After high-throughput sequencing, we obtained thousands of 16S rDNA reads for each sample. When we analyzed these sequences, firstly, we performed quality filter to remove the low quality sequences, then we randomly resampled to even the samples, then we usually defined OTUs at 97% similarity. Some OTUs only have one single sequence, we could not confirm that if they are very rare species or wrong sequences because some unknown reasons. Some papers retained these single-sequence OTUs, while other papers removed them. So, if these single-sequence OTUs should be removed? And why? I will be very appreciate if you can give some references about the question. Thank you very much.
Hi Guangshan,
To further clean the read set, you can apply an error correction step to the original reads, this may be reduce the number of OTU you find.
As further step, did you screen for chimeras (as obtained from spurious fusion of sequence during the PCR step) ?
Another possible source for them is a contamination (DNA extraction kits and so on are not necessary bacteria-free).
On your protocol, do you define OTUs (at 97%) across all the samples after evening the number of reads? I do prefer to call OTUs using all the reads (after quality filtering), so that any low count OTUs that is present in more than one sample will be identified more easily. In your way, low abundance OTUs may not even reach the OTU selection step.
In general, I think removing or keeping single sequence OTU, very much depends on what you are looking for: in many application you probably will keep only OTUs above 1% of abundance in any of the samples;
if you want to look at fine diversity or low abundance species, you will probably need them.
In my opinion, as far you know what you want and you treat all the samples in the same way, you can discuss easily both way in a paper.
Luca
Hi Guangshan,
To further clean the read set, you can apply an error correction step to the original reads, this may be reduce the number of OTU you find.
As further step, did you screen for chimeras (as obtained from spurious fusion of sequence during the PCR step) ?
Another possible source for them is a contamination (DNA extraction kits and so on are not necessary bacteria-free).
On your protocol, do you define OTUs (at 97%) across all the samples after evening the number of reads? I do prefer to call OTUs using all the reads (after quality filtering), so that any low count OTUs that is present in more than one sample will be identified more easily. In your way, low abundance OTUs may not even reach the OTU selection step.
In general, I think removing or keeping single sequence OTU, very much depends on what you are looking for: in many application you probably will keep only OTUs above 1% of abundance in any of the samples;
if you want to look at fine diversity or low abundance species, you will probably need them.
In my opinion, as far you know what you want and you treat all the samples in the same way, you can discuss easily both way in a paper.
Luca
Hi Guangshan,
While I was still using the OTU approach I used to remove singletons based on the results we obtained from sequencing mock communities. even after the removal of singletons, the error rate was still very high. As Luca suggested using an abundance filter reduces potential bias, but also important information might get lost.
Currently, I would suggest you to follow this SOP:
https://f1000research.com/articles/5-1492/v1
It uses DADA2 to get real sequences, thus avoiding the blurry OTU definition, and PhyloSeq for the analysis.
have a look at it.
-Mina
EDIT: there is a new version of this workflow:
https://f1000research.com/articles/5-1492/v2
Hi Sheik, Luca, and Mina,
Thank you very much for your detailed answers and good advices. I have known how to deal with the singletons. Firstly, we should know the aim of our study and then to decide the processing methods, focusing on the domainant or rare species? Secondly, we should perform more strictly quality filtering in order to reduce the singletons. Finally, new SOPs, like Mina mentioned above, should be another better way. Thanks again.
regards,
Guangshan
Hi,
yes, go for this. However, as many filter you can apply I do not think you will be able to get rid of all the noise in th data, but this is one of the joy of NGS!
Mina is right, clustering methods with fixed similarity thresholds are showing their limits. I personally like DADA2 as well as swarm (https://github.com/torognes/swarm ).
Luca
Hello Guangshan,
I find the below link useful for determining the future of singleton in my project. Hope it will be useful to you too.
http://www.drive5.com/usearch/manual/singletons.html
Hi Toral,
Thank you very much for your answer and it's very useful for us.
regards,
Guangshan
- Badges
- Science method