Obtained a strange pickle file

[cyychong]

Hi, I ran pyTME on one of my tomograms and got a pickle file where the peaks match but it has a strange kind of "inverted" look (please see image attached).

I have gotten pyTME to work on my old data before and this time I ran it on a set of tomograms newly reconstructed on IMOD 5.2.2. I noticed that the min and max densities differ from my template + if I fiddle with those to better match it to my template MRC file, then I get the "correct" pickle file. I was wondering if there is way to somehow normalize the tomogram densities on pyTME?

Thanks in advance for any help!

Here is the command I ran:
Running the following command using gpuid:0
match_template.py --target ./tom/20241106_33_ts_004_rec_deconv.mrc --target_mask ./tom/20241106_33_ts_004_rec_deconv_mask-clipped.mrc --template /fast/pytlion/ref_msk/ref_cyl.mrc --template_mask /fast/pytlion/ref_msk/msk_cyl.mrc -a 15 --wedge_axes 0,2 --tilt_angles ./tom/20241106_33_ts_004.tsv --no_centering --use_gpu --memory_scaling 0.85 --temp_directory ./tmp/ --gpu_indices 0 --output 20241106_33_ts_004_rec_deconv.pickle

---

•                             pytme v0.2.5                                 *
  

---

Target

• Initial Shape: (500, 1440, 1022)
• Sampling Rate: (6.8, 6.8, 6.8)
• Final Shape: (500, 1440, 1022)

Target Mask

• Initial Shape: (500, 1440, 1022)
• Sampling Rate: (6.8, 6.8, 6.8)
• Final Shape: (500, 1440, 1022)

Template

• Initial Shape: (36, 36, 36)
• Sampling Rate: (6.8, 6.8, 6.8)
• Final Shape: (36, 36, 36)

Template Mask

• Inital Shape: (36, 36, 36)
• Sampling Rate: (6.8, 6.8, 6.8)
• Final Shape: (36, 36, 36)

---

Template Matching

• Angular Sampling: 15.0 [1992 rotations]
• Center Template: False
• Scramble Template: False
• Invert Contrast: False
• Extend Fourier Grid: False
• Extend Target Edges: True
• Interpolation Order: 3
• Setup Function: <function 'tme.matching_scores.flcSphericalMask_setup'>
• Scoring Function: <function 'tme.matching_scores.corr_scoring'>

Computation

• Backend: <class 'tme.backends.cupy_backend.CupyBackend'>
• Compute Devices: CPU [1], GPU [1]
• Use Mixed Precision: False
• Assigned Memory [MB]: 12856.0 [out of 15125.0]
• Temporary Directory: ./tmp/
• Target Splits: 0:2, 1:2, 2:2 [N=8]

Filters

• Smooth Pass: True
• Pass Format: sampling_rate
• Wedge Axes: (0, 2)
• Tilt Angles: ./tom/20241106_33_ts_004.tsv
• Extend Filter Grid: False

Analyzer

• Analyzer: <class 'tme.analyzer.MaxScoreOverRotations'>
• score_threshold: 0
• number_of_peaks: 1000
• min_distance: 12
• use_memmap: False

---

Distributing 8 splits on 1 job each using 1 core.
Running Template Matching. This might take a while ...

---

Runtime real: 941.826s user: 941.826s.

[maurerv]

Hi @cyychong,

It sounds like the contrast of your template is inverted relative to your tomogram, i.e., the template consists of positive densities, whereas your template of interest in the tomogram consists of negative densities. You can use the flag --invert_target_contrast to rescale the tomogram accordingly :)

[cyychong]

Hi @maurerv , thanks for your quick reply! I did think of that first but the flag didn't work. I also tried inverting the contrast of my template itself but that didn't change the outcome either.

[maurerv]

Are the scores actually negative at positions you expect your template to match? And when you say the flag did not work, do you mean match_template failed with some error, or that the scores looked the same?

Just as a side note, if your mask is not spherical, you probably get better results -s FLC (https://kosinskilab.github.io/pyTME/quickstart/matching/summary.html#scoring-table), but that would not affect the inversion issue you are describing

[cyychong]

Thanks for the comment about spherical mask! I hadn't caught that -- definitely makes more sense to use the FLC scoring function, rather than the one for spherical mask.

In the pickle file image I attached above, the regions where the pixels are blue give negative scores (e.g. -77), whilst the red regions = 0. In-between colours e.g. yellow and orange give values closer to 0 i.e. -44, -30..

The flag did not work --> I meant that match_template ran to completion with no errors but I still get the orange-background-blue-peaks pickle file. From this "inverted" pickle file, the regions with "blue peaks" actually do match my densities so it seems like the template matching is working?

One of the hypothesis we have been testing is whether the range of the tomogram density values (i.e. if you header a tomogram and read the min/max density values) might be causing this because that was the only difference we could find between the tomogram that produced the "inverted" pickle file versus one that produced the correct pickle file. However, I am having a hard time reproducing this error today and now I seem to be getting the correct blue-background-red-peaks pickle file. I don't have any other clues to follow and was thinking to just keep an eye out for it in case it happens again?

[cyychong]

Hi @maurerv , we finally solved this issue! We obtained this orange-background-blue-peaks pickle file when my target mask file is in map mode 0, whilst all the other MRC files are in either map mode 1 (my target) or map mode 2 (template, template mask).

[maurerv]

Hi @cyychong, thanks for digging into this! Sorry for the delay on my end. I will investigate this and get back to you :)