defocus handedness WinWarp vs LinuxWarp

[Jesse Hansen]

hi all,

I am processing the same data in parallel in WinWarp vs LinuxWarp. I'm trying to treat them as identical as possible, including using AreTomo for both tilt series alignmnent.  

I have noticed two differences in the output:

1) The physical handedness appears to be flipped (Z flip/mirror) between WinWarp and LinuxWarp.  My understanding is that this should only happen if the tilts are processed in reverse order (ascending vs descending tilt order), however I checked the rawtilt files output from both Win/Linux WARP and they are identical.  So I am I am thoroughly confused how this might happen.  It is inconsequential I suppose, unless you want to import particle coordinates from Win Warp into Linux Warp.  

2) the defocus handedness check appears to give opposite results between win/linux Warp.  Could this be related to my first point, though I don't see how that could be? In win Warp the correlation seems to be quite strongly negative, I checked several tilt series and they all gave values around -0.8 to -0.9.  However when processing the same data in Linux Warp the correlation was about +0.5 to +0.6.  So I'm unsure of whether or not to flip the handedness.  

any insight would be appreciated.   

Jesse Hansen

p.s. is there any way to stop the fishing spam in this forum?

[Alister Burt]

Hi Jesse,

1. comes from differences in when tilt angles are imported (and potentially flipped). In desktop warp, stacks and corresponding tilt angles are always written out with angles from the Mdoc. In Linux warp, it’s possible to flip the angles during ts_import which then propagates to stack writing etc

2. Something appears off with your estimate if you’re seeing such weak correlation, unsure what you’re doing

Out of interest, why are you so invested in identical processing between the 1.X desktop version and 2.X? The code has changed quite a lot between these versions…

Cheers,

Alister

Sent from mobile - apologies for brevity

On Aug 22, 2024, at 11:09, Jesse Hansen <bluu...@gmail.com> wrote:



--
You received this message because you are subscribed to the Google Groups "Warp" group.
To unsubscribe from this group and stop receiving emails from it, send an email to warp-em+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/warp-em/dc34caeb-6a88-4de9-b437-c8c45a0b4046n%40googlegroups.com.

[Jesse Hansen]

thanks Alister,

1) makes sense!  Thanks!   Silly me, coming from SPA background didn't fully appreciate this option.

2) Good to know.  Wonder what could cause that.  I just carefully checked the commands I used to reach that point and I think they look fine.  The averaged 2D tilts look fine and the metrics (WarpTools filter_quality) seem comparable to desktop warp.   I used the same motion correction and ctf estimation params as was done in desktop warp.  tilt series alignment job seems pretty standard, and moreover when I reconstruct tomos after this they look quite nice.  The only difference I can think of that it could it be due to the gold fiducials? I had removed them in Win Warp but now with Linux Warp I am unable to do so. 

Anyways, I guess I will just try with and without flipping defocus handedness and see which one limits me more in resolution in the end.  

I may have phrased that a bit odd, I'm not specifically testing the parallel processing pipelines.   Both pipelines work for me but there are some issues I encounter with each, and it's helpful to compare.  My goal is to switch to Linux Warp.  

thanks!

Jesse 

[Jesse Hansen]

update.  I reprocessed all the data in Linux Warp and now the defocus handedness check is giving a stronger correlation.   It is ~0.9.   However it's still weird to me that this is the opposite from what I was getting in desktop Warp (which was -0.9).  I will go ahead and ignore what desktop Warp says, but it's still intriguing to me... 

best

Jesse

[Jesse Hansen]

Another update.  I've tried ts_import with and without --dont_invert flag.  With --dont_invert we end up with a tomostar that has the -60 movie labelled as as the -60 tilt, whereas when I remove the --dont_invert flag I end up with the -60 movie labelled as the +60 deg tilt.   Adding/removing this flag alters the physical/geomtric handedness of the output tomos.  

What I was initially confused about was the connection between the physical handedness and the defocus handedness.  What I noticed when I checked the defocus handedness of data imported with or without the dont_invert flag is that one had a positive correlation and the other had a negative one.   Thus it seems that defocus handedness is directly connected to physical handedness.  Sorry, until now I wasn't exactly sure what the connection was between these two things. Hopefully this helps somebody out there.