DLPFC_analysis.Rmd

---
title: "DLPFC_analysis"
output: html_document
date: "2023-10-02"
---

```{r,warning=FALSE,error=FALSE,massage=FALSE,prompt=FALSE}
suppressPackageStartupMessages({
library(readr)
library(Seurat)
library(SeuratDisk)
library(SeuratObject)
library(anndata)
library(ggplot2)
library(factoextra) # clustering visualization
library(mclust)
library(zinbwave)
library(SingleCellExperiment)
library(aricode)
library(pscl)
library(knitr)
library(kableExtra)
library(Biobase)
library(Matrix)
library(MCMCpack)
library(Hmisc)
library(tidyverse)  # data manipulation
library(cluster)  # clustering algorithms
library(dendextend) # for comparing two dendrograms
library(dplyr)
library(GGally) # correlation plot
library(tidyr)
library(flextable)
library(NMF)
library(ggpubr) # MDS
library(RcppML) 
library(reshape2) 
library(cowplot) 
library(MAST)
library(smfishHmrf)
library(trendsceek)
library(sparklyr)
library(multinet)
library(RTriangle)
library(FactoMineR)
library(jackstraw)
library(CARD)
library(spacexr)
library(gridExtra)
library(pbmcapply)
library(Giotto)
library(smfishHmrf)
library(BayesSpace)
library(Banksy)
library(gridExtra)
})
```

### DLPFC dataset 151673

```{r}
library(BayesSpace)
library(ggplot2)
```

#### Read the dataset

```{r}
dlpfc_151673 <- getRDS("2020_maynard_prefrontal-cortex", "151673")
```

#### BayesSpace
##### Processing the data

A cleaned SingleCellExperiment object containing the dataset (after removing data specific to the spatialLIBD analyses) is available through BayesSpace. We preprocessed the data by performing PCA on the top 2,000 HVGs.

```{r}
set.seed(2023)
dec <- scran::modelGeneVar(dlpfc_151673)
top <- scran::getTopHVGs(dec, n = 2000)
dlpfc_151673 <- scater::runPCA(dlpfc_151673, subset_row=top)
## Add BayesSpace metadata
dlpfc_151673 <- spatialPreprocess(dlpfc_151673, platform="Visium", skip.PCA=TRUE)
```

We clustered the first 15 principal components, specifying 7 clusters to match the six brain layers plus one white matter region, and ran the MCMC algorithm for 50,000 iterations. We set our smoothing parameter gamma to 3, which we generally suggest for Visium datasets.

```{r}
q <- 7  # Number of clusters
d <- 13  # Number of PCs
```

```{r}
## Run BayesSpace clustering
set.seed(2023)
dlpfc_BayesSpace <- dlpfc_151673
dlpfc_BayesSpace <- spatialCluster(dlpfc_BayesSpace, q=q, d=d, platform='Visium',nrep=50000, burn.in = 1000, gamma=3, save.chain=TRUE)
```

###### Save the data

```{r}
#saveRDS(dlpfc_BayesSpace,"/dski/nobackup/xiaoyinl/DLPFC_analysis/BayesSpace_151673_t-distributed")
```

###### Read the data

```{r}
dlpfc_151673_BayesSpace <- readRDS("/dski/nobackup/xiaoyinl/DLPFC_analysis/BayesSpace_151673_t-distributed")
```

```{r}
BayesSpace_labels_151673 <- dlpfc_151673_BayesSpace$spatial.cluster
```

```{r}
## We recoded the cluster labels to match the expected brain layers
BayesSpace_labels_151673_relabel <- dplyr::recode(BayesSpace_labels_151673, 1, 4, 6, 3, 7, 5, 2)
```

###### Figure

```{r}
## View results
clusterPlot(dlpfc_151673_BayesSpace, label=BayesSpace_labels_151673_relabel, palette=NULL, size=0.05) +
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="BayesSpace")
```

```{r}
BayesSpace_clusters_151673 <- BayesSpace_labels_151673
```

#### Other clustering methods

Next, we ran several popular non-spatial clustering algorithms that are typically used in scRNA-seq analysis - k-means, Louvain, and mclust. For each of these algorithms, we clustered the first 15 principal components of the log-normalized expression matrix, specifying 7 clusters when possible. The Louvain algorithm was run on the k=10 shared nearest neighbor graph (weighted using Jaccard similarity). The mclust algorithm was run using the EEE multivariate mixture model.

