Where to find optimized wxc_scale/wxu_scale weighting factors after optimization
Hello, I ran a refinement optimizing geometry and B factor weighting factors using optimize_wxc=true optimize_wxu=true. I was given these statistics in the pdb header: Rwork Rfree Bonds Angles start .1588 .1838 .007 1.235 end .1454 .1780 .016 1.601 (1.37 A resolution) 1. I would like to find the optimized weighting factors in the log file but cannot. Instead I find the following (pasted below). However, the scale here doesn't make sense to me because many our greater than 0.5. How do I find the optimized values so that I can use them in further rounds of refinement? I'm sorry if this has been covered in previous posts but I could not find an answer. 2. Can anyone explain the concept of optimizing the B-factor? I don't understand how changing it brings you closer or further from an ideal model as oppose to xray. (which is how I make sense of refining the geometry) Thank you -Yarow ================================ xyz refinement =============================== Start r_work= 0.1462 r_free= 0.1784 bonds= 0.015 angles= 1.53 r_work= 0.1492 r_free= 0.1792 bonds= 0.005 angles= 1.08 scale= 1.00 r_work= 0.1484 r_free= 0.1788 bonds= 0.006 angles= 1.14 scale= 1.40 r_work= 0.1480 r_free= 0.1787 bonds= 0.007 angles= 1.19 scale= 1.80 r_work= 0.1476 r_free= 0.1785 bonds= 0.008 angles= 1.23 scale= 2.20 r_work= 0.1473 r_free= 0.1785 bonds= 0.008 angles= 1.26 scale= 2.60 r_work= 0.1471 r_free= 0.1784 bonds= 0.009 angles= 1.29 scale= 3.00 r_work= 0.1469 r_free= 0.1784 bonds= 0.010 angles= 1.32 scale= 3.40 r_work= 0.1467 r_free= 0.1783 bonds= 0.010 angles= 1.34 scale= 3.80 r_work= 0.1465 r_free= 0.1782 bonds= 0.011 angles= 1.36 scale= 4.20 r_work= 0.1464 r_free= 0.1782 bonds= 0.011 angles= 1.39 scale= 4.60 r_work= 0.1463 r_free= 0.1782 bonds= 0.011 angles= 1.40 scale= 5.00 r_work= 0.1461 r_free= 0.1781 bonds= 0.012 angles= 1.45 scale= 6.00 r_work= 0.1458 r_free= 0.1781 bonds= 0.013 angles= 1.49 scale= 7.00 r_work= 0.1456 r_free= 0.1781 bonds= 0.014 angles= 1.53 scale= 8.00 r_work= 0.1454 r_free= 0.1781 bonds= 0.015 angles= 1.57 scale= 9.00 r_work= 0.1453 r_free= 0.1780 bonds= 0.016 angles= 1.60 scale= 10.00 r_work= 0.1495 r_free= 0.1794 bonds= 0.005 angles= 1.05 scale= 0.85 r_work= 0.1501 r_free= 0.1797 bonds= 0.005 angles= 1.02 scale= 0.70 r_work= 0.1507 r_free= 0.1800 bonds= 0.004 angles= 0.98 scale= 0.55 Final r_work= 0.1453 r_free= 0.1780 bonds= 0.016 angles= 1.60
Hi Yarow,
I ran a refinement optimizing geometry and B factor weighting factors using optimize_wxc=true optimize_wxu=true.
I was given these statistics in the pdb header:
Rwork Rfree Bonds Angles start .1588 .1838 .007 1.235 end .1454 .1780 .016 1.601
(1.37 A resolution)
Looks ok. Did you update solvent using "ordered_solvent=true"? Did you remember to add have H atoms to your structure?
1. I would like to find the optimized weighting factors in the log file but cannot. Instead I find the following (pasted below).
I'm wondering why you are trying to find them? It is such a technical thing that changes from macro-cycle to macro-cycle...
However, the scale here doesn't make sense to me because many our greater than 0.5.
Hm... In the text below actually none of them is equal to 0.5. May be you meant ADP refinement?
How do I find the optimized values so that I can use them in further rounds of refinement?
See below.
