callumjcparr/guppy_basecaller repository overview

Once built locally you can run update for guppy basecaller to latest version

apt-get update apt-get upgrade

or if you wish to only upgrade specifically guppy apt-get --upgradeable

To list application release names that can be upgraded to newer version

This image also contains an install of ONT experimental basecaller trainer, taiyaki.

To use remember to source activate $ source /taiyaki/venv/bin/activate

Please refer to documentation from GitHub repository:

https://github.com/nanoporetech/taiyaki⁠

This has a walkthrough for training a basecaller that is unaware of modified bases. This would be useful for when truing to improve accuracy for PCR-based sequencing libraries where modification information is erased during PCR. There is an additional walkthrough for making a basecaller aware for your modification of choice. This requires more thought and work as you must define a ground truth for your data that will be used in the training, i.e. you must with some high confidence define which bases are modified and which are not. If you are not working with IVT samples you may consider other software such as tombo to first tell you which bases are non-canonical (are modified), which can be used to label your training set.

These two programs are bundled in this one image because that benefit each other with guppy necessary to first create a reference dataset for taiyaki. Models trained in taiyaki can in theory be easily transferred to guppy using python script to convert your model to a .json file for guppy.

UPDATE: version 3 of this image has an important modification of the qscore.py script found in taiyaki. This now allows you to output fastq files with GPU. This would previously throw an error forcing CPU use which is around 10x slower. Note fastq output is not yet compatible with modbase output thus can only be use with canonical AGCT alphabet model files.