Sequence length after denoise paired-end reads

[kyle_paddock]

Hi,

I have a question regarding sequence length after denoising. I have amplified the V4 region using 515F-806R primers. I trimmed primers and adapter sequences and denoised based on quality scores observed in the demux summary in Qiime2-2022.8.

qiime cutadapt trim_paired 
	--i-demultiplexed-sequences paired-end-demux.qza \
	--p-adapter-f "ATTAGAWACCCBDGTAGTCC" \
	--p-front-f "GTGCCAGCMGCCGCGGTAA" \
	--p-adapter-r "TTACCGCGGCKGCTGGCAC" \
	--p-front-r "GGACTACHVGGGTWTCTAAT" \
	--p-indels False \
	--p-discard-untrimmed True \
	--o-trimmed-sequences demux-trim-primers.qza \

qiime dada2 denoise-paired \
  --i-demultiplexed-seqs demux-trim-primers.qza \
  --p-trunc-len-f 231 \
  --p-trunc-len-r 209 \
  --o-table feature.table.qza \
  --o-representative-sequences denoise-seqs.qza \
  --o-denoising-stats denoise-stats.qza 

These parameters resulted in a fairly high retention of my sequence after filtering, denoising, and merging (~86%). I then tabulated sequences to see what the lengths were and was surprised to see the range from 231-420bp. I'm under the impression they should be around 253 bp. Granted my mean length is 253. Is there a parameter I did not include to trim merged sequences to the correct length? Am I joining forward and reverse reads in incorrect places?

There was a topic similar to this but it was unclear if there should be concern with sequences that are too long.

Thanks!

[lizgehret]

Hi @kyle_paddock,

Welcome to the forum!

This is a great question - and your impression is correct, your sequence lengths should be around 253 bp. A couple of good things to point out with your data set is that your mean length is 253.11, which is right where it should be! Additionally, the standard deviation is only about 3bp which means that most of your data is sitting right around that 253 bp mark.

However, you do seem to have some outliers on both ends (the min and max lengths) - especially that max length, which is much longer than the V4 region. The short answer is that these lengths can be related to non-target DNA, which you will most likely want to filter out unless they are of interest to you. You could try filtering out anything that's shorter than 240 and longer than 255 and see what the statistics look like after that, but that should remove those outliers you're seeing.

This is a great forum post that goes into more detail on this situation, for your reference.

Sequence length statistics

That variation is much more than usual, indeed. There is another related discussion somewhere on the forum (I can’t find it right now) where it was mentioned that this variation is usually related to non-target DNA… e.g., chloroplast + mitochondrial 16S is shorter than bacterial, and sometimes some eukaryote 18S can be hit by the primers. So unless if this is something that interests you, you should just filter out anything too long/short. I’d say filter anything < 240 nt or > 255 nt You are …

Hope this helps! Cheers

[kyle_paddock]

Thanks @lizgehret !

I have since filtered my sequences to include 240 < seq length < 255. However, I'm a little confused about how to regenerate a feature table with these new sequences.

I can find ways to filter sequences based on a feature table, but not the other way around. Do I need to rerun my new set of representative sequences back through dada2 denoise-paired to create a feature table?

Thanks for the help!

[lizgehret]

Hi @kyle_paddock,

Apologies for the delay in response! That's a great question - you can do this using qiime feature-table filter-features and use your filtered rep-seqs.qza as metadata. Here's what that would look like as a command:

qiime feature-table filter-features \
--i-table YOUR_ORIGINAL_FEATURE_TABLE \
--m-metadata-file rep-seqs.qza \
--o-filtered-table YOUR_FILTERED_FEATURE_TABLE

Hope this helps! Cheers