devtools::install_github("EddieLv/SP_toolbox")
Here I developed 3 functions for DBIT-seq data: 1.shiny_st() # Edit your spot in web freely! 2.seurat2scanpy() # Transform seurat.object to scanpy.anndata totally (including counts, data, dimreduc, neighbor, spaitla, etc.)! 3.run_st_neighbor_genger() # Cell communication algorithm based on spatial neighbor!
An interactive Shiny app for manually annotating, cropping, and labeling spatial transcriptomics spots with a tissue image as background.
seurat_result <- shiny_st(
seurat, # Seurat object (required)
assay = "SCT", # Assay to use for gene expression query
slot = "data", # Slot to use ("data", "counts", "scale.data")
image = NULL, # Name of the image in seurat@images (required, e.g. "sample1")
python_env = NULL, # Path to Python executable for AI filtering (optional)
script = NULL, # Path to filter_pixel_AI.py for AI filtering (optional)
tooltip = NULL, # meta.data column name to show as tooltip on hover (optional)
crop = TRUE # Crop the view to cells retained in the object
)Return value: a modified Seurat object. The app runs in the foreground — assign the result after quitting.
library(SP.toolbox)
# Check the image name in your Seurat object
Images(seurat_obj)
# [1] "sample1"
# Launch the app and save the result
seurat_obj <- shiny_st(seurat_obj, assay = "SCT", slot = "data", image = "sample1")For a slide containing several tissues, subset the cells first. With the default
crop = TRUE, shiny_st() frames only the retained tissue, matching the behavior
of SpatialPlot(crop = TRUE):
one_tissue <- subset(seurat_obj, cells = cells_from_one_tissue)
one_tissue <- shiny_st(one_tissue, image = "sample1", crop = TRUE)Set crop = FALSE to keep showing the complete slide image.
After calling shiny_st(), a browser window opens showing the tissue H&E image with spots overlaid.
The sidebar on the left contains all controls.
| Control | Description |
|---|---|
| Select feature | Color spots by any meta.data column (e.g. orig.ident, seurat_clusters, celltype) |
| Select shape | Switch between square (22) and circle (21) spot shapes |
| Gene input + √ | Type a gene name and click √ to color spots by its expression level |
Use the image alignment controls in the sidebar to register spots onto the tissue:
| Control | Range | Description |
|---|---|---|
| Flip by vertical | — | Mirror spots top ↔ bottom |
| Flip by horizontal | — | Mirror spots left ↔ right |
| Rotate by angle + √ | −360 to 360 | Rotate all spots around the center |
| Move spots horizontally + √ | −96 to 96 | Shift all spots left / right |
| Move spots vertically + √ | −96 to 96 | Shift all spots up / down |
| Shrink spots horizontally + √ | 0 to 5 | Scale spot positions along the x-axis |
| Shrink spots vertically + √ | 0 to 5 | Scale spot positions along the y-axis |
| Control | Range | Description |
|---|---|---|
| Spot.alpha | 0–1 | Spot transparency (0 = invisible, 1 = fully opaque) |
| Spot.size | 0–1 | Spot rendering size |
Use Select feature to subset + the text input to display only certain groups before lasso selection.
Type comma-separated values (e.g. typeA,typeB) and click Select.
Click Back to all celltypes! to restore all spots.
- Click the lasso icon in the top-right toolbar of the plot panel.
- Draw a freehand boundary around the spots to annotate.
- In Set label for selected spots, type a label (e.g.
tumor_region). - Click Confirm — the label is written to the currently selected feature column.
- Repeat Steps 1–4 for additional regions or labels.
Tip: set Select feature to the column you are annotating so newly labeled spots change color immediately.
Requires python_env and script to be provided at launch:
seurat_obj <- shiny_st(
seurat_obj,
image = "sample1",
python_env = "~/miniconda3/envs/daily/bin/python",
script = "~/Biosoftwares/SP_toolbox/filter_pixel_AI.py"
)In the app:
- Adjust the threshold slider (0–100) to control filtering sensitivity.
