update vignettes

vignettes/Session_2_Tidy_spatial_analyses.Rmd

@@ -359,6 +359,8 @@ spatial_data_gated =
 spatial_data_gated |> select(.cell, .gated)
 ```
 
+This is not meant to be executed. It is an example of how to save a gate to a file.
+
 ```{r, eval=FALSE}
 tidygate_env$gates |> saveRDS("<PATH>")
 spatial_data_gates = tidygate_env$gates
@@ -520,7 +522,7 @@ spe_regions_aggregated
 
 ```{r}
 library(tidySummarizedExperiment)
-
+library(tidyprint)
 spe_regions_aggregated
 
 ```
@@ -845,7 +847,7 @@ To to practice the use of `tidyomics` on spatial data, we propose a few exercise
 
 We assume that the cells we filtered as non-alive or damaged, characterised by being enriched uniquely for mitochondrial, genes, and genes, linked to up apoptosis. It is good practice to check these assumption. This exercise aims to estimate what genes are differentially expressed between filtered and unfiltered cells. Then visualise the results.
 
-Use `tidyomic`/`tidyverse` tools to label dead cells and perform differential expression within each region. Some of the comments you can use are: `mutate`, `nest`, `map`, `aggregate_cells`, `tidybulk:::test_differential_abundance`, 
+Use `tidyomic`/`tidyverse` tools to label dead cells and perform differential expression within each region. Some of the commands you can use are: `mutate`, `nest`, `map`, `aggregate_cells`, `tidybulk:::test_differential_expression`, 
 
 A hint:
 

vignettes/Session_3_imaging_assays.Rmd

@@ -520,6 +520,8 @@ The selected subset of genes can then be passed to the subset.row argument (or e
 ```{r}
 tx_spe_sample_1 =  
   tx_spe_sample_1 |> 
+
+  # We use fixed PCA as we have a limited number of features
   fixedPCA( subset.row=top.hvgs )
 ```
 
@@ -545,7 +547,7 @@ cluster_labels =
   tx_spe_sample_1 |> 
    scran::clusterCells(
      use.dimred="PCA", 
-     BLUSPARAM=bluster::NNGraphParam(k=20, cluster.fun="louvain")
+     BLUSPARAM=bluster::NNGraphParam(k=30, cluster.fun="louvain")
     ) |> 
    as.character()
 
@@ -558,9 +560,9 @@ Now we add this cluster column to our `SpatialExperiment`
 ```{r}
 tx_spe_sample_1 = 
   tx_spe_sample_1 |> 
-  mutate(clusters = cluster_labels)
+  mutate(my_clusters = cluster_labels)
 
-tx_spe_sample_1 |> select(.cell, clusters)
+tx_spe_sample_1 |> select(.cell, my_clusters)
 ```
 
 As we have done before, we caclculate UMAPs for visualisation purposes.
@@ -575,11 +577,11 @@ tx_spe_sample_1 =
   tx_spe_sample_1 |>
   runUMAP() 
 ```
-Now, let's visualise the clusters in UMAP space.
+Now, let's visualise the my_clusters in UMAP space.
 
 ```{r, fig.width=7, fig.height=8}
 tx_spe_sample_1 |> 
-  plotUMAP(colour_by = "clusters") +
+  plotUMAP(colour_by = "my_clusters") +
   scale_color_discrete(
     colorRampPalette(brewer.pal(9, "Set1"))(30)
   )
@@ -624,7 +626,7 @@ Plot ground truth in tissue map.
 ```{r, fig.width=7, fig.height=8}
 
 tx_spe_sample_1 |>
-  ggspavis::plotCoords(annotate = "clusters") +
+  ggspavis::plotCoords(annotate = "my_clusters") +
   guides(color = "none")
 
 ```
@@ -645,6 +647,9 @@ tx_spe_sample_1 |>
 :::
 
 
+
+
+
 ### 4. Neighborhood analyses
 
 hoodscanR [Liu et al., 2025](https://www.bioconductor.org/packages/release/bioc/vignettes/hoodscanR/inst/doc/Quick_start.html)
@@ -663,12 +668,73 @@ Algorithm:
 
