bpreveal 5.2.1

BPReveal is a suite of tools for deep learning on genomics data.

BPReveal

BPReveal is a suite of tools for building and interpreting sequence-to-profile models of biological data. The model architectures are based on chrombpnet, which is in turn based on BPNet. It incorporates PISA analysis, which extracts pairwise interactions between every base in a model's input and each one of its outputs.

You can find the documentation here.

Components

BPReveal is designed as a set of command-line tools that can be flexibly combined for many analysis tasks. A precise specification for each tool can be found in the documentation.

You can find a tutorial for training a model on transcription factor data in mouse in this document

Organization

BPReveal follows a traditional Unix-style directory structure:

• bin contains the executables. This is added to your path when you use one of the provided scripts to build a conda environment.
• doc contains examples of using the package along with a formal specification of all the file types and programs.
  • doc/demos contains demonstrations of fun things you can do with BPReveal.
  • doc/presentations contains (you guessed it!) presentations related to BPReveal.
• pkg is a silly folder that is added to your python search path, letting you import bpreveal.utils without having to mess about with $PYTHONPATH. You can ignore it.
• src contains all the programs.
  • src/schematools contains the json schema that are used to validate the input to the programs.
  • src/internal contains source code that must be compiled using the f2py routines in numpy.
  • src/tools contains a hodgepodge of scripts that are useful for specific tasks. These are not actively maintained and tested, but may be useful.
• test contains the files that are used in the demonstration notebooks. Note that test is not committed to the git repository - it's many gigabytes of data files. If you're on the Stowers network, you can find this directory at /n/projects/cm2363/bpreveal/test.

Documentation

The readthedocs documentation is kept current, but you can easily get your own local copy. The BPReveal project uses Sphinx to generate documentation. If you want to build the documentation, make sure that INSTALL_DEVTOOLS=true in the conda install script, then run make html, make man, or make latexpdf in the doc directory. Documentation will be in doc/_build/html/index.html, doc/_build/man, or doc/_build/latex/bpreveal.pdf. If you use the buildConda scripts, then it will put the man pages on MANPATH but you need to make man in the doc/ directory to generate the pages.

If you're at Stowers, then the documentation will be included in the repositories that I maintain at /n/projects/cm2363/public-bpreveal//doc. You can use man bpreveal to get a list of the available man pages.

Installing

There are two ways to install BPReveal: pip or conda. For most users, pip will be easier, but the conda install offers more flexibility and is better if you're working on the BPReveal source code.

To install BPReveal with pip, download the wheel file from one of the releases on GitHub, and run pip install bpreveal-x.y.z-cp312-cp312-linux_x86_64.whl. I would recommend setting up a conda environment or other virtual environment, because BPReveal pulls in a lot of other dependencies. You will need Python 3.12 in this environment, which you can specify when you create it: conda create -n bpreveal python=3.12

To install BPReveal as a conda environment, EDIT and then run one of the buildConda scripts in the root directory. If you're using the Cerebro cluster at Stowers, run buildCondaCerebro.slurm. If you're installing BPReveal on a local machine, run buildCondaLocal.zsh. If you're at Stowers, you can just activate one of the conda environments I maintain at /n/projects/cm2363/public-bpreveal/(version)/env.

License

BPReveal is released under the GNU GPL, either version 2 of that license or (at your option) any later version. A copy of the license can be found in the COPYING file.