Development

As well as providing the useful functionality of creating consensus calls from basecall data, medaka demonstrates a framework for both training and inference. The code exploits the keras deep learning library.

A simple method presented to inspire further ideas is to align input sequences and a truth sequence to a common baseline (e.g. reads and a reference to a draft assembly), and extract ‘feature vectors’ for input into a neural network classifier.