Low % of reads pass merging and chimera filter after DADA2 for rSSU (Fungi)

[PoitrasAMFresearch (Jérémie Poitras)]

Hi all,

FYI, I am a beginner in bioinformatics. I am using QIIME2 (2024.2) in a conda environment. I am working with 147 samples of AMF rSSU amplicons of a median length of 469bp (should be 550bp in theory). I had a max of 448bp and a min of 281bp. Out of my 147 samples, I have 5 replicates/treatment combination.
I removed the sequences beginning with my primers sequences (WANDA & AML2) with Cutadapt, but a few samples still had them even after cutadapt. For 4 samples, 124; 147; 233; 337 sequences were found beginning with the forward primer sequence. About 45 other samples have 1-3 reads beginning with the forward primer sequence. No sample has any reads beginning with the reverse primer sequence.

My quality plots looked very good as I benefited from the new MiSeq i100 of the sequencing platform I dealt with. See my quality plots photo attached.

So I decided to not trim or trunc. After denoising with DADA2 with these settings:

qiime dada2 denoise-paired \
  --i-demultiplexed-seqs demux-trimmed.qza \
  --p-trunc-len-f 0 \
  --p-trunc-len-r 0 \
  --p-n-threads 0 \
  --p-max-ee-f 2 \
  --p-max-ee-r 3 \
  --o-table table.qza \
  --o-representative-sequences rep-seqs.qza \
  --o-denoising-stats denoising-stats.qza

I end up with very variable % of input that are non-chimeric: between 5.59 and 69.8% (with a median of 21.17%). Filtering only removed 0.1 to 4% of the inputs. It seems like the bottleneck is the merging of my reads (pair-end). I’m loosing between 5% and 91% (median of 68%). For the sequences that merged well (<20% lost after merging), I loose about 30-40% when removing chimera to end up with good final % of input kept (between 40 and 70%). Those might only represent a third of my samples.

The count of reads kept after all those denoising filters are between 1,278 and 99,385 inputs with a median of 26,607. Only 6 samples are <10,000. I would probably get rid of those with sampling depth later (assuming it makes sense based on rarefaction plots).

My questions are:

1. Do you spot any beginner mistake that I am oblivious to?

2. Should I be concerned with the low % of reads that passed the filters of DADA2 even knowing that I am working with root samples that were inoculated with low AMF abundance/diversity?

3. Should I modify my pipeline or any settings to increase the retention of reads?

4. Do you need more information to answer the previous questions? I am happy to provide more.

Thank you for your input and your help, it would be much appreciated.

Jérémie

[cherman2 (Chloe Herman)]

Hi @PoitrasAMFresearch,

@PoitrasAMFresearch

My questions are:

1. Do you spot any beginner mistake that I am oblivious to?

These are not beginner mistakes but I think that the AMF sequences are a little tricky so lets discuss some issues and steps forwards!

@PoitrasAMFresearch

Should I be concerned with the low % of reads that passed the filters of DADA2 even knowing that I am working with root samples that were inoculated with low AMF abundance/diversity?

I personally would spend some time messing with parameters to get better % of sequences through. We would expect for their to be a smaller number of sequences since its low abundance but I would expect that you still get a higher percentage though the filter.

@PoitrasAMFresearch

Should I modify my pipeline or any settings to increase the retention of reads?

@SoilRotifer pointed out to me that the more your reads overlap the more chance there will be a mismatch, which will contribute to merge failures if the sequence hits the mismatch threshold. He recommends trimming 20-30 bases regardless of quality.

I would be point out that there is a chance that your AMF SSU region is too long to be covered by 2x300 seequnecing and in those cases there is nothing to do about the failure to merge.

I would mess around with truncating and see what that does for your % of sequence passed. I personally only like to mess with one parameter at a time. Once you have that sorted and have maximized the amount of sequences that are merge-able, I would look over this post to address your chimeric detection issue:

Trimming output problem

Thanks for sharing these @Liviacmg. I think for those samples, in which you're losing more sequences, appears to be due to losing proportionally more due quality filtering rather than chimera removal. The drop from denoised to non-chimeric isn't as much of a drop as the quality filtering. This could probably be helped by truncating a little more. Another option is to adjust the --p-min-fold-parent-over-abundance to 8 or 16. I wouldn't suggest any higher than that. This is based on the followin…

I hope this helps and let us know if you have any follow-up questions.