(scRNA-seq) I have a dataset with 4 conditions (experiments) should I be anchoring aka integrating data for each comparison when looking for DEGs?

Question

I have an experiment with 4 conditions in it two of them are controls. When I am correcting for batch effects (either with PCA or CCA) should I restrict my comparisons between particular groups? So first I should do control 1 with control 2 then hopefully there will be no DEGs between them. Next I could do Control 2 vs Condition 1 then condition 1 vs condition2 and control 2 vs condtion 2. Sorry if this is confusing but I am trying to avoid going into the experimental details in the interest of keeping this post brief.

As a sort of follow up question can anyone suggest the "standard" or best seurat method for correcting batch effects. I've found some tutorials that recommend using CCA and others suggest PCA.

Answer 1

Batch correction/data integration is a tricky subject for scRNA-seq, see a few recent papers for benchmarks: https://www.nature.com/articles/s41592-021-01336-8 and https://genomebiology.biomedcentral.com/articles/10.1186/s13059-019-1850-9

Batch correction methods may not even impact differential expression. For example, if you correct the PCs, via a popular method like harmony, then this will impact the t-SNE/UMAP plots but will not impact the actual expression values in the counts matrix used to detect DEGs.

For detecting DEGs, the most common methods are simple stats tests like Wilcoxan Rank test that largely ignore batch effects. However, these are likely not the best approaches. I suggest this recent paper (https://www.nature.com/articles/s41467-021-25960-2). Here they state that the best methods for detecting DEGs consider biological replicates, like is done with bulk RNA-seq. By considering replicates, this is effectively like batch correction