Forum:What counts as a valid biological replicate in single cell RNAseq?

Question

[david.ferenbach]

I've run an experiment where I collected orans from 3x healthy control mice, and 3x post-injury mice - and ended up generating around 3000 individual single cells from each sample. For consistency and cost reasons, we pooled our 3x single cell sorts together into 2 lanes of the same 10X chip - so generated 2 samples (but each derived from 3 original mice).

Does this represent a publication-quality dataset by 2022 standards?

And (hoping that it does) - is it valid to use these datasets to identify any novel cell types which are gained/lost in the injured organ?

Finally - what would the best tests be to identify the different populations? and would it be essential to confirm this in a second RNAseq dataset? and would it also be essential/preferable to provide additional biological verification using completely separate approaches - e.g ISH, antibody-co-staining etc..

Any tips would be hugely helpful!

Answer 1

[benformatics]

I think the hard part here would be the validation on the wet-lab side.

Sure you can identify "novel" cell populations in your comparative single-cell genomic experiments. The reality is you could (and almost most definitely will) have a wide range of non-novel cells that change their transcriptional profile between the normal vs. injury condition. It is probable that many of these have been profiled (on a non-transcriptomic level) and even more likely is that the novel transcriptional signatures are actually just known cell-types that are activated/primed in different directions. If they were actual new cell-types you would need to prove this by looking at expressed proteins, cell-receptors, etc...

If you manage to discover something new, I don't think this can be published in a "big" journal without:

1. very very solid 'additional biological verification using completely separate approaches'
2. the novelty and promise of the new cell types would have to be substantial (e.g. you identify a subspecific subset of normally "X" cells or new "Y" cells that are clearly regenerating tissue etc...)

If you just want to publish, the longer you wait the lower the "hotness" becomes as there already a lot of injury/healthy single cell mouse comparisons published (this was from a 30 second google search):

https://rupress.org/jem/article/218/8/e20210040/212391/Single-cell-analysis-of-the-cellular-heterogeneity

https://www.pnas.org/doi/10.1073/pnas.2005477117

https://www.nature.com/articles/s41419-022-04864-z

Answer 2

[ATpoint]

Did you hashtag the mice, so can you separate them in silico? Towards the "quality standard", that entirely depends on the reviewers you get. If you find interesting biology that you properly validate then probably nobody will question your dataset. But if it is all spurious findings, without validation, no biological replication due to pooling mice and not hashing them -- they might go off on that. Nobody can predict.

It is hard to comment more without knowing details. Basically, you would do integration and clustering, and then see whether clusters appear to be specific for the injury condition. If available use published gene sets to annotate the clusters. See what markers they express and whether the markers of the "new" clusters in the injury condition suggest that a different celltype gets recruited to the injury site or starts homing there.

Answer 3

[Istvan Albert]

The reality is that the less we know about a system the more valuable any insight is - thus even in RNA-Seq no replicates at all were acceptable early on, today 3 would be the absolute minimum and probably 4-6 would be preferable.

In you case since you are pooling the RNA it is a sort of replicate, but won't show the individual variability.

As ATpoint mentioned what the reviews think and matters the most. If your results provide evidence towards a previously uncharacterized mechanism then they would judge it differently than if it attempts to refine an existing framework.