Hi Tom,

thanks for your second take on it.

yes, you are right, I should have been a bit more clear on about what I am after. But I think now you described it quite correctly.

I am not trying to test a fitting algorithm neither am I developing one. I am indeed just after creating a simulated cryo-EM map based on a provided PDB model.
I wish to simulate a cryo-EM map that comes close to the experimental map of a given resolution X (corresponding to an experimental reconstruction that achieved a FSC resolution of X).

I have been talking to Steven Ludtke about it. He was mentioning a strategy to generate a more realistic cryo-EM map:

Quoting Steven Ludtke:
>>>
1) pdb2mrc or similar - generate a map at more or less full resolution ie, sampling =1 A/pix -> res = 2 A

2) apply a simulated CTF to the map. This is tricky. You don't want to apply a single defocus CTF to the map because any real map will consist of the average over many defoci (with phase flipping). So to do this well on a 3-D map, you need to compute theoretical CTF over a range of defoci, take the absolute value, then average them together before applying to the 3-D map. This is a critical step, because the low resolution filtration of the CTF has a profound impact on the behavior of many standard algorithms
3) add noise at an appropriate level. This must be done after applying the CTF because the noise is not really CTF filtered to a significant degree. If you applied an envelope function (B-factor or more complicated) as part of step 2, then you will need to find an appropriate envelope function for the noise as well (probably not the same one).  
 
perform the process above twice (same CTF, regenerate noise), then compute the FSC between the two maps. Check the FSC curve shape and cutoff to make sure it's reasonable. If not, adjust the noise model appropriately and try again. When you're happy with the FSC, average together the 2 simulated half-maps and that's your simulated noisy CryoEM map.
>>>

Somewhere along the way one should be able to control the resulting resolution.
Steve also mentioned this can be taken even further starting on simulated particle images and reconstructing those.

This is most likely all overkill to what I am trying to achieve. Additionally, the point you made about the visual appearance of such maps:

" The effects of Gaussian smoothing or simply truncating beyond a certain frequency will create a map that has much higher signal-to-noise then a real low resolution map.  You won't see the low resolution noise characteristic of an actual low resolution map.  I don't know if that would be visually apparent."

This is exactly the point I was trying to bring up. And I was wondering whether it is possible to simulate a true low resolution map. As heard from Steve Ludtke, it is and people are doing it.
Apart from this, would one even notice a difference between such a simulated low resolution map and a gaussian filtered map when looking at it in Chimera, (yes I am simply talking about to the human eye). Certainly, if the simulated resolution is quite low, one should not see any difference. However, at higher resolution the lower signal-to-noise may translate to the visual appearance of the map and renders features of the map different from the gaussian filtered high resolution map.
The effect might still be marginal and does not justify any of the more advanced efforts for simulating realistic low resolution maps, this is of course different when considering fitting algorithms that much mroe readily pick up on the signal-to-noise of the maps at any frequence regime.

Anyway, for what is worth, all this helped me quite a bit understanding the matter much better, thanks again for your input.

Best regards

Dennis



 




Quoting Tom Goddard <goddard@sonic.net>:

Hi Dennis,
 
  You need to be clearer about what you are trying to achieve
 
"So I guess my question is whether there is a good way to mimick/visualize a lower resolution experimental map using a higher resolution experimental map without considering a sophisticated noise model."
 
If you wanted to mimic a low resolution map for purposes of testing an algorithm (e.g. fitting, segmentation, ...) then the signal to noise at various frequencies would be important.  If you just want it to look like a low resolution map to the human eye (why?), well I don't know how much having very different signal-to-noise as a function of frequency will visually alter the appearance.  The effects of Gaussian smoothing or simply truncating beyond a certain frequency will create a map that has much higher signal-to-noise then a real low resolution map.  You won't see the low resolution noise characteristic of an actual low resolution map.  I don't know if that would be visually apparent.
 
Tom


On Jan 10, 2024, at 12:40 AM, Guillaume Gaullier <guillaume.gaullier@kemi.uu.se> wrote:

Hi Dennis,
 
A low-pass filter is what you want then: this doesn't dampen signal and noise, simply discards everything past a certain spatial frequency you choose. Just like the low-pass filter applied on-the-fly when using a map as a reference for a 3D refinement or classification job.
I don't know how this can be done in ChimeraX though.
 
Guillaume

 

From: Dennis Dannecker via ChimeraX-users <chimerax-users@cgl.ucsf.edu>
Sent: Wednesday, January 10, 2024 7:15:09 AM
To: Tom Goddard
Cc: chimerax-users@cgl.ucsf.edu
Subject: [chimerax-users] Re: Question regarding gaussian filtering
 
Hi Tom,

thank you vey rmuch for the fast and elaborate response.

I believe I understand it much better now and see why the resolution
per se is not what I am after.

My intention is to visualize/model the appearance of a lower
resolution map using a higher resolution map. While this should in
principal be possible (the other way around not) but would then
require an appropriate noise model.

