CellBender

CellBender is a software package for eliminating technical artifacts from high-throughput single-cell RNA sequencing (scRNA-seq) data.

The current release contains the following modules. More modules will be added in the future:

• remove-background:

  This module removes counts due to ambient RNA molecules and random barcode swapping from (raw) UMI-based scRNA-seq count matrices. Also works for snRNA-seq and CITE-seq.

Please refer to the documentation for a quick start tutorial.

WARNING:

The release tagged v0.3.1 included a bug which caused output count matrices to be incorrect. The bug, introduced in #303, compromised output denoised count matrices (due to an integer overflow) and would often show up as negative entries in the output count matrices. The bug also existed on the master branch until #347.

We recommend using v0.3.2 or higher. The codebase has been revitalized and updated in v0.4.0, and we recommend using that or higher for all new analyses. The algorithm remains the same as in the paper for versions v0.3.0, v0.3.2, and v0.4.0. Completely identical, just codebase updates. Version v0.5.0 will use a very slight tweak in MKCP estimation to improve compute efficiency and reduce memory usage, but again the algorithm is the same.

As of CellBender v0.4.0, CellBender is compatible with python 3.10+, and can be installed via

$ pip install cellbender

(and we recommend installing in its own conda environment to prevent conflicts with other software).

CellBender is run as a command-line tool, as in

(cellbender) $ cellbender remove-background \
      --cuda \
      --input my_raw_count_matrix_file.h5 \
      --output my_cellbender_output_file.h5

See the usage documentation for details.

A GPU-enabled docker image is available from the Google Container Registry (GCR) as:

us.gcr.io/broad-dsde-methods/cellbender:latest

Available image tags track release tags in GitHub, and include latest, 0.3.0, 0.3.2, and 0.4.0.

A workflow written in the workflow description language (WDL) is available for CellBender remove-background.

For Terra users, a workflow called cellbender/remove-background is available from the Broad Methods repository.

There is also a version available on Dockstore.

Create a conda environment and activate it:

$ conda create -n cellbender python=3.12
$ conda activate cellbender

Clone this repository to your local machine.

(cellbender) $ git clone https://github.com/broadinstitute/CellBender.git

From the repo root, run

(cellbender) $ make install

Ensure that your installation is appropriate for your hardware (i.e. that the relevant CUDA drivers get installed and that torch.cuda.is_available() returns True if you have a GPU available.

This installs CellBender with dev dependencies (in editable -e mode):

This can be achieved via

(cellbender) $ pip install --no-cache-dir -U git+https://github.com/broadinstitute/CellBender.git@<SHA>

where <SHA> must be replaced by any reference to a particular git commit, such as a tag, a branch name, or a commit sha.

If you use CellBender in your research (and we hope you will), please consider citing our paper in Nature Methods:

Stephen J Fleming, Mark D Chaffin, Alessandro Arduini, Amer-Denis Akkad, Eric Banks, John C Marioni, Anthony A Phillipakis, Patrick T Ellinor, and Mehrtash Babadi. Unsupervised removal of systematic background noise from droplet-based single-cell experiments using CellBender. Nature Methods, 2023. https://doi.org/10.1038/s41592-023-01943-7

If you're unable to access that, see also our preprint on bioRxiv.