###### K-means

```{r}
d <- 15
pca_data_selected_dimension_df <- reducedDim(dlpfc_151673, "PCA")[, seq_len(d)]
```

```{r}
## K-means with 7 clusters
set.seed(2023)
k_means_clusters_151673 <- kmeans(pca_data_selected_dimension_df, centers = 7)$cluster
```

```{r}
## We recoded the cluster labels to match the expected brain layers
kmeans_labels_151673_relabel <- dplyr::recode(k_means_clusters_151673, 1, 2, 6, 3, 5, 4, 7)
```

```{r}
## View results
clusterPlot(dlpfc_151673, label=kmeans_labels_151673_relabel, palette=NULL, size=0.05)+
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="K-means")
```

###### Hclust

```{r}
## hclust 
set.seed(2023)
# Dissimilarity matrix
PCA_results_matrix <- dist(pca_data_selected_dimension_df, method = "euclidean")
# Ward's method
PCA_results_hclust_ward <- hclust(PCA_results_matrix, method = "ward.D2" )
# Cut tree into 7 groups
hclust_clusters_151673 <- cutree(PCA_results_hclust_ward, k = 7)
```

```{r}
## We recoded the cluster labels to match the expected brain layers
hclust_labels_151673_relabel <- dplyr::recode(hclust_clusters_151673, 2, 1, 7, 3, 5, 6, 4)
```

```{r}
## View results
clusterPlot(dlpfc_151673, label=hclust_labels_151673_relabel, palette=NULL, size=0.05)+
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="Hierarchical clustering")
```

###### Louvain

```{r}
dlpfc_151673_counts_df <- counts(dlpfc_151673)
dlpfc_151673_counts_matrix <- as.matrix(dlpfc_151673_counts_df)
sce_louvain <- SingleCellExperiment(assays=list(
  counts=as(dlpfc_151673_counts_matrix,"dgCMatrix")))
pca_data_selected_dimension_matrix <- as.matrix(pca_data_selected_dimension_df)
reducedDims(sce_louvain)[["PCA"]] <- pca_data_selected_dimension_matrix
sce_louvain
```

```{r}
## Louvain with 7 clusters
set.seed(2023)
g.jaccard = scran::buildSNNGraph(sce_louvain, use.dimred="PCA", type="jaccard")
louvain_clusters_151673 <- igraph::cluster_louvain(g.jaccard, resolution = 0.8)$membership
```

```{r}
## We recoded the cluster labels to match the expected brain layers
louvain_labels_151673_relabel <- dplyr::recode(louvain_clusters_151673, 3, 1, 7, 5, 6, 2, 4)
```

```{r}
## View results
clusterPlot(dlpfc_151673,label=louvain_labels_151673_relabel, palette=NULL, size=0.05)+
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="Louvain")
```

