References request: deciding what duplicate samples to include to avoid batch effects with previous studies

Question

Suppose there is cohort with many samples to assay. In phase I of the project, subjects with a certain disease incidence and healthy controls were assayed. Now, in phase II of the project, we are going to assay samples with another disease and analyze them with healthy controls from phase I of the project. From past experience, we know that we have to assay healthy controls from phase I (whose bio samples are still available for assay) to avoid batch effects between phase I and phase II of the project. However, I don't know how much duplicate samples I should include and how the duplicates should be distributed.

My question is certainly related to experiment design, and perhaps there have been extensive discussions on this kind of design, but I don't know the exact term of this kind of experiment design such that I couldn't find anything relevant to my case in the literature. Can anyone point me to references related to this kind of experiment design?

Answer 1

at least six biological replicates should be used, rising to at least 12 when it is important to identify SDE genes for all fold changes.

See this paper for an explanation of why six replicates is good for differential expression analysis:

https://dx.doi.org/10.1261/rna.053959.115

When looking at conditions, it's necessary to think about what exactly needs to be compared (ideally prior to doing any experiments or sequencing, because that reduces the cost and frustration of re-doing things).

I recommend you consult with a statistician to find out the appropriate mixture of samples for your experiment, because there can be subtle differences between studies that mean a particular design does not work.

One example from our own investigations is that we were doing cDNA sequencing of cell lines with and without mitochondrial genomes, and it was only after doing two years worth of sequencing and looking at the data that we realised we'd never done any sequencing runs that only involved cell lines without mitochondrial DNA - our rationale for doing that was to reduce batch effects. This meant we couldn't perfectly distinguish between barcode spillover and a very small proportion of mitochondrial DNA in our "mitochondria-free" samples.