Seurat v5 and how to correctly integrate across multiple experiments
0
0
Entering edit mode
14 days ago

I am using Seurat v5 to combine data from my own experiments, data from a publication and data from an open portal.

I have all the libraries imported at Seurat libraries and I have ran merge() on the dataset to get a single Seurat objetct. With v5, the concept of layers has been introduced. When I read the vignette for integrative analysis in Seurat the example given is that of different technologies assaying the same cell types. My data is different experiments on the same technology (10x scRNA) and likely the nature of each sample is different given that I am collecting tumours from patients across pediatric high-grade glioma.

When I merge my data, I end up with 60 ish layers (one for each 10x cell library). Then, following the vignette steps:

obj <- NormalizeData(obj)
obj <- FindVariableFeatures(obj)
obj <- ScaleData(obj)
obj <- RunPCA(obj)

The steps are ran separately on each layer which, to me, seems counterintuitive to the reason for running integration.

Could someone give some insight / suggest the correct sequence of merge, integrate, merge layers, dimensionality reduction etc for my use case?

Seurat scRNA R • 201 views
ADD COMMENT
0
Entering edit mode

In my experience integration methods are also often used for different samples/batches across the same technology. The crucial thing is to evaluate if and how your samples are indeed affected by sample/batch-specific effects. You can use UMAP plots before and after integration and check for things such as: before integration, do your cells cluster by sample/batch, or by cell type, or by condition? After integration, do the cells now cluster by cell type? Before-after integration plots may also help you to detect clusters of low-quality cells, etc. because they often come together. Nonetheless, there are many integration methods and the strength of the integration (i.e. how much "variability" which can also be biological variability they remove) varies a lot. It also depends on your biological question, what you want to find, why do you want to integrate, is it just to help annotate cell types, is it to clearly group together some conditions, etc.

ADD REPLY

Login before adding your answer.

Traffic: 2672 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6