Installation

For installation instructions in a cluster computing environment, see here

Only tested on Ubuntu 16.04 and 18.04

If required, install CUDA for GPU-support

Necessary for training cell classification, highly recommended for inference. For compatibility, we require CUDA 10.1 and cuDNN >=7.6, instructions can be found here. If you need other versions of CUDA (for other work), please see here for how to link CUDA versions to conda environments.

Alternatively, CUDA can be installed into your conda environment (see below).

Set up conda environment and install dependencies

Recommended, allthough not necessary Conda environments are used to isolate python versions and packages.

Once conda environment is set up, everything must be carried out in this environment

  • Download miniconda installation file here

  • Run miniconda installation file, e.g.:

bash ~/Downloads/Miniconda3-latest-Linux-x86_64.sh
  • Create and activate new minimal conda environment

conda create --name cellfinder python
conda activate cellfinder

Optional: Install CUDA and cuDNN into the conda environment:

This avoids the need to install CUDA and cuDNN system-wide, but the GPU drivers are still required

conda install cudatoolkit=10.1 cudnn

Install cellfinder

pip install cellfinder

If you use a conda environment, remember to activate it (conda activate cellfinder) before using cellfinder.

Download atlas and trained classification models (optional)

When cellfinder runs, the appropriate machine learning models and reference atlas will be downloaded (if not previously done so). If you would like to download in advance to save time (or if you will not have an internet connection) please use cellfinder_download.

Currently, the only supported atlas is the Allen reference mouse brain, originally from here. The atlas will download into ./cellfinder/atlas.

In the future, other atlases will be available, and you will be able to choose which is downloaded.

If you want to modify the cellfinder download, use:

  • --atlas-install-path Supply a path to download the atlas elsewhere. This should also update the default registration.conf file so that the correct atlas is sourced. Alternatively, use this command to tell cellfinder where an existing atlas is, to save it being downloaded twice. (Requires 20GB disk space)

  • --atlas download path The path to download the atlas into. (Requires 1.2GB disk space). Defaults to /tmp.

For developers

If you want to edit cellfinder, run tests and build the documentation etc, then use:

git clone --branch dev https://github.com/SainsburyWellcomeCentre/cellfinder
cd cellfinder
pip install -e .[dev]