Zoom Transcript

---

https://www.youtube.com/c/phenixtutorials

---

thanx Claudia

---

And the paper :) https://journals.iucr.org/d/issues/2019/10/00/di5033/

---

Can you please make it full screen?

---

Do you need CCP4 to install PHENIX on Windows?

---

OK thanks

---

In “VIEW OPTIONS” you can make the view of Paul’s screen bigger if you want

---

Does the current version of Phenix support the new MacOS?

---

Yes

---

Yes, I have installed on macOS 11 Big Sur. I’m still testing it, but what I have tested is working. Please let us know if you find an issue!

---

Will you generally support the new Apple Silicin M1 processor(s) (it's an ARM as far as I understood)….

---

The current installers should run, but will run in translation mode (Rosetta 2). We are getting one of the new Apple Silicon machines for more testing.

---

We will be looking into building universal installers that work natively on both architectures.

---

Do u plan to integrate data processing programs in the phenix package in future

---

Currently, we distribute DIALS, but there is no GUI associated with it. It is run via the command-line.

---

If you report a bug, please make sure you’ve given your real email address in the preferences! If you don’t, we can’t ask questions or work with you to solve the p;

---

…problem

---

We are shipping DIALS 2.2 (https://dials.github.io).

---

No

---

no

---

no

---

Yes

---

no

---

yes

---

Use the icons in the participants list

---

Stop the counting :)

---

no

---

That one button :)

---

no

---

Something like this will appear, and you just have to click yes or no

---

can you share start file for this workshop?

---

to use hands on practice on phenix

---

Do you have Phenix 1.18.2 installed?

---

Stop your video for a while?

---

yes

---

The tutorial dataset can be created by clicking on “New project".

---

Then “Set up tutorial data”

---

The phenix interface has tutorial data integrated that you can use for the project.

---

P9-sad is the first Experimental Phasing dataset.

---

thanks

---

yes

---

yes

---

can you make your screen biger, its hard to see

---

tnx

---

there, you can zoom in on the shared screen

---

Say, to 150% for example, and will be much larger :)

---

How do you (we) know which fields are mandatory and need to be filled and which ones can be skipped?

---

what about the Se white line?

---

how will we plan experiments in case of soaking

---

what is minimum I/siga which is sufficient to solve the structure?

---

Soaking: choose the inputs according to your anomalous scatterer: type of anomalous scatterer, number of sites, wavelength etc

---

Soaking: typical number of sites: 1 site per 10 or 20 protein atoms.

---

Is there a similar planner for SIR/MIR?

---

Another way of checking which fields are necessary is to go to the online documentation, the instructions for running a program via the command line will often say which parameters are mandatory and which are optional

---

How would you reach na I/sigI of 98 in practice?

---

By the way, for those of you zooming, If you need to recenter the zoomed region to see better the phenix interface you just need to put the cursor there, click and move it till is like you want it

---

Radiation damage problems?

---

It is surprising because I was always told that for S-SAD you need data better than 2A. That doesn’t seem to be true..

---

Sulfur

---

He wrote S-SAD not Se-SAD

---

how do we find whether the Met’s are ordered or not….

---

How strong the isormorphous signal that we normally get to do SIR or MIR successfully? Compare with SAD, does SIR/MIR require higher data accuracy?

---

what about disulphide bridges

---

What if we have large number of Sulphurs (disulfide bonds) ie 56 sulphur in 800 residues. According to formular No. of sites should be less?

---

thank you a lot!

---

In practice, I found that high multiplicity helps improve the odds of successful phasing. Is this mostly due to an improved I/sigma or is there more/other contribution? one has to strike a balance between collecting more data to improve I/sigma and not letting radiation damage ruin the prospects of solving the structure. In difficult cases, I just collect until I see clear signs of crystal decay and truncate my data later in various ways to find the proper cutoff by looking at CCano. Is there a better, real-time metric during data collection to guide our experiment on the fly?

---

Many thanks

---

How do you calculate the sites?

