Can densMAP (in development version) yield reproducible clustering?

[NicolaasVanRenne]

densMAP seems to work (hallelujah!) but the clustering is not reproducible. If I run the same pipeline twice over, the clustering is similar yet different. Is there a way to make it reproducible?

thanks,

Nicolaas

ps command I run is:
RunUMAP(MySeuratObject, dims = 1:UMAP.dimensions, densmap=TRUE)

[k3yavi]

Hi @NicolaasVanRenne ,

I think it's unrelated to Seurat and rather the feature/issue of generating low dimensional embeddings using UMAP as it can result in different looking (although topologically similar) clusters in multiple runs of the same data. Although you can try setting the seed like set.seed(12) and check if that helps.

[NicolaasVanRenne]

Thanks for replying @k3yavi ,

However, I always apply set.seed as you mentioned. It does not work to make densMAP reproducible. Getting reproducible plots is important to me - as I would prefer my analyses to be reproducible...

Maybe @lldelisle can help here? As she wrote the code as I understood from #4630

Any thoughts?

Nicolaas

[lldelisle]

Hi,
I am sorry. I cannot really help. Apparently the seed is set in python by:

seurat/R/dimensional_reduction.R

Lines 1268 to 1270 in 421f472

	if (!is.null(x = seed.use)) {
	py_set_seed(seed = seed.use)
	}

And it is set to 42 by default:

seurat/R/dimensional_reduction.R

Line 1681 in 421f472

seed.use = 42L,

...
Sorry...