Skeletons#

This tutorial will show you how to load and save skeletons.

Skeletons are probably the most common representation of neurons and are stored as a series of connected nodes (the "skeleton"). In NAVis, skeletons are represented by the navis.TreeNeuron class.

You can either construct these manually (see bottom of this page) or use one of the built-in functions to them from one of the various file formats:

From SWC files#

SWC is a common format for storing neuron skeletons. Thus NAVis provides functions to both read and write SWC files. To demo these, we will be using supplemental data from Bates, Schlegel et al. (Current Biology, 2020). If you want to follow along, please download Supplemental Data S1 (link). If you do, make sure to adjust the filepaths in the examples according to where you saved it to.

I extracted the archive with the supplemental data inside my downloads folder.

It contains a bunch of CSV files with meta data but the important file for us is the "skeletons_swc.zip". Now you could extract that zip archive too but NAVis can actually read directly from (and write to) zip files!

import navis
skeletons = navis.read_swc(
    'mmc2/skeletons_swc.zip',
    include_subdirs=True
)
skeletons

Let's say you are looking at a huge collection of SWC files and you only want to sample a few of them:

Load only the first 10 skeletons

sample = navis.read_swc(
    './mmc2/skeletons_swc.zip',
    include_subdirs=True,
    limit=10
)
sample

We can also point navis.read_swc() at single files instead of folders or zip archives:

For this I extraced the skeletons_swc.zip archive

s = navis.read_swc('./mmc2/swc/CENT/11519759.swc')
s

You can even use URLs or FTP servers directly:

# From URL:
s = navis.read_swc('https://v2.virtualflybrain.org/data/VFB/i/jrch/jup2/VFB_00101567/volume.swc')

# From an FTP folder:
nl = navis.read_swc('ftp://download.brainlib.org:8811/biccn/zeng/pseq/morph/200526/', limit=3)


# !!! tip
#     [`read_swc`][navis.read_swc] is super flexible and can handle a variety of inputs (file names, folders, archives, URLs, etc.).
#     Importantly, it also let you customize which/how neurons are loaded. For example:
#      - the `limit` parameter can also be used to load only files matching a given pattern
#      - the `fmt` parameter lets you specify how to parse filenames into neuron names and ids
#     Many of the other `navis.read_*` functions share these features!

To SWC files#

Now let's say you have skeletons and you want to save them to disk. Easy!

# Write a single neuron:
navis.write_swc(s, './mmc2/my_neuron.swc')

# Write a whole list of skeletons to a folder and use the neurons' `name` property as filename:
navis.write_swc(sample, './mmc2/{neuron.name}.swc')

# Write directly to a zip file:
navis.write_swc(sample, './mmc2/skeletons.zip')

# Write directly to a zip file and use the neuron name as filename:
navis.write_swc(sample, './mmc2/{neuron.name}.swc@skeletons.zip')

See navis.write_swc for further details!

From NMX files#

NMX is a xml-based format used e.g. by pyKNOSSOS to store skeletons plus meta data. NAVis supports reading (but not writing) this format. If you want to follow along download this dataset by Wanner et al. (2016). Just like the SWCs, I extracted the archive to my downloads folder:

Read a single file

s = navis.read_nmx('./WannerAA201605_SkeletonsGlomeruli/Neuron_id0001.nmx')
s

Read all files in folder

nl = navis.read_nmx('./WannerAA201605_SkeletonsGlomeruli/')
nl

navis.plot2d(nl[:10], method='2d', radius=False)

Out:

Plot neurons:   0%|          | 0/10 [00:00<?, ?it/s]

Plot neurons:  70%|#######   | 7/10 [00:00<00:00, 67.39it/s]


(<Figure size 640x480 with 1 Axes>, <Axes: xlabel='x', ylabel='y'>)

From Neuroglancer#

Among other formats, neuroglancer supports a "precomputed" format for skeletons (see specs here. This binary format is more compact than uncompressed SWC files but is not used outside of neuroglancer as far as I know. That said: NAVis lets you read and write skeletons from/to precomputed format using navis.read_precomputed and navis.write_precomputed. Note that these functions work on both precomputed skeletons and meshes.

Manual construction#

What if you have some obscure data format for which NAVis does not have a read function? The data underlying a navis.TreeNeuron is a simple SWC table - so as long as you can produce that from your data, you can create your own skeletons.

Here's a quick & dirty example:

import pandas as pd

# Create a mock SWC table for a 2-node skeleton
swc = pd.DataFrame()
swc['node_id'] = [0, 1]
swc['parent_id'] = [-1, 0]   # negative indices indicate roots
swc['x'] = [0, 1]
swc['y'] = [0, 1]
swc['z'] = [0, 1]
swc['radius'] = 0

swc

This SWC can now be used to construct a TreeNeuron:

s = navis.TreeNeuron(swc, name='my_neuron', units='microns')
s

There are a few other ways to construct a navis.TreeNeuron (e.g. using a graph) - see the docstring for details.

Also note that all NAVis neurons can be stored to disk using pickle - see the pickling tutorial.

Hopefully the above has given you some entry points on how to load your data. See also the I/O API reference.

Please also keep in mind that you can also convert one neuron type into another - for example by skeletonizing MeshNeurons (see also the API reference on neuron conversion).

Total running time of the script: ( 1 minutes 15.090 seconds)

Download Python source code: tutorial_io_00_skeletons.py

Download Jupyter notebook: tutorial_io_00_skeletons.ipynb

Gallery generated by mkdocs-gallery