- Click Power BY AI — background spots are labeled
filtered, tissue spots are labeledexistin a new columnai.filter. - Set Filter mode to
onto switch spots to outline-only display for easier checking.
The toolbar in the top-right corner of the plot provides:
| Button | Function |
|---|---|
| lasso | Freehand spot selection mode |
| zoom | Zoom into a rectangular region |
| recover | Reset zoom to full view |
| download png | Save the current plot view as a PNG file |
Click Quit in the sidebar. The app closes and returns the modified Seurat object with:
- All Confirm-ed labels written back into the corresponding
meta.datacolumn. - An
ai.filtercolumn (exist/filtered) if AI filtering was used.
# Inspect annotations after quitting
table(seurat_obj$orig.ident)
table(seurat_obj$ai.filter) # only if AI filtering was run
# Save
saveRDS(seurat_obj, "seurat_annotated.rds")library(SP.toolbox)
seurat_obj <- shiny_st(
seurat_obj,
assay = "SCT",
slot = "data",
image = "sample1",
python_env = "~/miniconda3/envs/daily/bin/python",
script = "~/Biosoftwares/SP_toolbox/filter_pixel_AI.py",
tooltip = "celltype" # show celltype label on hover
)
saveRDS(seurat_obj, "seurat_annotated.rds")| Symptom | Likely cause | Fix |
|---|---|---|
| Background image not showing | ggplot2 ≥ 4.0 breaking change | Update: devtools::install_github("EddieLv/SP_toolbox") |
| Spots misaligned with image | Seurat ≥ 5.0 coordinate scaling change | Same — update the package |
Please set image! error |
image argument missing or wrong name |
Run Images(seurat_obj) to get the exact name |
| AI button has no effect | Wrong python_env or script path |
Verify both paths exist and the Python env has required packages |
(repackaged from https://github.com/JohnGenome/ST-mouse-kidney-development)
sample = "E16.5_slice18"
srat.merge = readRDS("/media/biogenger/D/Projects/CMY/Analysis/mouse_lung/All/Paper/Fig4/srat.merge.reanno.rds")
srat.merge = SubsetSTData(srat.merge, spots = rownames(srat.merge@meta.data)[srat.merge$orig.ident == sample])
srat.merge$celltype.re = droplevels(srat.merge$celltype.re)
table(srat.merge$celltype.re)
#Sox9.pro AT1.pre AT1 AT2.pre AT2 Matrix.FB myo.FB gCap
# 55 802 103 690 226 955 576 646
if (! dir.exists(ccc_path)) dir.create(ccc_path)
neighbor.res = run_st_neighbor_genger(srat.merge, coordx = "barcodeB", coordy = "barcodeA", label.col = "celltype.re", self.interaction = T, permut = T, nperm = 2000, max.distance = 2)
st_neighbor_heat_genger(neighbor.res$neighbor.df)
st_celltype_pair_neighbor_genger(srat.traj, neighbor.res = neighbor.res, image = "E16.5_slice18", celltype_pair = c("AT1.pre", "AT1"), label.col = "celltype.re", pt.size = 2, show.image = T, show.label = T)
### Load LR-DB and start CCC Analysis ###
lr_pair.df = read.table("/media/biogenger/D/LR_database/mouse_ligand_receptors_giotto.txt", header = T, stringsAsFactors = F)
