Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update clustering report to include leiden clustering #895

Merged
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
166 changes: 67 additions & 99 deletions analyses/cell-type-ewings/scripts/utils/clustering-functions.R
Original file line number Diff line number Diff line change
Expand Up @@ -10,152 +10,116 @@ source(jaccard_functions)
validation_functions <- file.path(module_base, "scripts", "utils", "tumor-validation-helpers.R")
source(validation_functions)

# Perform clustering -----------------------------------------------------------

# get louvain, jaccard clusters for a specified value of k (nearest neighbors)
get_clusters <- function(pcs, k) {
clusters <- bluster::clusterRows(
pcs,
bluster::NNGraphParam(
k = k,
type = "jaccard",
cluster.fun = "louvain"
)
)

return(clusters)
}

# define a function to perform clustersweep and get clusters across multiple values of k (5,40,5)
cluster_sweep <- function(sce) {
# first perform clustering across parameters
cluster_results <- bluster::clusterSweep(reducedDim(sce, "PCA"),
bluster::NNGraphParam(),
k = as.integer(seq(5, 40, 5)),
cluster.fun = "louvain",
type = "jaccard"
)

# turn results into a data frame
cluster_df <- cluster_results$clusters |>
as.data.frame() |>
# add barcode column
dplyr::mutate(barcodes = colnames(sce)) |>
# combine all cluster results into one column
tidyr::pivot_longer(
cols = ends_with("jaccard"),
names_to = "params",
values_to = "cluster"
) |>
# separate out parameters, nn, function, and type into their own columns
dplyr::mutate(
nn_param = stringr::word(params, 1, sep = "_") |>
stringr::str_replace("k.", "k_"),
cluster_fun = stringr::word(params, 2, sep = "_") |>
stringr::str_remove("cluster.fun."),
cluster_type = stringr::word(params, -1, sep = "_") |>
stringr::str_remove("type.")
) |>
# remove combined params column
dplyr::select(-params)

return(cluster_df)
}

# cluster statistics functions -------------------------------------------------


# get silhouette width and cluster purity for each cluster
# calculates values across all nn_param options used to determine clustering
# all_cluster_results must have nn_param column
# calculates values across all parameters used to determine clustering
# all_cluster_results must have cluster_params column
get_cluster_stats <- function(sce,
all_cluster_results) {
pcs <- reducedDim(sce, "PCA")

# split clustering results by param used
split_clusters <- all_cluster_results |>
split(all_cluster_results$nn_param)
split(all_cluster_results$cluster_params)

# for each nn_param get cluster width and purity
all_stats_df <- split_clusters |>
purrr::map(\(df){
sil_df <- bluster::approxSilhouette(pcs, df$cluster) |>
as.data.frame() |>
tibble::rownames_to_column("barcodes")
tibble::rownames_to_column("cell_id")

purity_df <- bluster::neighborPurity(pcs, df$cluster) |>
as.data.frame() |>
tibble::rownames_to_column("barcodes")
tibble::rownames_to_column("cell_id")

# join into one data frame to return
stats_df <- sil_df |>
dplyr::left_join(purity_df, by = "barcodes")
dplyr::left_join(purity_df, by = "cell_id")

return(stats_df)
}) |>
dplyr::bind_rows(.id = "nn_param")
dplyr::bind_rows(.id = "cluster_params") |>
dplyr::left_join(all_cluster_results, by = c("cell_id", "cluster_params"))

return(all_stats_df)
}

# calculate cluster stability for a single set of clusters using ari
# bootstrap and get ari for clusters compared to sampled clusters
# re-clusters and gets ari across 20 iterations
get_ari <- function(pcs,
clusters,
k) {
ari <- c()
for (iter in 1:20) {
# sample cells with replacement
sample_cells <- sample(nrow(pcs), nrow(pcs), replace = TRUE)
resampled_pca <- pcs[sample_cells, , drop = FALSE]

# perform clustering on sampled cells
resampled_clusters <- get_clusters(resampled_pca, k)

# calculate ARI between new clustering and original clustering
ari[iter] <- pdfCluster::adj.rand.index(resampled_clusters, clusters[sample_cells])
}

ari_df <- data.frame(
ari = ari,
k_value = k
)
}

# get cluster stability for each nn_param cluster results are available for
# get cluster stability for each unique combination of params used for clustering
# must have `cluster_params` column
get_cluster_stability <- function(sce,
all_cluster_results) {
pcs <- reducedDim(sce, "PCA")

# split clustering results by param used
cluster_df_list <- all_cluster_results |>
split(all_cluster_results$nn_param)

split(all_cluster_results$cluster_params)
# for each parameter, get ari values
cluster_stability_df <- cluster_df_list |>
purrr::imap(\(df, k_value){
# make sure k is numeric and remove extra k_
k <- stringr::str_remove(k_value, "k_") |>
as.numeric()

get_ari(pcs, df$cluster, k)
purrr::map(\(df){

# make sure we set objective function to available options
objective_function <- dplyr::if_else(!is.na(unique(df$objective_function)),
unique(df$objective_function),
"CPM")


# run stability
rOpenScPCA::calculate_stability(sce,
cluster_df = df,
algorithm = unique(df$algorithm),
nn = unique(df$nn),
resolution = unique(df$resolution),
objective_function = objective_function)

