How good is good enough (analysis of mock community)?

In comparing analysis of sequence data prepared from mock community, how closely in the final analysis (i.e. post curation) should the data output reflect the true identity of the mock community? What are the acceptable tolerances of experimental error in data creation and analysis?
Is it possible to ascribe or apply a statistical perspective to adequately rank experimental vs actual community composition?

My mock is equimolar and has 20 species.

Kind regards,

Hello Brindha,
Most of the times Pat and Sarah answer, but I want to contribute to the community as well. So here are my 2 cents.

That is a very good question. I have been dealing with those same issues and doing a ton of reading. One paper I can highly recommend is

“Ironing out the wrinkles in the rare biosphere through improved OTU clustering” By Huse et al.

Now, if you adapt a OTU bases analysis, the resulting number of OTUs are a function of (what I have understood thus far)

  1. Your sequencing error rate
  2. Cutoff for clustering
  3. Number of 16s operons in each of your organisms (some bacteria have more than ONE 16s operons. Did you know that?)
  4. Variability of the 16s sequences
  5. The precise algorithm used (Complete linkage, Average Linkage or Single Linkage)

Ultimately, it is a interplay of ALL of these factors which determines the number of OTUs you obtain. So I do not think there is a straightforward answer about HOW MANY OTUs should you get. If you use a Phylogeny based approach, you WILL get better results for a mock community, but there is no guarantee you will get those same results on your actual samples (it is database dependent). Ultimately, what I have realized is that OTUs are just a “computational” way of aggregating sequences together and comparing it between samples.

i.e., OTU’s != Bacterial strains.

Thanks for the question and for contributing, quantrix.

My philosophy is that the level to which the output resembles the expectation is not what we’re after. The number of OTUs out the back end can be affected by the number of reads that went into the analysis. More reads, more garbage, more OTUs. We focus on the error rates. Generally, we see error rates below 0.0002 or 0.02%. I would worry if you are at 0.1% or above.

Also, good for you for getting a mock community sequenced! Be sure to report your error rate in your paper’s methods section.


Hi Pat,

I’m currently analysing some mock community 16S sequencing data and have a couple of questions.

The way I see it, you have two things you need to take into account in order to check that your methods are not altering the true biological signal in your sample (in this case the mock community):

  1. Presence/Absence: Check to see if after your experiments (and analysis) you have the same number of species as the mock communities.
  2. Abundance: Check to see if PCR amplification and other stuff is not altering the true abundance of your mock community.

What do you mean when you say to check the error rates? Of both presence/absence and the abundance profile?

If’ve checked and in my data there are 19 OTUs and they account for all species in the mock community except one (I used the BEI 782 mock community B even concentration).
However, I’m not quite sure how to analyse the abundance part of this data, if you could point in the right direction I would appreciate it.



  1. Presence/Absence: Check to see if after your experiments (and analysis) you have the same number of species as the mock communities.

Actually… I don’t really care about this. If you sequence 1000 reads you could get dead on 20 OTUs (or whatever) like you might expect. If you did 10,000 reads, you might get 50 OTUs. The difference is the sampling depth. The thing in common is the error rate. If we reduce the error rate, then we know we’ll still get extra OTUs, but the rate of extra OTUs is likely to be the same across you samples.

Now… if your PCR primers are biased (and they all are) then you are likely to not get things to amplify. We see this with the V4 primers - they don’t really amplify P. acnes. The typical 8F primer doesn’t amplify Bifidobacteria. Etc… So you have to pick your primers to match the expected biodiversity of your community.

  1. Abundance: Check to see if PCR amplification and other stuff is not altering the true abundance of your mock community.

This is a PCR issue, not a sequencing issue and the field has long suspected that there are biases in PCR using multi template DNA pools. To quantify this, you would really need to use qPCR to quantify the individual genomes in your template and then run the samples through your pipeline and see if what you get out is the same as the input. Because of errors in DNA quantification, pipetting, etc. you can’t take it on faith that what is billed as an “even mixture” is really even. In the end, however, what is one to do if the input and output don’t match? We get over this by acknowledging there are biases and asserting that the biases equally botch all of the samples.

Hope this helps.

Hi Pat,

Thanks for your helpful input.

I’ll calculate the error rates using multiple subsamplings at different sampling depths and see what I get.

Regarding the abundance part, I also suspect this bias is due to PCR amplification and I’ll keep this in mind when analyzing the data.

In case others are interested, I found this article that might be useful: