Re: [COOT] German- McClure Alpha

[Paul Emsley]

On 13/10/2020 15:40, Sorbhi Rathore wrote:

— The RNA segment that I am trying to fit is ~130 bp.



That should be fine. A bit slow maybe.


— If I try to fit it without restraints, the RNA helix doesn't go into 
the density.



Maybe try fitting to a blurred map?


When I drag or remove the restraints it collapses (base pairing, 
planarity is destroyed.)



(i) Too fast or too extensive drag, try 10 micro-yanks instead of one 
big yank


(ii) alpha too high

(iii) map to model weight too high


But that will be a thing of the past in 0.9.1.


Paul.



To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1

This message was issued to members of www.jiscmail.ac.uk/COOT, a mailing list 
hosted by www.jiscmail.ac.uk, terms & conditions are available at 
https://www.jiscmail.ac.uk/policyandsecurity/

Re: [COOT] German- McClure Alpha

[Paul Emsley]

On 13/10/2020 17:36, Frank von Delft wrote:

That video is rather dramatic...



:-)



did it do that on its own, or was there dragging-by-mouse going on?



"No, she went of her own accord."


(no dragging)


The jelly/morphiness seems to allow a greater radius of convergence than 
rigid-body refinement.



The tools that I've been developing for 0.9 have been predicated on 
having a reasonable guess at the fold - mainchain only will do. If you 
have that [1], then you can place and build a model pretty 
lickety-split. (I mean for proteins, not RNA.)





The "youtube tutorial" - is that linked on the coot page?  I looked 
but couldn't obviously find it.  (That's the mrc page, 
https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/ 
)





The two relevant tutorials are the SBGrid one

https://www.youtube.com/watch?v=hhB8qUBBnJQ

and the RNA one

https://www.youtube.com/watch?v=Zzim3SXnDSk

presumably Sorbhi meant the latter.

Also related is Ana's CCP4 presentation:

https://www.youtube.com/watch?v=HJr-hJEF8oA


These are all a bit out of date as I've moved the GUI around and 
improved the refinement. I intend to do a modern version - that includes 
demonstrating proportional editing. I need to get 0.9.1 released first 
though.



Paul.


[1] as opposed the output of buccaneer or map_to_model, I mean.





To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1

This message was issued to members of www.jiscmail.ac.uk/COOT, a mailing list 
hosted by www.jiscmail.ac.uk, terms & conditions are available at 
https://www.jiscmail.ac.uk/policyandsecurity/

Re: [COOT] German- McClure Alpha

[Frank von Delft]

That video is rather dramatic...  did it do that on its own, or was 
there dragging-by-mouse going on?


The "youtube tutorial" - is that linked on the coot page?  I looked but 
couldn't obviously find it.  (That's the mrc page, 
https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/ 
)



On 12/10/2020 14:28, Robert Nicholls wrote:

Hi Sorbhi,

The Geman-McClure weight (alpha) controls robustness to outliers. As 
this value is raised, more restraints are treated as outliers and 
down-weighted. This will allow local conformational differences where 
there are dissimilarities between target and reference structures.


So if you really really want the restraints to be used (e.g. if you're 
trying to force something to be more helical) then you can use a lower 
alpha. Though if there may be some real differences between the 
restraints and the true structure - e.g. if you're using restraints 
based on a homologue - then you might want to use as slightly higher 
alpha so that regions displaying larger discrepancies from the 
restraint targets will not be influenced by the restraints. If you're 
using self-restraints then it's a balancing act - you want the 
structure to be stabilised, but you also want to allow it to deform 
enough to allow local structure to find the correct place in the map.


In Refmac values around 0.01-0.02 are typical. In Coot it varies - 
potential use cases during model building can be more varied than in 
refinement. The overall strength of the restraint weights also plays a 
big part, so it's a case of trial and error. Maybe Paul can advise 
about his preferred strategy? Fortunately in Coot it's easy to try 
different values, and the real time visual feedback of refinement 
performance should make it intuitive to see whether the alpha should 
be increased or decreased, especially when looking at more difficult 
regions.


Paul's got some videos of this in action, e.g.: 
https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/files/movies/refine-example-nov.ogv


Regards,
Rob


On 12 Oct 2020, at 13:44, Sorbhi Rathore > wrote:


Dear all,

Could anyone please suggest the best way to decide which German- 
McClure Alpha value to use during refinement (both at a higher 
resolution and lower resolution areas for a cryo-EM map)?

