Basset

Deep convolutional neural networks for DNA sequence analysis.

Basset provides researchers with tools to:

1. Train deep convolutional neural networks to learn highly accurate models of DNA sequence activity such as accessibility (via DNaseI-seq or ATAC-seq), protein binding (via ChIP-seq), and chromatin state.
2. Interpret the principles learned by the model.

Read more about the method in the manuscript here:

DR Kelley, J Snoek, JL Rinn. Basset: learning the regulatory code of the accessible genome with deep convolutional neural networks. Genome Research 26 (7), 990-999.

As well as follow up work here:

DR Kelley, YA Reshef, M Bileschi, D Belanger, CY McLean, J Snoek. Sequential regulatory activity prediction across chromosomes with convolutional neural networks. Genome Research 28 (5), 739-750

Which has an associated repo where continued development on this toolkit now occurs. You can now run Basset-style peak prediction using Basenji, and I recommend using that software because I can better support it. See here..

---

Installation

Basset has a few dependencies because it uses both Torch7 and Python and takes advantage of a variety of packages available for both.

First, I recommend installing Torch7 from here. If you plan on training models on a GPU, make sure that you have CUDA installed and Torch should find it.

For the Python dependencies, I highly recommend the Anaconda distribution. The only library missing is pysam, which you can install through Anaconda or manually from here. You'll also need bedtools for data preprocessing. If you don't want to use Anaconda, check out the full list of dependencies here.

Basset relies on the environmental variable BASSETDIR to orient itself. In your startup script (e.g. .bashrc), write

    export BASSETDIR=the/dir/where/basset/is/installed

To make the code available for use in any directory, also write

    export PATH=$BASSETDIR/src:$PATH
    export PYTHONPATH=$BASSETDIR/src:$PYTHONPATH
    export LUA_PATH="$BASSETDIR/src/?.lua;$LUA_PATH"

To download and install the remaining dependencies, run

    ./install_dependencies.py

Alternatively, Dr. Lee Zamparo generously volunteered his Docker image.

To download and install additional useful data, like my best pre-trained model and public datasets, run

    ./install_data.py

---

Documentation

Basset is under active development, so don't hesitate to ask for clarifications or additional features, documentation, or tutorials.

• File specifications
  • BED
  • Table
  • HDF5
  • Model
• Preprocess
  • preprocess_features.py
  • seq_hdf5.py
  • basset_sample.py
• Learning
  • basset_train.lua
  • basset_test.lua
  • basset_predict.lua
• Visualization
  • basset_motifs.py
  • basset_motifs_infl.py
  • basset_sat.py
  • basset_sat_vcf.py
  • basset_sad.py

---

Tutorials

These are a work in progress, so forgive incompleteness for the moment. If there's a task that you're interested in that I haven't included, feel free to post it as an Issue at the top.

• Preprocess
  • Prepare the ENCODE and Epigenomics Roadmap compendium from scratch.
  • Prepare new dataset(s) by adding to a compendium.
  • Prepare new dataset(s) in isolation.
• Train
  • Train a model.
• Test
  • Test a trained model.
• Visualization
  • Study the motifs learned by the model.
  • Execute an in silico saturated mutagenesis
  • Compute SNP Accessibility Difference profiles.