Zoom Transcript

---

We’ve started the recording.

---

Thank you

---

resolution recommendations for anisotropic or tls refinement?

---

for high anisotropic data try Staraniso website and check if anisotropic map are better than isotropic map

---

for me tls is just a way of decrease R factor for things you have not yet modelled, so bad

---

is it possible to run xtriage on serial data?

---

how does the meaning of the occupancy differ in case of ligand. For non-covalent ligands what should be the occupancy fixed at at the beginning of the refinement.

---

At which point is better to run TLS parameter?

---

Is there any way if I have two molecules of ligand in my binding pocket but not appear as the same occupancy so is phenix can tell us or optimize the occupancy of the two copy of ligand

---

Occupancy for covalent ligands can be set to something like 20 or similar to the bonded protein atom. It may be that the protein is too high so starting at 20 or 30 is fine

---

How would you approach refining a structure with multiple ligand populations at the same position (i.e. multiple intermediates in the same crystal with different occupancy)?

---

If the data are very anisotropic (i.e. 2.5 Angstroms, 4 and 4 along x, y z axis), but have been elliptically truncated, is there anything especial that one needs to think of when refining? Model is close to final.

---

you can refine two ligand occupancies

---

read this http://phenix-online.org/newsletter/CCN_2015_07.pdf#page=12

---

There are two types of options for simulated annealing (Cartesian and Torsion angles) what is the difference and when to use ?

---

To

---

Two different ligands in the pocket is similar to example 5

---

Of course, the accuracy of the answer depends on the resolution

---

bioarx https://www.biorxiv.org/content/10.1101/2020.03.26.010587v1

---

journal https://www.cell.com/structure/fulltext/S0969-2126(20)30287-2

---

Great talk Pavel!

---

thanks very useful

---

Hi Pavel, Tanks for this very nice talk (basic teaching, but very nice), as usual. I’d like to ask about anisotropy and how phenix.refine deals with it. Apart my specific question, I think that anisotropy is not well understood by most people, so general & specific comments should be good for all. Anisotropic scaled dataset (from Staraniso) can have anisotropic cut-off reflections marked as non measured or not be present in the mtz (as far as I remember) . How phenix.refine deals with those reflections during the refinement: filled / unfilled? Any difference between Staraniso anisotropic cut-off data and normal spherical cut-off data? (I know that phenix.refine output filled/unfilled map) Thanks, Lionel

---

For a 0.9 A dataset, can I use Phenix or do I need to use packages like Shelx?

---

Thank you for the wonderful talk.

---

newsletter http://phenix-online.org/newsletter/

---

Phenix will work well for 0.9Å data

---

Is it possible to subscribe to the newsletter? (couldn't find a way...) Or get some way of reminder when a new newsletter comes out?

---

it is announced in the phenix jan

---

I announce it on the phenixBB

---

Thank you, Nigel.

---

Yes Paul point is very relevant. To use in phasing please do use original data, not truncated

---

There are special features for hi-res refinements that you should become aware of

---

Hi Pavel, Tanks for this very nice talk (basic teaching, but very nice), as usual. I’d like to ask about anisotropy and how phenix.refine deals with it. Apart my specific question, I think that anisotropy is not well understood by most people, so general & specific comments should be good for all. Anisotropic scaled dataset (from Staraniso) can have anisotropic cut-off reflections marked as non measured or not be present in the mtz (as far as I remember) . How phenix.refine deals with those reflections during the refinement: filled / unfilled? Any difference between Staraniso anisotropic cut-off data and normal spherical cut-off data? (I know that phenix.refine output filled/unfilled map) Thanks, Lionel

---

Super! Thank you very much! Understood!

---

For high resolution refinement, the documentation has some useful info:

---

http://cci.lbl.gov/ceres/goto_entry/7a6b_11669/10_2020

---

wrong link

---

https://www.phenix-online.org/documentation/reference/refinement.html#refinement-at-high-resolution-higher-than-approx-1-0-angstrom

---

Thank you, Dorothee!

---

Can I make Phenix refinement run faster?

---

I have question about merge data sets in phenix,, how you decide what data can be merged, and hoe is the scaling of data sets done?

---

When it comes to high resolution data (0.84A) I have realized that the B factors are not correctly refined and as a result a lot of positive peaks (green map) are observed for instance on main chain atoms. The data is very high accuracy and I/sigI in the highest resolution shell about 5 (the detector couldn't have been moved closer) How can we solve this issue?

