Seema, I agree completely with Eleanor. You need to take a step back
here. When you say that 'Rfree got stuck at 29-30' what makes you so
sure that this isn't the correct Rfree? Who told you that there's a
problem to be solved in the first place?
If you look at our paper (Acta Cryst., 1998. D54, 547-557) you'll see
how to estimate the expected Rfree; then you can compare that with the
Rfree you actually obtained and use that objective information to
decide if any remedial action is necessary. If we take your values,
say Rwork = 0.205, Rfree = 0.295 we can compute the ratio ('y' in the
paper) Rfree/Rwork = 1.44 (note that it's the ratio that matters NOT
the difference!). From this we compute 'z' = (y^2-1)/(y^2+1) = 0.35
(see Fig 2). Now look at the constellation of points (purple squares)
representing structures in the PDB in the resolution range 3.0-2.5
Ang. You''ll see that your z value 0.35 falls pretty well in the
middle of those. To pinpoint the exact value of z (hence y, and hence
Rfree) that you would expect to get, you would need make use of the
ratio no of atoms/no of reflections.
You can also see that, depending on the resolution and the obs/param
ratio, the z value could go as low as 0.1 or indeed as high as 0.5.
The latter corresponds to y = 1.73, which for Rwork = 0.20 corresponds
to expected Rfree = 0.35 (in other words an Rfree much less than this
would lead one to conclude there's something wrong); it all depends on
the resolution and the observation/parameter ratio.
So I really don't think you have anything to worry about (except maybe
referees who don't understand the statistics of cross-validation!).
As others have pointed out, you are likely to do more damage to your
structure by cutting out perfectly good data in a vain attempt to
solve a non-existent problem!
Cheers
-- Ian
On Mon, May 23, 2011 at 9:54 AM, Eleanor Dodson <[log in to unmask]> wrote:
> I dont think there is an Rfree problem..
> At 2.7A you expect quite a big difference between R and Rfree
>
> Reducing the resolution will a) probably makethe Rfree/R difference greater,
> and b) degrade the quality of your maps and model.
>
> Eleanor
>
> On 05/21/2011 02:28 AM, Seema Mittal wrote:
>>
>> Hi Ethan,
>>
>> You are absolutely right. As a matter of fact, I had initially processed
>> the data to 2.7A and it looked pretty decent with R symm less than 10%.
>> The maps looked good too.
>>
>> The problem arose during second round of refinement. The Rfree got stuck
>> at around 29-30 while the Rfactor kept decreasing to about 20-21. The
>> bond length and angle values are fine too.
>>
>> I cut down the resolution to 3A hoping to improve the data quality by
>> removing some noise. But, it did not work. i also tried to put restrains
>> on the backbone B factors with limited success.
>>
>> Any thoughts on how i can resolve this Rfree issue?
>>
>> Thanks much,
>> Seema
>>
>>
>>
>> On May 20, 2011, at 5:38 PM, Ethan Merritt wrote:
>>
>>> On Friday, May 20, 2011 02:28:26 pm Mittal, Seema wrote:
>>>>
>>>> Hi All,
>>>>
>>>> I am currently working on a 3A resolution dataset. The scaled file
>>>> shows the following statistics (scroll down to the end of this
>>>> email). It is P212121 space group with R merge of 8.8%.
>>>
>>> Your data statistics look fine. In fact, it looks to me that your
>>> crystal is
>>> probably yielding good data to considerably better resolution than 3A.
>>> Why did you choose to cut it there?
>>>
>>>> My question is : Is there a way to selectively use only the data with
>>>> I/Sigma value of 2 and more for refinement?
>>>
>>> That is a bad idea. By removing data you are throwing away information.
>>> Noisy data is still better than no data.
>>>
>>> good luck with your [probably better than 3A] structure,
>>>
>>> Ethan
>>>
>>>
>>>> And how do i achieve this using refmac? I am aware that this would
>>>> come at the cost of compromising data completeness. Any
>>>> suggestions/help would be greatly appreciated.
>>>>
>>>>
>>>> Thanks much,
>>>> Seema Mittal
>>>> Department of Biochemistry & Molecular Pharmacology
>>>> 970L Lazare Research Building
>>>> University of Massachusetts Medical School
>>>> 364 Plantation Street
>>>> Worcester, MA 01605
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> Shell I/Sigma in resolution shells:
>>>> Lower Upper % of of reflections with I / Sigma less than
>>>> limit limit 0 1 2 3 5 10 20 >20 total
>>>> 50.00 6.46 2.0 3.8 5.3 6.2 7.6 12.5 34.3 65.0 99.3
>>>> 6.46 5.13 0.7 2.2 3.9 5.3 8.2 15.7 36.6 63.4 100.0
>>>> 5.13 4.48 1.3 2.8 4.0 5.8 9.3 13.8 27.3 72.7 100.0
>>>> 4.48 4.07 0.7 1.7 4.0 5.4 7.9 13.9 35.4 64.1 99.5
>>>> 4.07 3.78 1.8 3.6 5.1 6.9 11.8 20.8 49.6 47.3 96.9
>>>> 3.78 3.56 1.5 3.8 6.7 8.7 13.3 26.7 65.4 30.8 96.2
>>>> 3.56 3.38 0.8 3.2 7.1 8.9 12.9 31.1 76.6 20.0 96.6
>>>> 3.38 3.23 2.0 4.8 8.1 14.8 23.4 44.8 84.7 12.7 97.5
>>>> 3.23 3.11 4.1 9.2 13.8 18.4 29.6 51.0 86.0 11.0 96.9
>>>> 3.11 3.00 2.4 8.6 13.9 18.8 30.6 53.9 92.4 4.5 96.9
>>>> All hkl 1.7 4.3 7.1 9.8 15.3 28.0 58.1 39.9 98.0
>>>>
>>>>
>>>> Shell Lower Upper Average Average Norm. Linear Square
>>>> limit Angstrom I error stat. Chi**2 R-fac R-fac
>>>> 50.00 6.46 511.7 20.0 8.8 1.098 0.065 0.073
>>>> 6.46 5.13 284.6 10.1 6.3 1.047 0.062 0.064
>>>> 5.13 4.48 500.9 17.0 8.8 1.007 0.062 0.069
>>>> 4.48 4.07 446.1 17.4 9.2 1.032 0.069 0.070
>>>> 4.07 3.78 307.1 14.5 8.4 1.065 0.089 0.092
>>>> 3.78 3.56 243.4 13.8 7.9 1.033 0.108 0.112
>>>> 3.56 3.38 182.3 12.0 8.3 1.083 0.132 0.134
>>>> 3.38 3.23 136.5 10.4 7.7 1.048 0.155 0.151
>>>> 3.23 3.11 107.4 9.2 7.3 1.096 0.184 0.163
>>>> 3.11 3.00 91.0 8.7 7.3 1.044 0.215 0.201
>>>> All reflections 287.7 13.5 8.0 1.055 0.088 0.082
>>>>
>>>
>>> --
>>> Ethan A Merritt
>>> Biomolecular Structure Center, K-428 Health Sciences Bldg
>>> University of Washington, Seattle 98195-7742
>>>
>
|