Xenium Breast Cancer Dataset 2

Python code

# apply Tangram to impute genes
import os, sys
import anndata as ad
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import scanpy as sc
import torch
import tangram as tg

tg.__version__

#############################################################################
# spatial transcriptomics data
ad_sp = ad.read_h5ad('./xenium_breast2_sp.h5ad')
ad_sp.obsm["spatial"] = ad_sp.obs[["x_centroid", "y_centroid"]].copy().to_numpy()

# single cell rna sequencing reference
ad_sc = ad.read_h5ad('./xenium_breast2_sc.h5ad')
sc.pp.normalize_total(ad_sc)

gene_names = ad_sp.var_names
markers = gene_names.tolist()
print(len(markers))

tg.pp_adatas(ad_sc, ad_sp, genes=markers)
assert ad_sc.uns['training_genes'] == ad_sp.uns['training_genes']

ad_map = tg.map_cells_to_space(
    adata_sc=ad_sc,
    adata_sp=ad_sp,
    device='cpu')

tg.project_cell_annotations(ad_map, ad_sp, annotation='celltype')
annotation_list = list(pd.unique(ad_sc.obs['celltype']))

ad_map.uns['train_genes_df']
ad_ge = tg.project_genes(adata_map=ad_map, adata_sc=ad_sc)
print(ad_ge)
ad_ge.write('xenium_breast2_tangram.h5ad')

# training score
tg.plot_training_scores(ad_map, bins=20, alpha=.5)
plt.savefig("xenium_breast2_train_score.png",dpi=300)

# cell type mapping
tg.plot_cell_annotation_sc(ad_sp, annotation_list,spot_size= 50, scale_factor=0.1, perc=0.001)
plt.savefig("xenium_breast2_tangram_cell_annotation.png",dpi=300)

####################################################################
## scale gene expression to scRNA range
ad_ge.var_names=[i.upper() for i in list(ad_ge.var_names)]
ad_sc = ad.read_h5ad('./xenium_breast2_sc.h5ad')
ad_sc.var_names=[i.upper() for i in list(ad_sc.var_names)]
ad_sc.var_names.is_unique
duplicates = ad_sc.var_names[ad_sc.var_names.duplicated()]
print("Duplicate var_names:", duplicates)
ad_sc = ad_sc[:, ~ad_sc.var_names.duplicated()].copy()

genes_adata1 = ad_ge.var_names
genes_adata2 = ad_sc.var_names

common_genes = genes_adata1.intersection(genes_adata2)
ad_sp_filtered = ad_ge[:, common_genes].copy()
ad_sc_filtered = ad_sc[:,common_genes].copy()

X1 = ad_sp_filtered.X
X2 = ad_sc_filtered.X

import scipy.sparse
if scipy.sparse.issparse(X1):
    X1 = X1.toarray()

if scipy.sparse.issparse(X2):
    X2 = X2.toarray()

min_vals_1 = np.min(X1, axis=0)
max_vals_1 = np.max(X1, axis=0)
min_vals_2 = np.min(X2, axis=0)
max_vals_2 = np.max(X2, axis=0)

scaled_X = (X1 - min_vals_1) / (max_vals_1 - min_vals_1)
scaled_X = scaled_X * (max_vals_2 - min_vals_2) + min_vals_2

ad_sp_filtered.X = scaled_X
ad_sp_filtered.write_h5ad('xenium_breast2_sc_scale.h5ad')

R code

# apply highSpaClone to infer spatial CNV and detect tumor subclones
library(highSpaClone)
library(anndata)
library(ggplot2)
library(dplyr)

data=read_h5ad("./xenium_breast2_sc_scale.h5ad")
counts=Matrix::t(data$X)
colnames(counts)=data$obs$cell_id

location=data.frame(id=factor(data$obs$cell_id),
                    x=data$obs$x_centroid,
                    y=data$obs$y_centroid)
rm(data)
gc()
rownames(location)=location$id
location <- location[match(colnames(counts),rownames(location)),]
annotation <- read.csv('./xenium_breast2_label.csv')
xenium_breast2_obj <- list(counts=data$counts, location=data$location, annotation=annotation)
saveRDS(xenium_breast2_obj, file='./xenium_breast2_obj.rds')

# load data
data <- readRDS('./xenium_breast2_obj.rds')

# create highSpaClone object
obj <- createObject(counts=data$counts,
                         location=data$location,
                         annotations_file = data$annotation)

# smooth gene expression by chromosome
smooth_obj <- smooth_expr(obj = obj, parallel = T, use_chunk = T)

# CNV inference and subclone detection
cnv_obj <- FindClone(smooth_obj,
                      ref=c("Bcell", "Tcell", "Mast", "Myeloid"),
                      tumor=c('DCIS_Basal','DCIS_LumHR','DCIS_LumSec','IDC_LumHR'),
                      K=2,
                      lambda=1000)

# subclone spatial visualization
colors <- c("#D57358", "#8a508f")
spatialplot(cnv_obj, colors=colors, point_size = 0.05)

# CNV heatmap
log.cnv <- log2(cnv_obj@cnv.data)
log.cnv[log.cnv > -0.3 & log.cnv < 0.3] <- 0

xenium_breast2 <- list(
  cnv.data   = log.cnv,
  annotation = cnv_obj@cluster,
  chr_pos    = cnv_obj@chr_pos
)

cnv_heatmap(
  cnv.obj=xenium_breast2,
  groupby = "cell.label",
  min_value  = -2.5,  
  max_value  = 2.5,
  auto_size = F,
  width_in = 7,
  height_in = 5,
  show    = TRUE,       
  save    = TRUE,         
  outfile = "xenium_breast2_heatmap.pdf" 
)