### Spatial clustering algorithms

#### BANKSY

```{r}
library(Banksy)
library(SummarizedExperiment)
library(SpatialExperiment)
library(scuttle)
library(scater)
library(cowplot)
library(ggplot2)
```

```{r}
### Banksy
#### Preprocessing
pca_data_selected_dimension_transpose <- t(pca_data_selected_dimension_df) # PCA results
pca_data_selected_dimension_transpose_df <- as.data.frame(pca_data_selected_dimension_transpose)
expr_banksy <- as.matrix(pca_data_selected_dimension_transpose_df)
location_banksy <- data.frame(row=dlpfc_151673@colData$array_row, 
                              col=dlpfc_151673@colData$array_col)
rownames(location_banksy) = colnames(dlpfc_151673)
location_banksy
```

```{r}
bank_object <- BanksyObject(own.expr = expr_banksy,
                           cell.locs = location_banksy)
head(bank_object)
```

```{r}
set.seed(2023)
bank_object <- ClusterBanksy(bank_object, method = 'leiden', pca = FALSE, resolution = 0.9)
```

```{r}
features <- clust.names(bank_object)
feature.types <- rep('discrete', 1)
main <- c('BANKSY')
plotSpatialFeatures(bank_object, by = features, type = feature.types, main = main, pt.size = 0.9, main.size = 15)
```

```{r}
bank_object_meta_data_df <- as.data.frame(bank_object@meta.data)
spatial_cluster_Banksy_clusters <- bank_object_meta_data_df$clust_M1_lam0.2_k50_res0.9
```

```{r}
## We recoded the cluster labels to match the expected brain layers
Banksy_labels_151673_relabel <- dplyr::recode(spatial_cluster_Banksy_clusters, 3, 7, 2, 6, 1, 5, 4)
```

```{r}
## View results
clusterPlot(dlpfc_151673,label=Banksy_labels_151673_relabel, palette=NULL, size=0.05)+
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="BANKSY")
```

#### HMRF

We also ran one recently published spatial clustering algorithms Giotto (HMRF). To analyze these samples we adapted the respective tutorials (Giotto) and provide this code. For ease of reproduction, we include the obtained cluster labels directly with this package.

```{r}
HMRF_labels_151673 <- read.csv(system.file("extdata", "2020_maynard_prefrontal-cortex", "151673.Giotto_HMRF.csv", package = "BayesSpace"))$HMRF_km_Delaunay_k7_b.9
```

```{r}
## We recoded the cluster labels to match the expected brain layers
HMRF_labels_151673_relabel <- dplyr::recode(HMRF_labels_151673, 1, 5, 6, 4, 2, 3, 7)
```

```{r}
## View results
clusterPlot(dlpfc_151673,label=HMRF_labels_151673_relabel, palette=NULL, size=0.05)+
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="Giotto")
```

### 3D clustering

#### STitch3D

```{r}
clusters_151673_STitch3D <- read.csv("/dski/nobackup/xiaoyinl/DLPFC_analysis/STitch3D_dataset/results_DLPFC/cluster_results_151673_whole.csv")
clusters_151673_STitch3D$Cluster = clusters_151673_STitch3D$Cluster+1
clusters_151673_STitch3D
```

```{r}
subset_dlpfc_151673 <- dlpfc_151673[, dlpfc_151673@colData$barcode %in% clusters_151673_STitch3D$barcode]
```

```{r}
STitch3D_labels_151673 <- clusters_151673_STitch3D$Cluster
```

```{r}
## We recoded the cluster labels to match the expected brain layers
STitch3D_labels_151673_relabel <- dplyr::recode(clusters_151673_STitch3D$Cluster, 5, 4, 3, 2, 7, 6,1)
```

```{r}
## View results
clusterPlot(subset_dlpfc_151673, label=STitch3D_labels_151673_relabel, palette=NULL, size=0.05)+
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="STitch3D")
```

```{r}
Manual_annotation_STitch3D <- subset_dlpfc_151673@colData$layer_guess_reordered
```

```{r}
Manual_annotation_STitch3D_df <- data.frame(Manual_annotation_STitch3D)
Manual_annotation_STitch3D_df$Manual_annotation_STitch3D <- dplyr::recode(Manual_annotation_STitch3D_df$Manual_annotation_STitch3D, 
                                   "Layer1" = "1",
                                   "Layer2" = "2",
                                   "Layer3" = "3",
                                   "Layer4" = "4",
                                   "Layer5" = "5",
                                   "Layer6" = "6",
                                   "WM" = "7")
Manual_annotation_STitch3D_df
```

```{r}
ARI_STitch3D <- adjustedRandIndex(STitch3D_labels_151673, Manual_annotation_STitch3D_df$Manual_annotation_STitch3D)
NMI_STitch3D <- NMI(STitch3D_labels_151673, Manual_annotation_STitch3D_df$Manual_annotation_STitch3D)
ARI_STitch3D #  0.5746382
NMI_STitch3D #  0.7007096
```

