Possible proposal: output 1 matrix per "channel"?

[cmdcolin]

Currently, the matrix outputted by bilby encoder is a (12*X) x Y matrix, where:

• X is the number of BAM files inputted
• Y is the 'window size'. 12 is the number of "channels" that are predicted (6 per strand)

and the output file has a series of 12 channels, stacked against each other in a single matrix.

I was wondering if instead we output 12 different X - Y matrices (e.g. 1 matrix pr "channel")?

That way, the M->N transition would have one matrix, M->M would be another matrix, N->N would be another, etc.

This could be done through code changes to directly output, or as a post processing step on the current output

Possible points of reference:

• Using different channels containing different attributes is used as input for DeepVariant prediction https://github.com/google/deepvariant/blob/r1.6.1/docs/images/inference_flow_diagram.svg
  This is similar to the different channels we have.
• Can also sort of imagine a comparison to the different R,G,B channels in image analysis