Re: [ccp4bb] To win or not twinning?

[John R Helliwell]

Dear Napo,
I imagine this article, although rather wide ranging in its content, will
be of assistance:-
http://journals.iucr.org/d/issues/2005/06/00/ic5050/
More generally you might consult Chayen, Helliwell and Snell
"Macromolecular Crystallisation and Crystal Perfection" Published by OUP
and IUCr. [Apologies to CCP4bb colleagues as this involves a commercial
product ie my own book.]
All best wishes,
John

On Wed, Nov 30, 2016 at 7:01 AM, Napoleao Fonseca Valadares <
n...@ifsc.usp.br> wrote:

> Dear CCP4ers,
>
> I'd like to kindly ask your advice. Sorry for the long e-mail.
>
> I have crystals of a 12.3 KDa protein that grow in hexagon-like patterns,
> link for the crystal image:
> http://fullonline.org/science/cryst01.jpg
>
> XDS, Phenix and Pointless always suggest that the data sets for these
> crystals belong to the space group P622. However, Phenix, Phaser and
> Pointless indicate that twinning is present.
>
> "Bad looking" diffraction images, diffracted to 1.6 A (collected 6 months
> ago):
> http://fullonline.org/science/dataset1_image37.png
> http://fullonline.org/science/dataset1_image7.png
>
> Best data set, diffracted to 2.01 A (collected a month ago):
> http://fullonline.org/science/dataset2_image51.png
>
> The second data set present better looking images, a better XDS ISa value
> (around 24) and diffracted to 2.2 A. The "bad looking" data set diffracted
> to 1.6 A, but I decided to stop working with it (XDS ISa around 10).
>
> There is a template with 60% identity, I used XDS to try to process the
> data in all trigonal/hexagonal space groups from P3 to P6(3)22, and spend a
> lot of time trying molecular replacement procedures in Phaser and Morda,
> and refining the candidate solutions. Used Zanuda too, trying to figure out
> the space group (and read as much as possible in this CCP4 list looking for
> similar cases).
>
> Unit cells and typical MR results:
>
> P1: 56.430   77.718   77.673 119.99  89.99  89.97
> SOLU SET  RFZ=9.5 TFZ=* PAK=0 LLG=91 RF++ TFZ=18.2 PAK=0 LLG=311
> TFZ==18.4 (&
> TFZ==17.0) LLG+=(311 & 726) LLG=5751 TFZ==64.6 PAK=0 LLG=5751 TFZ==64.6
>
> P3: 77.675   77.675   56.409  90.00  90.00 120.00
> RFZ=5.5 TFZ=8.5 PAK=0 LLG=86 TFZ==9.4 LLG=2575 TFZ==46.5
>
> P6: 77.534   77.534   92.986  90.00  90.00 120.00
> RFZ=11.2 TFZ=16.2 PAK=2 LLG=437 TFZ==25.7 LLG=437 TFZ==25.7
>
> P6(3): 77.675   77.675   56.409  90.00  90.00 120.00
>RFZ=8.8 TFZ=11.5 PAK=0 LLG=66 TFZ==7.6 LLG=66 TFZ==7.6
>
> P622: 77.660   77.660   56.400  90.00  90.00 120.00
>   RFZ=4.8 TFZ=5.4 PAK=1 LLG=22 TFZ==4.7 LLG=24 TFZ==4.8
>
>
> In P1 (LLG=5751 TFZ==64.6) there are 12 molecules in the asymmetric unit,
> and in P3 (LLG=2575 TFZ==46.5) 4 molecules. Packing looks good, in P1 the
> ASU looks like two superposed hexagonal donuts formed by 6 molecules each.
> Refining in P1, without adding waters or TLS, yield r_work = 0.2964 and
> r_free = 0.3428, and it is hard to decrease these values.
> Refining in P3, I managed to get r_work = 0.2934 and r_free = 0.3399, but
> looks like it's not getting any better than this.
> Refining in P6 yields horrible r_free values (>0.50).
>
> Trying to refine MR solutions in any other other space groups yield Rfree
> 0.41 or more.
>
> If in the P3 space group I use the twin operator -k,-h,-l (estimated twin
> fraction of 0.490) suggested by Phenix Xtriage, the values miraculously go
> down to r_work = 0.1923 r_free = 0.2202, without waters (r_free without
> using a twin law = 0.3399). Adding 120 water molecules and doing some
> refining yields 0.1711 r_free = 0.1935 (the asymmetric unit contains 420
> residues and the resolution is 2.01 A).
> I've been reading about twinning refinement and how it can drop the
> Rvalues, but from what I understood if I use it improperly I may compromise
> the refinement quality.
>
> I would like advice on:
> 1 - References that can teach how to look at protein diffraction images
> and understand what I am seeing. The basics like recognizing bad data and
> what usually leads to deformity in the spots (for example, elliptical or
> duplicated) would be of great help.
>
> 2 - Should I look for other space groups? What else could be tried? Is
> this a case where a twin law should be used in the refinement? If yes, what
> can I do to confirm the need for a twin law in the refinement?
>
> Thank you all in advance.
> Regards,
>  Napo
>



