refinement protocol

Submitted by mbycliaw on Fri, 2012-03-16 21:55

Forums

Frealign

Hi,

I am new for frealign. I took the dataset in EMAN2 workshop as a test case for practice. It is a groel dataset, which has already refined to 8.4A resolution. I used this model as initial model to do orientation search (mode -3; DANG=1), then I start from refinement resolution at 25A, 22A, 19A, 17A, 15A, 13A. (5 cycles for each resolution) But the FSC shown it goes down to about 12.5A at very beginning of refinement. then only increase resolution to 12.3A for total 30 cycles refinement. The model looks like resolution around 13 to 15A. I did try to reduce the PBC to 5 for all resolution or grduately reduce it, I also tried put RBFACT=-100. I also turn FCREF and FBEAUT to T. However, still no improvement.

I must miss something. 

data_input              GROEL
raw_images1             GROEL_stack
thresh_reconst          90.0
thresh_refine           15.0
pixel_size              2.11999988556
dstep                   10.0
WGH                     0.07
CS                      4.1
kV1                     300.0
radiusO                 140.5
radiusI                 0.0
PBC                     5
BOFF                    60.0
DANG                    1
MODE                    1
XSTD                    0.0
RBFACT                  0.0
FLIP                    F
FMAG                    F
FDEF                    F
FASTIG                  F
IEWALD                  0
res_reconstruction      6.36
res_low_refinement      50.0

The similar thing happen to my virus dataset. Also refine to 8.9A by EMAN2 and in frealign also go to 12.3A. The model also looks like around 15A resolution.

So I must miss something. Help!

niko Sun, 2012-03-18 14:44

From your post it is not clear how you know what the actual resolution of your map is. How do you know your EMAN reconstruction has a resolution of 8.4 Å? The FSC might give you an over-optimistic estimate due to over-refinement of your data. Please make sure the density of your reconstruction shows detail at the resolution you think you are at.

Please also remember that for most Frealign reconstructions, the appropriate FSC threshold is not 0.5 but 0.143 (see [1]). What is the Frealign resolution at FSC = 0.143?

Also, you should set the high-resolution limit of your refinement to an appropriate value. Many people set this value to the highest possible resolution (4.24 Å in your case). However, the signal at high resolution is so weak that it does not contribute much to the alignment while the noise might actually make the alignment worse. Try limiting the high-resolution limit to something like 12 – 15 Å.

Finally, when you look at the Frealign reconstruction, please apply an appropriate negative B-factor and resolution filter to sharpen the map and filter it at the correct resolution limit (e.g. FSC = 0.143). You can use the bfactor program from our web page to do this.

References


mbycliaw Sun, 2012-03-18 21:35

Hi Niko,

I judge the resolution not only by the FSC. The 8.4A from EMAN2 demo dataset show most of helix as tube, which I think it should be around that resolution range.

Great! I always look at FSC = 0.5. So I need to change for frealign. Ah! bfactor program. Learn one more thing!

Do you mean refinement resolution or reconstruction resolution? I did set refinement resoltuion to 12-15A.(res_refinement) But some people told me always set the reconstruction resolution to Nyguist frequency.(Res_recon) Do you mean it is wrong?

One more thing is how do we determine the low resolution limit? I saw your previous structures. they are 200, 50, 30. I can not find out what is the rule?

niko Sun, 2012-03-18 21:38

Yes, I meant the res_refinement value. The reconstruction should be calculated to the max resolution (Nyquist), unless you think you can make the program run faster by calculating the reconstruction to a lower resolution (usually this makes only sense with very larger images, e.g. 500 x 500 pixels).

Also make sure that the low-resolution limit in the refinement is set to 200 Å or so. In your previous post it looked like you set this to 50 Å.

mbycliaw Mon, 2012-03-19 04:41

Hi Niko,

Do you mean in refinement set low resolution to 200 and in reconstruction set to the size of particle?

Also I saw the the mparameter in BPV the low resolution is set to 20.0. But the size of BPV is about 570A in diameter, not 20 A. So what is the unit of the size you mean?

One more thing is in frealign there is a parameter is BFACTOR. Do we normally need to set it? of course, bfactor can fit the bfactor, which frealign can not do. If we set a value in frealign, BFACTOR = -100, does it affect the result of bfactor?

similar question is how to define the resolution range to fit b factor in bfactor program? I found change low resolution give different b factor.