Xenium Data Analysis

Here we will be showing how we preprocessed the external Xenium datasets we used for validation. We use (Bell et al. 2024) & (Fiorini et al. 2025) datasets. The preprocessing steps traditional scRNAseq processing and Harmony integration. The same pipeline was applied to both with differences in the clusters annotation due to differences in the gene panels used.

1 Setup and data import

library(Seurat)
library(tidyverse)
library(SeuratWrappers)
library(reticulate)
library(qs)
library(spacexr)
library(harmony)
library(clustree)
library(SeuratExtend)
library(UCell)
library(parallel)
source("utils.R")

samples_info <- read.table("data/samples_manifest.txt") %>%
  `colnames<-`(c("sample_id", "xeniumranger_outDir", "ecDNA_state"))
proseg_outDir <- "outputs/proseg/"

# Importing samples -------------------------------------------------------

xenium_samples <- lapply(samples_info$sample_id, function(x){
  
  xenium_obj <- ProsegToSeurat(paste0("outputs/proseg/", x))
  xenium_obj$sample_id <- x
  
  rctd_res <- readRDS(paste0("outputs/RCTD/", x, "_rctd.rds"))
  annotations.df <- rctd_res@results$results_df
  annotations <- annotations.df$first_type
  names(annotations) <- rownames(annotations.df)
  xenium_obj <- AddMetaData(xenium_obj, data.frame(predicted_cellType = annotations))
  
  xenium_obj <- RenameCells(xenium_obj, add.cell.id = x)
  names(xenium_obj@images) <- x
  xenium_obj@images[[1]]@key <- x
  
  return(xenium_obj)
}) %>%
  reduce(merge)

2 Data processing and normalization

It seems that VR35 has a lot of regions with non-identified cell types due to low transcriptional reads. We will remove it and remove all cells with NA predicted celltype

# Data filteration --------------------------------------------------------
xenium_samples <- subset(xenium_samples, subset = sample_id != "VR35")
xenium_samples <- xenium_samples[,!is.na(xenium_samples$predicted_cellType)]

# Data processing ---------------------------------------------------------
xenium_samples <- NormalizeData(xenium_samples)
xenium_samples <- FindVariableFeatures(xenium_samples, nfeatures = 414)
xenium_samples <- ScaleData(xenium_samples, features = rownames(xenium_samples))
xenium_samples <- RunPCA(xenium_samples, features = rownames(xenium_samples))
xenium_samples <- RunUMAP(xenium_samples, dims = 1:15)

# Harmony integration -----------------------------------------------------
xenium_samples <- JoinLayers(xenium_samples)
xenium_samples <- RunHarmony(
  object = xenium_samples,
  group.by.vars = "sample_id",      
  reduction.use = "pca",      
  dims.use = 1:20
)

xenium_samples <- RunUMAP(xenium_samples, reduction = 'harmony', dims = 1:20)
xenium_samples <- FindNeighbors(xenium_samples, reduction = 'harmony', dims = 1:20)
xenium_samples <- FindClusters(xenium_samples, resolution = 0.2)
DimPlot(xenium_samples, group.by = "sample_id") + DimPlot_scCustom(xenium_samples, group.by = "predicted_cellType")
qsave(xenium_samples, "outputs/xenium_obj_harmony_integrated.qs")

3 Cell Type Deconvolution

We will use RCTD from spaceXr (Cable et al. 2022) in the doublet mode to estimate the cell type in each segmented cell. We will use the same scRNAseq reference that can be downloaded from Zenodo here.

ref <- qread("scRef_filtered_integrated_annotated.qs")
ref <- subset(ref, subset = main_annotation_scvi != 'T_Cells(Hi_Ribo)')
counts <- GetAssayData(ref, assay = "RNA", slot = "counts")
ref$`Cell Type` <- as.character(ref$`Cell Type`)
cluster <- as.factor(ref$`Cell Type`)
names(cluster) <- colnames(ref)
nUMI <- ref$nCount_RNA
names(nUMI) <- colnames(ref)
levels(cluster) <- gsub("/", "-", levels(cluster))
reference <- Reference(counts, cluster, nUMI, n_max_cells = max(table(cluster)))
rm(list = c("ref", "counts","cluster","nUMI"))