I am using coot 0.9.1 pre EL(ccpem).

Best wishes,
Sorbhi



To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1






To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1






To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1

This message was issued to members of www.jiscmail.ac.uk/COOT, a mailing list 
hosted by www.jiscmail.ac.uk, terms & conditions are available at 
https://www.jiscmail.ac.uk/policyandsecurity/

Re: [COOT] German- McClure Alpha

[Sorbhi Rathore]

— The RNA segment that I am trying to fit is ~130 bp.
— If I try to fit it without restraints, the RNA helix doesn't go into the 
density. When I drag or remove the restraints it collapses (base pairing, 
planarity is destroyed.)
— Even if I reduce the map weight it has same issues. (I have tried different 
map weights, but I did not see much improvement)
— This stretch of the map varies in resolution from ~4-6Å.


Best wishes,
Sorbhi

> On Oct 13, 2020, at 4:12 PM, Paul Emsley  wrote:
> 
> 
> 
> On 13/10/2020 14:59, Sorbhi Rathore wrote:
>> Hi Paul and Rob,
>> 
>> Many thanks for your prompt reply. 
>> 
>> One more question. Would you use the same settings for fitting RNA?
> 
> my thinking would be the same.
> 
> 
> 
>> (I have tried the exact settings shown in your YouTube video tutorial. 
>> However, in the low resolution regions of the map it is extremely tricky to 
>> move the RNA chain, especially when there is a conformational change)
> 
> Can you explain this in more detail? Why is it tricky? How many bp? If you 
> refined your fragment without GM restraints and without dragging on an atom, 
> what would happen? What happens when you reduce the map weight? Fitting RNA 
> with a conformational change is one of the things that Coot now does well (if 
> I may be so bold).
> 
> 
> 
> Paul
> 
> 
> 
> 
>> 
>> 
>>> On Oct 12, 2020, at 8:11 PM, Paul Emsley >> > wrote:
>>> 
>>> 
>>> 
>>> On 12/10/2020 13:44, Sorbhi Rathore wrote:
 Dear all,
 
 Could anyone please suggest the best way to decide which German- McClure 
 Alpha value to use during refinement (both at a higher resolution and 
 lower resolution areas for a cryo-EM map)?
 I am using coot 0.9.1 pre EL(ccpem).
 
>>> 
>>> A carefully-worded question.
>>> 
>>> For cases, where the model doesn't initially fit the map:
>>> 
>>> In 0.9, there was the danger or tearing the protein apart with too much 
>>> yank and too much alpha (low weight). That's why I used to be conservative 
>>> - using an alpha of 0.03 or so (occasionally 0.1 or 0.001).
>>> 
>>> With proportional editing in 0.9.1, I find I yank less (instead, using 
>>> several micro yanks) and because the shift is distributed over a far larger 
>>> range and is ameliorated by distance from the yanked atom, I find that 
>>> protein either doesn't tear or usually re-anneals  very pleasingly even 
>>> with a high alpha.
>>> 
>>> Proportional editing radius is changed by Ctrl-middle-mouse scroll.
>>> 
>>> For cases where the model more or less fits the map, I use 1 or (.3 or 10 
>>> (if I want to practically turn them off without deleting them))
>>> 
>>> Paul.
>>> 
>>> 
>>> 
>> 




To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1

This message was issued to members of www.jiscmail.ac.uk/COOT, a mailing list 
hosted by www.jiscmail.ac.uk, terms & conditions are available at 
https://www.jiscmail.ac.uk/policyandsecurity/

Re: [COOT] German- McClure Alpha

[Paul Emsley]

On 13/10/2020 14:59, Sorbhi Rathore wrote:

Hi Paul and Rob,

Many thanks for your prompt reply.

One more question. Would you use the same settings for fitting RNA?



my thinking would be the same.


(I have tried the exact settings shown in your YouTube video tutorial. 
However, in the low resolution regions of the map it is extremely 
tricky to move the RNA chain, especially when there is a 
conformational change)



Can you explain this in more detail? Why is it tricky? How many bp? If 
you refined your fragment without GM restraints and without dragging on 
an atom, what would happen? What happens when you reduce the map weight? 
Fitting RNA with a conformational change is one of the things that Coot 
now does well (if I may be so bold).