-- 
Professor John R Helliwell DSc

Re: [ccp4bb] To win or not twinning?

[Eleanor Dodson]

Not sure about protein diffraction images - I would recommend going to any
course where Zbysek Dauter was teaching - he is the master..
And he lectured at severa;l CCP4 Study Weekends. It would be worth looking
up those. I think they are online at the CCP4 website or in Acta Cryst.

Eleanor

On 1 December 2016 at 20:24, Napoleao Fonseca Valadares <n...@ifsc.usp.br>
wrote:

> Hi,
>
> Thank you all for the answers.
>
> Eleanor, you nailed it, thank you! Your observation that the cell in P6
> had a different C axis led me to reprocess the data using a cell of the
> same size of the other space groups. I have commented that processing in P6
> before yielded r_free > 0.50, but now the r_free is 0.3232 without using a
> twin law, and using the twin law h,-h-k,-l which was recommended by Phenix
> Xtriage, yields r_work = 0.1674 and r_free = 0.1960. Thank you for the
> other suggestions and reference, I followed them.
>
> I reviewed all my analyses and only P6 had the wrong cell size. I don't
> know why.
>
> - P3 yields r_work = 0.1711 and r_free = 0.1935 with 120 water molecules
> (the asymmetric unit contains 420 residues and the resolution is 2.01 A).
> - P6 yields r_work = 0.1674 and r_free = 0.1960 with 94 water molecules
> (the asymmetric unit contains 210 residues and the resolution is 1.91 A).
>
> I guess I should stick with P6, since it has higher symmetry, right?
>
> Regardless of how I process the data, in all space groups, pointless
> always says it is sure that the space group P622. That is probably due to
> the twinning, that causes reflections to lay over each other and gives the
> appearance of higher symmetry.
>
> A question, does it makes sense to use TLS and a twin law to refine the
> data at this resolution (1.9 A)?
>
> Carlos, thank you for the references.
>
> Carlos Frazao, the PDB header generated by Phenix indicates that the twin
> fraction is still 0.490 after refinement in P3 and P6.
>
> Jacob Keller, I have tried all trigonal and hexagonal space groups, except
> for those I reported (P1, P3 and now P6), all others yield unrefinable
> solutions (R_free > 0.41). However, there are a lot of explanations for it,
> for example, the data might be processed using the wrong cell size, or the
> MR solution might just be wrong.
>
> Thank you all for the help!
>
> I'm still looking for references that can teach me how to look at protein
> diffraction images and understand what I am seeing. The basics like
> recognizing bad data and what usually leads to deformity in the spots (for
> example, elliptical or duplicated) would be of great help.
>
> Regards,
> Napo
>
>
> ------
>
> *De: *"Eleanor Dodson" <eleanor.dod...@york.ac.uk>
> *Para: *CCP4BB@JISCMAIL.AC.UK
> *Enviadas: *Quarta-feira, 30 de Novembro de 2016 14:10:16
> *Assunto: *Re: [ccp4bb] To win or not twinning?
>
>
> First - there is a useful document about possible twin laws in the CCP4
> documentation. It references the SHELX description of twinning for small
> molecules.
> http://www.ccp4.ac.uk/html/twinning.html
>
> It is possible in a trigonal system that you could have two twin operators
> but the twinning stats should lok odd in that case
>
> For a trigonal space group there are several options, but your technique
> seems to have correctly found SG P3.
> As Jacob Keller says - check POINTLESS to see if P322 etc are options?
> Or ask Zanudu..
>
> Just to comment - I think you must have the cell dimensions wrong for the
> P63 case? I cant see how the c axis can be longer?
>
>  I like the data processing POINTLESS/AIMLESS/ CTRUNCATE..
>
> POINTLESS lists all axial reflections so you can make an educated guess at
> screw axes (or let POINTLESS do it for you...)
> POINTLESS analyses each symmetry operator separately. So if the -k, -h, -l
> was slightly less convincing than others, AND the twin tests suggested
> twinning, you might expect that to be a twin operator and not a symmetry
> operator..
>
> ctruncate tests each possible twin operator and scores them. (And so does
> XTRIAGE of course..)
>
>
> On 30 November 2016 at 14:00, Keller, Jacob <kell...@janelia.hhmi.org>
> wrote:
>
>> What about spacegroups in PG 32, e.g., p3212?
>>
>>
>>
>> JPK
>>
>>
>>
>> *From:* CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] *On Behalf Of
>> *Napoleao Fonseca Valadares
>> *Sent:* Wednesday, November 30, 2016 2:01 AM
>> *To:* CCP4BB@JISCMAIL.AC.UK
>> *Subject:* [ccp4bb] To win or not twinning?
>>
>>
>>
>> Dear CCP4ers,
>>
>> I'd like to kindly ask your advice. Sorry for the long e-mail.
>>
>&