slurm_jobs <- Slurm_lapply(
  njobs = length(samples_info$sample_id),
  job_name = "RCTD",
  tmp_path = getwd(),
  plan = "none",
  sbatch_opt = list(partition = 'standard', account = '',
                    mem = '128G', "cpus-per-task" = "16", time = "24:00:00"),
  export = c("reference", "proseg_outDir"),
  overwrite = TRUE,
  X = as.list(samples_info$sample_id),
  FUN = function(x) {
    
    library(spacexr)
    
    expected_counts_path <- file.path(proseg_outDir, paste0(x, "/expected-counts.csv.gz"))
    expected_counts <- read.csv(expected_counts_path, header=TRUE, sep=",")
    cell_metadata_path <- file.path(proseg_outDir, paste0(x, "/cell-metadata.csv.gz"))
    cell_metadata <- read.csv(cell_metadata_path, header=TRUE, sep=",")
    
    mask <- is.finite(cell_metadata$centroid_x) &
      is.finite(cell_metadata$centroid_y)
    cell_metadata <- cell_metadata[mask, ]
    expected_counts <- expected_counts[mask, ]
    
    coords_df <- cell_metadata[c("centroid_x", "centroid_y")]
    names(coords_df) <- c("x", "y")
    rownames(coords_df) <- rownames(expected_counts)
    
    query <- SpatialRNA(coords_df, round(t(expected_counts)))
    RCTD <- create.RCTD(query, reference, UMI_min = 10, max_cores = parallel::detectCores())
    RCTD <- run.RCTD(RCTD, doublet_mode = "doublet")
    saveRDS(RCTD, paste0("outputs/RCTD/", x, "_rctd.rds"))
  })

sbatch(slurm_jobs)

rctd_res <- lapply(list.files("outputs/RCTD/", pattern = "rctd.rds"), readRDS)
rctd_res <- lapply(rctd_res, function(x){
  annotations.df <- x@results$results_df
  annotations <- annotations.df$first_type
  names(annotations) <- rownames(annotations.df)
  return(data.frame(annotations))
}) %>%
  bind_rows()

xenium.obj <- AddMetaData(xenium.obj, rctd_res)

3 Clusters Annotation

# Annotation --------------------------------------------------------------

res <- seq(from = 0.1, to = 1, by = 0.1)
for(x in res){
  xenium_samples <- FindClusters(xenium_samples, resolution = x, cluster.name = paste0("res.",x))
}
clustree(xenium_samples, prefix = 'res.', layout = "sugiyama")
xenium_samples <- FindClusters(xenium_samples, resolution = 0.5)

seurat_clusters_markers <- FindAllMarkers(xenium_samples, assay = 'Xenium', group.by = 'seurat_clusters')
seurat_clusters_markers$score <- seurat_clusters_markers$avg_log2FC * seurat_clusters_markers$pct.1
xenium_samples@misc$seurat_clusters_markers <- seurat_clusters_markers

## I will remove cluster 14 as it has low transcript count
xenium_samples <- subset(xenium_samples, subset = seurat_clusters != '14')
xenium_samples <- RunUMAP(xenium_samples, reduction = 'harmony', dims = 1:20)
seurat_clusters_markers <- FindAllMarkers(xenium_samples, assay = 'Xenium', group.by = 'seurat_clusters')
seurat_clusters_markers$score <- seurat_clusters_markers$avg_log2FC * seurat_clusters_markers$pct.1
xenium_samples@misc$seurat_clusters_markers <- seurat_clusters_markers

annotation <- c("0" = "Tumor",
                "1" = "Fibro",
                "2" = "Macrophages",
                "3" = "Fibroblasts",
                "4" = "Fibroblasts",
                "5" = "T_Cells",
                "6" = "Endothelial",
                "7" = "Tumor",
                "8" = "Pericytes",
                "9" = "Endocrine",
                "10" = "Mast_Cells",
                "11" = "Plasma_Cells",
                "12" = "Tumor",
                "13" = "Tumor")
xenium_samples <- RenameIdents(xenium_samples, annotation)
xenium_samples$annotation <- Idents(xenium_samples)