#### BASS

##### Input the dataset

**Dataset** 

3,639 (151673), 3,673 (151674), 3,592 (151675), and 3,460 (151676) spots along with their spatial locations

```{r}
load("/dski/nobackup/xiaoyinl/BASS-Analysis/data/spatialLIBD_p3.RData")
```

##### Running the algorithm

```{r,message=FALSE}
library(BASS)
library(Matrix)
# hyper-parameters
# We set the number of cell types to a relatively large
# number (20) to capture the expression heterogeneity.
C <- 20
# number of spatial domains
R <- 7
```

```{r}
set.seed(2023)
# Set up BASS object
BASS <- createBASSObject(cntm, xym, C = C, R = R,
  beta_method = "SW", init_method = "mclust", 
  nsample = 10000)
```

```{r,warning=FALSE}
# Data pre-processing:
# 1.Library size normalization followed with a log2 transformation
# 2.Select top 3000 spatially expressed genes with SPARK-X
# 3.Dimension reduction with PCA
BASS <- BASS.preprocess(BASS, doLogNormalize = TRUE,
  geneSelect = "sparkx", nSE = 3000, doPCA = TRUE, 
  scaleFeature = FALSE, nPC = 20)
```

```{r}
# Run BASS algorithm
BASS <- BASS.run(BASS)
```

```{r}
# post-process posterior samples:
# 1.Adjust for label switching with the ECR-1 algorithm
# 2.Summarize the posterior samples to obtain the spatial domain labels
BASS <- BASS.postprocess(BASS)
```

```{r}
zlabels <- BASS@results$z # spatial domain labels
```

```{r}
labels_151673 <- zlabels[[1]]
labels_151673
```

```{r}
labels_151674 <- zlabels[[2]]
labels_151674
```

```{r}
labels_151675 <- zlabels[[3]]
labels_151675
```

```{r}
labels_151676 <- zlabels[[4]]
labels_151676
```

```{r}
write.csv(labels_151673,"/dski/nobackup/xiaoyinl/DLPFC_analysis/BASS_labels/BASS_labels_151673")
```

```{r}
write.csv(labels_151674,"/dski/nobackup/xiaoyinl/DLPFC_analysis/BASS_labels/BASS_labels_151674")
```

```{r}
write.csv(labels_151675,"/dski/nobackup/xiaoyinl/DLPFC_analysis/BASS_labels/BASS_labels_151675")
```

```{r}
write.csv(labels_151676,"/dski/nobackup/xiaoyinl/DLPFC_analysis/BASS_labels/BASS_labels_151676")
```

##### 151673

```{r}
BASS_labels_151673 <- read.csv("/dski/nobackup/xiaoyinl/DLPFC_analysis/BASS_labels/BASS_labels_151673")
```

```{r}
BASS_labels_151673_chosen <- BASS_labels_151673$x
```

```{r}
## We recoded the cluster labels to match the expected brain layers
BASS_labels_151673_recode <- dplyr::recode(BASS_labels_151673_chosen, 7, 5, 3, 4, 6, 1, 2)
```

```{r}
## View results
clusterPlot(dlpfc_151673,label=BASS_labels_151673_recode, palette=NULL, size=0.05)+
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="BASS")
```