Re: [ccp4bb] To win or not twinning?

[Napoleao Fonseca Valadares]

Hi, 

Thank you all for the answers. 

Eleanor, you nailed it, thank you! Your observation that the cell in P6 had a 
different C axis led me to reprocess the data using a cell of the same size of 
the other space groups. I have commented that processing in P6 before yielded 
r_free > 0.50, but now the r_free is 0.3232 without using a twin law, and using 
the twin law h,-h-k,-l which was recommended by Phenix Xtriage, yields r_work = 
0.1674 and r_free = 0.1960. Thank you for the other suggestions and reference, 
I followed them. 

I reviewed all my analyses and only P6 had the wrong cell size. I don't know 
why. 

- P3 yields r_work = 0.1711 and r_free = 0.1935 with 120 water molecules (the 
asymmetric unit contains 420 residues and the resolution is 2.01 A). 
- P6 yields r_work = 0.1674 and r_free = 0.1960 with 94 water molecules (the 
asymmetric unit contains 210 residues and the resolution is 1.91 A). 

I guess I should stick with P6, since it has higher symmetry, right? 

Regardless of how I process the data, in all space groups, pointless always 
says it is sure that the space group P622. That is probably due to the 
twinning, that causes reflections to lay over each other and gives the 
appearance of higher symmetry. 

A question, does it makes sense to use TLS and a twin law to refine the data at 
this resolution (1.9 A)? 

Carlos, thank you for the references. 

Carlos Frazao, the PDB header generated by Phenix indicates that the twin 
fraction is still 0.490 after refinement in P3 and P6. 

Jacob Keller, I have tried all trigonal and hexagonal space groups, except for 
those I reported (P1, P3 and now P6), all others yield unrefinable solutions 
(R_free > 0.41). However, there are a lot of explanations for it, for example, 
the data might be processed using the wrong cell size, or the MR solution might 
just be wrong. 

Thank you all for the help! 

I'm still looking for references that can teach me how to look at protein 
diffraction images and understand what I am seeing. The basics like recognizing 
bad data and what usually leads to deformity in the spots (for example, 
elliptical or duplicated) would be of great help. 

Regards, 
Napo 

- Mensagem original -

> De: "Eleanor Dodson" <eleanor.dod...@york.ac.uk>
> Para: CCP4BB@JISCMAIL.AC.UK
> Enviadas: Quarta-feira, 30 de Novembro de 2016 14:10:16
> Assunto: Re: [ccp4bb] To win or not twinning?

> First - there is a useful document about possible twin laws in the
> CCP4 documentation. It references the SHELX description of twinning
> for small molecules.
> http://www.ccp4.ac.uk/html/twinning.html

> It is possible in a trigonal system that you could have two twin
> operators but the twinning stats should lok odd in that case

> For a trigonal space group there are several options, but your
> technique seems to have correctly found SG P3.

> As Jacob Keller says - check POINTLESS to see if P322 etc are
> options?

> Or ask Zanudu..

> Just to comment - I think you must have the cell dimensions wrong for
> the P63 case? I cant see how the c axis can be longer?

