Harmony batch correction vs. regressing out donor effect in Seurat

Question

[Lucy]

Hi,

I am performing combined analysis of three scRNA-seq samples from different donors generated using 10x Genomics technology. I am trying to decide whether to use Harmony (https://www.biorxiv.org/content/10.1101/461954v2) to remove donor effects or whether to regress out donor during the sctransform step (https://satijalab.org/seurat/v3.0/sctransform_vignette.html) in the Seurat pipeline.

What are the relative advantages and disadvantages of the two approaches?

Best wishes,

Lucy

[igor]

You might also find this recent discussion helpful: about batch correction in scRNA-seq

Answer 1

[jared.andrews07]

You should not use SCTransform to regress out batch effects. Rather, you should use one of Seurat's integration methods. Seurat also has a number of wrappers around different integration methods, including Harmony.

Personally, fastMNN has worked well for me, but it's usually worth trying a few methods, as they don't all perform similarly across all datasets.

[Lucy]

Thank you. What is the reason to not use SCTransform to regress out batch effects?

[jared.andrews07]

The authors don't recommend it, as it's only meant to remove minor variation/biases due to differences in cell cycle, reads in mitochondrial genes, etc. rather than batch effects that affect every cell in a much more systemic manner. Their docs are a bit of a mess, so I can't find a specific reference, but if you dig around their github issues page, you can find threads that talk about it. Additionally, you can read their preprint about sctransform and/or their paper on integration specifically that shed more light on this.

[Lucy]

Great, thanks for the help!

[saeedabdurahman]

Hi @Jared.andrews07,

It is an interesting response for me. I have now tried both Harmony and Seurat Integration. But how did you try FastMNN? For eg; is there a package which I can use? Or It will be great if you can share a public github link that was useful for you or any notebook that you have prepared yourself?

Thanks in advance!

[jared.andrews07]

See the Seurat wrappers page.

[623202215]

Hi,

I have 3 patients with normal and tumor tissue sample(10× technology), there are six samples in total. I want combine them and find whether there is some difference between normal and tumor sample in specific cell type (such as immune cell). I am struggle to choose integrate method. For the seurat integrate method, SCtransform method, harmony method, which one is more suitable for my case, further, should I regress out donor when perform scaleData in seurat? Thanks a lot, hope to get your suggestions!

Best, Wei

[jared.andrews07]

Please post this as a new question.

[623202215]

Hi jared,

I have post a new question, hope to get your suggestions!Thanks a lot! Here is the new post

Best, Wei