Paul






On Oct 12, 2020, at 8:11 PM, Paul Emsley > wrote:



On 12/10/2020 13:44, Sorbhi Rathore wrote:

Dear all,

Could anyone please suggest the best way to decide which German- 
McClure Alpha value to use during refinement (both at a higher 
resolution and lower resolution areas for a cryo-EM map)?

I am using coot 0.9.1 pre EL(ccpem).



A carefully-worded question.

For cases, where the model doesn't initially fit the map:

In 0.9, there was the danger or tearing the protein apart with too 
much yank and too much alpha (low weight). That's why I used to be 
conservative - using an alpha of 0.03 or so (occasionally 0.1 or 0.001).


With proportional editing in 0.9.1, I find I yank less (instead, 
using several micro yanks) and because the shift is distributed over 
a far larger range and is ameliorated by distance from the yanked 
atom, I find that protein either doesn't tear or usually re-anneals  
very pleasingly even with a high alpha.


Proportional editing radius is changed by Ctrl-middle-mouse scroll.

For cases where the model more or less fits the map, I use 1 or (.3 
or 10 (if I want to practically turn them off without deleting them))


Paul.








To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1

This message was issued to members of www.jiscmail.ac.uk/COOT, a mailing list 
hosted by www.jiscmail.ac.uk, terms & conditions are available at 
https://www.jiscmail.ac.uk/policyandsecurity/

Re: [COOT] German- McClure Alpha

[Sorbhi Rathore]

Hi Paul and Rob,

Many thanks for your prompt reply. 

One more question. Would you use the same settings for fitting RNA?
(I have tried the exact settings shown in your YouTube video tutorial. However, 
in the low resolution regions of the map it is extremely tricky to move the RNA 
chain, especially when there is a conformational change)

Best wishes,
Sorbhi


> On Oct 12, 2020, at 8:11 PM, Paul Emsley  wrote:
> 
> 
> 
> On 12/10/2020 13:44, Sorbhi Rathore wrote:
>> Dear all,
>> 
>> Could anyone please suggest the best way to decide which German- McClure 
>> Alpha value to use during refinement (both at a higher resolution and lower 
>> resolution areas for a cryo-EM map)?
>> I am using coot 0.9.1 pre EL(ccpem).
>> 
> 
> A carefully-worded question.
> 
> For cases, where the model doesn't initially fit the map:
> 
> In 0.9, there was the danger or tearing the protein apart with too much yank 
> and too much alpha (low weight). That's why I used to be conservative - using 
> an alpha of 0.03 or so (occasionally 0.1 or 0.001).
> 
> With proportional editing in 0.9.1, I find I yank less (instead, using 
> several micro yanks) and because the shift is distributed over a far larger 
> range and is ameliorated by distance from the yanked atom, I find that 
> protein either doesn't tear or usually re-anneals  very pleasingly even with 
> a high alpha.
> 
> Proportional editing radius is changed by Ctrl-middle-mouse scroll.
> 
> For cases where the model more or less fits the map, I use 1 or (.3 or 10 (if 
> I want to practically turn them off without deleting them))
> 
> Paul.
> 
> 
> 




To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1

This message was issued to members of www.jiscmail.ac.uk/COOT, a mailing list 
hosted by www.jiscmail.ac.uk, terms & conditions are available at 
https://www.jiscmail.ac.uk/policyandsecurity/

Re: [COOT] German- McClure Alpha

[Paul Emsley]

On 12/10/2020 19:11, Paul Emsley wrote:



On 12/10/2020 13:44, Sorbhi Rathore wrote:

Dear all,

Could anyone please suggest the best way to decide which German- 
McClure Alpha value to use during refinement (both at a higher 
resolution and lower resolution areas for a cryo-EM map)?

I am using coot 0.9.1 pre EL(ccpem).



A carefully-worded question.

For cases, where the model doesn't initially fit the map:

In 0.9, there was the danger or tearing the protein apart with too 
much yank and too much alpha (low weight). That's why I used to be 
conservative - using an alpha of 0.03 or so (occasionally 0.1 or 0.001).




I meant 0.1 or 0.01


Also the above presumes that you have pressed the "Estimate" button when 
selecting the map weight. (Actually I often divide that number by 2 or 
so - it doesn't make much difference but tightens up the final geometry 
a bit).


As Rob says, it's a bit "trial an error" (where the trial lasts a few 
seconds) - really I'm asking "is the protein moving like I want it to 
move?" One cannot interactively change alpha (that might be a useful 
thing to add to the refinement).