---

on the subject of systematic errors, what are the impacts of ice rings? it is hard to preserve xtals in the summer without ice.

---

I have to go. Where do we find the recording afterwards?

---

In the phenix website

---

https://www.phenix-online.org/user-workshops.html

---

It’ll take a couple of days to upload the videos and the chat transcript.

---

Just to clarify... when we say I/sigma(I), are we talking mean(I)/sd(I) ?

---

How does the “add to group” work? thanks

---

Yes, I/sigma(I) here refers to the mean. So you can get sufficiently accurate data by merging data from a number of isomorphous crystals, and it’s the final accuracy that matters.

---

Thanks Randy. I guess you need to collect from at least 4-5 crystals. I/sigma(I) of 98 is very high...

---

What is meant by NCS copies?

---

Could you please explain a bit more what “Data accuracy” means? What are the things which we should be looking at?

---

what are those values or parameters we need to see and conform that we got the sub-structure solution

---

If you run Auto_sol and the FOM is around .209 should you kill it or keep it running?

---

I suppose data precision is being used as proxy for data accuracy here?

---

what is systematic error and how it is different than common error? what is the link of systematic error and SAD

---

Is it possible to solve the structure with a low FOM?

---

And also are there more chances to solve the structure using MAD than SAD?

---

connectivity and map skew

---

coz of less noice in ryt side

---

There are continuous density

---

solvent is flat on right

---

connectivity

---

Contiguous density

---

Clear difference between solvente and non-solvente regions.

---

error is less

---

continuity

---

continuous main chains, side chains visible

---

In a right map

---

clear separation of protein and solvent density

---

Continuous density, shape of density, can see aas

---

Solvent region obvious

---

the density is continuing

---

connected features

---

Clear solvent boundary

---

protein density easily discernable

---

Can autosol also accomodate a molecular replacement solution (pdb file) into its phase determination in cases where you're just using it to determine some heavy atom binding sites?

---

How to use Autosol if the protein has significant number of non-natural residues

---

say 3 nonnatural every 7 residues

---

what are those values or parameters we need to see and conform that we got the sub-structure solution in phenix log

---

yes

---

Thanx

---

Following up on the latest question, how do we know that the dataset we have collected has sufficient anomalous signal? DANO/sigDANO? CC(ano)?

---

Thanks

---

Thanks

---

thanks for the great lecture

---

in the P9 example, a cis peptide in front of non-proline was built at residue 19. How stringently are the geometric restraints used at this stage? Can one enforce a tighter restraint?

---

So, we have a problem with a beamline scientist where we collect data, who insists on trimming the data way too much and he refuses to deliver the raw data. Is there any way around this to solve the structure?

---

The “beta-blip” is the first tutorial dataset under “Molecular Replacement”.

---

how do we know ASU number?

---

Can % identity simply be found through aligning the sequence files?

---

Could we put 2 protein sequence into one file?

---

Is simple one-component interface suitable for solving twinned crystals if the crystal is almost identical to its PDB model, or is this irrelevant at this point?

---

You can check the probability associated to the Matthews Coefficient that is plot by phaser once running

---

Monika from Matthews co efficient

---

If we use half model for search, do we use identity 50%?

---

Is there is use for searching for one component at a time rather than both at once?

---

ok then follow up question, how to calculate Matthews co efficient

---

For a homomer, would you have a sequence file with only one sequence or the entire homomer? Also, would you make an ensemble for every component of the homomer?

---

We will be having time for questions later on during the general discussion time so if we do not reply to all of the questions in the chat, we will have a look to them later

---

Jacob the sequence identity that you get from the alignment is good enough for Phaser. There are more sophisticated ways and I will show one later.

---

Ensembling also will be discussed now :)

---

Alternatively, could you search for Blip run MR then use it as a partial solution for MR with beta?

---

Thank you massimosammito :)

---

Monika, Phaser computes the Matthews Coefficient and plot the probability disitribution in the Graph tabs of the results

---

ok, so does that mean we don't need to know the number before running the phaser?