> I like the data processing POINTLESS/AIMLESS/ CTRUNCATE..

> POINTLESS lists all axial reflections so you can make an educated
> guess at screw axes (or let POINTLESS do it for you...)
> POINTLESS analyses each symmetry operator separately. So if the -k,
> -h, -l was slightly less convincing than others, AND the twin tests
> suggested twinning, you might expect that to be a twin operator and
> not a symmetry operator..

> ctruncate tests each possible twin operator and scores them. (And so
> does XTRIAGE of course..)

> On 30 November 2016 at 14:00, Keller, Jacob <
> kell...@janelia.hhmi.org > wrote:

> > What about spacegroups in PG 32, e.g., p3212?
> 

> > JPK
> 

> > From: CCP4 bulletin board [mailto: CCP4BB@JISCMAIL.AC.UK ] On
> > Behalf
> > Of Napoleao Fonseca Valadares
> 
> > Sent: Wednesday, November 30, 2016 2:01 AM
> 
> > To: CCP4BB@JISCMAIL.AC.UK
> 
> > Subject: [ccp4bb] To win or not twinning?
> 

> > Dear CCP4ers,
> 

> > I'd like to kindly ask your advice. Sorry for the long e-mail.
> 

> > I have crystals of a 12.3 KDa protein that grow in hexagon-like
> > patterns, link for the crystal image:
> 
> > http://fullonline.org/science/cryst01.jpg
> 

> > XDS, Phenix and Pointless always suggest that the data sets for
> > these
> > crystals belong to the space group P622. However, Phenix, Phaser
> > and
> > Pointless indicate that twinning is present.
> 

> > "Bad looking" diffraction images, diffracted to 1.6 A (collected 6
> > months ago):
> 
> > http://fullonline.org/science/dataset1_image37.png
> 
> > http://fullonline.org/science/dataset1_image7.png
> 

> > Best 

Re: [ccp4bb] To win or not twinning?

[Eleanor Dodson]

First - there is a useful document about possible twin laws in the CCP4
documentation. It references the SHELX description of twinning for small
molecules.
http://www.ccp4.ac.uk/html/twinning.html

It is possible in a trigonal system that you could have two twin operators
but the twinning stats should lok odd in that case

For a trigonal space group there are several options, but your technique
seems to have correctly found SG P3.
As Jacob Keller says - check POINTLESS to see if P322 etc are options?
Or ask Zanudu..

Just to comment - I think you must have the cell dimensions wrong for the
P63 case? I cant see how the c axis can be longer?

 I like the data processing POINTLESS/AIMLESS/ CTRUNCATE..

POINTLESS lists all axial reflections so you can make an educated guess at
screw axes (or let POINTLESS do it for you...)
POINTLESS analyses each symmetry operator separately. So if the -k, -h, -l
was slightly less convincing than others, AND the twin tests suggested
twinning, you might expect that to be a twin operator and not a symmetry
operator..

ctruncate tests each possible twin operator and scores them. (And so does
XTRIAGE of course..)


On 30 November 2016 at 14:00, Keller, Jacob 
wrote:

