compared to how thick the sample is. For neuronal branches that
are only a few 100 nm thick, setting the parameter to 800 pixels
should be enough. Select “Generate Tomogram” and view once
finished. If the whole volume of your target is not contained within
the tomogram, reset the tomogram thickness to a higher value and
repeat until the entire object is captured in the reconstruction.
Select “Done”.
3.7.8 Post-processing
If there is too much empty space after reconstruction, use the
“Volume Trimming” section to set the X, Y, and Z ranges for
trimming the volume. If the tomograms will be used for automated
segmentation, uncheck the “convert to bytes” box. EMAN2 (used
in the “Segmentation” section below) typically displays the
non-converted images better. Check the “Rotate Around X Axis”
box in the “Reorientation” section. The rotated volume will open
displaying the x-y orientation automatically within 3dmod and
EMAN2. This reorientation will not overwrite the original file
(which will be denoted by a “_full” suffix), but it does double the
amount of data produced in the reconstruction process. Select
“Trim Volume” and view the volume once finished (see Figs. 4d
and 6a). Select “Done”. Choose the “Clean Up” tab. Delete any
intermediate/excess files (see Note 17). Select “Done”.
3.8 Automated
Segmentation
with EMAN2
This section outlines a basic workflow for using EMAN2.31’s convolutional neural networks for automatically segmenting both
membranes and microtubules from cryotomograms of neurons
[7] although the segmentation of other cellular features would
follow the same basic protocol (see Note 18). The training process
is supervised and each cellular component will need a separate,
specifically trained network for identifying/segmenting this component. That being said, using the default preprocessing settings
(a bandpass filter) to remove high and low frequency noise from the
image, worked well in these examples. Exploring different frequency ranges within your image may prove fruitful for increased
fidelity (see Note 18). In this protocol, all options are left as default,
unless otherwise noted.
1. Open the EMAN2 project manager using the command
“e2projectmanager.py”. The command terminal window will
provide updates to the progress of each step described below.
2. Select the “Tomo” workflow mode.
3. Select “Raw Data”, then “Import Tomograms”. Browse and
select the desired tomogram file. Select “Launch”. In the
“Segmentation” tab, select “Preprocess tomograms”. Browse
to the imported tomogram. It should have a .hdf file extension.
Select “Launch”.
40
Ryan K. Hylton et al.
are only a few 100 nm thick, setting the parameter to 800 pixels
should be enough. Select “Generate Tomogram” and view once
finished. If the whole volume of your target is not contained within
the tomogram, reset the tomogram thickness to a higher value and
repeat until the entire object is captured in the reconstruction.
Select “Done”.
3.7.8 Post-processing
If there is too much empty space after reconstruction, use the
“Volume Trimming” section to set the X, Y, and Z ranges for
trimming the volume. If the tomograms will be used for automated
segmentation, uncheck the “convert to bytes” box. EMAN2 (used
in the “Segmentation” section below) typically displays the
non-converted images better. Check the “Rotate Around X Axis”
box in the “Reorientation” section. The rotated volume will open
displaying the x-y orientation automatically within 3dmod and
EMAN2. This reorientation will not overwrite the original file
(which will be denoted by a “_full” suffix), but it does double the
amount of data produced in the reconstruction process. Select
“Trim Volume” and view the volume once finished (see Figs. 4d
and 6a). Select “Done”. Choose the “Clean Up” tab. Delete any
intermediate/excess files (see Note 17). Select “Done”.
3.8 Automated
Segmentation
with EMAN2
This section outlines a basic workflow for using EMAN2.31’s convolutional neural networks for automatically segmenting both
membranes and microtubules from cryotomograms of neurons
[7] although the segmentation of other cellular features would
follow the same basic protocol (see Note 18). The training process
is supervised and each cellular component will need a separate,
specifically trained network for identifying/segmenting this component. That being said, using the default preprocessing settings
(a bandpass filter) to remove high and low frequency noise from the
image, worked well in these examples. Exploring different frequency ranges within your image may prove fruitful for increased
fidelity (see Note 18). In this protocol, all options are left as default,
unless otherwise noted.
1. Open the EMAN2 project manager using the command
“e2projectmanager.py”. The command terminal window will
provide updates to the progress of each step described below.
2. Select the “Tomo” workflow mode.
3. Select “Raw Data”, then “Import Tomograms”. Browse and
select the desired tomogram file. Select “Launch”. In the
“Segmentation” tab, select “Preprocess tomograms”. Browse
to the imported tomogram. It should have a .hdf file extension.
Select “Launch”.
40
Ryan K. Hylton et al.