---

You can also get the Matthews Coefficient from xtriage

---

If you see that your initial assumption does not match with the best probability output from phaser, is worth to stop the run and restart with the correct number. Although this parameter might be difficult to predict and it is worth to try several runs with different possibilities in borderline cases

---

What is the minimum recommended %sequence identity number for Phaser to work?

---

How small can be the search model

---

Maria I will say 30% indicates at least homology

---

Thank you.

---

As Claudia is explaining, the eLLG is a complicated combination of sequence identity, model completeness and data resolution, so none of these in isolation tells you whether MR will work.

---

The model can be small but it should at least be 10% of the total scattering in the crystal and only if the resolution is very good. eLLG is the key parameter to look at to guide you in choosing the model

---

Some models with less than 15% sequence identity can work, especially when used as trimmed ensembles (see Massimo’s presentation in a few minutes), but the success rate at this end of the spectrum will be low!

---

phenix dot tools are super handy :)

---

https://toolkit.tuebingen.mpg.de/tools/hhpred

---

Really relevant this point Massimo is making: the RMSD for the ensemble is already set by Ensembler! No need to type a seq id

---

Eta Isiorho, phaser has sophisticated algorithms to choose the correct order in which the two molecules (beta and blip) should be searched, based on the eLLG. Unless, you have a serious reason, I will just configure the run with both molecules and let Phaser choose the order automatically

---

Randy already pointed out at the very beginning about how powerful is the fact that after you get a component place correctly you increase the signal a lot. Therefore, you want to get your best (larger, less RMSD) model placed first

---

So that your background is better before placing the next one

---

And usually that decision is taken automatically by Phaser

---

In the Graph tab of the result of a Phaser run in Phenix GUI you can find a nice plot for the intensity distribution that might worth to look at to check if twinning might be occurring as Airlie was mentioning

---

https://youtu.be/QlWahDGBTbE

---

That is the tutorial Airlie is referring to

---

Q1) When I run Phaser, I get low LLGs and the run is taking forever. What does it mean? What can I do?

---

When I run phaser, I get high LLGs but a poor map and no structure solution

---

Eta, look at the TFZ scores, and is it a single solution?

---

I want to know when we do the MR for a complex, could we put the two protein sequences into one file instead of two separate file?

---

Mathews Coefficient gives more than 5 probable oligomeric state. What does it mean? Should I try to run phaser with all oligomeric state

---

Q2) In a multicomponent search, which starts out well with a clear signal, at a certain point for a new component Ido not get an increase in the LLG and zscores are low. What does this mean?

---

In case of twinning, how to determine the real space group? (P1 in case of the orange juice boxes)

---

How does phaser deal with a anistropic scaled dataset ( from StarAniso) ?

---

The sequences are used to calculate the molecular weight. You have to specify what is the full content of the crystal. So put a fasta file with the whole thing.

---

I looked at the TFZ, it’s ~7 but 6 MR solutions

---

You should not use anisotropically truncated files from staranlso in phaser, as Airlie mentioned all the data should be available to phaser to do its best correction

---

LC it is better not process data with Staraniso for using it in Phaser, as we have sophisticated way to work with anisotropy

---

A good solution is indicated by a TZ> 8

---

What would be the best strategy for highly helical structures? (In my case a collagen triple helix, with potentially pseudotranslational tNCS due to the internal rotation axis)

---

Phaser can use StarAniso data, but it will do a better job of correcting the data for anisotropy and tNCS if it gets a complete set of data, so we really prefer data that have not been through StarAniso.

---

Ok, Thanks.

---

For a 11-mer of identical subunits, should the sequence file contain the 11 sequences or 1 sequence?

---

After a phaser run, with a high LLG (1455) and a high TFZ (39.9), I get these warnings: "the top solution from a FTF did not pack" and "the top solution from a TF rescoring did not pack". What does it mean? Do I have to worry about them?

