Samples missing from DADA2 table but not denosing stats

[saatkinson (Samantha)]

Hello,

I am using QIIME2 v. 2018.8 and I am denoising my paired end read data with DADA2. I am seeing a discrepancy between the number of samples in the denoised-stats file (78, the number that should be there), and the number of samples in the feature table (73, missing 5 samples). Here is the command I ran to produce the files:

qiime dada2 denoise-paired --i-demultiplexed-seqs demux-seqs.qza --p-trunc-len-f 251 --p-trunc-len-r 218 --p-n-threads 15 --o-representative-sequences rep-seqs.qza --o-table table.qza --o-denoising-stats denoising-stats --verbose

I know the sequences aren't the best, with a lot that have low sequence counts/non-chimeric output. I thought it might be removing samples with low merged/non-chimeric sequences, but looking at the denoise stats I have 18 that have zero non-chimeric sequences and only 5 are missing. If you could help me determine why it's removing samples/why those 5 specifically I would really appreciate it.

The 5 missing samples are: ES1-1-B_10-10-17_8, ES1-1-B_12-08-17_67, ES1-1-D_04-06-18_186, ES1-3-A_09-29-17_-3, ES1-3-D_02-02-18_123

Here's the command I ran to visualize the stats file and the table:

qiime metadata tabulate --m-input-file denoising-stats.qza --o-visualization vis-denoising-stats.qzv

​qiime feature-table summarize --i-table table.qza --o-visualization vis-table.qzv --m-sample-metadata-file MappingFILE_Round2_20180814.txt

I'm also attaching the files if they're helpful.

table.qza (25.2 KB)
denoising-stats.qza (12.8 KB)
rep-seqs.qza (25.2 KB)
vis-table.qzv (327.5 KB)
vis-denoising-stats.qzv (1.2 MB)

Thanks,
Samantha

[Nicholas_Bokulich]

Hi @saatkinson,

@saatkinson

The 5 missing samples are: ES1-1-B_10-10-17_8, ES1-1-B_12-08-17_67, ES1-1-D_04-06-18_186, ES1-3-A_09-29-17_-3, ES1-3-D_02-02-18_123

It looks like those 5 samples are the only 5 that have 0 reads post-filtering:

I think the feature table must be made after denoising, dropping all samples that have 0 sequences, and then chimera removal happens after that — @benjjneb is that correct?

@saatkinson

I know the sequences aren’t the best, with a lot that have low sequence counts/non-chimeric output.

Those 5 samples (and many others) cannot be revived any way you slice it, since the input read counts are so low. However, you are losing many reads at the filtering step, which leads me to believe you might be able to save more reads in your other samples by adjusting your trimming/truncation parameters. Let us know if you have any questions about that.

Good luck!

[benjjneb (Ben Callahan)]

@Nicholas_Bokulich

I think the feature table must be made after denoising, dropping all samples that have 0 sequences, and then chimera removal happens after that — @benjjneb is that correct?

Technically samples are dropped if they have zero reads after filtering (and before denoising) but basically you are right.