northstar

Single cell type annotation guided by cell atlases, with freedom to be queer.

Brief description

northstar is a Python package to identify cell types within single cell transcriptomics datasets. northstar's superpower is that it learns from cell atlases but still allows queer cells to make their own cluster if they want to.

Also, northstar was heavily developed during Pride Month.

Atlas resources

Curated averages and subsamples from several atlases: https://northstaratlas.github.io/atlas_landmarks/

If you want us to add you cell atlas, open an issue on: https://github.com/northstaratlas/atlas_landmarks/issues

Documentation

https://northstar.readthedocs.io

Installation

pip install northstar

To automatically download and use our online atlas collection at https://northstaratlas.github.io/atlas_averages/, you will need to call:

pip install 'northstar[atlas-fetcher]'

Dependencies

• numpy
• scipy
• pandas
• scikit-learn
• anndata
• python-igraph>=0.8.0
• leidenalg>=0.8.0

It is recommended that you install python-igraph and leidenalg using pip. However, any installation (e.g. conda) that includes recent enough versions of both packages should work.

Optional deps to use our online atlases:

• requests
• loompy
• scanpy
• pynndescent (only useful if you use scanpy as well)

If you have scanpy installed, northstar will use it to speed up a few operations (PCA, graph construction). You can turn this off in two ways:

1. Uninstall scanpy is you don't need it for anything else, or
2. Set the environment variable NORTHSTAR_SKIP_SCANPY to anything except empty string, e.g. in a notebook:

import os
os.environ['NORTHSTAR_SKIP_SCANPY'] = 'yes'
import northstar as ns

(rest of the notebook/script)

Hot-swapping between the two modes (w or w/o scanpy) is not currently supported.

Usage

See the paper below or the documentation for detailed instructions and examples. The simplest way to use northstar is to classify a new single cell dataset using one of the available atlases, e.g. Darmanis_2015 on brain cells:

import northstar

# Choose an atlas
atlas_name = 'Darmanis_2015'

# Get a gene expression matrix of the new dataset (here a
# random matrix for simplicity)
N = 200
L = 50
new_dataset = pd.DataFrame(
    data=np.random.rand(L, N).astype(np.float32),
    index=<gene_list>,
    columns=['cell_'+str(i+1) for i in range(N)],
    )

# Initialize northstar classes
model = northstar.Averages(
        atlas='Darmanis_2015',
        n_neighbors=5,
        n_pcs=10,
        )

# Run the classifier
model.fit(new_dataset)

# Get the cluster memberships for the new cells
membership = model.membership

Citation

If you use this software please cite the following paper:

Fabio Zanini*, Bojk A. Berghuis*, Robert C. Jones, Benedetta Nicolis di Robilant, Rachel Yuan Nong, Jeffrey Norton, Michael F. Clarke, Stephen R. Quake. Northstar enables automatic classification of known and novel cell types from tumor samples. Scientific Reports 10, Article number: 15251 (2020), DOI: https://doi.org/10.1038/s41598-020-71805-1

License

northstar is released under the MIT license.

NOTE: The module leidenalg to perform graph-based clstering is released under the GLP3 license. You agree with those licensing terms if you use leidenalg within northstar.