Idents(xenium_samples) <- xenium_samples$seurat_clusters
sub_annotation <- c("0" = "Tumor1",
                    "1" = "Fibro1",
                    "2" = "Macrophages",
                    "3" = "Fibro2",
                    "4" = "Fibro3",
                    "5" = "T_Cells",
                    "6" = "Endothelial",
                    "7" = "Tumor2",
                    "8" = "Pericytes",
                    "9" = "Endocrine",
                    "10" = "Mast_Cells",
                    "11" = "Plasma_Cells",
                    "12" = "Tumor3",
                    "13" = "CyclingTumor")
xenium_samples <- RenameIdents(xenium_samples, sub_annotation)
xenium_samples$sub_annotation <- Idents(xenium_samples)

xenium_samples$annotation <- factor(xenium_samples$annotation, levels = c("Tumor", "Fibroblasts", "Pericytes", "Endocrine", "Endothelial",
                                                                          "Plasma_Cells", "Mast_Cells", "Macrophages", "T_Cells"))
Idents(xenium_samples) <- xenium_samples$annotation

4 Projecting Fibro2 and Fibro3 signature on fibroblasts of Xenium data

fibro2_v_fibro2_DEGs <- read.csv("/outputs/Fibroblasts_Analysis/stroma2_v_stroma3_DEGs.csv", row.names = 1)

fibro2_sig <- fibro2_v_fibro2_DEGs %>% 
  filter(padj < 0.05 & log2FoldChange > 0.5) %>%
  rownames() %>% 
  intersect(rownames(xenium_samples))
fibro2_sig <- c("ACTA2", "LOXL2", "COL5A2", "THBS2", "MFAP5", "INHBA")

fibro3_sig <- fibro2_v_fibro2_DEGs %>% 
  filter(padj < 0.05 & log1p(baseMean) > 3 & log2FoldChange < -0.5) %>%
  rownames() %>% 
  intersect(rownames(xenium_samples))
fibro3_sig <- c("CRISPLD2", "STEAP4", "CCL19", "PTN","TNC", "MYC","EDNRB","TFPI","EGFR","PTGDS","RSPO3","FBLN1","PDGFRA","DPT", "RSPO1","C7","MEDAG","MAMDC2", "OGN")

fibro <- subset(xenium_samples, subset = annotation == 'Fibroblasts')
fibro <- AddModuleScore_UCell(fibro,features = list(fibro3_sig = fibro3_sig, fibro2_sig = fibro2_sig), ncores = detectCores())
fibro@meta.data <- mutate(fibro@meta.data, 
                          fibro2_score_scaled = (fibro2_sig_UCell - min(fibro2_sig_UCell)) / (max(fibro2_sig_UCell) - min(fibro2_sig_UCell)),
                          fibro3_score_scaled = (fibro3_sig_UCell - min(fibro3_sig_UCell)) / (max(fibro3_sig_UCell) - min(fibro3_sig_UCell))) %>%
  mutate(fibro_combined_score = log((fibro2_score_scaled + 1e-3) / (fibro3_score_scaled + 1e-3)))
xenium_samples <- AddMetaData(xenium_samples, fibro@meta.data %>% select(fibro2_sig_UCell, fibro3_sig_UCell, fibro2_score_scaled, fibro3_score_scaled, fibro_combined_score))

5 Saving Object

qsave(xenium_samples, "outputs/xenium_obj_harmony_integrated.qs")

6 Session Info

sessionInfo()

References

Bell, Alexander TF, Jacob T Mitchell, Ashley L Kiemen, Melissa Lyman, Kohei Fujikura, Jae W Lee, Erin Coyne, et al. 2024. “PanIN and CAF Transitions in Pancreatic Carcinogenesis Revealed with Spatial Data Integration.” Cell Systems 15 (8): 753–69.

Cable, Dylan M, Evan Murray, Luli S Zou, Aleksandrina Goeva, Evan Z Macosko, Fei Chen, and Rafael A Irizarry. 2022. “Robust Decomposition of Cell Type Mixtures in Spatial Transcriptomics.” Nature Biotechnology 40 (4): 517–26.

Fiorini, Elena, Antonia Malinova, Daniel Schreyer, Davide Pasini, Michele Bevere, Giorgia Alessio, Diego Rosa, et al. 2025. “MYC ecDNA Promotes Intratumour Heterogeneity and Plasticity in PDAC.” Nature, 1–10.