Running the benchmarks, I hit something I think is worth discussing. By default, mag uses Filtlong for lr QC filtering. That's fine, since you can easily switch to nanoq or chopper. But when short reads are also supplied, they're wired in as a reference. This can be a problem: as the Filtlong README notes, in that case Phred quality is ignored and instead derived from k-mer matching the short reads. If short-read coverage is poor, or there are subtle differences between the long and short-read metagenomes (even with hybrid assembly disabled), the filtering can be suboptimal, and we don't have a parameter to turn this off. It's a risky default, because we're making assumptions that can lead to real data loss. I think we have two options for this: • Make this opt-in. Add a parameter for that (e.g.

--longread_filtering_by_shortreads

) and leave it disabled by default • Change the default tool to chopper, so long-read preprocessing becomes lighter (it also replaces NanoLyse). Document how Filtlong behaves.

I vote 1

Option 3 is doing both

I guess could also work tbf, just more work 😬

Change default to chopper is fine I think & document when filtlong could be the better choice.

PR: github.com/nf-core/mag/pull/1059