> What about spacegroups in PG 32, e.g., p3212?
>
>
>
> JPK
>
>
>
> *From:* CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] *On Behalf Of 
> *Napoleao
> Fonseca Valadares
> *Sent:* Wednesday, November 30, 2016 2:01 AM
> *To:* CCP4BB@JISCMAIL.AC.UK
> *Subject:* [ccp4bb] To win or not twinning?
>
>
>
> Dear CCP4ers,
>
> I'd like to kindly ask your advice. Sorry for the long e-mail.
>
> I have crystals of a 12.3 KDa protein that grow in hexagon-like patterns,
> link for the crystal image:
> http://fullonline.org/science/cryst01.jpg
>
> XDS, Phenix and Pointless always suggest that the data sets for these
> crystals belong to the space group P622. However, Phenix, Phaser and
> Pointless indicate that twinning is present.
>
> "Bad looking" diffraction images, diffracted to 1.6 A (collected 6 months
> ago):
> http://fullonline.org/science/dataset1_image37.png
> http://fullonline.org/science/dataset1_image7.png
>
> Best data set, diffracted to 2.01 A (collected a month ago):
> http://fullonline.org/science/dataset2_image51.png
>
> The second data set present better looking images, a better XDS ISa value
> (around 24) and diffracted to 2.2 A. The "bad looking" data set diffracted
> to 1.6 A, but I decided to stop working with it (XDS ISa around 10).
>
> There is a template with 60% identity, I used XDS to try to process the
> data in all trigonal/hexagonal space groups from P3 to P6(3)22, and spend a
> lot of time trying molecular replacement procedures in Phaser and Morda,
> and refining the candidate solutions. Used Zanuda too, trying to figure out
> the space group (and read as much as possible in this CCP4 list looking for
> similar cases).
>
> Unit cells and typical MR results:
>
> P1: 56.430   77.718   77.673 119.99  89.99  89.97
> SOLU SET  RFZ=9.5 TFZ=* PAK=0 LLG=91 RF++ TFZ=18.2 PAK=0 LLG=311
> TFZ==18.4 (&
> TFZ==17.0) LLG+=(311 & 726) LLG=5751 TFZ==64.6 PAK=0 LLG=5751 TFZ==64.6
>
> P3: 77.675   77.675   56.409  90.00  90.00 120.00
> RFZ=5.5 TFZ=8.5 PAK=0 LLG=86 TFZ==9.4 LLG=2575 TFZ==46.5
>
> P6: 77.534   77.534   92.986  90.00  90.00 120.00
> RFZ=11.2 TFZ=16.2 PAK=2 LLG=437 TFZ==25.7 LLG=437 TFZ==25.7
>
> P6(3): 77.675   77.675   56.409  90.00  90.00 120.00
>RFZ=8.8 TFZ=11.5 PAK=0 LLG=66 TFZ==7.6 LLG=66 TFZ==7.6
>
> P622: 77.660   77.660   56.400  90.00  90.00 120.00
>   RFZ=4.8 TFZ=5.4 PAK=1 LLG=22 TFZ==4.7 LLG=24 TFZ==4.8
>
>
> In P1 (LLG=5751 TFZ==64.6) there are 12 molecules in the asymmetric unit,
> and in P3 (LLG=2575 TFZ==46.5) 4 molecules. Packing looks good, in P1 the
> ASU looks like two superposed hexagonal donuts formed by 6 molecules each.
> Refining in P1, without adding waters or TLS, yield r_work = 0.2964 and
> r_free = 0.3428, and it is hard to decrease these values.
> Refining in P3, I managed to get r_work = 0.2934 and r_free = 0.3399, but
> looks like it's not getting any better than this.
> Refining in P6 yields horrible r_free values (>0.50).
>
> Trying to refine MR solutions in any other other space groups yield Rfree
> 0.41 or more.
>
> If in the P3 space group I use the twin operator -k,-h,-l (estimated twin
> fraction of 0.490) suggested by Phenix Xtriage, the values miraculously go
> down to r_work = 0.1923 r_free = 0.2202, without waters (r_free without
> using a twin law = 0.3399). Adding 120 water molecules and doing some
> refining yields 0.1711 r_free = 0.1935 (the asymmetric unit contains 420
> residues and the resolution is 2.01 A).
> I've been reading about twinning refinement and how it can drop the
> Rvalues, but from what I understood if I use it improperly I may compromise
> the refinement quality.
>
> I would like advice on:
> 1 - References that can teach how to look at protein diffraction images
> and understand 

Re: [ccp4bb] To win or not twinning?

[Keller, Jacob]

What about spacegroups in PG 32, e.g., p3212?

JPK

From: CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] On Behalf Of Napoleao 
Fonseca Valadares
Sent: Wednesday, November 30, 2016 2:01 AM
To: CCP4BB@JISCMAIL.AC.UK
Subject: [ccp4bb] To win or not twinning?

Dear CCP4ers,

I'd like to kindly ask your advice. Sorry for the long e-mail.

I have crystals of a 12.3 KDa protein that grow in hexagon-like patterns, link 
for the crystal image:
http://fullonline.org/science/cryst01.jpg

XDS, Phenix and Pointless always suggest that the data sets for these crystals 
belong to the space group P622. However, Phenix, Phaser and Pointless indicate 
that twinning is present.

"Bad looking" diffraction images, diffracted to 1.6 A (collected 6 months ago):
http://fullonline.org/science/dataset1_image37.png
http://fullonline.org/science/dataset1_image7.png

Best data set, diffracted to 2.01 A (collected a month ago):
http://fullonline.org/science/dataset2_image51.png

The second data set present better looking images, a better XDS ISa value 
(around 24) and diffracted to 2.2 A. The "bad looking" data set diffracted to 
1.6 A, but I decided to stop working with it (XDS ISa around 10).

