BPCells is a package for high performance single cell analysis on RNA-seq and ATAC-seq datasets. It can analyze a 1.3M cell dataset with 2GB of RAM in under 10 minutes. This makes analysis of million-cell datasets practical on a laptop.
BPCells provides:
Additionally, BPCells exposes its optimized data processing infrastructure for use in scaling 3rd party single cell tools (e.g. Seurat)
BPCells is easiest to install directly from github:
remotes::install_github("bnprks/BPCells/r")
Before installing, you must have the HDF5 library installed and accessible on your system. HDF5 can be installed from your choice of package manager.
You will also need a C/C++ compiler either gcc >=8.0 (>=9.1 recommended), or clang >= 7.0 (>= 9.0 recommended). This corresponds to versions from late-2018 and newer. Older versions may work in some cases so long as they have basic C++17 support, but they are not officially supported.
Obtaining the HDF5 dependency is usually pretty straightforward on Linux
sudo apt-get install libhdf5-dev
sudo yum install hdf5-devel
conda install -c anaconda hdf5
apt
or yum
) when possible,
as it appears to give a slightly more reliable installation experience.Compiling R packages from source on Windows requires installing R tools for Windows. See Issue #9 for more discussion.
For MacOS, installing HDF5 through homebrew seems to be most reliable: brew install hdf5
.
Mac-specific troubleshooting:
sessionInfo()
, and seeing if it lists ARM or x86 under "Platform". BPCells tries to print informative error messages during compilation to help diagnose the problem. For a more
verbose set of information, run Sys.setenv(BPCELLS_DEBUG_INSTALL="true")
prior to remotes::install_github("bnprks/BPCells")
. If you still can't solve the issue with that additional information, feel free to file a Github issue, being
sure to use a collapsible section for the verbose installation log.
BPCells is an open source project, and we welcome quality contributions. If you are interested in contributing and have experience with C++, along with Python or R, feel free to reach out with ideas you would like to implement yourself. I'm happy to provide pointers for how to get started, my time permitting.
If you are unfamiliar with C++ it will be difficult for you to contribute code, but detailed bug reports with reproducible examples are still a useful way to help out. Github issues are the best forum for this.
If you maintain a single cell analysis package and want to use BPCells to
improve your scalability, I'm happy to provide advice. We have had a couple of labs
try this so far, with promising success. Email is the best way to get in touch
for this (look in the DESCRIPTION
file on github for contact info). Python
developers welcome, though the full python package will likely not be
available until after summer 2023.