shdam
commented
3 weeks ago Hi there,
Correcting batch effects in longitudinal data is a bit challenging. The best approach is to include the same representative sample (i.e., one or more technical replicates that have every expected cell type) in every batch/time point, which works as an anchor sample. Then, you can correct using an anchor, as described at the bottom of the GitHub Readme.
Without including an anchor sample, I am unsure if a method exists that can handle batch effects appropriately.
Always bear in mind the minimum requirement for batch correction:
The only requirement is that at least one condition from each batch is present in at least one other batch (ref)
Where "condition" can be interpreted as timepoint or a representative sample.
Even when closely controlling experiment conditions, it is my experience that there is always some form of batch effects, but the effect is not always impactful to the results. Without replicates or overlapping conditions, it is very hard to assess.
I hope this answer is helpful, and best of luck with the analysis!
Best regards, Søren
gavin-oliver-xilis
I have a flow cytometry dataset of 10 samples/5 markers measured across 10 timepoints in which I am interested in monitoring marker expression across the timepoints longitudinally.
While we aimed to control conditions as closely as possible, I am curious about the possibility of batch effect assessment/correction.
Since the variable of interest here (time) is confounded with the batch (10 timepoints), is the possibility of batch effect assessment and correction with CyCombine precluded?