Autocomplete CATMAID Neurons

This examples will illustrate how to use the Google autoseg to autocomplete a partially reconstructed CATMAID neuron.

Note

For this to work, you need to have CATMAID API write and import access

First initialize connections to the manual and autoseg CATMAID instances set up - make sure to replace HTTP_USER, HTTP_PW and API_TOKEN with your corresponding credentials:

import pymaid
manual = pymaid.CatmaidInstance('https://neuropil.janelia.org/tracing/fafb/v14',
                                api_token='API_TOKEN',
                                http_user='HTTP_USER',
                                http_password='HTTP_PW',
                                caching=False,
                                max_threads=20)

auto = pymaid.CatmaidInstance('https://spine.itanna.io/catmaid/fafb-v14-seg-li-200412.0',
                              api_token='API_TOKEN',
                              http_user='HTTP_USER',
                              http_password='HTTP_PW',
                              caching=False,
                              max_threads=20)

Next get the neuron to autocomplete (exchange the 16 for the skeleton ID of your neuron).

x = pymaid.get_neuron(16, remote_instance=manual)

Now get everything that overlaps with this neuron in autoseg. See fafbseg.google.find_autoseg_fragments() on how to fine tune this step.

import fafbseg
ol = fafbseg.google.find_autoseg_fragments(x, autoseg_instance=auto)

Now visualise the large overlapping fragments alongside the manual tracings:

x.plot3d(color='w')
ol[ol.n_nodes > 40].plot3d()

You can use the 3D viewer to select which fragments actually overlap with you neuron. For this, turn on the legend by pressing L (this might take a second) and then pressing P to enable picking. Now you can click on the legend entries to show/hide neurons.

Hide autoseg fragments you do not want to use to autocomplete your neuron and then run this:

import numpy as np
visible = ol[np.isin(ol.skeleton_id, navis.get_viewer().visible)]

Before we can start the merge process, we have to stitch all autoseg fragments to form a single virtual neuron for upload:

y = navis.stitch_neurons(visible, method='NONE')

If you want to have a final look this is how you can co-visualize the manual tracings and the to-be-merged autoseg fragments:

x.plot3d(color='w', clear=True)
y.plot3d(color='r')

Once you are ready start the upload process as described in Merging. (see also fafbseg.move.merge_neuron() for additional parameters):

resp = fafbseg.move.merge_neuron(y, target_instance=manual, tag='YOURTAG')

Gotchas

When looking for overlapping autoseg fragments, you can end up finding the autoseg version of your original neuron - x in above example. This happens if somebody has merged a Google skeleton into x.

This is problematic because fafbseg uses the skeleton ID to identify where new and old nodes originate from but now we have two neurons with the same skeleton ID. merge_neuron() will throw in error:

ValueError: Duplicate skeleton IDs found. Try manually assigning unique skeleton IDs.

To resolve this, you need to manually change the skeleton ID of y - ideally to that of the Google fragment that got merged into it: look for an annotation like Merged: Google: 5819659900 and then change the skeleton ID:

y.skeleton_id = '5819659900'
y._clear_temp_attributes()