This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Doubt about single Cell Data integration

Hi all,

I am conducting a Single Cell Analysis with data comming from different batches. Cells have been sorted so that I mostly have Melanocytes in my dataset. I performed the QC filtering for each of the batches separately and I could see how the majority of my cells are actually Melanocytes via SingleR.

Now I would like to analyze my data, so my first step was to look at the UMAP i created when merging my Seurat objects, only to find that my cells where separated by batch, which is something I knew that would happen and now I would like to integrate my Seurat objects into a single one.

My problem here is that my batch distribution is this:

  • Batch1: Untreated Melanocytes
  • Batch2: Treated Melanocytes (Treatment 1)
  • Batch3: Treated Melanocytes (Treatment 1)
  • Batch4: Treated Melanocytes (Treatment 2)

So that the batch effect is also the condition. I was also thinking about working only with the Melanocytes as the 90% of the cells are Melanocytes and then perform the integration.

Is my set-up a good for integration ?

10x single cell harmonization integration

For integration it might work but you cannot do any meaningful DE analysis with such a, frankly, terrible design. What was the point of the experiment in the first place?

The point is to see the differences between treated (with Treatment 1), treated (with Treatment2) and control Melanocytes. Also just to clarity if necessary, by different bathces I mean that the cells have been sorted differently and that then they have been sequenced in different runs so each run/batch has different Melanocyte composition.

I understand. Batch is fully confounded by condition, so in terms of differential expression you (formally) cannot do any comparison. Any differences can, and a lot of it will, be by the batch. As said, it's a most suboptimal experimental design.

Yes ... I understand this is the worst case scenario, but the data was sequenced before I even arrived at my job and they would like to find differences, I was thinking about:

  • Perform QC in each sample and get the cell type annotation by cell with singleR (already done).
  • Subset only the cell type I am interested in and re-work the data.
  • Merge the different datasets and cluster them.
  • Try to perform the harmonization/integration if required.
  • Start with marker indentification.

What do you think ?

You can do that but as said, any observed differences can and probably largely will be batch.

0 answers

No answers yet.

Log in to answer this question.