I run DADA2 and got significantly different ASV numbers at different machines with the same data and setting.

11112 · May 13, 2021, 8:11am

As titled, I confirm the DADA2 version (1.18.0) and QIIME2 version (2021.2.0) are the same at these machine. What causes the results? Thank you!
My DADA2 setting is qiime dada2 denoise-paired --p-trim-left-f 0 --p-trim-left-r 0 --p-trunc-len-f 0 --p-trunc-len-r 0 --p-trunc-q 20 --p-n-threads 30 --p-min-fold-parent-over-abundance 1

colinbrislawn · May 19, 2021, 2:34am

Hello @11112,

I wonder if some of the defaults are different between DADA2 in R and the Qiime2 plugin.

Here’s all the settings that Q2-DADA2 uses:

github.com

qiime2/q2-dada2/blob/master/q2_dada2/_denoise.py#L222-L232


def denoise_paired(demultiplexed_seqs: SingleLanePerSamplePairedEndFastqDirFmt,
                   trunc_len_f: int, trunc_len_r: int,
                   trim_left_f: int = 0, trim_left_r: int = 0,
                   max_ee_f: float = 2.0, max_ee_r: float = 2.0,
                   trunc_q: int = 2, min_overlap: int = 12,
                   pooling_method: str = 'independent',
                   chimera_method: str = 'consensus',
                   min_fold_parent_over_abundance: float = 1.0,
                   n_threads: int = 1, n_reads_learn: int = 1000000,
                   hashed_feature_ids: bool = True
                   ) -> (biom.Table, DNAIterator, qiime2.Metadata):

You may have found this page already, but you can also view all the dada2 settings on the docs:
https://docs.qiime2.org/2021.4/plugins/available/dada2/denoise-paired/

The settings --p-pooling-method and --p-chimera-method have a huge impact on results, so it’s worth double checking that these are the same.

Let us know what you find, or if you would like to post some test data that shows this difference!

11112 · May 19, 2021, 5:48am

Dear @colinbrislawn
Thanks for your reply!
I checked the two settings and found they are the same.
I keep trying to find the other different default values.

Mehrbod_Estaki · May 19, 2021, 5:57am

Hi @11112 ,
Are you comparing running q2-dada2 under identical settings on 2 different machines? Or as @colinbrislawn was hinting at comparing q2-dada to stand-alone DADA2 in R?

Can you also let us know what these ‘differences’ are exactly with some more detail?

One thing that may be happening, though I’d have to check how QIIME 2 handles this under the hood, is that if you transferring your files to other systems, and during the transfer the order of your samples are changing? Say in one run you import your data as S1, S2, S3, and in other S3, S2, S1, then the first 1,000,000 reads in --p-n-reads-learn, that is drawn by DADA2 for its training model, may reflect different reads. But still, even with all that I wouldn’t expect drastic differences, so I am interested to hear some more more details on this.

11112 · May 20, 2021, 7:17am

Thanks for your reply! @Mehrbod_Estaki
I found the qiime2 version and the plugin version which I added are not consistent.
This may cause the difference. I will keep checking.

system · June 20, 2021, 1:17pm

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.