Skip to content

Latest commit

 

History

History
113 lines (92 loc) · 4.58 KB

README.md

File metadata and controls

113 lines (92 loc) · 4.58 KB
Raw

Spatial Transcriptomics analysis

Data access:

We are analyzing an FFPE sample with pathologist annotations from a ductal carcinoma patient obtained from the 10x Genomics Datasets portal. We preprocessed these data using the pipelines in ST-preprocess repository. The resulting Seurat object, stored in 738811QB_spatial_SCTnormalised_clones.rds file, can be accessed through the following link.

Using beyondcell

Read the ST object

First, we load the Seurat object.

library("beyondcell")
library("Seurat")
library("ggplot2")
set.seed(1)
path_to_st <- "~/Downloads/738811QB_spatial_SCTnormalised_clones.rds"
# Read spatial transcriptomics experiment.
st <- readRDS(path_to_st)

Compute the BCS

Next, we generate a geneset object containing the SSc signatures. The SSc collection reflects the transcriptional differences between sensitive and resistant cancer cell lines before drug treatment. Thus, these signatures are helpful in predicting drug response in treatment-naive samples like the one we are analyzing.

# Generate geneset object with SSc signatures.
gs <- GetCollection(SSc, include.pathways = FALSE)

Then, we can compute the Beyondcell Scores (BCS).

# Set Assay.
DefaultAssay(st) <- "SCT"
# Compute BCS.
bc <- bcScore(st, gs, expr.thres = 0.1)

Compute the TCs

After obtaining our beyondcell object, we group the spots into Therapeutic Clusters (TCs) based on their predicted response to the SSc collection.

# Run the UMAP reduction. 
bc <- bcUMAP(bc, k.neighbors = 20, res = 0.2)
# Run the bcUMAP function again, specifying the number of principal components 
# you want to use.
bc <- bcUMAP(bc, pc = 10, k.neighbors = 20, res = 0.2)

When we compare the TCs with the pathologist annotations, we can observe that the fibrous tissue corresponds to TC0, the invasive carcinoma to TC1, and the necrosis to TC3. Moreover, TC2 seemed composed of immune cells and fibrous tissue in close contact with cancer cells.

# Visualize the TCs and the pathologist annotations.
pathologist_colors <- 
  c(Fibrous_tissue = "#D2362A", Invasive_carcinoma = "#296217", 
    Necrosis = "#F9D94A", Immune_cells = "#932CE7", Fat = "#1400F5")
bcClusters(bc, idents = "bc_clusters_res.0.2", spatial = TRUE, pt.size = 1.5) |
  bcClusters(bc, idents = "Pathologist", spatial = TRUE, pt.size = 1.5) +
  scale_fill_manual(values = pathologist_colors)

Compute ranks

We are interested in identifying drugs that specifically target cancer cells. We compute a signature ranking using the bcRanks function to do so. Then, we plot this ranking using the bc4Squares function. As we are interested in the drugs that specifically target TC1, we will focus on the orange area of the plot, which depicts the drugs with intermediate Switch Points (indicating a heterogeneous response in the sample) and positive residual's mean (indicating higher sensitivity in TC1 compared to the rest of spots).

# Prioritize drugs for each TC.
bc <- bcRanks(bc, idents = "bc_clusters_res.0.2", extended = FALSE)
# Find drugs to specifically target TC1 (cancer cells).
bc4Squares(bc, idents = "bc_clusters_res.0.2", lvl = "1")

Plot drug sensitivity predictions

According to the previous plot, lapatinib is the best drug to target TC1 specifically. Lapatinib is a ERBB2 inhibitor approved by the FDA to treat HER2+ breast cancers. Accordingly, we observe a high expression of this biomarker in cancer cells, which correlates with lapatinib sensitivity.

# Plot the predicted sensitivity to lapatinib and ERBB2 expression.
bcSignatures(bc, signatures = list(values = "sig-21463"), spatial = TRUE, pt.size = 1.5) |
  bcSignatures(bc, genes = list(values = "ERBB2"), spatial = TRUE, pt.size = 1.5)

Support

Additional information can be found in the package's documentation and in the analysis workflow or visualization tutorials. If you have any questions regarding the use of beyondcell, feel free to submit an issue.