search

kstreet13 / slingshot

Functions for identifying and characterizing continuous developmental trajectories in single-cell data.
259 stars 42 forks source link

What assay to use after Seurat integration #200

Closed nickhir closed 1 year ago

nickhir commented 1 year ago

Hi,

I have 8 different experiments that I have integrated using Seurat. I now want to perform trajectory analysis with slingshot, but I am unsure what assay to use (either seurat@RNA$integrated or seurat@assays$integrated).

In issue #59 you mentioned that you "fell most comfortable using the integrated counts". However, in the tutorial that you mention in the same issue ´, I am pretty sure that you are using the "unintegrated" counts (i.e. the RNA assay), because you are specifically exporting the RNA assay back to a single cell experiment and not the "integrated" assay.

## convert back to singleCellExperiment
sce <- as.SingleCellExperiment(integrated, assay = "RNA")

Do you have any new insights or recommendations regarding this issue?

Any help is much appreciated!

kstreet13 commented 1 year ago

Hi @nickhir,

This is a very good question and there's certainly room for debate, so I can only give my recommendation, not a definitive answer.

So, we did pull a little slight-of-hand in the tutorial code, but it still generally represents my recommended workflow. We used Seurat to calculate the UMAP dimensionality reduction based on the integrated/normalized expression values. This UMAP was used for Slingshot. But since we needed the original raw counts for tradeSeq, we exported the RNA assay when converting back to a SingleCellExperiment. That way, we had the UMAP based on normalized counts for Slingshot and the raw counts for tradeSeq. In general, that is what I would recommend (although I now generally use PCA for Slingshot, and just UMAP for visualization).

Hope this helps and let me know if anything is still unclear! Best, Kelly

nickhir commented 1 year ago

Thank you very much for your answer. This definitely cleared things up for me!