lr_pair.df %>% head()
# ligand receptor
#1 A2m Lrp1
#2 Aanat Mtnr1a
#3 Adam12 Itga9
#4 Adam12 Itgb1
#5 Adam12 Sdc4
#6 Adam15 Itga5
corrected_umi.ls = apply(srat.traj@assays$SCT@data, 1, sum)
corrected_umi.ls = corrected_umi.ls[corrected_umi.ls > 0]
lr_pair.df$mouseLigand.exist = ifelse(lr_pair.df$mouseLigand %in% names(corrected_umi.ls), "ligand", "no")
lr_pair.df$mouseReceptor.exist = ifelse(lr_pair.df$mouseReceptor %in% names(corrected_umi.ls), "receptor", "no")
lr_pair.df$lr.exist = paste0(lr_pair.df$mouseLigand.exist, "_", lr_pair.df$mouseReceptor.exist)
table(lr_pair.df$lr.exist)
lr_pair.df = lr_pair.df %>% dplyr::filter(lr.exist == "ligand_receptor")
lr_pair.df = lr_pair.df[ , 1:2]
colnames(lr_pair.df) = c("ligand", "receptor")
lr_pair.df$LR = paste0(lr_pair.df$ligand, "_", lr_pair.df$receptor)
result.df = run_st_neighbor_chat_genger(srat.merge, neighbor.res = neighbor.res, lr.df = lr_pair.df, assay = "SCT", slot = "data", label.col = "celltype.re", celltype_pair = c("AT1.pre", "myo.FB"), nperm = 2000)
result.df %>% head()
result.df %>% dplyr::filter(pval.both < 0.05 & min.expr.both > 0.1) %>% arrange(pval.both)
result.df %>% dplyr::filter(pval.ind < 0.05 & min.expr.ind > 0.1) %>% arrange(pval.ind)
# ligand receptor LR celltype_pair ligand_celltype_both_mean receptor_celltype_both_mean ligand_celltype1_mean receptor_celltype2_mean min.expr.both min.expr.ind pval.ligand.both pval.receptor.both
#1 A2m Lrp1 A2m_Lrp1 AT1.pre_myo.FB 0.022416981 0.30527435 0.019948741 0.3688052 0.02241698 0.01994874 0.9950 0.9720
#2 Aanat Mtnr1a Aanat_Mtnr1a AT1.pre_myo.FB 0.004674503 0.00000000 0.006085798 0.0000000 0.00000000 0.00000000 0.3920 0.0000
#3 Adam12 Itga9 Adam12_Itga9 AT1.pre_myo.FB 0.821827418 0.78610743 0.796703430 0.9556919 0.78610743 0.79670343 1.0000 0.0000
#4 Adam12 Itgb1 Adam12_Itgb1 AT1.pre_myo.FB 0.821827418 0.82841966 0.796703430 0.8369752 0.82182742 0.79670343 1.0000 0.9985
#5 Adam12 Sdc4 Adam12_Sdc4 AT1.pre_myo.FB 0.821827418 0.18802907 0.796703430 0.1787223 0.18802907 0.17872231 1.0000 0.6115
#6 Adam15 Itga5 Adam15_Itga5 AT1.pre_myo.FB 0.035670005 0.08477808 0.033044533 0.1134542 0.03567000 0.03304453 0.8035 1.0000
# pval.both pval.ligand.ind pval.receptor.ind pval.ind
#1 0.98350 0.989 0.0010 0.49500
#2 0.19600 0.154 0.0000 0.07700
#3 0.50000 1.000 0.0000 0.50000
#4 0.99925 1.000 0.8695 0.93475
#5 0.80575 1.000 0.7785 0.88925
#6 0.90175 0.854 0.3355 0.59475
srat.merge$Vegfa.expr = FetchData(srat.merge, vars = "Vegfa", slot = "data")[[1]]
srat.merge$Nrp2.expr = FetchData(srat.merge, vars = "Nrp2", slot = "data")[[1]]
Spatial2Featureplot_genger(srat.merge, orig.ident = "E16.5_slice18", image = "E16.5_slice18", show.image = F, featureA = "Vegfa.expr", featureB = "Nrp2.expr", show.label = T, theme.dark = F)
Feel free to ask any questions, open a PR if you feel something can be done differently!
Created by EddieLv