---

Mmm I’m also interested in the question about merging of datasets in phenix! :)

---

How do you generate Rfree flags if not already present in the mtz file?

---

@Maria: It may be that the B-factors are in a wrong minimum. You could reset them (f.ex. use phenix.pdbtools) and refine again. Also, are you using isotropic or anisotropic B-factors?

---

@Mark:

---

You can generate Rfree flags

---

https://www.phenix-online.org/documentation/reference/reflection_file_editor.html#dealing-with-r-free-flags

---

Is it appropriate to truncate data to try for a lower Rfactor/Rfree? If not, what are some strategies for getting lower Rfactor/Rfree?

---

@Mark phenix.refine will also do it if you set the option

---

@Dorothee, @Nigel - thanks!

---

@Alex: it is not advisable to truncate data for the sole purpose of lowering Rwork/Rfree. Truncating should be only done if the data don’t have enough information content. Instead, I suggest 1) investigate if the data have some underlying issues (twinning, radiation damage, etc) 2) think about the refinement strategy (what parameters are refined) 3) look at the maps: are there unmodeled regions or are the maps “clean”?

---

@Alex (continued): You should also look at model quality and model-vs-data fit. R-factors are not the sole measures for refinement outcome.

---

@Dorothee - thank you!

---

On the topic of NCS and whether we should apply it - is there a percentage of residues on each chain that we should reach to determine whether or not we apply NCS restraints?

---

coment on optimize xray?

---

i have a dataset where i see 2 molecules in the asymmetric unit but they are distant apart? I tried solving the data in different space group to see of its correct but data fots very well in this space group..Appreciate for any suggestions

---

fits*

---

“Is it appropriate to truncate data to try for a lower Rfactor/Rfree? If not, what are some strategies for getting lower Rfactor/Rfree?” — No, bad idea.

---

How can one refine one particular bond lengh in a ligand, during refinement run?

---

“resolution recommendations for anisotropic or tls refinement?” — Indivisual anisotropic: 1.7A or better, TLS otherwise.

---

@Lionel, re STARANISO: refine with whatever let’s you to progress as far as you can, but switch to original data towards the end.

---

“At which point is better to run TLS parameter?”: towards the end of refinement. Or definitely not in the begining!

---

“resolution recommendations for anisotropic or tls refinement?” - remember that individual anisotropic add 5 more parameters per atom, so a lot of data is needed. Consider that you’d like 3 or more reflections per parameter in refinement, so with N atoms that would be 3x9xN reflections (approximately) for anisotropic displacements and coordinate refinement.

---

“On the topic of NCS and whether we should apply it - is there a percentage of residues on each chain that we should reach to determine whether or not we apply NCS restraints?” — No black/white answer here. Try both and see what works best.

---

@Dorothee Liebschner: It is refined anisotropically but I also tried isotropically. I have reseted B-factors to some values (like3, 6 etc) but at the end the map shows positive peaks on many atoms.

---

On the topic of NCS and whether we should apply it - is there a percentage of residues on each chain that we should reach to determine whether or not we apply NCS restraints?” - adding any restraints from NCS should help, but the fewer the restraints the less impact on refinement. The default cutoff is 85% sequence identity. If you have chains that are more different then you might need to change the cutoff.

---

“how does the meaning of the occupancy differ in case of ligand. For non-covalent ligands what should be the occupancy fixed at at the beginning of the refinement.” — occupancy reflects the probability of the ligand to be there, which ranges from 0 to 1. It is likely the ligand is partially occupied in which case occupancy should be refined.

---

“How do you generate Rfree flags if not already present in the mtz file?” — you should do it first thing once you got the data. Reflection file editor in Phenix GUI is perhaps the best place to do it!

---

@Maria: I see. If you want, I can have a look at your data. If you can share, you can send them to [email protected]. I would need all relevant files to do refinement: model, data, ligand restraints, non-default options you used

---

“How can one refine one particular bond lengh in a ligand, during refinement run?” — Not sure what this means. Try to come back with this questions again at Q&A session!

