Running Salmon in alignment mode on UMI-deduplicated BAM file

[grst]

Is the bug primarily related to salmon (bulk mode) or alevin (single-cell mode)?
bulk

Describe the bug
I am working with UMI-tagged Lexogen QuantSeq data. Since salmon does not (yet?) support handling UMIs with bulk RNA-seq directly (see #306), I am using umi_tools + STAR to generate a deduplicated transcriptome BAM file and run Salmon in alignment mode as implemented in the nf-core/rnaseq pipeline.

Unfortunately, Salmon does not seem to handle the deduplicated BAM well. A lot of genes have zero reads which shouldn't.

For instance, for ENSMUSG00000029657 I get the following results (last column denotes counts in all cases)

# Salmon on transcriptome BAM, without umit_tools dedup: quant.genes.no_umi.sf
ENSMUSG00000029657.15   3803.74 3650.23 17.3078 438
# Salmon on deduplicated transcriptome BAM: quant.genes.sf
ENSMUSG00000029657.15   1947.36 1614.62 0       0   
# Feature counts on genome BAM, without umi_tools dedup: 
ENSMUSG00000029657.15   [...]  7266     415
# Feature counts on deduplicated genome BAM:
ENSMUSG00000029657.15   [...]  7266    289

Here's a scatterplot of log1p(counts) of the salmon quant results for a single sample with and without umi_tools dedup

To Reproduce
Run Salmon quant on the aligned transcriptome BAM file. I provide subsampled versions of both the deduplicated and non-deduplicated BAM files. If you need the full BAM files, LMK and we can arrange a transfer.

• POOLAM1-40.Aligned.toTranscriptome.out.subsample.bam.gz
• POOLAM1-40.umi_dedup.transcriptome.sorted.subsample.bam.gz
• salmon_featurecounts_results.tar.gz

Specifically, please provide at least the following information:

• Which version of salmon was used? 1.5.1
• How was salmon installed (compiled, downloaded executable, through bioconda)? bioconda
• Which reference (e.g. transcriptome) was used? gencode.vM25.primary_assembly.annotation.gtf
• Which read files were used? Lexogen QuantSeq 3' UMI
• Which which program options were used? --noLengthCorrection

Expected behavior
Correctly quantify results on deduplicated BAM.

Desktop (please complete the following information):

• OS: CentOS 7
• Version 3.10.0-1160.11.1.el7.x86_64 #1 SMP Fri Dec 18 16:34:56 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux

Additional context
There's already an issue with RSEM described in the UMI-tools repository (CGATOxford/UMI-tools#465), maybe that's related.

CC @chripla

[grst]

Closing this as it's not an issue with Salmon.
The main problem was that the nf-core/rnaseq pipeline didn't call umi_tools with the --paired flag.

Some more fine-tuning of the UMI-tools output might be necessary to make sure unpaired reads are properly counted by Salmon. For this, see CGATOxford/UMI-tools#465.