Installation

Requirements

This program requires just a few libraries. I provide a buildCondaLocal.zsh script that you can run on your own machine to automatically build an environment, or you can install the dependencies yourself with conda or pip.

python >= 3.12

This project uses features that were introduced in Python 3.11 heavily, so earlier Python versions will not work.

tfmodisco-lite

Used to identify motifs. (You may need to install cmake to build the wheel for modiscolite.)

pyBigWig

Used to read in data files.

pysam

used to read in fasta files.

pyBedTools

Used to read in region files. You will also need to make sure you have BedTools installed to use pyBedTools.

tensorflow 2.16

Does the heavy work of machine learning. Tensorflow 2.16 added a bunch of crummy backwards-incompatible behavior and some of the internal API had to be restructured, this was done in BPReveal 4.2.0.

tensorflow-probability

Used to create the multinomial loss function.

matplotlib

Used to make pretty plots.

tqdm

Used to create progress bars.

h5py

Used to read and write data files.

jsonschema

Used to validate the json files given to the Main CLI programs.

gcc, gfortran, meson

You’ll need a C compiler and a Fortran compiler to build the Jaccard and ushuffle libraries, and as of Python 3.12 you also need the meson build system.

While not strictly necessary, the following packages are very useful:

jupyterlab

For interactive data wrangling.

pandoc

Used to export Jupyter notebooks as pdfs.

pydot, graphviz

These are only necessary to use showModel.py, which is deprecated and will be removed in BPReveal 6.0.0.

Additionally, for development, there are a few libraries that I find useful:

flake8, pydocstyle, pylint

Used to check the code for style issues.

sphinx, sphinx_rtd_theme, sphinx-argparse, sphinx-autodoc-typehints

Used to generate the documentation you’re reading right now.

coverage

Used to check code coverage.

Setup

Once you have the requisite packages, you need to do a bit more setup to get BPReveal fully installed. If you use the buildCondaLocal.zsh script, you can skip this because the script takes care of this part.

  1. In the src/ directory of BPReveal, run make clean && make to build the Jaccard and ushuffle libraries. This step also generates the json schemas that are used to check your inputs for bugs.

  2. Add the bin/ directory to your path. This will be something like:

    export PATH=/path/to/bpreveal/bin:$PATH

  3. Add the pkg/ directory to your python path. The easy way to do this is with conda develop:

    conda develop /path/to/bpreveal/pkg

  4. Set Tensorflow to use its old internal Keras version:

    export TF_USE_LEGACY_KERAS=1

  5. (Optional) Build the documentation in the doc/ directory:

    make html latexpdf