---

“Is there any way if I have two molecules of ligand in my binding pocket but not appear as the same occupancy so is phenix can tell us or optimize the occupancy of the two copy of ligand” — Good question. Yes, it is possible, this topic is covered here: http://phenix-online.org/newsletter/CCN_2015_07.pdf#page=12

---

Links to residues are applied automatically... at which level? How to generate and modify the distance, angles, dihedrals between protein and ligands in general? Also, in general, do we need a cif file for the ligands and one cif to specify the connectivity with the particular protein residue it is bound to?

---

“For a 0.9 A dataset, can I use Phenix or do I need to use packages like Shelx?
” — to add on this.. This may need some parameter tweaking. So please reach us out if you need any help and we will make sure you have best parameter setup for refinement.

---

Is it always recommended to use at some point Amber restrains or are there more particular cases better suited for it?

---

“Links to residues are applied automatically... at which level? How to generate and modify the distance, angles, dihedrals between protein and ligands in general? Also, in general, do we need a cif file for the ligands and one cif to specify the connectivity with the particular protein residue it is bound to?” — automatic linking totally relies on input model geometry. If atoms are close enough and it makes chemical sense, then the program will try to link them. If not, it will leave them alone. In rare instances it may link what’s not supposed to be linked. So better check. *geo file out of refinement lists all the links and geometry restraints so it’s the best place to check!

---

Thanks Pavel. What if the atoms to be linked are far and I want to force the link between them. How do I control the geomentry of the bond?

---

“Thanks Pavel. What if the atoms to be linked are far and I want to force the link between them. How do I control the geomentry of the bond?” — yes, you can do it by defining custom bonds in refinement: http://phenix-online.org/documentation/reference/refinement.html#definition-of-custom-bonds-and-angles

---

If ou make a modification in .geo file, rather then cif file, hoe do you include the .geo file in the next refinement cycle using gui version?

---

@vserrano: the .geo file is only informative output. It cannot be used to change restraints. You will have to define custom bonds in the edits file.

---

i have a dataset where i see 2 molecules in the asymmetric unit but they are distant apart? I tried solving the data in different space group to see of its correct but data fits very well in this space group, lower Rfactor/Rfree..Appreciate for any suggestions

---

Thanks.

---

“i have a dataset where i see 2 molecules in the asymmetric unit but they are distant apart?” - one of the molecules probably needs to have a symmetry operation applied to put it close to the other. I think you can do this in Coot.

---

@Paul Adams thanks ..any option you can elaborate on

---

protein molecule with 12 chains, each chain has ribose 5-phosphate bound, but density of the ligand (R5P) is different in each chain, mean ligand can be ribose and/or can be ribofuranose, how one can go with eLBOW and refinement?

---

“Can I make Phenix refinement run faster?” — In general it does a lot of things to save you from doing them yourself manually. Do you have an example where runtime overhead is unreasonable vs outcome?

---

What is your recommendation to generate bond information to create a data_link cif file between a compound and a protein residue?

---

Can you point us to the most updated list of ligand restraints available on the latest version of phenix?

---

“Can you point us to the most updated list of ligand restraints available on the latest version of phenix?” — Look in your Phenix sources in phenix/modules/chem_data

---

at what stage of refinement after molecular replacement should one start looking for ligand density

---

Which file should be provided as you mentioned for the two same ligand in the pocket but different occupancy?

---

“at what stage of refinement after molecular replacement should one start looking for ligand density” — Keep eye on the pocket constantly from start to end. But model the ligand towards the very end when all else is as good as possible.

---

and the ligand (R5P) binds through anomeric carbon

---

Thanks

---

Thanks Nigel, great talk

---

Thanks Nigel. Pavel also replied in the chat. I will get in touch if I have troubles.

---

How can we fix the occupancy of two same ligand in the same bonding pocket?

---

How we be 100% sure that the ligand density we are looking for is the exact molecule that you think it is..in my case I doubt the ligand density with cryosolution ingredient..i checked the crystals mass spec after washing with the mother liquor but it gives mass of the ligand..is it acceptable?

---

Thank you!

---

@Nigel it is R5P with different confirmations, density is not clear for Ribose and ribofuranose I will connect via email for details

---

Can you comment on Polder maps for ligands?

---

What is the best way to prove the existence of a ligand in the structure?

---

Yes you right we can fix the occupancy in coot but we run in phenix its change the occupancy half and half again

---

@Giang: https://www.phenix-online.org/documentation/reference/polder.html

---

“What is the best way to prove the existence of a ligand in the structure?” — Use Polder OMIT map and its interpretation.

