baskerville-tf

Superseded by baskerville (PyTorch). This TensorFlow version is maintained for borzoi reproducibility.

Sequential regulatory activity predictions with deep convolutional neural networks.

baskerville-tf provides researchers with tools to:

1. Train deep convolutional neural networks to predict regulatory activity along very long chromosome-scale DNA sequences
2. Score variants according to their predicted influence on regulatory activity across the sequence and/or for specific genes.
3. Annotate the specific nucleotides that drive regulatory element function.

---

Documentation

Documentation page: https://calico.github.io/baskerville-tf/index.html

• Document page for transfer learning to hg38 tracks
• Document page for transfer learning to mm10 tracks

---

Installation

git clone git@github.com:calico/baskerville-tf.git
cd baskerville-tf
pip install .

The package installs as baskerville-tf and imports as baskerville. The command-line scripts live in src/baskerville/scripts; put them on your path:

export BASKERVILLE_DIR=/home/<user_path>/baskerville-tf
export PATH=$BASKERVILLE_DIR/src/baskerville/scripts:$PATH
export PYTHONPATH=$BASKERVILLE_DIR/src/baskerville/scripts:$PYTHONPATH

---

Contacts

Dave Kelley (codeowner)