#### Evaluation
##### Ground truth

```{r}
Manual_annotation <- dlpfc_151673@colData$layer_guess_reordered
```

```{r}
Manual_annotation_df <- data.frame(Manual_annotation)
# Convert factor to character
Manual_annotation_df$Manual_annotation <- as.character(Manual_annotation_df$Manual_annotation)
# Recode including NA values
Manual_annotation_df$Manual_annotation <- dplyr::recode(Manual_annotation_df$Manual_annotation, 
                                   "Layer1" = "1",
                                   "Layer2" = "2",
                                   "Layer3" = "3",
                                   "Layer4" = "4",
                                   "Layer5" = "5",
                                   "Layer6" = "6",
                                   "WM" = "7",
                                   .missing = "7")
# Convert back to factor if needed
Manual_annotation_df$Manual_annotation <- as.factor(Manual_annotation_df$Manual_annotation)
Manual_annotation_df
```

```{r}
ground_truth_label = Manual_annotation_df$Manual_annotation
```

```{r}
clustering_summary_df <- data.frame(
  k_means_clusters_151673,
  hclust_clusters_151673,
  louvain_clusters_151673,
  BayesSpace_labels_151673,
  HMRF_labels_151673,
  BASS_labels_151673_chosen
)
```

```{r}
ARI_k_means_PCA_results <- adjustedRandIndex(ground_truth_label, clustering_summary_df$k_means_clusters_151673)
ARI_hclust_PCA_results  <- adjustedRandIndex(ground_truth_label, clustering_summary_df$hclust_clusters_151673)
ARI_louvain_PCA_results  <- adjustedRandIndex(ground_truth_label, clustering_summary_df$louvain_clusters_151673)
ARI_giotto_PCA_results <- adjustedRandIndex(ground_truth_label,clustering_summary_df$HMRF_labels_151673)
ARI_BayesSpace_PCA_results  <- adjustedRandIndex(ground_truth_label, clustering_summary_df$BayesSpace_labels_151673)
#ARI_Banksy_PCA_results <- adjustedRandIndex(ground_truth_label,clustering_summary_df$spatial_cluster_Banksy_clusters)
ARI_BASS_results <- adjustedRandIndex(ground_truth_label,clustering_summary_df$BASS_labels_151673_chosen)
```

```{r}
ARI_results_df <- data.frame(
    ARI_k_means_PCA_results, 
    ARI_hclust_PCA_results, 
    ARI_louvain_PCA_results, 
    ARI_giotto_PCA_results, 
    ARI_BayesSpace_PCA_results, 
    ARI_BASS_results
  )
ARI_results_df
```

```{r}
NMI_k_means_PCA_results <- NMI(ground_truth_label, clustering_summary_df$k_means_clusters_151673)
NMI_hclust_PCA_results  <- NMI(ground_truth_label, clustering_summary_df$hclust_clusters_151673)
NMI_louvain_PCA_results  <- NMI(ground_truth_label, clustering_summary_df$louvain_clusters_151673)
NMI_giotto_PCA_results <- NMI(ground_truth_label,clustering_summary_df$HMRF_labels_151673)
NMI_BayesSpace_PCA_results  <- NMI(ground_truth_label, clustering_summary_df$BayesSpace_labels_151673)
#NMI_Banksy_PCA_results <- NMI(ground_truth_label,clustering_summary_df$spatial_cluster_Banksy_clusters)
NMI_BASS_results <- NMI(ground_truth_label,clustering_summary_df$BASS_labels_151673_chosen)
```

```{r}
NMI_results_df <- data.frame(
    NMI_k_means_PCA_results, 
    NMI_hclust_PCA_results, 
    NMI_louvain_PCA_results, 
    NMI_giotto_PCA_results, 
    NMI_BayesSpace_PCA_results, 
    NMI_BASS_results
  )
NMI_results_df
```

```{r}
library(ggplot2)
library(RColorBrewer)

ari_data <- data.frame(
