Compare Two Sigle-cell Sample for Differentially Expressed Genes (DEGs). How?

Question

[wangziwei0010]

I have two tumor samples (A & B) sequenced by 10X separately, each sample contains 1000 cells.

sample A is from cohort 1, and sample B is from cohort 2.

Now I want to compare 1000 cells in A with 1000 cells in B.

I know I can use wilcoxon test for the purpose and get something like 1000 DEGs.

But How could I know if the difference comes from sample instead of sequencing runs or cohort-driven batch effect?

Or is there any method to alleviate/consider the batch effect?

Answer 1

[ATpoint]

But How could I know if the difference comes from sample instead of sequencing runs or cohort-driven batch effect?

You cannot. It's perfectly confounded with the effect you want to test.

Or is there any method to alleviate/consider the batch effect?

No, single-cell obeys to the basic experimental design principles as any other assay.

[wangziwei0010]

Hi ATpoint,

Thank you for your though provoking answer. Big fans!

Actually I have some normal cells in each sample, and I found good mixture of those normal cells, while tumor cells are still separated by sample.

I'm thinking of performing the differential expression analysis anyway, and say 'look, the normal cells are mixed, which means individual heterogeneity, instead of cohort, dirves the variation!'

How do you think of that?

Wang

Answer 2

[teofoskol]

There are few ways, but you can't be 100% sure. 1st you can try to normilize with SCTransform (suppossed to remove some of the batch effect), try some intergration techniques ( ex. Harmony) for batch effect removal. Just be sure to not overdue it because you may also lose the biological relevant features.

[ATpoint]

It's confounded. SCtransform is not doing anything in such situations and hatmony operates on a per-cell embeddings level and not per gene as OP needs for differentiation expression. Especially the latter is one of the most common misunderstandings in single-cell.