2. Can anyone explain the concept of optimizing the B-factor? I don't understand how changing it brings you closer or further from an ideal model as oppose to xray. (which is how I make sense of refining the geometry)
Start here ("On atomic displacement parameters (ADP)"): http://www.phenix-online.org/newsletter/CCN_2010_07.pdf and http://www.phenix-online.org/presentations/latest/pavel_refinement_general.p... then may be have a look: http://www.phenix-online.org/presentations/latest/pavel_phenix_refine.pdf
================================ xyz refinement ===============================
Start r_work= 0.1462 r_free= 0.1784 bonds= 0.015 angles= 1.53 r_work= 0.1492 r_free= 0.1792 bonds= 0.005 angles= 1.08 scale= 1.00 r_work= 0.1484 r_free= 0.1788 bonds= 0.006 angles= 1.14 scale= 1.40 r_work= 0.1480 r_free= 0.1787 bonds= 0.007 angles= 1.19 scale= 1.80 r_work= 0.1476 r_free= 0.1785 bonds= 0.008 angles= 1.23 scale= 2.20 r_work= 0.1473 r_free= 0.1785 bonds= 0.008 angles= 1.26 scale= 2.60 r_work= 0.1471 r_free= 0.1784 bonds= 0.009 angles= 1.29 scale= 3.00 r_work= 0.1469 r_free= 0.1784 bonds= 0.010 angles= 1.32 scale= 3.40 r_work= 0.1467 r_free= 0.1783 bonds= 0.010 angles= 1.34 scale= 3.80 r_work= 0.1465 r_free= 0.1782 bonds= 0.011 angles= 1.36 scale= 4.20 r_work= 0.1464 r_free= 0.1782 bonds= 0.011 angles= 1.39 scale= 4.60 r_work= 0.1463 r_free= 0.1782 bonds= 0.011 angles= 1.40 scale= 5.00 r_work= 0.1461 r_free= 0.1781 bonds= 0.012 angles= 1.45 scale= 6.00 r_work= 0.1458 r_free= 0.1781 bonds= 0.013 angles= 1.49 scale= 7.00 r_work= 0.1456 r_free= 0.1781 bonds= 0.014 angles= 1.53 scale= 8.00 r_work= 0.1454 r_free= 0.1781 bonds= 0.015 angles= 1.57 scale= 9.00 r_work= 0.1453 r_free= 0.1780 bonds= 0.016 angles= 1.60 scale= 10.00 r_work= 0.1495 r_free= 0.1794 bonds= 0.005 angles= 1.05 scale= 0.85 r_work= 0.1501 r_free= 0.1797 bonds= 0.005 angles= 1.02 scale= 0.70 r_work= 0.1507 r_free= 0.1800 bonds= 0.004 angles= 0.98 scale= 0.55 Final r_work= 0.1453 r_free= 0.1780 bonds= 0.016 angles= 1.60
The optimal wxc_scale = 10.00, that corresponds to Final r_work= 0.1453 r_free= 0.1780 bonds= 0.016 angles= 1.60 All the best! Pavel.
Hi Yarow,
I ran a refinement optimizing geometry and B factor weighting factors using optimize_wxc=true optimize_wxu=true.
I was given these statistics in the pdb header:
Rwork Rfree Bonds Angles start .1588 .1838 .007 1.235 end .1454 .1780 .016 1.601
(1.37 A resolution)
Looks ok. Did you update solvent using "ordered_solvent=true"? Did you remember to add have H atoms to your structure?
Thanks Pavel, No, I did not have these turned on. It appears that "ordered_solvent=true" is essentially a water checker in later stages of refinement (correct me if I am wrong). I will try turning this on. What are the advantages of adding H atoms if you are not going to refine them? (Bear with me--this is my first structure).
1. I would like to find the optimized weighting factors in the log file but cannot. Instead I find the following (pasted below).
I'm wondering why you are trying to find them? It is such a technical thing that changes from macro-cycle to macro-cycle...
I was attempting to use the best weighting factors in subsequent rounds of refinement so that I did not have to optimize each time (because it takes a long time to run). Is this a good strategy?
However, the scale here doesn't make sense to me because many our greater than 0.5.
Hm... In the text below actually none of them is equal to 0.5. May be you meant ADP refinement?
How do I find the optimized values so that I can use them in further rounds of refinement?
See below.
