Unexpected particle positions after template matching

[Lorenzo Gaifas]

Dear all,

We are trying to use template matching in Warp, and we encountered some unexpected results when visualizing the output.

Points and orientations were extracted from the relative .star file. Since in the file they are normalized between 0 and 1, we multiplied them by the size of the volume in each dimension. As you can see from the attached figure `volume.png`, this operation yields points that are within the confines of the cylindrical space occupied within the cross-correlation volume, so we expected this operation to be correct.

While the peaks in the cross-correlation volume clearly match the expected location of the structure in the tomogram, the particle positions in the starfile don't seem to correspond to the peaks in the CC volume (as shown in `peaks.png`).

On the other hand, the few points that ARE on the surface of interest seem to all be aligned perpendicularly to the surface, as we would expect (as (poorly) visible in in `zoomed.png`).

The overall particle positions (correctly within the volume, some aligned as expected) seem to imply that we're visualizing them correctly, but the fact that most of the points do not gather on the highest peaks does not match our expectations.

Does anyone have any experience/advice with this? There aren't many parameters to change.

Best,

Lorenzo

[Pranav Shah]

Hi Lorenzo,

we dont see the attached images. Could you please send another email with the images attached?

Best,

Pranav

[Lorenzo Gaifas]

Dear all,

It seems that the attachments did not come through to the group email. Here's another attempt at sending them.

Best,

Lorenzo

--
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/1d452a85-34b0-493b-9d7e-881e84c5dc68n%40googlegroups.com.

[teg...@gmail.com]

Dear Lorenzo,

To rule out any intermediate processing issues, can you open the correlation volume in Cube (dtegunov/cube: Particle picking for tomograms (github.com)), import Warp's result using a volume size of 1x1x1, and post a screenshot here? Please also note that Warp intentionally gives you (likely) too many positions with their correlation scores, so you can adjust the threshold and kick out false positives quickly rather than repeating the matching procedure. You can adjust the correlation score range in Cube's interface.

Cheers,

Dimitry

[alist...@gmail.com]

Hey Dimitry,

Here is the data opened in Cube - the highest peaks do indeed lie on our densely packed surface but there are only a couple of particles rather than the expected densely packed lattice positions

Cheers,

Alister

[Alister Burt]

Picture didn't want to attach, attempt 2...

To view this discussion on the web visit https://groups.google.com/d/msgid/warp-em/83748aa3-04a5-4ae6-8147-423be5b97564n%40googlegroups.com.

[teg...@gmail.com]

Thanks! Is it a particularly non-spherical particle? What did you use as diameter, compared to the inter-peak distances in the correlation volume? You can repeat the 'peak picking' part of the template matching procedure with a smaller diameter using the same dialog with 'reuse old correlation volumes if available' enabled.

Cheers,

Dimitry

[Alister Burt]

Very non-spherical indeed

We set the diameter to roughly the length of the long axis... will switch it to the short axis and possibly change where we center the template... cheers!

Alister

To view this discussion on the web visit https://groups.google.com/d/msgid/warp-em/1816ac9d-9d39-42c5-97c9-7284d781bd52n%40googlegroups.com.

[Lorenzo Gaifas]

Dear all,

We are still having some issues related to the diameter parameter in template matching.

As you might remember, if we set the diameter to a high value, the resulting points are very distanced.

However, if the diameter is set to a low enough value to match the spacing of our particles, the CC volumes look much, much worse.

Re-running the peak-picking part on the good cross-correlation volume obtained by calculation with a large diameter only partially solves the problem, because in doing that we lose orientation information.

As it stands, we have to use the 2-steps approach, and guess the orientations ourselves based on surface definitions.

It appears that the diameter parameter controls two different things: the distance between picked particles, and some sort of mask size for the cross correlation calculation, is this correct?

If so, is it possible to separately define these two parameters? Could you also let us know what is the meaning of the "use X% of template" parameter?

Best,

Lorenzo

To view this discussion on the web visit https://groups.google.com/d/msgid/warp-em/CAOt3kHqjr7QOCvJQYbE0V78xzFrZtb0FCme9kHeLCZOm%3DM1ibQ%40mail.gmail.com.

[teg...@gmail.com]

Hi Lorenzo,

look in the nightly build and you shall find!

Cheers,

Dimitry