Thank you. I am having trouble understanding how the batch correction would affect the DE. For example, let's pretend I did batch correction when selecting highly variable genes and then I clustered the data using the leiden algorithm. Then I did DE between the clusters and used the markers and identified the cell types. Then if I decide to do DE between the control and treated for each cell type why would the batch correction affect the data? I use scanpy and I checked that my adata.X is the same after normalizing (before selecting HVGs) and after doing the leiden clustering.
Help with scanpy batch correction
Hello
I am a bit confused about hatch corrections. Scanpy says you can specify batch key when choosing highly variable genes. From what I understand using the batch key may help with identifying the cell types. If I have 2 conditions (control and treated) can I use condition as the batch key? If I do that and then assign the cell types can I do DE between the 2 conditions for each cell type or will it affect the results? It seems like the batch key affected the highly variable genes and PCA but it does not affect the adata.X which is used for DE.
Thank you
1 answer
Batch correction or "blocking" for a group in gene selection helps that the clustering and visualization (UMAP, dimreductions) are dominated by celltype differences (so genes varaible between cells) rather than variable due to the group. Usually, in most cases, you want to block for any group differences during clustering and dimensionality reduction. So per-cell analysis. Later, in DE, which is per-gene you of course preserve that difference and explicitely test for the difference, adjusting for other pontential confounders, such as true batches, or other sources of unwanted variation.
No it would not affect, because the labels control vs treated are predefined by your experimental design. Just the celltypes is what you have to figure out, and even this you could do by something like SingleR with a reference dataset so the whole clustering and UMAP thing could be replaced. But usually people do cluster-markerDE-annotation, that's fine.
Log in to answer this question.
More posts like this
-
Which key should I use with Harmony?
written by bioinfo •
-
How to use harmony with scanpy?
written by bioinfo •
-
How to run trajectory analysis with only one cell type using palantir?
written by bioinfo •
-
How to assess which cell population is most affected by treatmant in single cell rna seq data?
written by abbas89 •
-
Can I perform DE analysis with Seurat if I only have one sample per condition?
written by bioinfo •
-
DESeq2 multiple group comparisons
written by Bhumi •
-
How to interpret highly variable genes plot in Scanpy?
written by Pratik
-
Cell lineage identification from pathway anlaysis
written by rob.costa1234
-
Why DE genes are not among highly variable genes between groups
written by Björn •
-
Advice for the following PCA analysis
written by Mozart