There is a template with 60% identity, I used XDS to try to process the data in 
all trigonal/hexagonal space groups from P3 to P6(3)22, and spend a lot of time 
trying molecular replacement procedures in Phaser and Morda, and refining the 
candidate solutions. Used Zanuda too, trying to figure out the space group (and 
read as much as possible in this CCP4 list looking for similar cases).

Unit cells and typical MR results:

P1: 56.430   77.718   77.673 119.99  89.99  89.97
SOLU SET  RFZ=9.5 TFZ=* PAK=0 LLG=91 RF++ TFZ=18.2 PAK=0 LLG=311 TFZ==18.4 
(&
TFZ==17.0) LLG+=(311 & 726) LLG=5751 TFZ==64.6 PAK=0 LLG=5751 TFZ==64.6

P3: 77.675   77.675   56.409  90.00  90.00 120.00
RFZ=5.5 TFZ=8.5 PAK=0 LLG=86 TFZ==9.4 LLG=2575 TFZ==46.5

P6: 77.534   77.534   92.986  90.00  90.00 120.00
RFZ=11.2 TFZ=16.2 PAK=2 LLG=437 TFZ==25.7 LLG=437 TFZ==25.7

P6(3): 77.675   77.675   56.409  90.00  90.00 120.00
   RFZ=8.8 TFZ=11.5 PAK=0 LLG=66 TFZ==7.6 LLG=66 TFZ==7.6

P622: 77.660   77.660   56.400  90.00  90.00 120.00
  RFZ=4.8 TFZ=5.4 PAK=1 LLG=22 TFZ==4.7 LLG=24 TFZ==4.8


In P1 (LLG=5751 TFZ==64.6) there are 12 molecules in the asymmetric unit, and 
in P3 (LLG=2575 TFZ==46.5) 4 molecules. Packing looks good, in P1 the ASU looks 
like two superposed hexagonal donuts formed by 6 molecules each.
Refining in P1, without adding waters or TLS, yield r_work = 0.2964 and r_free 
= 0.3428, and it is hard to decrease these values.
Refining in P3, I managed to get r_work = 0.2934 and r_free = 0.3399, but looks 
like it's not getting any better than this.
Refining in P6 yields horrible r_free values (>0.50).

Trying to refine MR solutions in any other other space groups yield Rfree 0.41 
or more.

If in the P3 space group I use the twin operator -k,-h,-l (estimated twin 
fraction of 0.490) suggested by Phenix Xtriage, the values miraculously go down 
to r_work = 0.1923 r_free = 0.2202, without waters (r_free without using a twin 
law = 0.3399). Adding 120 water molecules and doing some refining yields 0.1711 
r_free = 0.1935 (the asymmetric unit contains 420 residues and the resolution 
is 2.01 A).
I've been reading about twinning refinement and how it can drop the Rvalues, 
but from what I understood if I use it improperly I may compromise the 
refinement quality.

I would like advice on:
1 - References that can teach how to look at protein diffraction images and 
understand what I am seeing. The basics like recognizing bad data and what 
usually leads to deformity in the spots (for example, elliptical or duplicated) 
would be of great help.

2 - Should I look for other space groups? What else could be tried? Is this a 
case where a twin law should be used in the refinement? If yes, what can I do 
to confirm the need for a twin law in the refinement?

Thank you all in advance.
Regards,
 Napo

Re: [ccp4bb] To win or not twinning?

[frazao]

Hi,


The significant difference of refinement R values of the two models, 
single vs twinned crystal, indicates that the later is a far superior 
model of your crystal, corroborated its lower difference in Rfree-Rwork.



However, I am curious about the refined twinning fraction, did it 
decreased from the initial Xtriage estimate of 0.49?



Carlos




On 11/30/2016 07:01 AM, Napoleao Fonseca Valadares wrote:

Dear CCP4ers,

I'd like to kindly ask your advice. Sorry for the long e-mail.

I have crystals of a 12.3 KDa protein that grow in hexagon-like 
patterns, link for the crystal image:

http://fullonline.org/science/cryst01.jpg

XDS, Phenix and Pointless always suggest that the data sets for these 
crystals belong to the space group P622. However, Phenix, Phaser and 
Pointless indicate that twinning is present.