---

How about the bias from the model provided then? With simulated annealing in addition?

---

“Yes you right we can fix the occupancy in coot but we run in phenix its change the occupancy half and half again” — phenix.refine is supposed to refine the occupancy. So if it refines it to 50/50 then perhaps this is what it is!

---

what to do in case where a part of the ligand doesn’t have density but the rest has very good density and fits well?

---

for ligand ID tool, where to find in phenix?

---

“How about the bias from the model provided then? With simulated annealing in addition?” — If you have not built the ligand in the first place then there is no bias.

---

thank you

---

Thanks everyone

---

Thanks!

---

Thanks Nigel for your talk! You also answered my initial question about SS bonds (it worked now :) )

---

also very interested in tools that helps with interpreting extraneous densities connected to protein residues, i.e., possible covalent modifications?

---

If you have extraneous densities, it could be covalent mods but we don’t have an easy solution

---

I have some ideas so I’d be happy to take a look

---

@Silvia Glad it worked

---

Ligand identification tool is under the “ligands” tab.

---

@Najeeb I can help with your two ligand situation

---

I am solving a structure where a ligand is a disaccharide. Previously solved structure of the same protein demonstrated the shift in certain regions upon binding of a monosaccharide. My protein exhibits the exact conformational shift expected for the binding of the ligand, however the electron density is not there in sufficient area to fit the disaccharide. The area of green density only roughly corresponds to a monosaccharide moiety. MS experiments confirmed the disaccharide is stil intact. Resolution is 2. The ligand was soaked, not co-crystallized. Any suggestions?

---

I have a structure at 1.3A resolutions and most of it looks good and I am at the end of model refinement but I the clashscore is high.

---

is there any easy why to deal with clashscores?

---

@Jossaf: Did you look at the clashes in Coot and did you try to fix them manually?

---

yes and some of them does not look like clashes because they have good defined electron density maps

---

@Nika Fit the monosaccharide and refine. The density may appear. Also do a Polder map of the monosaccharide to see if you can see the second

---

@Nika you can also then fit the second and do a Polder of the second and so what happens

---

Thanks!

---

and most of the time sidechain and water molecule should be where they are

---

Please educate us about the ethics of publishing structures looking at the ascertainity around the ligand density..I often have seen criticism over published structure with false ligand density..as a newbie I really want to hear the experienced

---

@jossaf are there other errors (rama outliers, rotamer outliers, etc) near where there are clashes? sometimes clusters of errors can help identify areas that have a problem that needs to be fixed

---

@Jossaf: Then fix first those where the density is not so good. The Phenix GUI orders the clashes according to overlap. This allows finding the worst overlaps.

---

@Rajnandani Run Polder to see what it says. Read the Twilight paper to see what Rupp says should be there.

---

@Rjnandani: 10.1107/S0907444912044423 and follow up papers

---

thanks

---

@Rajnandani BTW, we are working on a suite of tests so we would like to see your example if you’d like to share

---

@vincent in some parts yes, but in some no

---

@dorothee thank I will try to re-look them

---

@Nigel W. Moriarty Sure

---

@Rajnandani: yes, examples for unclear ligands would be helpful

---

Can you explain 'RSRZ' outliers in PDB validation?

---

I have a question about a amount of water in a structure, I have 1.2A resolution, my protein has 400 AA and is highly soluble in sodium buffers at pH 5. What is a good amount of water? I am adding water manually with COOT after phenix.refine and I have 340 H2O molecules is this enough? I am looking the electron density map at 1.40 rmsd.

---

Phenix can incorporate it? [email protected]

---

PDB validation reports always list RSRZ>2.

---

rule of thumb: 2 waters/residue

---

“Can you explain 'RSRZ' outliers in PDB validation?” — This is a flawed metric as it is implemented in PDB. Tom and others have reported issues with this metric and never heard back from PDB. So.. personally I think as long as usual CC is good you are good.

---

I totally agree with you regarding RSRZ outliers.

---

@paul thank you very much

---

clap clap

---

Thanks Christopher, very nice talk!

---

👏

---

Thank you Christopher, awesome talk!

---

could you comment on adjusting the wxc or wxc_scale factors to improved overall structural bonds and angles.

---

Where do you stand on the replacement of the refined hydrogen atoms when determining the clashscore?

