ONT flag default settings

[mbhall88]

I'm raising this so it is on our radar and we don't forget, but the default behaviour when the --ont flag is given seems to need more investigation.
Currently, when this flag is set, the following parameters are set/overridden

mykrobe/src/mykrobe/cmds/amr.py

Lines 344 to 361 in 6a7e7f6

	if args.ont:
	args.expected_error_rate = 0.15
	args.ploidy = "haploid"
	args.ignore_minor_calls = True
	logger.warning("Setting ploidy to haploid (because --ont flag used)")
	logger.warning("Setting ignore_minor_calls to True (because --ont flag was used)")
	logger.warning(f"Setting expected_error_rate error rate to {args.expected_error_rate} (because --ont flag was used)")
	args.model = "kmer_count"
	# If the user didn't specify the conf_percent_cutoff, then set it
	# depending on whether or not the --ont flag was used
	if args.conf_percent_cutoff == -1:
	if args.ont:
	args.conf_percent_cutoff = 90
	logger.warning("Setting conf_percent_cutoff to 90 (was not specified, and --ont flag was used)")
	else:
	args.conf_percent_cutoff = 100
	logger.warning("Setting conf_percent_cutoff to 90 (was not specified, and --ont flag was not used)")

And this also runs the simulations.

As part of mbhall88/head_to_head_pipeline#75 and mbhall88/head_to_head_pipeline#76 I have been trying some different error rates. In order to stop mykrobe from overriding the error rate I was giving though I couldn't provide the --ont flag. So I provided -e 0.08 --ploidy haploid but didn't realise there were the other settings that this flag also changed. The results from the -e 0.08 --ploidy haploid settings were considerably better than using the default --ont settings. But what I had accidentally discovered was that if I also provided the other default parameters such as --model kmer_count --conf_percent_cutoff 90, which are normally set when --ont is passed, I get much worse results than just setting the error rate and ploidy. By "worse" I mean I get a lot more FNs. It does indeed improve the FPs (actually make FPR better than Illumina), but at a big cost to the FNR.
So it seems like either the kmer count model, or the simulations of confidence thresholds (or both) need further work.

[mbhall88]

Providing some concrete data to support this change. Here are some figures and associated PDF for the relevant part of my thesis. This is also relevant for #133

mykrobe-params.pdf

In the end, the parameters I used to run mykrobe for Nanore data are

mykrobe predict -e 0.08 --ploidy haploid ...

[iqbal-lab]

To discuss. Essentially we drop our FNs for isoniazid, ethambutol and rifampicin significantly. There is a price, which is elevated FPs for isoniazid.