Overlapping clusters for different biological conditions: Seurat, UMAP

Question

[bio_info]

Hello! I am currently working on some published scRNA-seq datasets and have implemented a Seurat pipeline . I have 141,583 cells after QC steps and have performed the necessary steps of normalisation, scaling, dimentionality reduction etc. I have constructed a UMAP but it does not look at all, with clusters for the 'control' and 'disease' condition overlapping. I have 13 clusters here.

I have tried clustering with default parameters and have played with the resolution, taken reduced dims etc. Can someone give some advice how I can fix this? This is the code I am running. The conditions 'Crohns' and 'Control' should not be overlapping.

# Compute neighbors and clusters -----
sobj <- FindNeighbors(martin2019, dims = 1:10)

sobj <- FindClusters(martin2019, resolution = 0.2)

# Run UMAP -----
sobj <- RunUMAP(sobj, dims = 1:10, metric = "euclidean")

# Visualise UMAP -----
DimPlot(sobj, reduction = "umap", raster = F, label = T)

[Bastien Hervé]

What is the need of PBMC in your analysis ? If none, removing them from the beginning will allow your clusters to be more specific to your remaining cell types.

Can we have a look at your Elbow plot after your dimension reduction ? How many features are you using to do your PCA ?

Playing with the resolution will not affect your network, just the number of communities.

[bio_info]

I guess I was confused about requiring PBMCs in the same Seurat object. I have removed the PBMCs and now the clustering looks a lot better. I am taking 10 dimensions and computed 5000 features.

Thank you for the suggestion!