scRNA seq integration

Question

[friguiahlem8]

Hello, I'm working on different replicate of 2 types of conditions (patients and control ) my question : should we created a seurat object for each replicate and don't merge then and then apply the normlization and batch effect seperately or we merge 2 conditions and then apply the batch effect ? second question : How to choose the appropriate method for batch effect (harmony or seurat or ... ) My ultimate objective is to determine differencial gene expression across the 2 conditions

Answer 1

[rpolicastro]

You usually process and normalize each sample separately and then integrate them afterwards.

The best integration method often varies by sample source and batch intensity, so it sometimes requires trying a few methods to find the one that works best for you. I've tended to have the best luck across all sample types and dataset sizes with scVI.

If your replicates are biological you'll want to use pseduobulk for differential expression.

[friguiahlem8]

thank you for your response. -Why it's better to process the normlization seperately and then create a list that combines all the seurat objects ?

• How we can judge if the integration method is good or not ?
• so before the pseudobulk , I should first apply the integration , determine the clusters and annotate them ?