Clarify barrnap use

[hjarnek]

Hello Daniel & Daniel et al...

Could you please clarify exactly how barrnap is currently being used to filter out off-target co-amplification? What does the decision logic look like?

Please refer to this thread for details: tseemann/barrnap#119

I interpret the developer's response as meaning that barrnap might not currently be suitable for metabarcoding - but it may depend on the usage.

Best regards,
Joel

[d4straub]

Thanks for critically looking at the pipeline!

By default barrnap is only used as information, i.e. nothing is filtered. Only when --filter_ssu is actively chosen, ASVs are actually filtered. This is also mentioned in in the output docs.

In ampliseq version 2.17.0, Barrnap v0.9 is classifying ASVs into all four kingdoms, summarizing the results, and assigns the kingdom with the lowest evalue to each ASV. When filtering is chosen, the ASVs are filtered based on that assignment (kingdom with the lowest evalue). So there is no mapping to all kingdoms at once.

This was mainly implemented in PRs #396, #548, #722.

This implementation might not be perfect, but seemed to do the job. If there is indeed indication that this is improper, we need to change that ofc. It would be certainly more desirable if barrnap would compare each seqeuence/ASV to all kingdoms at once and barrnap itself would provide a decision instead of us comparing e-values.

[hjarnek]

To me, your description sounds like the best approach available at the moment. But it sounded like Torsten was open to implementing this logic you're talking about into barrnap! Probably better if you communicate directly with him, e.g. in the linked GH thread, about how you think this feature should look like.

Cheers