charlesxu90 / rodan

A fully convolutional architecture for basecalling nanopore RNA sequencing data

RODAN

A fully convolutional architecture for basecalling nanopore RNA sequencing data

Published in BMC Bioinformatics

Generated Taiyaki RNA data: https://doi.org/10.5281/zenodo.4556884

RNA training and validation data: https://doi.org/10.5281/zenodo.4556950

RNA test data: https://doi.org/10.5281/zenodo.4557004

Requirements

• Python 3
• torch >= 1.4.0 <= 1.8.0
• numpy
• h5py
• ont-fast5-api
• fast-ctc-decode
• pyyaml
• tensorboard
• pytorch-ranger (only for training)

Installation

Create a python virtual environment.

python3 -m venv virtualenv
source virtualenv/bin/activate
git clone https://github.com/biodlab/RODAN.git
cd RODAN
pip install -r requirements.txt

Basecalling

To basecall (must be run from root directory):

./basecall.py /path/to/fast5files > outfile.fasta

Basecall will recursively search in the specified directory for all fast5 files which can be single or multi fast5 files.

Training

To train, download the RNA training data from the above link.

mkdir runs
pip install pytorch-ranger
./model.py -c rna.config -n NAME -l

Parameters

-c for configuration file
-l for label smoothing
-n the name for the run, the model weights, configuration, and results will be saved in the runs directory
-v verbose

Test data

Five samples of human RNA fast5 data is provided in test-data.tgz.

Memory errors

If you run out of memory, reduce the batch size with the basecaller with "-b 100" or lower. The default is 200.

License

MIT License.