Rfree refuse to come down
HI, I have a few questions about a dataset that I am working on currently. So I ran Xtraige and it tells me that other than pseudoNCS, there is no other problems like twinning etc. And then I used autoMR and Phaser from Phenix to run molecular replacement with a the structure of the same protein (but that is with DNA bound). The TFZ is more than 9 after from both the runs. Autobuild managed to build the 4 molecules in my ASU (Rfactor/Rfree = 28/32 for a resolution of 2.4A). But because there are some clashes and outliers in the Ramachandran Plot etc, I started to manually refine them in coot. In addition, there is a loop that is unable to fit into the density, so I chopped about 10 residues off each molecule in the ASU. However, as I started to refine the model from AutoBuild, my Rfree begins to climb. So currently, I am at Rfactor/Rfree=27/36. I don't know if there is anything wrong with I had done. So many thanks in advance for your advices on this!! Warmest Regards, Kelly Hew Nanyang Technogical University School of Biological Sciences Division of Structural Biology & Biochemistry Address : #07-01, IMCB, 61 Biopolis Drive, Proteos, Singapore 138673 Telephone (O): +65 65869673 Telephone (HP): +65 98713553
A couple of things: TFZ of 9 is on the lowish side therefore one could question the validity of the MR solution, are all protein molecules correctly positioned for example. Are there domains and subdomains that have moved in your structure relative to your search model? with 4-fold NCS you can use the redundancy to 1) improve the maps by iterative NCS averaging; 2- use NCS contraints (then restraints later on) during refinement. Fred. PS refinement of a model does not mean "bring R-free down", the R-free going up is a symptom that should be investigated and the causes of it then treated... #HEW KAI LI KELLY# wrote:
HI,
I have a few questions about a dataset that I am working on currently.
So I ran Xtraige and it tells me that other than pseudoNCS, there is no other problems like twinning etc. And then I used autoMR and Phaser from Phenix to run molecular replacement with a the structure of the same protein (but that is with DNA bound). The TFZ is more than 9 after from both the runs. Autobuild managed to build the 4 molecules in my ASU (Rfactor/Rfree = 28/32 for a resolution of 2.4A). But because there are some clashes and outliers in the Ramachandran Plot etc, I started to manually refine them in coot. In addition, there is a loop that is unable to fit into the density, so I chopped about 10 residues off each molecule in the ASU.
However, as I started to refine the model from AutoBuild, my Rfree begins to climb. So currently, I am at Rfactor/Rfree=27/36. I don't know if there is anything wrong with I had done. So many thanks in advance for your advices on this!!
Warmest Regards, Kelly Hew
Nanyang Technogical University School of Biological Sciences Division of Structural Biology & Biochemistry Address : #07-01, IMCB, 61 Biopolis Drive, Proteos, Singapore 138673 Telephone (O): +65 65869673 Telephone (HP): +65 98713553
------------------------------------------------------------------------
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
This may be a stupid question, but what are the things I should look out for to change if my Rfree goes up?
Kelly
On 2 Mar, 2012, at 3:17 PM, "Vellieux Frederic"
A couple of things:
TFZ of 9 is on the lowish side therefore one could question the validity of the MR solution, are all protein molecules correctly positioned for example. Are there domains and subdomains that have moved in your structure relative to your search model?
with 4-fold NCS you can use the redundancy to 1) improve the maps by iterative NCS averaging; 2- use NCS contraints (then restraints later on) during refinement.
Fred.
PS refinement of a model does not mean "bring R-free down", the R-free going up is a symptom that should be investigated and the causes of it then treated...
#HEW KAI LI KELLY# wrote:
HI,
I have a few questions about a dataset that I am working on currently.
So I ran Xtraige and it tells me that other than pseudoNCS, there is no other problems like twinning etc. And then I used autoMR and Phaser from Phenix to run molecular replacement with a the structure of the same protein (but that is with DNA bound). The TFZ is more than 9 after from both the runs. Autobuild managed to build the 4 molecules in my ASU (Rfactor/Rfree = 28/32 for a resolution of 2.4A). But because there are some clashes and outliers in the Ramachandran Plot etc, I started to manually refine them in coot. In addition, there is a loop that is unable to fit into the density, so I chopped about 10 residues off each molecule in the ASU.
