@chase mateusiak has brought up an interesting issue regarding RSEM and custom alignment settings. In older versions of the pipeline, RSEM handled the STAR alignment of the reads internally, such that it was simply not possible to provide any custom arguments to STAR when using this quantifier. But in the current pipeline version, STAR is also run externally for RSEM, while the required arguments to ensure compatibility are provided by the

generateStarAlignArgs()

function in the subworkflow’s config. This made me wonder if the following warnings/errors during the parameter validation in

utils_nfcore_rnaseq_pipeline

are still applicable: • Using

params.aligner == 'star_rsem' && params.extra_star_align_args

will trigger the

rsemStarExtraArgumentsWarn()

, but

generateStarAlignArgs()

will nonetheless include them in the

final_args_map

. It may not be advisable, but to me it seems that is factually possible to override the RSEM STAR arguments via the pipeline’s extra parameter? •

params.with_umi  && params.aligner == 'star_rsem'

will result in an

rsemUmiError()

. Technically, UMI deduplication is now possible as well. Jonathan has already explained recently why he intends to uphold the current restrictions. However, both, umi-tools and umicollapse consider the alignment position during deduplication, such that outputting the unaligned reads as part of the BAM with

--outSAMunmapped Within

will hardly have negative consequences for the deduplication - it must only be corroborated that the unaligned reads are retained.

--outFilterType BySJout

removes low-quality spliced alignments spanning poorly supported, novel junctions, particularly with short overhangs. This will just increase the number of unmapped reads slightly. Considering that UMIs are very helpful to obtain reliable deduplication, I think that the error could perspectively be downgraded to a warning as well?

You're probably right on

rsemStarExtraArgumentsWarn

, though I'd probably do a quick check to config the param is passed. PRs gratefully received. I mostly didn't have time to assess the custom args inherited from RSEM to make sure the UMI dedup would work fine, I didn't want to be dealing with support requests stemming from incompatibilities. If you can test and show that works I'm content to remove the restriction.

sorry all -- i hadn't joined this channel and posted in #CE8SSJV3N

I can look into this in more detail. I just wanted to first learn what your intention is, so if you think it is desirable to extent the RSEM route with more flexibility.

Absolutely, I have no axe to grind here, just being cautious