Consider storing per-base coverage info in uint16_t

[iqbal-lab]

1. Genome sequencing depth is generally <60x for big genomes, <100x for bacterial genomes and <10000x for viral genomes. Repetitive copy number is (roughly) highest in big, and lowest in viral genomes.
2. We cannot draw good inference in repeat regions, so either they will be low confidence or even masked.

So, we might as well store coverage with a cap, and above that ceiling we don't count. eg use uint16_t, with max 65536.

We did something like this in cortex here: https://github.com/iqbal-lab/cortex/blob/59658afe054e3a4b3854dc57954547be5383a9e9/src/cortex_var/many_colours/element.c#L421

• except that was bigger than uint16, but the principle is the same

This could save us a lot of RAM.

[iqbal-lab]

This has been discussed previously, just wanted an issue so we don't lose it. There might easily be more sophisticated things we might do, either using SDSL or using uint16_t during initial quasimap, and then a larger int when re-mapping (but not at all sure the latter is worth it)

[ffranr]

Implemented here: https://github.com/iqbal-lab-org/gramtools/commit/442ddfe7d68fca86ad97a123792673edff815b04

Leaving issue open for a few days.

[iqbal-lab]

hey this is exciting