 - The K-means clustering algorithm finds recurring neighbours
 
+
+Before we scan cellular neighborhoods, each cell needs a biological label rather than an unsupervised cluster ID. We annotate cells with `SingleR`, using the human prefrontal cortex reference prepared in Session 1. Our Xenium panel reports mouse gene symbols, whereas the Zhong reference uses human symbols; for this workshop we harmonise identifiers by converting both to upper case so that orthologous symbols in the panel can be matched. In your own analyses, use a reference matched to the species and tissue under study.
+
+```{r, message=FALSE, warning=FALSE}
+library(scRNAseq)
+library(SingleR)
+
+# Get reference (same workflow as Session 1)
+brain_reference <- fetchDataset("zhong-prefrontal-2018", "2023-12-22")
+
+brain_reference =
+  brain_reference |>
+  scuttle::aggregateAcrossCells(ids = paste(brain_reference$sample, brain_reference$cell_types, sep = "_"))
+
+brain_reference = brain_reference[, brain_reference |> assay() |> colSums() > 0]
+brain_reference = brain_reference[, !brain_reference$cell_types |> is.na()]
+
+brain_reference =
+  brain_reference |>
+  logNormCounts()
+```
+
+Prepare the imaging data for annotation, as we did for the spatial data in Session 1.
+
+```{r, message=FALSE, warning=FALSE}
+tx_spe_sample_1_annot =
+  tx_spe_sample_1[
+    !grepl("NegControl.+|BLANK.+", rownames(tx_spe_sample_1)),
+  ]
+rownames(tx_spe_sample_1_annot) <- toupper(rownames(tx_spe_sample_1_annot))
+tx_spe_sample_1_annot <- logNormCounts(tx_spe_sample_1_annot)
+
+rownames(brain_reference) <- toupper(rownames(brain_reference))
+
+genes_for_annotation <-
+  grep("(^MT-)|(^mt-)|(\\.)|(-)", rownames(brain_reference), value = TRUE, invert = TRUE) |>
+  intersect(rownames(tx_spe_sample_1_annot))
+```
+
+
+We can now annotate each cell with `SingleR`. With a targeted Xenium panel we only use genes present in both datasets (`genes_for_annotation`). In `SingleR`, the `genes` argument does **not** take a character vector of gene names: it must be `"de"` (automatic marker detection within the reference) or a **named list** of markers per cell type. To limit analysis to the panel, pass those genes via row subsetting and/or `restrict`.
+
+```{r, message=FALSE, warning=FALSE}
+singler_pred <- SingleR(
+  test = tx_spe_sample_1_annot[genes_for_annotation, ],
+  ref = brain_reference[genes_for_annotation, ],
+  labels = brain_reference$cell_types,
+  genes = "de",
+  de.method = "wilcox",
+  hint.sce = FALSE
+)
+```
+
+```{r}
+tx_spe_sample_1 =
+  tx_spe_sample_1 |>
+  mutate(cell_type = singler_pred$labels)
+
+tx_spe_sample_1 |> select(.cell, my_clusters, cell_type)
+```
+
 In order to perform neighborhood scanning, we need to firstly identify k (in this example, k = 100) nearest cells for each cells. The searching algorithm is based on Approximate Near Neighbor (ANN) C++ library from the RANN package.
 
 ```{r}
 tx_spe_neighbours = 
   tx_spe_sample_1 |> 
-  readHoodData(anno_col = "clusters") |> 
+  readHoodData(anno_col = "cell_type") |> 
   findNearCells(k = 100)
 
 ```
@@ -689,9 +755,9 @@ We can then perform neighborhood analysis using the function scanHoods. This fun
 pm <- scanHoods(tx_spe_neighbours$distance)
 
  # We can then merge the probabilities by the cell types of the 100 nearest cells. We get the probability distribution of each cell all each neighborhood. 
-hoods <- mergeByGroup(pm, tx_spe_neighbours$cells)
+hoods <- mergeByGroup(pm, group_df = tx_spe_neighbours$cells)
 
-hoods[1:2, 1:10]
+hoods[1:2, ]
 ```
 
 
@@ -715,19 +781,38 @@ We can then merge the neighborhood results with the `SpatialExperiment` object u
 ```{r}
 tx_spe_sample_1 =  tx_spe_sample_1 |> mergeHoodSpe(hoods)
 
+tx_spe_sample_1 <- calcMetrics(tx_spe_sample_1, pm_cols = colnames(hoods))
+
+# Entropy and perplexity statistics are added to the object
 tx_spe_sample_1
 ```
 
-We can see what are the neighborhood distributions look like in each cluster using `plotProbDist`. Here we only plot 10 clusters
+Perplexity of 1 means the cell is located in a very distinct neighborhood, perplexity of 2 means the cell is located in a mixed neighborhood, and the probability is about 50% to 50%.
+
+```{r}
+plotTissue(tx_spe_sample_1, size = 1.5, color = perplexity) +
+  scale_color_scico(palette = "tokyo")
+```
+k-means algorithm based on neighbour composition
+
+```{r}
+tx_spe_sample_1 <- clustByHood(tx_spe_sample_1, pm_cols = colnames(hoods), k = 10)
+```
+
+
+
+We can see what are the neighborhood distributions look like in each cluster using `plotProbDist`. 
 
 ```{r, fig.width=10, fig.height=10}
 tx_spe_sample_1 |> 
   plotProbDist(
     pm_cols = colnames(hoods),
     by_cluster = TRUE, 
     plot_all = TRUE, 
-    show_clusters = as.character(seq(10))
+    show_clusters = as.character(seq(99))
     )
+
+
 ```
 
 The clusters can then be plot on the tissue using `plotissue`

vignettes/Solutions.Rmd

@@ -148,9 +148,9 @@ plotSpotQC(
 
 ```{r, fig.width=7, fig.height=8, eval=FALSE}
 
-res_spatialLIBD = split(data.frame(res$mat), colData(spatial_data_gene_name)$sample_id ) 
+res_spatialLIBD = split(data.frame(res$mat), spatial_data_gene_name$sample_id, drop = TRUE ) 
 
-lapply(res_spatialLIBD, function(x) plotCorrelationMatrix(as.matrix(x[,-10]))) 
+lapply(res_spatialLIBD, function(x) plotCorrelationMatrix(as.matrix(x))) 
 
 ```
 
@@ -288,7 +288,7 @@ spatial_data |>
   filter(in_tissue, sample_id=="151673") |> 
 
   # Gate based on tissue morphology
-  tidySpatialExperiment::gate(alpha = 0.1) |> 
+  tidySpatialExperiment::gate(alpha = 0.1,colour = "spatialLIBD") |> 
 
   # Plot
   scater::plotUMAP(colour_by = ".gated")