Seurat: FeaturePlot fails when first cell has NA value

[milanmlft]

Hello Seurat devs,

I ran into an unexpected problem when trying to use FeaturePlot to visualize pseudotime values (which I store in the @meta.data slot of my Seurat object) on a reduced dimension embedding. Some cells might contain NA values if they are not part of a particular trajectory. This is in general not a problem as FeaturePlot will just color those cells in a dark grey color. However, when the first cell/row of @meta.data contains an NA value, then FeaturePlot throws the following error:

Error in if (all(data.plot[, feature] == data.plot[, feature][1])) { : missing value where TRUE/FALSE needed

This is rather unexpected and seems to be a problem that could occur with any variable where the first cell has a missing value.

Is there an easy fix for this or would you consider implementing a more permanent bug fix for this behavior?

I created a reproducible example below using pbmc_small and some random pseudotime values. I tried to simply reorder the meta.data data.frame so that the NA values would appear last, but this did not solve the problem.

Thanks a lot for your help!

Example

library(Seurat)
pbmc_small

# create random pseudotime variable
set.seed(42)
pseudotime <- sample(1:dim(pbmc_small)[2])

# introduce random NAs
na_idx <- which(pseudotime %in% sample(pseudotime, 15))
pseudotime[na_idx] <- NA

# add pseudotime to seurat
names(pseudotime) <- rownames([email protected])
pbmc_small$pseudotime <- pseudotime

# FeaturePlot with pseudotime (works)
FeaturePlot(pbmc_small, features = "pseudotime", reduction = "tsne")


# when first cell has pseudotime == NA
pbmc_small$pseudotime2 <- pbmc_small$pseudotime
pbmc_small$pseudotime2[1] <- NA
# does not work
FeaturePlot(pbmc_small, features = "pseudotime2", reduction = "tsne")

# reorder meta.data so that first cell is not NA anymore
[email protected] <- [email protected][order([email protected]$pseudotime2),]
# still doesn't work
FeaturePlot(pbmc_small, features = "pseudotime2", reduction = "tsne")

[yuhanH]

Thanks for the bug report. We will fix it in the next version.
Right now, you could do is to reorder the cell input rather than the meta.data table.

new.cell.order <- rownames([email protected])[order([email protected]$pseudotime2)]
FeaturePlot(pbmc_small, features = "pseudotime2", reduction = "tsne", cells = new.cell.order )

[mojaveazure]

This should be fixed in the development version of Seurat. To install the development version of Seurat, please see the instructions here.

[riselinuh]

Hi,
I am using seurat V3 to analyze v2 chemistry data. I have the same problem. Could someone advise me what to do next? Thank you.

head(x = Top[[]])
orig.ident nCount_RNA nFeature_RNA percent.mt
AAACCTGAGAAACCAT-1 Top 0 0 NaN
AAACCTGAGAAACCGC-1 Top 0 0 NaN
AAACCTGAGAAACCTA-1 Top 2 2 0.000000

VlnPlot(Top, features = c("nFeature_RNA", "nCount_RNA", "percent.mt"), ncol = 3)
Error in if (all(data[, feature] == data[, feature][1])) { :
missing value where TRUE/FALSE needed