So the gaussian filtering only dampens the amplitudes of the
frequencies in Fourier space. The dampening can be controlled by the
standard deviation of the used gaussian function. The dampening
affects signal and noise equally.

Does this correspond to a gaussian lowpass filter? Because I was
thinking about removing the high frequencies and setting the cutoff at
different frequencies then corresponding to different "resolutions".
While the resulting map probably has a higher signal-to-noise ratio at
lower frequencies (lower than the cutoff) than an experimental map of
a resolution that this map tries to depict it may still be a good
visual approximation.

So I guess my question is whether there is a good way to
mimick/visualize a lower resolution experimental map using a higher
resolution experimental map without considering a sophisticated noise
model.

Thanks again for your help.

Best
Dennis



Quoting Tom Goddard <goddard@sonic.net>:

> Hi Dennis,
>
>   I don't know the answer to your question, but here are some
> thoughts about it.  The Gaussian convolution filter just scales the
> Fourier space coefficients by a Gaussian making the higher
> frequencies have lower amplitudes.  That is reversible, the high
> frequencies can be scaled back up.  So although the real-space
> smoothed map looks lower resolution, it still has the same high
> resolution content.  The noise and the signal have both been scaled
> down at high resolutions so the signal-to-noise is just as good at
> those high resolutions.  If you did this smoothing on two half-maps
> used to compute the resolution by Fourier shell correlation it would
> still give 2 Angstroms, the smoothing did not change the FSC
> resolution!  If you were for example testing a fitting method it
> could easily extract that high resolution information and use it.
> In order to really lower the resolution you would need to add noise
> at high frequencies so the high-frequency signal is lost.
>
>   That said you could ask if the noise at high frequencies was not
> scaled down, only the signal, then what would the reduced resolution
> be.  We know at the FSC resolution the signal-to-noise ratio is some
> specific value that gives the cutoff correlation.  But you would
> need a model of the noise amplitude at other lower frequencies to
> know how attenuating just the signal but not the noise would change
> the resolution.  Maybe there is a pretty reliable noise model for
> these cryoEM maps, I'm not sure.
>
>       Tom
>
>
>> On Jan 9, 2024, at 2:41 PM, Dennis Dannecker via ChimeraX-users
>> <chimerax-users@cgl.ucsf.edu> wrote:
>>
>>
>> Hello,
>>
>> I would like to ask a question regarding the gaussian filtering.
>>
>> I filtered a given map (downloaded from the emdb with the "open"
>> command from chimerax) in chimerax using the following command:
>>
>> vop gaussian #1 sDev X
>>
>> with X taking different values ranging from 0.25 to 20 with different steps.
>>
>> Now I would like to know the resolutions these maps actually
>> correspond to after this gaussian filtering. The originally loaded
>> map is stated to have nominal resolution of 2 Angstroms (GSFSC).
>> sDev defines the standard deviation in Angstrom of the 3D Gaussian
>> function that is used for the filtering. How can one now calculate
>> the resolution these filtered maps correspond to? Is it possible to
>> calculate or state a resolution for these maps that is related to
>> the resolution given for the original map (here 2 Angstroms).
>>
>> Thank you very much.
>>
>> Kind regards
>>
>> Dennis
>>
>>
>>
>>
>> Dennis Dannecker
>> Master Student of Biochemistry
>> Department of Chemistry
>> Faculty of Mathematics and Natural Sciences
>> University of Cologne
>> Albertus-Magnus-Platz
>> 50923 Cologne, Germany
>> _______________________________________________
>> ChimeraX-users mailing list -- chimerax-users@cgl.ucsf.edu
>> To unsubscribe send an email to chimerax-users-leave@cgl.ucsf.edu
>> Archives:
>> https://mail.cgl.ucsf.edu/mailman/archives/list/chimerax-users@cgl.ucsf.edu/


--
Dennis Dannecker
Master Student of Biochemistry
Department of Chemistry
Faculty of Mathematics and Natural Sciences
University of Cologne
Albertus-Magnus-Platz
50923 Cologne, Germany

_______________________________________________
ChimeraX-users mailing list -- chimerax-users@cgl.ucsf.edu
To unsubscribe send an email to chimerax-users-leave@cgl.ucsf.edu
Archives: https://mail.cgl.ucsf.edu/mailman/archives/list/chimerax-users@cgl.ucsf.edu/

VARNING: Klicka inte på länkar och öppna inte bilagor om du inte känner igen avsändaren och vet att innehållet är säkert.
CAUTION: Do not click on links or open attachments unless you recognise the sender and know the content is safe.








När du har kontakt med oss på Uppsala universitet med e-post så innebär det att vi behandlar dina personuppgifter. För att läsa mer om hur vi gör det kan du läsa här: http://www.uu.se/om-uu/dataskydd-personuppgifter/

E-mailing Uppsala University means that we will process your personal data. For more information on how this is performed, please read here: http://www.uu.se/en/about-uu/data-protection-policy