}) |>
dplyr::bind_rows()

dplyr::bind_rows(.id = "cluster_params")
return(cluster_stability_df)
}

# Plotting ---------------------------------------------------------------------

# plot individual stats for clusters, either purity or width
plot_cluster_stats <- function(all_stats_df,
stat_column) {
ggplot(all_stats_df, aes(x = nn_param, y = {{ stat_column }})) +
stat_column,
plot_title) {
ggplot(all_stats_df, aes(x = nn, y = {{ stat_column }})) +
# ggforce::geom_sina(size = .2) +
ggbeeswarm::geom_quasirandom(method = "smiley", size = 0.1) +
facet_wrap(vars(resolution),
labeller = labeller(resolution = ~ glue::glue("{.}-res"))) +
stat_summary(
aes(group = nn),
color = "red",
# median and quartiles for point range
fun = "median",
fun.min = function(x) {
quantile(x, 0.25)
},
fun.max = function(x) {
quantile(x, 0.75)
}
) +
labs(
title = plot_title
)
}

# plot cluster stability
plot_cluster_stability <- function(stat_df,
plot_title){

ggplot(stability_df, aes(x = nn, y = ari)) +
geom_jitter(width = 0.1) +
facet_wrap(vars(resolution),
labeller = labeller(resolution = ~ glue::glue("{.}-res"))) +
labs(title = "Cluster stability") +
stat_summary(
aes(group = nn_param),
aes(group = nn),
color = "red",
# median and quartiles for point range
fun = "median",
Expand All @@ -165,7 +129,11 @@ plot_cluster_stats <- function(all_stats_df,
fun.max = function(x) {
quantile(x, 0.75)
}
) +
labs(
title = plot_title
)

}


Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -209,19 +209,44 @@ The annotations are shown below the heatmap.
- Density plot showing gene expression or gene set scores across all cells.
Each row is a cell type and the expression or score is plotted on the x-axis.

```{r}
# check that marker genes are expressed, otherwise turn off those plots
total_exp <- sum(classification_df[marker_gene_columns])
if(total_exp > 0){
show_marker_gene_plots <- TRUE
} else {
show_marker_gene_plots <- FALSE
message("No expression of marker genes in this library. No plots will be displayed in sections labeled 'Marker gene expression'.")
}

```

```{r}
# check that gene set scores aren't all 0, otherwise turn off those plots
total_score <- sum(classification_df[geneset_columns])
if(total_score > 0){
show_gene_set_plots <- TRUE
} else {
show_gene_set_plots <- FALSE
message("Genes present in provided gene sets are not expressed in this library. No plots will be displayed in sections labeled 'Gene set scores'.")
}

```


### Tumor vs. Normal

In this section we show just the cells that are considered tumor and normal, lumping all non-tumor cell types together.

**Marker gene expression**

```{r}
full_celltype_heatmap(classification_df, marker_gene_columns, "singler_tumor_normal")

```{r, eval=show_marker_gene_plots}
full_celltype_heatmap(classification_df, marker_gene_columns, "singler_tumor_normal")
```


```{r}
```{r, eval=show_marker_gene_plots}
plot_density(
classification_df,
"tumor_sum",
Expand All @@ -231,11 +256,11 @@ plot_density(

**Gene set scores**

```{r}
```{r, eval=show_gene_set_plots}
full_celltype_heatmap(classification_df, geneset_columns, "singler_tumor_normal")
```

```{r, fig.height=10}
```{r, fig.height=10, eval=show_gene_set_plots}
geneset_columns |>
purrr::map(\(column){
plot_density(
Expand All @@ -254,11 +279,11 @@ In this section we show all tumor cells and the top 5 most represented normal ce

**Marker gene expression**

```{r}
```{r, eval=show_marker_gene_plots}
full_celltype_heatmap(classification_df, marker_gene_columns, "singler_lumped")
```

```{r, fig.height=10}
```{r, fig.height=10, eval=show_marker_gene_plots}
marker_gene_columns |>
purrr::map(\(column){
plot_density(
Expand All @@ -273,11 +298,11 @@ marker_gene_columns |>

**Gene set scores**

```{r}
```{r, eval=show_gene_set_plots}
full_celltype_heatmap(classification_df, geneset_columns, "singler_lumped")
```

```{r, fig.height=10}
```{r, fig.height=10, eval=show_gene_set_plots}
geneset_columns |>
purrr::map(\(column){
plot_density(
Expand All @@ -295,12 +320,12 @@ Here we compare the marker gene expression and gene set scores for cells annotat

**Marker gene expression**

```{r}
```{r, eval=show_marker_gene_plots}
full_celltype_heatmap(classification_df, marker_gene_columns, "consensus")
```


```{r}
```{r, eval=show_marker_gene_plots}
plot_density(
classification_df,
"tumor_sum",
Expand All @@ -311,11 +336,11 @@ plot_density(

**Gene set scores**

```{r}
```{r, eval=show_gene_set_plots}
full_celltype_heatmap(classification_df, geneset_columns, "consensus")
```

```{r, fig.height=10}
```{r, fig.height=10, eval=show_gene_set_plots}
geneset_columns |>
purrr::map(\(column){
plot_density(
Expand Down
Loading