Using a deep learning CNN+RNN+CTC structure to establish end-to-end basecalling for the nanopore sequencer. Built with TensorFlow and python 2.7.

If you found Chiron useful, please consider to cite:

Teng, H., et al. (2017). Chiron: Translating nanopore raw signal directly into nucleotide sequence using deep learning. [bioRxiv 179531] (https://www.biorxiv.org/content/early/2017/09/12/179531)

Chiron requires the following packages:

• Tensorflow >= 1.0.1
• numpy >= 1.13.1
• h5py >= 2.6

Follow this guide for installing Tensorflow, if possible with GPU support. You can use pip for installing numpy and h5py:

pip install numpy
pip install h5py

To install Chiron, first clone the repository to your local machine:

git clone https://github.com/cangermueller/Chiron.git

To install the Chiron package for code development, execute the following command in the Chiron root directory:

cd Chiron
python setup.py develop

In this way, changes to the source code are visible right away.

To test the installation and train Chiron on a small dataset, decompress the example data and execute the example script:

cd ./examples
tar xf ./data.tar.gz
cd ./train
./train.sh

Chiron provides a train, call, and export command for training models, base calling, and exporting signal/label files from fast5 files.

chiron {train,call,export} [FLAGS]

You can get more information about the different commands by using the --help flag, e.g. chiron train --help, or reading the official Chiron guide.

• Christof Angermueller
• [email protected]
• https://cangermueller.com
• @cangermueller