2. Can anyone explain the concept of optimizing the B-factor? I don't understand how changing it brings you closer or further from an ideal model as oppose to xray. (which is how I make sense of refining the geometry)
Start here ("On atomic displacement parameters (ADP)"): http://www.phenix-online.org/newsletter/CCN_2010_07.pdf and http://www.phenix-online.org/presentations/latest/pavel_refinement_general.p... then may be have a look: http://www.phenix-online.org/presentations/latest/pavel_phenix_refine.pdf
Thank you!
================================ xyz refinement ===============================
Start r_work= 0.1462 r_free= 0.1784 bonds= 0.015 angles= 1.53 r_work= 0.1492 r_free= 0.1792 bonds= 0.005 angles= 1.08 scale= 1.00 r_work= 0.1484 r_free= 0.1788 bonds= 0.006 angles= 1.14 scale= 1.40 r_work= 0.1480 r_free= 0.1787 bonds= 0.007 angles= 1.19 scale= 1.80 r_work= 0.1476 r_free= 0.1785 bonds= 0.008 angles= 1.23 scale= 2.20 r_work= 0.1473 r_free= 0.1785 bonds= 0.008 angles= 1.26 scale= 2.60 r_work= 0.1471 r_free= 0.1784 bonds= 0.009 angles= 1.29 scale= 3.00 r_work= 0.1469 r_free= 0.1784 bonds= 0.010 angles= 1.32 scale= 3.40 r_work= 0.1467 r_free= 0.1783 bonds= 0.010 angles= 1.34 scale= 3.80 r_work= 0.1465 r_free= 0.1782 bonds= 0.011 angles= 1.36 scale= 4.20 r_work= 0.1464 r_free= 0.1782 bonds= 0.011 angles= 1.39 scale= 4.60 r_work= 0.1463 r_free= 0.1782 bonds= 0.011 angles= 1.40 scale= 5.00 r_work= 0.1461 r_free= 0.1781 bonds= 0.012 angles= 1.45 scale= 6.00 r_work= 0.1458 r_free= 0.1781 bonds= 0.013 angles= 1.49 scale= 7.00 r_work= 0.1456 r_free= 0.1781 bonds= 0.014 angles= 1.53 scale= 8.00 r_work= 0.1454 r_free= 0.1781 bonds= 0.015 angles= 1.57 scale= 9.00 r_work= 0.1453 r_free= 0.1780 bonds= 0.016 angles= 1.60 scale= 10.00 r_work= 0.1495 r_free= 0.1794 bonds= 0.005 angles= 1.05 scale= 0.85 r_work= 0.1501 r_free= 0.1797 bonds= 0.005 angles= 1.02 scale= 0.70 r_work= 0.1507 r_free= 0.1800 bonds= 0.004 angles= 0.98 scale= 0.55 Final r_work= 0.1453 r_free= 0.1780 bonds= 0.016 angles= 1.60
The optimal wxc_scale = 10.00, that corresponds to Final r_work= 0.1453 r_free= 0.1780 bonds= 0.016 angles= 1.60
All the best! Pavel.
Hi Yarrow,
I was given these statistics in the pdb header:
Rwork Rfree Bonds Angles start .1588 .1838 .007 1.235 end .1454 .1780 .016 1.601
(1.37 A resolution) Looks ok. Did you update solvent using "ordered_solvent=true"? Did you remember to add have H atoms to your structure? Thanks Pavel,
No, I did not have these turned on. It appears that "ordered_solvent=true" is essentially a water checker in later stages of refinement (correct me if I am wrong). I will try turning this on.
refinement means changing your structure, and changing the structure means some map improvement. Since solvent is added using the maps (mFo-DFc and 2mFo-DFc) it is a good idea to re-check it (solvent) once the maps are updated. The result of such re-check may be some more good water added, or removing some wrong waters originally added to noise density peaks.
What are the advantages of adding H atoms if you are not going to refine them? (Bear with me--this is my first structure).
See pages from 27 here: http://www.phenix-online.org/presentations/latest/pavel_phenix_refine.pdf Let me know if it is still not entirely clear why you would need to use H-atoms in refinement. After updating water and adding H atoms the R-factors may drop by 1-3%.
1. I would like to find the optimized weighting factors in the log file but cannot. Instead I find the following (pasted below). I'm wondering why you are trying to find them? It is such a technical thing that changes from macro-cycle to macro-cycle... I was attempting to use the best weighting factors in subsequent rounds of refinement so that I did not have to optimize each time (because it takes a long time to run). Is this a good strategy?
