We send some samples to sequence and we got several (fastq.gz) files for each sample. The files are distributed at two or three folders with different dates (more than a week apart). The dates of the folders where the data is in:


I asked the facility about why there are the same file in different folders, and they told us that:

we have split sequencing of that project on different flowcells, so all files are correct and you have to merge the individual (duplicate) files to get the requested total read numbers.

How can different flow cells be from the same experiment and be run with more than a week of difference? (I thought that all the flow cells are run at the same time in each sequencing)

Can be there some batch effect? How can I check if there is batch effect if we merge the files before trimming, mapping and counting?


1 Answer 1


Usually, a single sequencing library is created from each sample. Assuming you are using barcoded indexes to identify multiple samples, the index is attached to each library, and the libraries are pooled to make a library pool. A library pool can be frozen, and is very stable.

A pool can be sequenced multiple times by pipetting a portion of the pool and supplying it to the sequencing process. Each time, a sample of the molecules in the pool will be sequenced.

Generally, it is true that sequencing runs are very repeatable, and that you will not suffer any batch effects.

If you want to check for batch effects you would need to analyse all of your sequence files separately, and check this first, or in parallel, to your main analysis.

One thing you might want to check is whether the same library pool was sequenced multiple times, or whether separate pools or libraries were made for each sequencing run. In the second case, there may be more likely to be batch effects.

  • $\begingroup$ Oh, thanks. I think they submitted a portion of the library pool because the index are the same. However, why they didn't add more sample from the pool from the beginning? $\endgroup$
    – llrs
    Sep 30, 2019 at 10:27
  • 1
    $\begingroup$ Regarding your sample pooling question, probably this is not happening in your case but here is our workflow, which generates multiple output files per sample like you get: We use 10x and the most expensive part is the sequencing run. As single cells are delicate and not all "droplet formation" steps go smoothly, first we do "shallow sequencing" on NextSeq to see if there are enough number of good quality cells in a given sample (after mapping, read counting, single cell QC) and then continue with deeper sequencing (one or more runs) with NovaSeq until we reach saturation with number of reads. $\endgroup$
    – haci
    Sep 30, 2019 at 12:30
  • $\begingroup$ The number of sequences has a limit per sequencing run, even if you add more of the pool to the run - it is limited by the flow cell. For this reason, it is normal to only sequence a part of the pool in any one run - enough to fill the flow cell - and if you need more reads, then carry out more runs with additional flow cells. $\endgroup$
    – Jay Moore
    Sep 30, 2019 at 13:30
  • $\begingroup$ Another alternative would be to allocate some samples to one run, some samples to another run, and to fill up the flowcells that way. However, this would mean that, if there are any batch effects it would be much more difficult to separate them from sample effects. At least if all samples are included on every flow cell, the batch effects will not affect any between-samples comparison. $\endgroup$
    – Jay Moore
    Sep 30, 2019 at 14:22
  • $\begingroup$ @Jonathan Well, in our case not all samples where included in all the runs, that's partially why I ask about the batch effect. $\endgroup$
    – llrs
    Oct 2, 2019 at 7:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.