Training instructions

These are instructions for generating a Deepbinner model using your own reads.

Some reasons you might want to train your own models:

• You use a sequencing/barcoding kit other than the ones covered by Deepbinner's pre-trained models.
• You have a very particular use case (e.g. only barcodes 1–4).

Requirements

In addition to the normal Deepbinner requirements, training also requires TensorFlow running on a GPU. While you technically could train a model using a CPU, it would likely take far too long to finish.

The training-related commands also require a few more Python packages than are installed with Deepbinner: noise, edlib and mappy, so pip install these before continuing.

Steps

1. Generating a training set. This is the most involved part. Follow one of these instructions, as appropriate:
   • Creating a native barcoding training set
   • Creating a training set from another kit
2. Optionally: Refining a training set
3. Training the neural network