Seurat to Dataframe with ONLY counts, cells, and cluster labels?

Question

[Kind Katydid]

I'm trying to run MDSeq, which appears to require a plain old data frame - counts and cells, and with labels for conditions (in my case, it'll be clusters found from Seurat's clustering).

I'm completely lost as to what function, if any, could strip all data/metadata except for the 3 above, and then convert it into a dataframe?

Alternatively, I am using the below script to split it into dataframes for each cluster, could/should I combine these and retain/add cluster labels? Seems less elegant than the other approach.

cluster_list <- levels([e-mail])

for(cluster in cluster_list) {
  name <- paste("data_cluster_", cluster, sep="")
  clustersubset <- subset(data_transcripts_seurat, idents=cluster, slot="counts")
  assign(name, as.data.frame(clustersubset@assays$SCT@counts))
}

[Friederike]

Why not start with data_transcripts_seurat@assays$SCT@counts?

Something like:

counts.df <- as.data.frame(data_transcripts_seurat@assays$SCT@counts)

If you need the cells to be rows, you can use the t() function:

counts.df <- data_transcripts_seurat@assays$SCT@counts %>% as.matrix %>% t %>% as.data.frame

[Kind Katydid]

Thanks! That's a much better way to do it. Could you please also provide guidance on how the cluster assignment can also be transferred (correctly) to this dataframe?

[Friederike]

what does [e-mail] get you? Shouldn't those be the cluster assignments per cell? [Disclaimer: I do not use Seurat, so I'm basing this off your initial code]

[Kind Katydid]

Hi, yes I think it is! I did not realise this before (also very new to Seurat, and transcriptomics in general). How can I add this to counts.df? Is it with the attributes function?

Also, should I be concerned about the order of cells being different in counts.df and [e-mail]? Are there ways to make sure they are matched?

Answer 1

[Pratik]

4 months late I know... but anyways... I was trying to do something similar. Google helped and sort of pieced together this puzzle:

library(dplyr)
counts.df <- data_transcripts_seurat@assays$RNA@counts %>% as.matrix %>% t %>% as.data.frame
counts.df <- tibble::rownames_to_column(counts.df, "cellnames")
clusterassignemnts <- data.frame([e-mail])
clusterassignemnts <- tibble::rownames_to_column(clusterassignemnts, "cellnames")
counts.df <- merge(clusterassignemnts, counts.df, by = "cellnames")
rownames(counts.df) <- counts.df$cellnames
counts.df$cellnames <- NULL
colnames(counts.df)[1] <- c("clusters")

Note: I did choose RNA counts versus SCT, you could play around with that I guess.

Also there is probably a better simpler way, this is what i found

Hope this helps!

[Kind Katydid]

Thanks! That's a really good way to do what I intended to,

For me, however, I gave up and decided to just put the cluster into the object name as opposed to a column,

cluster_index <- levels([e-mail]) # counts_all is my main Seurat object

for(i in cluster_index) {
  name <- paste0("counts_", i)
  clustersubset <- subset(counts_all, idents=i, slot="counts")
  assign(name, as.data.frame(clustersubset@assays$RNA@counts))
  rm(name, clustersubset)
}