I see. Although you can do it, it is good to keep in mind that these values can change between refinement runs. You can get this value for coordinate refinement and unfortunately this values is not printed out for ADP refinement (because of a formatting bug that I keep forgetting to fix). Good luck! Pavel.
On 8/12/10 10:43 AM, Pavel Afonine wrote:
1. I would like to find the optimized weighting factors in the log file but cannot. Instead I find the following (pasted below).
I'm wondering why you are trying to find them? It is such a technical thing that changes from macro-cycle to macro-cycle...
Pavel, Having the optimized weight printed would be mildly handy. First, the weight is probably not changing wildly between cycle to cycle, so an approximately working wxc_scale can be gotten. Then, this number can be used for quicker test runs with optimize_wxc=false, and you would not worry that your weights are hugely non-optimal for your quick runs. Second, I recently realized that in the auto mode for optimize_wxc, my first cycle was not optimizing wxc, and I was getting huge jumps in geometry and R factors in the first cycle. If there was an obvious table showing wxc/wxu per cycle, it would take me shorter to realize what was going on. This first-cycle problem was serious, since it was breaking up all the sensible geometry that I spent days building in a low-resolution, large structure, which wasn't getting fixed in subsequent cycles (Obviously, setting optimize_wxc to every_cycle mode would have prevented this problem, but I bet you that most users aren't aware of such an option. You can also get the optimized wxc_scale by a simple multiplication if you dig into the log file, but that should not be necessary). My refinement work is not linear; I don't go from Refine_1 to _2 to _3. A lot of the time, I will have 3a, 3b, 3c, etc. quick test runs to see how I should proceed next. "Quick" is the operative word here, and a roundabout good wxc_scale value is valuable there. Best, Engin
Hi Engin,
Having the optimized weight printed would be mildly handy. First, the weight is probably not changing wildly between cycle to cycle, so an approximately working wxc_scale can be gotten. Then, this number can be used for quicker test runs with optimize_wxc=false, and you would not worry that your weights are hugely non-optimal for your quick runs.
I agree. The weight for coordinate refinement is reported in log file. The weight for ADP refinement is not reported because of a printing bug that I will fix.
Second, I recently realized that in the auto mode for optimize_wxc, my first cycle was not optimizing wxc, and I was getting huge jumps in geometry and R factors in the first cycle. If there was an obvious table showing wxc/wxu per cycle, it would take me shorter to realize what was going on. This first-cycle problem was serious, since it was breaking up all the sensible geometry that I spent days building in a low-resolution, large structure, which wasn't getting fixed in subsequent cycles (Obviously, setting optimize_wxc to every_cycle mode would have prevented this problem, but I bet you that most users aren't aware of such an option. You can also get the optimized wxc_scale by a simple multiplication if you dig into the log file, but that should not be necessary).
Ok, I will add the weight to the table of values that phenix.refine outputs by the end of each refinement. Thanks for always useful feedback! Pavel.
Thank you Pavel for the great piece of software, and for being very prompt and responsive to users' problems and suggestions.
Engin
----- Original Message -----
From: "Pavel Afonine"
Having the optimized weight printed would be mildly handy. First, the weight is probably not changing wildly between cycle to cycle, so an approximately working wxc_scale can be gotten. Then, this number can be used for quicker test runs with optimize_wxc=false, and you would not worry that your weights are hugely non-optimal for your quick runs.
I agree. The weight for coordinate refinement is reported in log file. The weight for ADP refinement is not reported because of a printing bug that I will fix.
Second, I recently realized that in the auto mode for optimize_wxc, my first cycle was not optimizing wxc, and I was getting huge jumps in geometry and R factors in the first cycle. If there was an obvious table showing wxc/wxu per cycle, it would take me shorter to realize what was going on. This first-cycle problem was serious, since it was breaking up all the sensible geometry that I spent days building in a low-resolution, large structure, which wasn't getting fixed in subsequent cycles (Obviously, setting optimize_wxc to every_cycle mode would have prevented this problem, but I bet you that most users aren't aware of such an option. You can also get the optimized wxc_scale by a simple multiplication if you dig into the log file, but that should not be necessary).
Ok, I will add the weight to the table of values that phenix.refine outputs by the end of each refinement. Thanks for always useful feedback! Pavel. _______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
participants (4)
-
amadrona＠uci.edu -
Engin Ozkan -
Engin Özkan
-
Pavel Afonine