denoising removes everything

[khemlalnirmalkar]

Hello folks,
I am dealing with an issue at the denoising step. My reads (only forward) are ~30 Phred scores and on avg ~20-30k reads (some >100K). But when i am denoising with DADA2, i am losing almost all reads. Please can you help me to understand why its happening, should I consider any other parameters? i attached reads and dada2 images

FYI: these are ion torrent 16S (v3) amplicon sequences,
Thanks

[colinbrislawn (Colin J Brislawn)]

Hello @khemlalnirmalkar,

Thanks for posting your DADA2 summary stats and the quality score data. Looks like most of your reads are removed during the initial quality filter step in the DADA2 pipeline.

Can you post the command you used to run the q2-dada2 plugin? Hopefully we can find some settings that persevere more of your dataset.

[khemlalnirmalkar]

Hi @colinbrislawn
This is the one i used
qiime dada2 denoise-single
--i-demultiplexed-seqs single-end-demux.qza
--p-trim-left 0
--p-trunc-len 210
--o-representative-sequences rep-seqs-dada2_210.qza
--o-table table-dada2_210.qza
--o-denoising-stats stats-dada2_210.qza

Thanks

[colinbrislawn (Colin J Brislawn)]

Got it!

Here's two thing to try:

Instead of dada2 denoise-single, try dada2 denoise-pyro, as that's designed for Ion Torrent reads like yours.

I like your choice of trim and trunc settings. You may also consider increasing the --p-max-ee from 2, which is the default, to 3 or 4. That would allow more noisy reads to pass the initial filter and continue through the pipeline.

[khemlalnirmalkar]

@khemlalnirmalkar

qiime dada2 denoise-single
--i-demultiplexed-seqs single-end-demux.qza
--p-trim-left 0
--p-trunc-len 210
--o-representative-sequences rep-seqs-dada2_210.qza
--o-table table-dada2_210.qza
--o-denoising-stats stats-dada2_210.qza

Thanks for the suggestions,
I tired this denoise-pyro but result is almost same
I tried to play with all options from this command but i got the same results
my Last command was
qiime dada2 denoise-pyro
--i-demultiplexed-seqs single-end-demux.qza
--p-trim-left 0
--p-trunc-len 200
--p-max-ee 4
--p-chimera-method none \
--p-n-reads-learn 100000
--o-representative-sequences rep-seqs-dada2_pyro.qza
--o-table table-dada2_pro.qza
--o-denoising-stats stats-dada2_pyro.qza

Please let me know if there is any other way
Thanks

[colinbrislawn (Colin J Brislawn)]

Just to check, did passing --p-max-ee 4 return the exact same results, or similar issues with not a lot of reads passing filters?

You could also try reducing the --p-trunc-len setting down to 190 or 180.

[khemlalnirmalkar]

Thanks for getting back to me
Got similar issues with not a lot of reads passing filters
I just tried with --p-trunc-len 180 with a command
--i-demultiplexed-seqs single-end-demux.qza
--p-trim-left 0
--p-trunc-len 180
--p-max-ee 4
--p-chimera-method none
--p-n-reads-learn 100000
--o-representative-sequences rep-seqs-dada2_pyro.qza
--o-table table-dada2_pro.qza
--o-denoising-stats stats-dada2_pyro.qza

and seems like it gave me a better result than earlier.
If you see sameID F6, F8, F9, i am still losing a good amount of reads. please can you suggest what other changes should i make to get a better result?

[colinbrislawn (Colin J Brislawn)]

OK, thanks for sharing these results with me.

I'm not sure what's our best path forward. As you can see from your quality plot, your overall read quality is relatively low, and will cause the expected errors to accumulate, then not pass filter. I guess you could increase your expected threshold further...

You face a difficult tradeoff between losing most of your data set, or using low quality reads.

Perhaps someone with more experience with Ion Torrent data can provide their perspective.

[khemlalnirmalkar]

Thanks @colinbrislawn
Finally decided to excluded samples with low reads but i will be happy to have suggestions from someone who faced a similar situation