However, as I started to refine the model from AutoBuild, my Rfree begins to climb. So currently, I am at Rfactor/Rfree=27/36. I don't know if there is anything wrong with I had done. So many thanks in advance for your advices on this!!
Warmest Regards, Kelly Hew
Nanyang Technogical University School of Biological Sciences Division of Structural Biology & Biochemistry Address : #07-01, IMCB, 61 Biopolis Drive, Proteos, Singapore 138673 Telephone (O): +65 65869673 Telephone (HP): +65 98713553
------------------------------------------------------------------------
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
Well, IMHO, one of the things to check is whether or not you are trying to refine far too many parameters for your data set (your observations), for example. You'd expect to see Rfree going up if you were to refine (at 2.4 A resolution) a structure with a full anisotropic temperature factor model (i.e. x, y, z and 6 Baniso values). However in your case I'd question first the molecular replacement solution. Make sure you have a valid solution... Did you see (in all 4 protein molecules) density indicative of the sequence changes between your MR search model and your protein ? This is indicative of a real solution... And I'd certainly start refinement using the NCS present, to relax it (restraints) as the refinement proceeds (smoothly) and possibly not use any NCS in the final stages of the refinement. You may need to carry out annealing initially and so forth. Fred. #HEW KAI LI KELLY# wrote:
This may be a stupid question, but what are the things I should look out for to change if my Rfree goes up?
Kelly
On 2 Mar, 2012, at 3:17 PM, "Vellieux Frederic"
wrote: A couple of things:
TFZ of 9 is on the lowish side therefore one could question the validity of the MR solution, are all protein molecules correctly positioned for example. Are there domains and subdomains that have moved in your structure relative to your search model?
with 4-fold NCS you can use the redundancy to 1) improve the maps by iterative NCS averaging; 2- use NCS contraints (then restraints later on) during refinement.
Fred.
PS refinement of a model does not mean "bring R-free down", the R-free going up is a symptom that should be investigated and the causes of it then treated...
#HEW KAI LI KELLY# wrote:
HI,
I have a few questions about a dataset that I am working on currently.
So I ran Xtraige and it tells me that other than pseudoNCS, there is no other problems like twinning etc. And then I used autoMR and Phaser from Phenix to run molecular replacement with a the structure of the same protein (but that is with DNA bound). The TFZ is more than 9 after from both the runs. Autobuild managed to build the 4 molecules in my ASU (Rfactor/Rfree = 28/32 for a resolution of 2.4A). But because there are some clashes and outliers in the Ramachandran Plot etc, I started to manually refine them in coot. In addition, there is a loop that is unable to fit into the density, so I chopped about 10 residues off each molecule in the ASU.
However, as I started to refine the model from AutoBuild, my Rfree begins to climb. So currently, I am at Rfactor/Rfree=27/36. I don't know if there is anything wrong with I had done. So many thanks in advance for your advices on this!!
Warmest Regards, Kelly Hew
Nanyang Technogical University School of Biological Sciences Division of Structural Biology & Biochemistry Address : #07-01, IMCB, 61 Biopolis Drive, Proteos, Singapore 138673 Telephone (O): +65 65869673 Telephone (HP): +65 98713553
------------------------------------------------------------------------
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
Maybe this tool I know from a previous colleague may be of some help: http://altair.sci.hokudai.ac.jp/g6/Research/Lafire_English.html Citations: "LAFIRE: software for automating the refinement process of protein structure analysis". YAO M., ZHOU Y., AND TANAKA I., Acta. Cryst., D62, 189-196 "New algorithm for protein model building: extending partial model in map segment". ZHOU Y., YAO M., AND TANAKA I., J. Appl. Cryst., 39, 57-63 By the way, this tool shouldn't be in phenix? ;) Regards, F. On 03/02/2012 04:36 PM, Vellieux Frederic wrote:
Well, IMHO, one of the things to check is whether or not you are trying to refine far too many parameters for your data set (your observations), for example. You'd expect to see Rfree going up if you were to refine (at 2.4 A resolution) a structure with a full anisotropic temperature factor model (i.e. x, y, z and 6 Baniso values).
