Seqkit backend and other improvements

[bsipos]

Dear rampart developers,

Karhide and I did some tweaks to the backend and the GUI adding the following features:

• Instead of parsing the PAF output generated by minimap2, the patched pipeline generates BAM, which is piped into seqkit bam to generate a rich TSV output which has all the information needed by the frontend pre-calculated, plus additional information which can be useful for diagnostics.
• The definition of "identity" (disregarding indels) was changed to "accuracy" (based on both substitutions and indels) as it is more widely used in the community.
• We patched the GUI to include histograms with additional information provided by seqkit bam - the changes to the GUI are more like a proof of concept, so feel free to suggest a more practical layout.
• The demux_map Snakemake pipeline was tweaked to allow for thread parameters here and there.
• Added a script under utils for quick (re)deployment of the pipeline in the artic-rampart conda environment.

Please merge if you think it is useful and let us know if you have ideas for improvement. Best, Botond

PS: We are contributing this code as individual developers and not as employees of Oxford Nanopore Technologies.

[rambaut]

Dear Botond, this looks great and we will need some time to look at this. One initial thought is I am against the use of 'accuracy' rather than 'identity' because the aim of this is to be able to assign reads to different genotypes on the basis of similarity of nucleotides. For most viruses at this level, indels are quite rare and are more likely to be sequencing errors.

[bsipos]

@rambaut If you wanted to use the number of matches instead of accuracy the Dump tool from the seqkit bam -T functionality can dump it to TSV along with the rest of the information needed. I would argue though, that the best way to assign the reads is the alignment score as it is a proxy for the likelihood of the read belonging to a particular reference. While the accuracy is essentially a length normalized alignment score.

On the test dataset using the accuracy gave me essentially the same results as the original approach, so it might not have a large impact in practice.