I have an experimental design where we change the genotype and the treatments. The replicates were collected at different times, so I am assmuing there is a batch effect. I have two questions here -

1. To build the DESeq object, I am using the ~Batch + treatment + genotype + treatment:genotype model. DESeq2 automatically converts the treatment and genotype to factor variables, but not the batch effect. So, how would it matter if Batch was considered as a factor or as a continuous variable?

2. When I do the PCA after removing batch effect using limma's removeBatcheffect, the samples still cluster into distinct groups as we would expect it to. So, do we still need to model for the batch effect? Is there anything else that determines if the batch effect would change the further downstream analysis drastically?

Thank you so much!

Just convert your categorical variables to factors ahead of running DESeq2. Just because DESeq2 recognizes characters as factors, you should still convert batch to a factor yourself. Running as numeric will give nonsensical results. Batch 3 =/= Batch 1 + Batch 2, etc.

Re: PCA and removing batch, can you show the plot? You mean they cluster by batch after removing batch? Or by some other grouping that is not related to batch? Could be other correlations, e.g. cage or parent, if working with animals. Good to collect lots of metadata so you can try to figure out the source.