However in your case I'd question first the molecular replacement solution. Make sure you have a valid solution... Did you see (in all 4 protein molecules) density indicative of the sequence changes between your MR search model and your protein ? This is indicative of a real solution... And I'd certainly start refinement using the NCS present, to relax it (restraints) as the refinement proceeds (smoothly) and possibly not use any NCS in the final stages of the refinement. You may need to carry out annealing initially and so forth.
Fred.
#HEW KAI LI KELLY# wrote:
This may be a stupid question, but what are the things I should look out for to change if my Rfree goes up?
Kelly
On 2 Mar, 2012, at 3:17 PM, "Vellieux Frederic"
wrote: A couple of things:
TFZ of 9 is on the lowish side therefore one could question the validity of the MR solution, are all protein molecules correctly positioned for example. Are there domains and subdomains that have moved in your structure relative to your search model?
with 4-fold NCS you can use the redundancy to 1) improve the maps by iterative NCS averaging; 2- use NCS contraints (then restraints later on) during refinement.
Fred.
PS refinement of a model does not mean "bring R-free down", the R-free going up is a symptom that should be investigated and the causes of it then treated...
#HEW KAI LI KELLY# wrote:
HI,
I have a few questions about a dataset that I am working on currently.
So I ran Xtraige and it tells me that other than pseudoNCS, there is no other problems like twinning etc. And then I used autoMR and Phaser from Phenix to run molecular replacement with a the structure of the same protein (but that is with DNA bound). The TFZ is more than 9 after from both the runs. Autobuild managed to build the 4 molecules in my ASU (Rfactor/Rfree = 28/32 for a resolution of 2.4A). But because there are some clashes and outliers in the Ramachandran Plot etc, I started to manually refine them in coot. In addition, there is a loop that is unable to fit into the density, so I chopped about 10 residues off each molecule in the ASU.
However, as I started to refine the model from AutoBuild, my Rfree begins to climb. So currently, I am at Rfactor/Rfree=27/36. I don't know if there is anything wrong with I had done. So many thanks in advance for your advices on this!!
Warmest Regards, Kelly Hew
Nanyang Technogical University School of Biological Sciences Division of Structural Biology & Biochemistry Address : #07-01, IMCB, 61 Biopolis Drive, Proteos, Singapore 138673 Telephone (O): +65 65869673 Telephone (HP): +65 98713553
------------------------------------------------------------------------
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
On Thu, Mar 1, 2012 at 11:36 PM, Vellieux Frederic
And I'd certainly start refinement using the NCS present, to relax it (restraints) as the refinement proceeds (smoothly) and possibly not use any NCS in the final stages of the refinement. You may need to carry out annealing initially and so forth.
Fred is correct, you absolutely want to use NCS at this resolution - if you use the torsion NCS parameterization (get a new-ish build for this), you probably will not need or want to turn off the restraints. The other thing to try is optimizing the X-ray/stereochemistry weight. This can take quite a long time (unless you have a large multi-core system), but it almost always yields a better result. You might also want to turn off the real-space refinement, sometimes it behaves weirdly. At 2.4A with an R-free of 0.32 you can probably start running the ordered solvent update, which will help a little bit. A more general suggestion is to look through the log file and identify where the R-factors begin to explode. This will give you a better idea where you need to change your strategy. Final suggestion: make sure you're running at least version 1.7.3, since there have been many improvements to refinement of more marginal cases. -Nat
participants (4)
-
#HEW KAI LI KELLY#
-
Francois Berenger
-
Nathaniel Echols
-
Vellieux Frederic