray/LASER/neutron, then you will get the idea for modeling and best fitting of the experiment data.

Oh, by the way. You said you improved your model manually. I assume by real-space refinement in O or coot?

This way you will train your phasing model to match the LLG

Hi Stefan,

Thank you so much for nice suggestions. As you said I am trying now MR-Rosetta , lets see how does it work. I did scalling in HKL2000 and Space group of my data is I41 with 4 NCS copies and 0.49 Twinning fraction. I got the twin law as -h,k,-l from extriage. Just to mention you that xtriage also show that data have psuedo-translation issue as well. Yah you are right i could get 2 potential MR slutions and by doing real space refinment in coot i could improve the model. When i say i improved the model it means in terms of LLG gain and crystal packing. I tracked the z-score and TFZ as well. Although i never got a high Z-score (it always remains below 5.0) but could get TFZ as 11.0. But when i go for refinement R-free doesnot drop much (i have included twin law in refinement). The best Rfree i could get was 0.41 but if i look at the maps they look crapy. But yah your suggestions are certainly nice and i hope to solve the issue. Please shade some light onpseudo translation. How one one overcome it? Do you have experiance of solving thestructure in lower symettry spce group (for twin data set)? Will update you after Rosetta.

Thank you so much

Raj

Hello Allan,

So I could obtaiin a best Rfree as 0.41 but map do not look good. I scalled my data in HKL2000. Space group is I41. Xtriage shows 0.49 fraction of pseudomeroheadral twinning with pseudo-translation issue. It suggested -h,k,-l twin law. I have NCS and there are total 4 copies in asymmetric unit. Actually i tried quite a long refinement also (50 cycles) in which Rfree goes down to 0.37 but again jumps in 40's and as i said maps are bad. Look forward for your suggestions.

Raj

Hello Christina,

Yah I am using phenix. The best Rfree i could get was 0.41 but map donot look gud also maps donot look good. I also tried to make omit map in phenix using create map program. While making omit map i provided .Sca and model as input. But resolve maps were random. Though whn i use omit maps obtained from sutobuild, they look little better but may be they could b biased too. I appreciate your suggestions. Thanks

Hi Harmer,

Thanks for the suggestion. Yah I used xtriage only for twin detection and am trying DM now, will update you once i finish it. Thanks

Hi Terwilliger,

Yah you are right the maps which i am getting after refinement look biased. Can you please tell me what is the best method to get reliable composite omit map? Autobuild or Creating them using create omit map program in phenix? I just got the feeling that when i use creat omit map option then resultant maps are random (no continuity) but one after autobuild look better. For create omit map progrm i just give .sca fil and pdb as input (no mtz, coz upon inclusion of mtz program show error and stop), Whereas in autobuild one can provide mtz and .sca both. Thanks

The best option is the scattering experiment.

Hi Dillip

What do you mean by scattering experiment?

It includes Small Angle X-ray Scattering (SAXS),Small Angle Neutron Scattering (SANS)Small Angle Light Scattering (SALS) which are the best options to reach at electronic, nuclear and micron level of the material under investigation.

X-Ray crystallography is also a kind of scattering experiment i guess and it is much stronger than SAXS which just gives envelop shape isn't it?

NO No ..

It is absolutely different from XRD technique and the structural parameters evaluated by SAXS is different from XRD.

Rajeshwer,

Does the data 'look like' I4(1)22? Do you have have enough frames to process your data in P1? P2 should be fine, too if you have less than 180 degree data.

The body center (I) can be generated bey either NCS or twinning.

and yes, I'd recommend d*trek or XDS for data processing if possible.

in a hurry

s.

EDIT:

Translational NCS can emulate a body center so your data does not have to be I41.

-h,k,-l twin law seems a little odd to me. You might have merged reflections that are twin related in order to come to I41 so the twin law you see is maybe not the actual law but a part of the actual symmetry or a mix between crystal symmetry and twin symmetry. You should definitely try the lowest space-group (P1) without merging assumed symmetry pairs. This way you should be able to identify the true symmetry/twin relation. Ask a small molecule crystallographer to help you if you know one. You might also want to go to the ccp41 bulletin board and ask there. It is more frequented by experts and many people there have strong hand-on expertise on twinning and MR.

One more thing, you really need to check if your reflections were all predicted by HKL2000 and if the predicted reflections were mostly there. Also zoom in the high res shells and see if the reflections split. HKL2000 is really good for that. If you see all of that you most likely have twinning by non-merohedry, in which case you really want to have different crystals.

good luck.

S.

Dear All,

Thank you so much for all the suggestions you guys have given. With your help we have been able to resolve an almost perfect twin and psuedo translated diffraction data. Now my R-factor is about 27 or so. therefore I am confident that it can be worked out successfully. I would like to tell you guys that the space group that we have predicted (I41) was correct and with the help of a combination of omit maps, DM, rosetta I could improve my model which finally yielded right solution in MR. So i appreciate your help and support. My special thanks to Stefan, Thomas Terwilliger, Nicholas Harmer, Robbie Joosten, Karsten Niefind and Christina.

Raj

The ZANUDA / MOLREP combination apparently is also very useful. I saw a convincing talk about it at the CCP4 Study Weekend (on video), but haven't had to use is myself yet.