---

true I have a problem in one loop

---

my b factors are around 100

---

thank you a lot

---

When are the talks gonna be online?

---

@Gihan In a few days

---

Question about refinement: I have twinned data and I did first refinement cycle using the twin law in Phenix. Would it be logical to use the refined reflection data from the first cycle in the second cycle of refinement and not use the twin option again in phenix? Can someone comment how the twinned data should be refined.

---

comment on Temperature factor variance for validation of structure

---

(hehe)

---

Hopefully, I can edit and upload the videos until next week.

---

Do you plan to implement a specific software for sugar conformation validation in Phenix similar to Privateer?

---

Can you comment a bit on the autobuild refinement feature? I am currently attempting to use it to rebuild flexible regions of my protein that don't always agree with my MR search model. In autobuild, what is the difference between the experimental map and initial map? Which map is best for rebuilding flexible regions that are too big or tricky to fix by hand on coot? Lastly, how do you preserve the numbering on your initial model when using autobuild? - I find that my model gets renumbered after the program works through the flexible regions that might have scantier density.

---

Is there anything different you need to do for the ‘final’ refinement when you

---

A particular problem that I recently encountered is how to resolve strong residual peaks (>5 sig) in the difference map after fitting a disulfide bridge between two cysteines. The geometry for the SS is okay to the eye and well fitted in density but there are still residual densities in the difference map. I just heard from someone that many times Zn2+ could be mistaken as -SS-. I am asking the experts here how often do you encounter this, if at all?

---

Re ready to upload

---

@Lucia Yes

---

Sorry, let me type my question again, is there anything different you need to do for the ‘final’ refinement when you’re ready to deposit?

---

@Jiang: It could be radiation damage as well

---

@Eta To deposit, you should use the mmCIF from refinement in a the PDB model deposition tool to add the sequence and then send the mmCIF to the PDB OneDep

---

@ Jiang: see for example 10.1073/pnas.97.2.623

---

thanks

---

Thank you

---

@Eta You should use a very recent version of Phenix to get the smoothest process

---

As completeness of dealing with all density is beneficial to the model that ends up as coordinates in the pdb file, are there thoughts on implementing the Squeeze procedure that exists for small molecules ? I realize that this has the potential to open cans and cans of worms but if done properly it may help. There are lots of structure with large unmodeled positive density in the data base.

---

When we see after the phenix.refine the increase in R factors does it mean that we used inappropriate refine strategy or we are overfitting data ?

---

I am at the end of refinement with R work 0.1554 and R free 0.1839

---

1 more question about ligand occupancy - If a region that indicates a bond between ligands shows 2Fo-Fc and red Fo-Fc density can that indicate a partial occupancy of that bond? Can a disappearance of the Fo-Fc density at lower occupancy further indicate that?

---

@Nigel, thanks for confirming that you will include a software for sugar validation in Phenix suite. As far as I know Privateer does not work with oligosaccharides made only by furanose units (at least for modified sugar which are not previously present in the PDB list of ligands). Will your software be able to overcome this issue? Do you know if another software can help me in the meanwhile?

---

It is a small molecule refinement technique.

---

Unmodelled (non-atomic) density is added back to the "model"

---

I found this reference for PLATON SQUEEZE (https://doi.org/10.1107/S2053229614024929)

---

this are starting R free is 0.1898

---

final

---

Alright, thanks! Will share the file to phenixBB to request more input.

---

How do you decide on occupancy, if you refine the same sugar in two conformations (4C1 and 1S3 sugar conformations). Do you experiment with numbers, say 0.0 and 0.5, and then 0.3 and 0.7? How do you decide on the correct ratio?

---

Thanks Paul and Billy, this is what I meant.

---

ok thank you

---

http://www.glycosciences.de/tools/pdb-care/

---

I typically use coot to make cif files for ligands. In case of Zn—his bonds or Zn—water or Zn—Asp and Zn-Glu does one need to restrain lengths and angles at ~2A res?

---

Many thanks

---

Many thanks for a wonderful workshop :)

---

Many thanks for a wonderful workshop :)

---

Thanks!

---

Thanks for this useful workshop!

---

Thanks!

---

thank you

---

thanks for the excellent workshop! keep it up

---

Thanks you. Good night :-)

---

Many thanks for the session. it was great.

---

Thanks a lot for the great workshop!

---

we can save it individually

---

Thank you for an excellent webinars.