More than 70% of ASVs appeared only once but most of them are not singletons

Hi, I tried to search for discussion under the same topic and I found the discussion below on low abundance ASVs after DADA2. However, I have a slightly different issue.

I ran DADA2 algorithm using QIIME2 and I am currently performing my analysis in R. I see that 72% of the ASVs appear only once across my samples (I have a few hundreds singletons but most of the ASVs that appear only once have >1 reads, so they are not singletons). My question is, how do you find out if these ASVs that appear only once are genuine or spurious? My samples are 16S from fine roots of a large tree (Brazil Nut).

I am a newcomer in this area, so your advice is greatly appreciated! Thank you!

Welcome to the forum!

Does that mean those ASVs appear in only 1 sample but have counts greater than 1 in that given sample?

Did you remove barcodes during demultiplexing? Did you remove primers with cutadapt before DADA2 (recommended), or did you trim them with DADA2?

Barcodes and degenerate primers in the sequence, as well as trimming primers instead of Cutadapt, may introduce technical variability across ASVs.

Hi @timanix thank you for responding to my question. Yes, the primers and the barcodes should had been removed with the following commands:

  1. qiime cutadapt trim-paired \
  2. --i-demultiplexed-sequences 02_import/demux_16S.qza \
  3. --p-front-f ^CCTACGGGNGGCWGCAG \
  4. --p-front-r ^GACTACHVGGGTATCTAATCC \
  5. --p-discard-untrimmed \
  6. --p-cores ${THREADS} \
  7. --o-trimmed-sequences 03_cutadapt/trimmed_16S.qza \
  8. --verbose \
  9. > 03_cutadapt/cutadapt_batch1.log 2>&1

I do expect many rare ASVs in my system. I was just surprised to have that many rare ASVs. This is the first time I am working with this type of sample, so I am not sure.

Hi @timanix and yes, most of them have more than 1 read per sample where they occur.

I've worked with roots before, and they usually contain many rare ASVs. I wonder if it is real diversity or due to some technical issues.

How many batches do you have? How many samples are in each batch, and how many are in the dataset in total? Do you have the same sample type, or are your samples from different types/niches/tissues? Did you apply absolutely identical parameters for each batch before merging (especially for DADA2)?

I just checked my dataset with roots. Out of all ASVs, 37.5% met the following criteria: total count > 10 and found in at least 2 samples. These 37.5% of retained ASVs accounted for 95% of the total frequency, meaning that the filtered-out rare ASVs (62.5%) accounted for only 5% of reads. If similar filtering significantly reduces total frequency, check again to be sure. If you are not losing much, it is probably PCR/amplification artifacts.

Also, could you rerun Cutadapt without any "^" at the beginning of the primer sequences?
One more question: did you use raw data, or did you perform any quality control (not recommended) before Cutadapt/DADA2?

Thank you @timanix.

  1. I have 2 batches
  2. In batch 1, I have 45 samples. In batch 2, I have 33 samples for Brazil Nut. The other samples in batch 2 come from other species (but this would be yet another issue as some of them have reads that are 1/5th of the Brazil Nut).
  3. All of the samples come from roots but some of the samples in batch 2 are from other tree species (as mentioned above).
  4. I applied exactly the same parameters for each batch before merging.

I would need some time to re-run it without ^ at the beginning of the primer sequences. Will get back to you once I have done it.

Really appreciate your advice. It has been very helpful.

Hi, I'd like to offer a different perspective.

The separate ASVs (e.g., ASV#1, ASV#5, and ASV#16) may be annotated as the same genus and/or clustered at the same tip of a phylogenetic tree.

If so, I think this means that a higher taxonomic level should be considered rather than simply using the ASV itself.

Hi @xiangyu,

There are a few issues here.

First, sparisty as a base issue may indicate problems with denoising and/or overlap in the sample. Whether or not ASVs are shared can be a basic QC question that needs to be addressed.

The second question is an issue of analytical resolution. The right answer here depends on your reserach question. In some cases collapsed resolution gives the necessary information. In others, lower level resolution is need, sometimes down to the ASV.

Best,
Justine

Thank you, @jwdebelius!

I overlooked the QC question because my understanding is that ASVs capture finer sequence-level variation than OTUs (e.g., 97% similarity). It therefore seems plausible that different samples could contain distinct ASVs (especially when each sample is denoised independently, see examples #1 and #2, dada2 tutorial, and Parameters - pooling_method) from the same genus or even the same species. For example, if two samples contain different ASVs that would be clustered into the same OTU, could that overestimate "species" diversity via ASV denoising?

Regarding the question,

My question is, how do you find out if these ASVs that appear only once are genuine or spurious?

my thought was that an ASV detected in only one sample could represent a genuine sequence variant without necessarily corresponding to a distinct "species". (In addition, interpreting ASVs without further taxonomic annotation seems less common). Am I thinking about this correctly?

My main concern was the second question. I encountered a similar issue months ago. Why most of ASVs (possibly more than 90% in 10,000+ ASVs) were detected in only one sample each? A similar pattern was reported in Too many ASVs occurring only once across samples - 16S amplicon · Issue #1233 · benjjneb/dada2 · GitHub. However, the approach discussed there is mainly intended to estimate the relative abundance of abundant taxa. I'm not sure if it is suitable for studying rare ASVs.