Interpretation of low "percentage of input non-chimeric" in DADA2 denoising stats

[Jun_Wu]

Hi QIIME 2 team and community,

I am processing paired-end 16S amplicon data in QIIME 2 (version 2025.4, conda installation on Ubuntu/WSL). After demultiplexing, primer trimming with cutadapt trim-paired, and running DADA2 with the following parameters:

qiime dada2 denoise-paired \
  --i-demultiplexed-seqs demux_trimmed.qza \
  --p-trim-left-f X --p-trim-left-r Y \
  --p-trunc-len-f 240 --p-trunc-len-r 245 \
  --p-n-threads 0 \
  --o-representative-sequences rep-seqs.qza \
  --o-table table.qza \
  --o-denoising-stats denoise-stats.qza

I summarized the results in denoise-stats.qzv (file attached here).

From the exported TSV, I noticed:

• Input reads per sample: ~40k–65k

• Filtered reads: ~87–91% of input (looks good)

• Merged reads: ~70–80% of input (also looks reasonable)

• Non-chimeric reads: only ~25–38% of input remain

So, the step with the largest drop is chimera removal.

---

My questions:

1. Is it normal that the percentage of input non-chimeric is only ~25–38%?

2. Are there recommended thresholds or “typical ranges” for this metric in QIIME 2 workflows?

3. Could this indicate that my truncation lengths (240/245) are too permissive and retaining low-quality tails that lead to false chimeras? Or is this more likely due to inherent properties of the dataset (PCR artifacts, etc.)?

4. Would adjusting truncation parameters (e.g., trimming more aggressively at the 3′ ends) usually help improve non-chimeric retention, without sacrificing merging efficiency?

Any guidance or examples from your experience would be greatly appreciated!

Thanks a lot in advance,

[cherman2 (Chloe Herman)]

Hi @Jun_Wu,
Welcome to the forum!

@Jun_Wu

Is it normal that the percentage of input non-chimeric is only ~25–38%?

I wouldn't say this is normal. As a quick sanity check, if filtering, merging or chimeric detection cause sequence loss over 50% I consider that cause for concern.

@Jun_Wu

Are there recommended thresholds or “typical ranges” for this metric in QIIME 2 workflows?

In this post, Mike Robeson explains changing the --p-min-fold-parent-over-abundance so that is an option: Trimming output problem - #14 by SoilRotifer

@Jun_Wu

Could this indicate that my truncation lengths (240/245) are too permissive and retaining low-quality tails that lead to false chimeras? Or is this more likely due to inherent properties of the dataset (PCR artifacts, etc.)?

Its unlikely that your trimming and truncation were too permissive and that your sequences were still passing filtering and merging checks. If you needed to trim or truncate more I would expect to see signs of this in the other checks. I would try messing with the --p-min-fold-parent-over-abundance parameter like Mike suggests in the post I linked and see where that gets you. If you are still seeing chimera issues after that, let us know and we can brainstorm some more!

Hope that helps