Using existing clustering reference

[vcglinnaeus]

Is it possible to assign a cluster instead of using the clustering algorithm in the package? I love Seurat and it is easy to use but for cases where biologists have prepared a clustered data as a reference, how can I use them. I tried reassigning it to active.idents, but I guess it doesnt work like that?

[f6v]

You could assign clusters/cell types by doing so:

your_cell_types <- ... # load from file
your_seurat_obj$cell_type <- your_cell_types

And then you can subset Seurat object:

your_seurat_obj[, your_seurat_obj$celltype == "Neuron"]

[vcglinnaeus]

This one has no association with clusters yet. I am trying to do a pca, via elbow joint i can see that that optimal k is 15. but i need the clusters to assume a k equivalent to the given clustering (data from wet lab). Thank you. And also I love Seurat it helped me a lot learning sc

[f6v]

If you specifically want to assign clusters, couldn't you use Idents?

Idents(your_seurat_obj) <- ... # vector of identities

[timoast]

@vcglinnaeus the elbow plot is used for determining the number of PCs to include in downstream analysis steps, not the number of clusters present in the data. I'm not completely sure what you're asking, but you might find the reference-based mapping vignette useful: https://satijalab.org/seurat/articles/integration_mapping.html

[vcglinnaeus]

Good day, I have tried it. But the number of clusters from the seurat clustering doesnt match mine. Well I tried specifying k = 33 (for example) but im not getting a cluster of k=33 but of 10,11 or 12 instead. It is like following the optimal k as observed from the elbow joint.

…

On Tue, Mar 16, 2021, 6:00 PM Igor Filippov ***@***.***> wrote: If you specifically want to assign clusters, couldn't you use Idents? Idents(your_seurat_obj) <- ... # vector of identities — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#4233 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQEB5H4LCO4JO7G3RLSQ4GLTD6FCTANCNFSM4ZHINJ4Q> .

[timoast]

Seurat does not use k-means clustering, but uses a graph-based community detection method. The k.param parameter in FindNeighbors refers to the number of nearest neighbors to find. Please see the documentation for FindNeighbors and FindClusters for more information.