Reduce bfactor sharpening noise

[Zach Pearson]

comment:9Changed 9 hours ago by Tristan Croll
Comparing against sharpening by Refmac, it's pretty clear that the factor of three should be in there. Looks like the Wikipedia entry is wrong - see equations 5 and 6 at ​​https://www3.cmbi.umcn.nl/bdb/theory/. Will discuss that in this morning's group meeting.
A side note: the B=-33.33 map sharpened by ChimeraX has some weird noise running through it that isn't in the Refmac one (the small jagged spots - they're quite high intensity and scattered in groups throughout the map). Something a bit funky about the Numpy FFT?

comment:10Changed 9 hours ago by Tristan Croll
... and the more I read that, the more I'm confused as to who's wrong. But wrong or not, that's how everyone seems to define it (all the major packages seem to agree on what a B=-100 sharpening value means).

comment:11Changed 4 hours ago by Tristan Croll
The high-resolution noise in the B=-33.33 map is probably not a bug, but rather lack of a feature. Not absolutely sure about Refmac (but CCP4 and Phenix tend to follow each other pretty closely), but the documentation for the equivalent in Phenix, phenix.auto_sharpen (​​https://phenix-online.org/documentation/reference/auto_sharpen.html) says that it only applies the sharpening out to the "headline" resolution of the map (as specified by the user). By default everything past that point is blurred with a B-factor of 200 (if that option is turned off it's left as-is - i.e. still not sharpened). Makes sense - most cryo-EM maps are quite heavily over-sampled, making a lot of very low-intensity, high-res "information" that you don't want to magnify.

comment:12Changed 82 minutes ago by Tom Goddard
Ok, so Wikipedia differs by a factor of 3 from ​​https://www3.cmbi.umcn.nl/bdb/theory/ and if the latter source is correct that would mean ChimeraX sharpening 3 times to much when B-factor is specified. Given your comparison to Refmac that appears to be the situation and is easy to fix. Zach can do that, and make a comment in the code as to the url to the B-factor equation being used.
The not sharpening beyond the map resolution makes sense -- usually the maps are oversampled by a factor of 3x. So when ChimeraX sharpens all the way out to the full oversampled resolution it is going to produce lots of nasty noise. I'd be surprised if the simply truncate the sharpening abruptly, ie have a discontinuous jump in the Fourier space scaling at a certain radius. I think that is likely to cause artifacts and I would suspect there is actually some roll-off in the transition. We may need to look at the Phenix code to see exactly what it is doing. We could add another option to specify the sharpening cutoff if we know how to transition the Fourier space scaling.

comment:13 in reply to: 15Changed 43 minutes ago by Tristan Croll
Relevant code used by phenix.auto_sharpen is at ​https://github.com/cctbx/cctbx_project/blob/fcff45919ee0b5cdf9f12cc9ea9d2f9efa9a0681/cctbx/maptbx/segment_and_split_map.py#L2810. Description of the k_sharpen parameter is at ​https://phenix-online.org/documentation/reference/auto_sharpen.html:

Steepness of transition between sharpening (up to resolution ) and not sharpening (d < resolution). Note: for blurring, all data are blurred (regardless of resolution), while for sharpening, only data with d about resolution or lower are sharpened. This prevents making very high-resolution data too strong. Note 2: if k_sharpen is zero or None, then no transition is applied and all data is sharpened or blurred.

If in doubt: this was all written by Tom Terwilliger (tterwilliger@…).
​https://phenix-online.org/documentation/reference/%22https://i.ytimg.com/vi/S56rYoZWxnM/sddefault.jpg%22<​https://phenix-online.org/documentation/reference/auto_sharpen.html>
Auto-sharpening cryo-EM or crystallographic maps with auto_sharpen - Phenix<​https://phenix-online.org/documentation/reference/auto_sharpen.html>
Specific limitations and problems: Maps produced with the extract-unique option of map_box should not be sharpened with auto-sharpen. These maps are closely masked around the density of a single molecule and are set to zero in much of the map, so the information about noise in the map that normally is available is missing and auto-sharpen does not work properly.
phenix-online.org