I´m trying to figure out if there is a way to deal with technical replicates when coming from different batches of sequencing.
I do not want to remove the batch effect from the data, I want to include it in the statistical model. So, is there a way I can use these technical replicates?
I came across this post using duplicateCorrelation with limma+voom for RNA-seq data mentioning the function duplicateCorrelation from limma, which seems to deal with technical replicates.
Would I be able to apply this to my technical replicates knowing that they come from different batches?
In this case, different batches mean the samples are run at a different time-point, so same protocols for library prep. and same sequencing technique are used.
I am not entirely clear what you mean by "technical" replicates in this case. Are they resequencing of the same RNA samples or are they fresh RNA samples from the same organisms? You say that the library preparation protocol was the same at both times, which implies that library prep had to be redone at each time, so it is not just resequencing of the same libraries.
If you were just resequencing the same samples or the same libraries then it is common to simply add the read counts from the technical replicates, for example by edgeR::sumTechReps().
Have you made an MDS plot to check that there actually is a batch effect?
In general, limma handles batches in the same way whether they arise from technical or experimental causes. duplicateCorrelation and batch effects in the design matrix are both options, depending on the design of your experiment.
The situation I have is a little tricky: I have a sequencing run where controls and treatments are sequenced in 2 different batches. And here different batches mean different lib. prep and sequencing. And I want to re-sequence a few of the same RNA samples to get an estimate of the batch effect (as for now it is impossible to dissociate from the treatment effect). The problem is that I can´t re-sequence everything, so I will re-sequence a few so in the end I will have samples in my dataset twice, and this is what I meant with technical replicates.
So, I was wondering what´s the best way to deal with these technical replicates.
Hi Gordon,
Thank you for your reply!
The situation I have is a little tricky: I have a sequencing run where controls and treatments are sequenced in 2 different batches. And here different batches mean different lib. prep and sequencing. And I want to re-sequence a few of the same RNA samples to get an estimate of the batch effect (as for now it is impossible to dissociate from the treatment effect). The problem is that I can´t re-sequence everything, so I will re-sequence a few so in the end I will have samples in my dataset twice, and this is what I meant with technical replicates.
So, I was wondering what´s the best way to deal with these technical replicates.