Paul





To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1

This message was issued to members of www.jiscmail.ac.uk/COOT, a mailing list 
hosted by www.jiscmail.ac.uk, terms & conditions are available at 
https://www.jiscmail.ac.uk/policyandsecurity/

Re: [COOT] German- McClure Alpha

[Paul Emsley]

On 12/10/2020 13:44, Sorbhi Rathore wrote:

Dear all,

Could anyone please suggest the best way to decide which German- 
McClure Alpha value to use during refinement (both at a higher 
resolution and lower resolution areas for a cryo-EM map)?

I am using coot 0.9.1 pre EL(ccpem).



A carefully-worded question.

For cases, where the model doesn't initially fit the map:

In 0.9, there was the danger or tearing the protein apart with too much 
yank and too much alpha (low weight). That's why I used to be 
conservative - using an alpha of 0.03 or so (occasionally 0.1 or 0.001).


With proportional editing in 0.9.1, I find I yank less (instead, using 
several micro yanks) and because the shift is distributed over a far 
larger range and is ameliorated by distance from the yanked atom, I find 
that protein either doesn't tear or usually re-anneals  very pleasingly 
even with a high alpha.


Proportional editing radius is changed by Ctrl-middle-mouse scroll.

For cases where the model more or less fits the map, I use 1 or (.3 or 
10 (if I want to practically turn them off without deleting them))


Paul.





To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1

This message was issued to members of www.jiscmail.ac.uk/COOT, a mailing list 
hosted by www.jiscmail.ac.uk, terms & conditions are available at 
https://www.jiscmail.ac.uk/policyandsecurity/

Re: [COOT] German- McClure Alpha

[Robert Nicholls]

Hi Sorbhi,

The Geman-McClure weight (alpha) controls robustness to outliers. As this value 
is raised, more restraints are treated as outliers and down-weighted. This will 
allow local conformational differences where there are dissimilarities between 
target and reference structures.

So if you really really want the restraints to be used (e.g. if you're trying 
to force something to be more helical) then you can use a lower alpha. Though 
if there may be some real differences between the restraints and the true 
structure - e.g. if you're using restraints based on a homologue - then you 
might want to use as slightly higher alpha so that regions displaying larger 
discrepancies from the restraint targets will not be influenced by the 
restraints. If you're using self-restraints then it's a balancing act - you 
want the structure to be stabilised, but you also want to allow it to deform 
enough to allow local structure to find the correct place in the map.

In Refmac values around 0.01-0.02 are typical. In Coot it varies - potential 
use cases during model building can be more varied than in refinement. The 
overall strength of the restraint weights also plays a big part, so it's a case 
of trial and error. Maybe Paul can advise about his preferred strategy? 
Fortunately in Coot it's easy to try different values, and the real time visual 
feedback of refinement performance should make it intuitive to see whether the 
alpha should be increased or decreased, especially when looking at more 
difficult regions.

Paul's got some videos of this in action, e.g.: 
https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/files/movies/refine-example-nov.ogv

Regards,
Rob


> On 12 Oct 2020, at 13:44, Sorbhi Rathore  wrote:
> 
> Dear all,
> 
> Could anyone please suggest the best way to decide which German- McClure 
> Alpha value to use during refinement (both at a higher resolution and lower 
> resolution areas for a cryo-EM map)?
> I am using coot 0.9.1 pre EL(ccpem).
> 
> Best wishes,
> Sorbhi
> 
> To unsubscribe from the COOT list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1 
> 



To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1

This message was issued to members of www.jiscmail.ac.uk/COOT, a mailing list 
hosted by www.jiscmail.ac.uk, terms & conditions are available at 
https://www.jiscmail.ac.uk/policyandsecurity/

[COOT] German- McClure Alpha

[Sorbhi Rathore]

Dear all,

Could anyone please suggest the best way to decide which German- McClure Alpha 
value to use during refinement (both at a higher resolution and lower 
resolution areas for a cryo-EM map)?
I am using coot 0.9.1 pre EL(ccpem).

Best wishes,
Sorbhi


To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT=1

This message was issued to members of www.jiscmail.ac.uk/COOT, a mailing list 
hosted by www.jiscmail.ac.uk, terms & conditions are available at 
https://www.jiscmail.ac.uk/policyandsecurity/