Use hostile for host read removal

[dialvarezs]

Description of feature

Instead of bowtie2

[d4straub]

In version 5.1.0 Kraken2 was dropped in favor of using nf-core/taxprofiler before nf-core/mag (as far as I remember the dicussion). Why now use hostile when nf-core/detaxizer could be used before nf-core/mag?
Also, why specifically hostile? What made you pick that tool?
Dont get me wrong, I like having host decontamination readily available in nf-core/mag and not need other pipelines.

[dialvarezs]

For me the key difference is that Kraken/Centrifuge was merely "informative", it wasn't really integrated in the downstream steps. Decontamination results is other hand are fed directly to the assembler. Also it's kinda required, at least for clinical samples with human/mice host.

The decision of Hostile is basically because it's essentially a wrapper of what we are using right now (Bowtie2 for short reads and Minimap2 for long reads). So, it can be considered a drop-in replacement and should simplify that part of the pipeline. And also it makes easier to select a good reference (Hostile provided the T2T reference masked at different levels).

[jfy133]

To clarify @d4straub in the long run I would like to drop host removal from mag (all preprocessing technically. ;)).

Difference between the taxonomic profiling is I know a lot of people still use host removal (it is a' critical' functionality that is in the middle of the pipeline), compared to the profiling which was an end point that wasn't super popular so it wasn't as 'bad' to drop it.

Also the original reason to touch this area of the code base was rhat there is an outstanding bug in regards to host removal which I was originally trying to address address. The switching to hostile would fix the bug, but also simplify the code base, with additional benefit of improved quality of host removal

[jfy133]

I.e. what Diego said while it took me 3h to write this comment 😅

[d4straub]

I see, thanks for the explanation, that makes sense to me.

[tpall]

It’s very simple to use the hostile even now, one has to download one of their reference genomes and/or bowtie2 index and set up --host_fasta and/or --host_fasta_bowtie2index parameters repectively in nf-core/mag. That's it.

[chrisgulvik]

Another important benefit of hostile over bt2 is its database. Reads with bt2 against the T2T alone is overly aggressive. Human genome segments can (and are!) present in some bacteria as well, and the current bt2 against T2T removes them (but shouldn't). I think for complex communities especially that too much is being discarded that shouldn't be. For 100 bp reads, especially after quality trimming shortening them, this is problematic. For longer reads like 300 bp the impact is much smaller. "Hostile removes 21%-23% more human short reads and 21-43 times fewer bacterial reads, typically in less time." source

To make that database, which is essentially a "masked" T2T, the authors fetched, organized, and aligned all Bacteria against the T2T reference file. This is a tremendous amount of work and compute resources, that most of us wouldn't want to re-do.

To summarize, I think hostile's "masked" T2T is of high value and worthy of replacing the current methods.