Why do I get better results with dada2 denoise-single although the data are paired-end sequences?

[fgara]

Dear all,

First of all, thanks a lot for all of your hard work with QIIME2!

I have a question please:
I imported my demultiplexed, free-from-adapters, paired-end data using the following command:

qiime tools import
--type 'SampleData[PairedEndSequencesWithQuality]'
--input-path $BATCHNAME"_manifest.tsv"
--output-path $BATCHNAME"_demux.qza"
--input-format PairedEndFastqManifestPhred33V2

Then I continued with the following command (by mistake):

qiime dada2 denoise-single
--i-demultiplexed-seqs $BATCHNAME"_demux.qza"
--p-trim-left 0
--p-trunc-len 200
--p-chimera-method 'consensus'
--o-representative-sequences $BATCHNAME"_rep_seqs_dada2_trunc200.qza"
--o-table $BATCHNAME"_table_dada2_trunc200.qza"
--o-denoising-stats $BATCHNAME"_stats_dada2_trunc200.qza"

Realizing that I should have used a different command, I then re-analyzed my data using:

qiime dada2 denoise-paired
--i-demultiplexed-seqs $BATCHNAME"_demux.qza"
--p-trim-left 0
--p-trunc-len 200
--p-chimera-method 'consensus'
--o-representative-sequences $BATCHNAME"_rep_seqs_dada2_trunc200.qza"
--o-table $BATCHNAME"_table_dada2_trunc200.qza"
--o-denoising-stats $BATCHNAME"_stats_dada2_trunc200.qza"

I then compared both results, hoping that my new results would look better.
But actually the results generated using denoise-single retain more sequences and show decent alpha rarefaction plots.

The results generated using denoise-paired lost a lot more sequences:

Does anyone know why is this so?
And is it okay for me to just use the denoise-single command for my paired-end data then?

Many thanks for your time and help

[jwdebelius (Justine Debelius)]

Hi @fgara,

I'd need to see your quality plot and your dada2 statistics to help diagnosis where the problem occurs. My guess is that reverse reads are either low quality and drop out during denoising, which can be solved by a shorter trim length, or that the reads are failing to merge, but its hard to know in your case specifically without checking your data.

Best,
Justine

[fgara]

Hi Justine,
Thank you for your reply.
Yes, I have uploaded it here:

11052020_demux.qzv

Thanks again!

[jwdebelius (Justine Debelius)]

Hi @fgara,

Can you also share the summary of your denosing statistics?

Best,
Justine

[fgara]

Hi Justine,

Yes, this is the file:

11052020_stats_dada2_pe_trunc200.qzv

Many thanks for your kind help!

[fgara]

Oh sorry I forgot to say, the last file is from the paired-end commands.
Many thanks!

[jwdebelius (Justine Debelius)]

Hi @fgara,

Check out the percent-of-input-merged column.

This typically indicates that some of your reads are too short to overlap. It looks like you've got some reads that merge well and some that are totally failing. Is there something different about these samples that might explain the difference?

Best,
Justine

[fgara]

Hi Justine,
Thank you for your reply!
Yes, it's quite bad... I will check again.

Many thanks!