umap plot with speckles

[doctorjb77]

I merged 8 objects and ran SCTranform and I'm getting these speckles on the umap that I have not seen before. What do these speckles signify?

[andrewwbutler]

Hi,

It's not really possible to say much without more information about the data and how you processed it. Very generally, something probably went wrong upstream of RunUMAP.

[no-response[bot]]

This issue has been automatically closed because there has been no response to our request for more information from the original author. With only the information that is currently in the issue, we don't have enough information to take action. Please reach out if you have or find the answers we need so that we can investigate further.

[jacksonloper]

@doctorjb77 did you ever figure it out? we get speckles too sometimes and can't figure out what it means

[JarneBelien]

@doctorjb77 I am getting a super similar result - did you ever figure out the reason?

[BenxiaHu]

@JarneBelien did you figure it out?

[JarneBelien]

@BenxiaHu I don't think there was anything particularly wrong because I did all the possible troubleshooting. I came across other people who sometimes also saw this. Solution: try different normalization and integration methods... After Harmony integration eg this did not occur anymore.

[BenxiaHu]

@JarneBelien thanks a lot. can you show the specific codes?

[JarneBelien]

@BenxiaHu I just checked my code again, and actually after all different integration methods (so also Seurat integration), the UMAP normalized, so I guess it was mainly due to the different samples/datasets not being properly integrated before visualisation.

For Seurat integration, I used the workaround appointing reference datasets to lower memory requirement as such:

Integration_Features <- SelectIntegrationFeatures(Split_Blood, nfeatures = 3000)

Split_Blood <- lapply(X = Split_Blood, FUN = function(x) {
    x <- RunPCA(x, features = Integration_Features, verbose = FALSE)
})

options(future.globals.maxSize = 20000*1024^2)

Split_Blood <- PrepSCTIntegration(Split_Blood, anchor.features = Integration_Features, verbose = FALSE)

Anchors <- FindIntegrationAnchors(object.list = Split_Blood, reference = c(1, 5, 10, 13, 17), reduction = "rpca", dims = 1:50, normalization.method = "SCT", anchor.features = Integration_Features)

For Harmony, you can just follow the vignette: https://portals.broadinstitute.org/harmony/articles/quickstart.html

Good luck!

[BenxiaHu]

thanks a lot. do you follow this one: https://satijalab.org/seurat/archive/v4.3/integration_introduction

[JarneBelien]

You can indeed follow that one. I followed a workaround that's less memory intensive, the one using recirpocal PCA: But that vignette is not supported anymore.

[BenxiaHu]

thanks a lot. I try this one: https://satijalab.org/seurat/archive/v4.3/integration_introduction

but still see many cell speckles.

[BenxiaHu]

organoid <- readRDS("snRNAseq.rds") organoid <- SplitObject(organoid, split.by = "SampleID")

organoid <- lapply(X = organoid, FUN = SCTransform) features <- SelectIntegrationFeatures(object.list = organoid, nfeatures = 3000) organoid <- PrepSCTIntegration(organoid,anchor.features = features) anchors <- FindIntegrationAnchors(organoid,normalization.method = "SCT",anchor.features = features) organoid <- IntegrateData(anchors, normalization.method = "SCT") organoid <- RunPCA(organoid, dims=1:50,verbose = FALSE) organoid <- RunUMAP(organoid, dims = 1:50,umap.method = "umap-learn", metric = "correlation",verbose = FALSE) organoid <- FindNeighbors(organoid, dims = 1:50, verbose = FALSE) organoid <- FindClusters(organoid,verbose = FALSE) DimPlot(organoid, raster=FALSE)

[BenxiaHu]

@JarneBelien do you see anything different from your code?

[JarneBelien]

I do not see an overt difference in the code no. You could try several things:

• make your preceding QC filtering more strict
• perform normalization and scaling in a different way, i.e. not through SCTranform
• try a different integration method eg Harmony or SCVI

Good luck!