Picking Ribosomes#

In the following we will use template matching to to identify and isolate ribosomes from a tomogram using a density map as a template.

The tomogram TS_037.rec is part of the dataset EMPIAR-10988 and can be downloaded from the section titled 2. Reconstructed cryo-electron tomograms acquired with defocus-only (DEF) on S. pombe cryo-FIB lamellae. The template used is EMD-3228, which you can download by clicking here. Alternatively, you can use wget:

wget https://ftp.ebi.ac.uk/pub/databases/emdb/structures/EMD-3228/map/emd_3228.map.gz

Template and Mask Generation#

EMD-3228 is at a different sampling rate than the binned tomogram. Although match_template.py can perform resampling on the fly, its generally recommended to provide a template with the correct sampling rate. This can either be achieved using the API (see Density.resample) or using preprocess.py. The right box size depends on the use-case, but generally speaking, larger boxes enable more accurate frequency operations and are essential for oscillating filters like the CTF. As a rule of thumb, you can set the box size to two times the minimum enclosing box.

preprocess.py \
    -m emd_3228.map.gz \
    --sampling_rate 13.48 \
    --box_size 70 \
    --invert_contrast \
    -o emd_3228_resampled.mrc

Note

The cisTEM tool simulate is a good alternative for template generation.

We can use the GUI to create a template mask. Since we inverted the template, do yourself a favor and use the Invert Contrast filter on the template. Alternatively, you can do this programatically, given you know the required mask dimensions

from tme import Density
from tme.matching_utils import create_mask

mask = create_mask(
    mask_type="ellipse",
    radius=(13,13,13),
    center=(34.50,34.50,34.50),
    sigma_decay = 1,
    shape=(70,70,70)
)
mask = Density(mask, sampling_rate=13.48)
mask.to_file("emd_3228_resampled_mask.mrc")

Your mask and template should look similar to the projection below

Template Matching#

For demonstration purposes we are going to process a subset of the data. However, the procedure for the full tomogram would be identical.

from tme import Density

dens = Density.from_file("TS_037.rec")
dens.data = dens.data[100:400, 450:750, 150:450]
dens.to_file("TS_037_subset.mrc")

Executing the following code will run template matching, using the common 40 Ångstrom lowpass filter to avoid bias towards high-frequency components. If you are performing this analysis on your laptop, make sure to increase the angular sampling rate from 6 to 15. If you are on an M-series MacBook you can add --backend jax to make use of the integrated GPU.

match_template.py \
    -m TS_037_subset.mrc \
    -i emd_3228_resampled.mrc \
    --template_mask emd_3228_resampled_mask \
    --lowpass 40 \
    -a 6 \
    -n 4 \
    --no_centering \
    --use_gpu \
    -o output_default.pickle

Note

You can inspect the results in the GUI by clicking the Import Pickle button.

The figure below shows a lowpass-filtered representation of the tomogram subset on the left and the corresponding template matching scores on the right. Overall, the majority of ribosomes appear to be accounted for. However, the peaks are fairly wide, and some represent erroneous matches to membranes and gold markers.

../../_images/particle_picking_default.png

Parameter Comparison#

The following outlines common filtering approaches we have tried so you do not have to. Due to the dynamic ranges of the scores and unique color mapping in napari, we can not directly determine which filtering method is optimal. However, we do note that peak shape and amplitude relative to the background can be optimized by going beyond the trivial example shown above.

Bandpass