---

Imagine, hypothetically, you have a two-subunit complex and you have a MR solution for one of them. How big should it be relatively to the "unknown" subunit in order to see the latter in electron density without the need for further phasing? Hypothetically, let's say data to 2.7A. And also hypothetically, P41212...

---

Q3) What does it mean when you get multiple solutions with good LLG and TFZ scores, but not a unique solution. Should I just take the top one?

---

Is there are feature in phenix where you can modify the .mtz file, for instance cut of a certain wedge from the reciprocal lattice and exclude it from the data?

---

Q4) Should I split my model into different domains? If yes, how?

---

I’m saving the chat :) so we can keep unanswered questions for the general discussion after our time slot in which all the speakers will be replying to both experimental phasing and molecular replacement related questions.

---

SCEDS

---

https://www.phenix-online.org/documentation/reference/sceds.html

---

When I ran the MR, I got a high LLG and TFZ over 8, but the map didn’t fit well, so was this solution right?

---

Q6) I have a convincing MR solution but my Rfree is too high

---

Silvia, the warning messages you refer to are telling that the translation with the highest LLG is not passing a packing test. This might or not be problematic depending on whether you are then getting solutions that are sensible or not. Is your structure a coiled coil or any other kind of highly repetitive structure? Do you get a single solution or multiple ones?

---

SAXS, EM, DLS, SEC suggest hexamer, but diffraction data suggesting hexamer with 70% solvent, 8mers 60%, 10mers 50% and 12mers 40%. and the final MR solution is 8mers, but when I generate symmetry mates in PyMol, shows dodecamer. Further unit cell ha 18 symmetries mates with dodecamer seen. hexamer-dodecamer equilibrium is reported, with dodecamer more active enzymatically.

---

I tried to use the twin law that I got from Xtriage to do refinement, but it didn’t work. Is there any better way to fix twinning problem?

---

LT is the solution a unique solution in Phaser? Are you sure you have find all the copies in the ASU? Or maybe there is a space for something more, or something else? When you say “the map does not fit well” if you referent to the rfree there are several reasons why that could happen including twinning.

---

Jose Artur Brito, you can compute eLLG for the first component search and compare with the ones you already get. This would not tell you whether you will see density or not, though. You might need to do some density modification and map improvement.

---

Jose, yours is a very difficult question to answer. I would say that the larger the better, although it really depends on the quality of the placed model and the bfactors of the missing one. I would say, to try density modification after placing the only model that you have to see if the density in the second region improve enough for tracing with auto build.

---

can one do something at this stage of structure solvinf about twinned data that display perfect twin in statistics, but also a space group in which no twin laws exist, such as space group 23/24?

---

is a 2 Ang model better than a 3 Ang of the same one?

---

Jan Gebauer do you mean a coiled-coil structure?

---

Sorry jan I read again, you mean collagen, but yes it suffers from the same problems than coiled coils

---

It's not a coiled coil (in terms of alpha helical) but very similar as it is a collagen triple helix (three polyproline type II helices...)

---

@Avradip all else being equal, yes. In practice, though, it can depend. If the domain you're interested is well resolved in the 3A structure but poorly resolved in the 2A one, the 3A may be better. If the two models are in quite different conformations, it would pay to try both (or just the common core of the two)

---

Thank you!

---

if I scale my data with StarAniso (anisotropic cut-off, I should use the full data (uncut anisotropic) for phaser and then the anisotropic cut data for refining?

---

yes LC

---

ok thanks

---

Yes, as a general advice (summary tips for the ‘’good’’ model search). Do try to find the models using HHPRED, and if there are multiple models available, after appropiatedly sculpting them, use them in phaser runs (both as single models but also with ensembles as Massimo showed)

---

What do you think/advise on using rmsd instead of identity % when the model if low id % but you pretty sure the structure is highly similar?

---

LC at the end what does it matter is the RMSD so that it means that if you know that number it will be a better estimation of the error

---

LC phaser will refine the value later on anyway

---

Thanks to all the participants

---

thanks to you

---

thank you.