Using counts matrix from Seurat object as input

[daniannr]

Hi,
I'm trying to use STgradient to find any correlation between gene expression and distance. I used Seurat for preprocessing and normalization and I'm wondering if I can extract the counts and coordinates from the Seurat object as input?
I tried creating an STlist using the matrices but it didn't work. See error below:

sGE_1014rep2 <- STlist(rnacounts= rep2_1014[["Spatial"]]$counts, spotcoords = GetTissueCoordinates(rep2_1014), samples = NULL)

Error in STlist(rnacounts = rep2_1014[["Spatial"]]$counts, spotcoords = GetTissueCoordinates(rep2_1014), : 
 Input not recognized. Please refer to documentation.

The counts matrix is 32285 x 2597, where the rows are genes and columns are Visium spots.
The coordinates matrix is 2597 x 2, where the rows are Visium spots and columns are x and y coordinates.

Additionally, when I specify samples = "rep2_1014", this error pops up:
Error in detect_input(rnacounts = rnacounts, spotcoords = spotcoords, : Number of sample names do not match number of RNA counts tables.

There's only one sample in the object.
Thanks!

[oospina]

Hello @daniannr
You would need to use data frames in a named list:

# Create data frame from counts
# Add column with gene names
# Put gene names as first column
# Remove row names
counts_df = as.data.frame(rep2_1014[["Spatial"]]@counts)
counts_df[['gene']] = rownames(counts_df)
counts_df = counts_df[, c( 'gene', grep('gene', colnames(counts_df), value=T, invert=T) ) ]
rownames(counts_df) = NULL

# Create data frame from coordinate data
# Add column with spot names (i.e., barcodes)
# Put barcodes as first column
# Remove row names
coords_df = GetTissueCoordinates(rep2_1014)
coords_df[['barcode']] = rownames(coords_df)
coords_df = coords_df[, c('barcode', 'imagerow', 'imagecol')]
rownames(coords_df) = NULL

# Make named list with count and coordinate data frames
# Names should match
counts_list = list(sample1=counts_df)
coords_list = list(sample1=coords_df)

# Create STlist
sGE_1014rep2 <- STlist(rnacounts=counts_list, spotcoords=coords_list, samples=NULL) 

Alternatively, it is possible to use the Seurat object directly. However, this option hasn't been tested much.

sGE_1014rep2 = STlist(rnacounts=rep2_1014)

Hope that helps!

Oscar

[daniannr]

Hi Oscar,

I just used the second solution and it worked! I read the STlist documentation but it didn't say that you can use a Seurat object directly.
I managed to get STgradient to work and note to other users who might encounter this issue – you would need to use the transform_data function before using STgradient since it searches for its input in the tr_counts slot.

An additional question for Oscar or other contributors:
Does the STgradient use a correlation matrix and is there a way to extract that?

Thanks!

[oospina]

Thanks a lot for suggesting those additions. We will put those once we test more using Seurat objects as input.
You probably realized this before, but if wanting to SCT counts from Seurat, for example, you could create a list of sparse matrices and put it in the x@tr_counts slot... Of course, use at your own risk as we haven't tested that before 😄

With regards to STgradient, it calculates an Euclidean distance matrix based on the spatial coordinates (not saved in object). The averages or finds the minimum distance of each spot/cell to the spots assigned to the reference cluster. Then calculates a Spearman's coefficient between the average/min distances and the normalized expression of each gene.

[oospina]

• Test Seurat as input to STlist
• Add option to STlist documentation (clarify implementation is for old Seurat object, i.e., not Seurat v5)
• Add note that transform_data should be run even if Seurat object has normalized data