"Bad looking" diffraction images, diffracted to 1.6 A (collected 6 
months ago):

http://fullonline.org/science/dataset1_image37.png
http://fullonline.org/science/dataset1_image7.png

Best data set, diffracted to 2.01 A (collected a month ago):
http://fullonline.org/science/dataset2_image51.png

The second data set present better looking images, a better XDS ISa 
value (around 24) and diffracted to 2.2 A. The "bad looking" data set 
diffracted to 1.6 A, but I decided to stop working with it (XDS ISa 
around 10).


There is a template with 60% identity, I used XDS to try to process 
the data in all trigonal/hexagonal space groups from P3 to P6(3)22, 
and spend a lot of time trying molecular replacement procedures in 
Phaser and Morda, and refining the candidate solutions. Used Zanuda 
too, trying to figure out the space group (and read as much as 
possible in this CCP4 list looking for similar cases).


Unit cells and typical MR results:

P1: 56.430   77.718   77.673 119.99  89.99  89.97
SOLU SET  RFZ=9.5 TFZ=* PAK=0 LLG=91 RF++ TFZ=18.2 PAK=0 LLG=311 
TFZ==18.4 (&
TFZ==17.0) LLG+=(311 & 726) LLG=5751 TFZ==64.6 PAK=0 LLG=5751 
TFZ==64.6


P3: 77.675   77.675   56.409  90.00  90.00 120.00
RFZ=5.5 TFZ=8.5 PAK=0 LLG=86 TFZ==9.4 LLG=2575 TFZ==46.5

P6: 77.534   77.534   92.986  90.00  90.00 120.00
RFZ=11.2 TFZ=16.2 PAK=2 LLG=437 TFZ==25.7 LLG=437 TFZ==25.7

P6(3): 77.675   77.675   56.409  90.00  90.00 120.00
   RFZ=8.8 TFZ=11.5 PAK=0 LLG=66 TFZ==7.6 LLG=66 TFZ==7.6

P622: 77.660   77.660   56.400  90.00  90.00 120.00
  RFZ=4.8 TFZ=5.4 PAK=1 LLG=22 TFZ==4.7 LLG=24 TFZ==4.8


In P1 (LLG=5751 TFZ==64.6) there are 12 molecules in the asymmetric 
unit, and in P3 (LLG=2575 TFZ==46.5) 4 molecules. Packing looks good, 
in P1 the ASU looks like two superposed hexagonal donuts formed by 6 
molecules each.
Refining in P1, without adding waters or TLS, yield r_work = 0.2964 
and r_free = 0.3428, and it is hard to decrease these values.
Refining in P3, I managed to get r_work = 0.2934 and r_free = 0.3399, 
but looks like it's not getting any better than this.

Refining in P6 yields horrible r_free values (>0.50).

Trying to refine MR solutions in any other other space groups yield 
Rfree 0.41 or more.


If in the P3 space group I use the twin operator -k,-h,-l (estimated 
twin fraction of 0.490) suggested by Phenix Xtriage, the values 
miraculously go down to r_work = 0.1923 r_free = 0.2202, without 
waters (r_free without using a twin law = 0.3399). Adding 120 water 
molecules and doing some refining yields 0.1711 r_free = 0.1935 (the 
asymmetric unit contains 420 residues and the resolution is 2.01 A).
I've been reading about twinning refinement and how it can drop the 
Rvalues, but from what I understood if I use it improperly I may 
compromise the refinement quality.


I would like advice on:
1 - References that can teach how to look at protein diffraction 
images and understand what I am seeing. The basics like recognizing 
bad data and what usually leads to deformity in the spots (for 
example, elliptical or duplicated) would be of great help.


2 - Should I look for other space groups? What else could be tried? Is 
this a case where a twin law should be used in the refinement? If yes, 
what can I do to confirm the need for a twin law in the refinement?


Thank you all in advance.
Regards,
 Napo


--
**
Dr. Carlos Frazao
Structural Biology Laboratory -
Macromolecular Crystallography Unit
ITQB NOVA, Av. da República
2780-157 Oeiras, Portugal

Phone:  (351)-214469666/609
FAX:(351)-214433644
e-mail: fra...@itqb.unl.pt
www.itqb.unl.pt