  Method = c(
    "BayesSpace", 
    "BANKSY",
    "K-means", 
    "Hierarchical",
    "Louvain", 
    "BASS",
    "HMRF",
    "STitch3D",
    "CARD",
    "RCTD"),
  ARI = c(
    0.5375717, 
    0.3265748,
    0.2669568, 
    0.2318858, 
    0.2938219, 
    0.5869484, 
    0.3446974,
    0.5746382,
    0.4167843,
    0.4964387),
  Category = c(
    "2D Spatial Clustering", 
    "2D Spatial Clustering", 
    "Classical Clustering", 
    "Classical Clustering", 
    "Classical Clustering", 
    "3D Spatial Clustering", 
    "2D Spatial Clustering", 
    "3D Spatial Clustering", 
    "2D Spatial Clustering", 
    "2D Spatial Clustering")
)

ari_data$Category <- factor(ari_data$Category, levels = c("Classical Clustering", "2D Spatial Clustering", "3D Spatial Clustering"))

colors <- c("Classical Clustering" = "#F4A582", 
            "2D Spatial Clustering" = "#FDDBC7", 
            "3D Spatial Clustering" =  "#D1E5F0")

ggplot(ari_data, aes(x = reorder(Method, ARI), y = ARI, fill = Category)) +
  geom_bar(stat = "identity") +
  theme_minimal() +
  labs(title = "ARI Evaluation", x = "Method", y = "ARI",fill="Method Category") +
  scale_fill_manual(values = colors) +
  theme(legend.position = "bottom")
```

```{r}
library(ggplot2)
library(RColorBrewer)

NMI_data <- data.frame(
  Method = c(
    "BayesSpace", 
    "BANKSY",
    "K-means", 
    "Hierarchical",
    "Louvain", 
    "BASS",
    "HMRF",
    "STitch3D",
    "CARD",
    "RCTD"),
  NMI = c(
    0.6564575, 
    0.4378206,
    0.3876779, 
    0.3789865, 
    0.3962948, 
    0.6847375, 
    0.4466828,
    0.7007096,
    0.5283489,
    0.5678436),
  Category = c(
    "2D Spatial Clustering", 
    "2D Spatial Clustering", 
    "Classical Clustering", 
    "Classical Clustering", 
    "Classical Clustering", 
    "3D Spatial Clustering", 
    "2D Spatial Clustering", 
    "3D Spatial Clustering", 
    "2D Spatial Clustering", 
    "2D Spatial Clustering")
)

NMI_data$Category <- factor(NMI_data$Category, levels = c("Classical Clustering", "2D Spatial Clustering", "3D Spatial Clustering"))

colors <- c("Classical Clustering" = "#F4A582", 
            "2D Spatial Clustering" = "#FDDBC7", 
            "3D Spatial Clustering" =  "#D1E5F0")
```

```{r}
ggplot(NMI_data, aes(x = reorder(Method, NMI), y = NMI, fill = Category)) +
  geom_bar(stat = "identity") +
  theme_minimal() +
  labs(title = "NMI Evaluation", x = "Method", y = "NMI",fill="Method Category") +
  scale_fill_manual(values = colors) +
  theme(legend.position = "bottom")
```

#### Comparison(Visualization)

```{r}
clusters_df <- as.data.frame(clusters)
clusters_df
```

```{r}
library(gridExtra)

plot1 <- clusterPlot(dlpfc, label=clusters_df$k.means, palette=NULL, size=0.05) +
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="k-means")

plot2 <- clusterPlot(dlpfc, label=clusters_df$hclust, palette=NULL, size=0.05) +
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="Hierarchical (Ward)")

plot3 <- clusterPlot(dlpfc, label=clusters_df$Louvain, palette=NULL, size=0.05) +
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="Louvain")

plot4 <- clusterPlot(dlpfc, label=clusters_df$mclust, palette=NULL, size=0.05) +
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="mclust")

plot5 <- clusterPlot(dlpfc, label=clusters_df$BayesSpace..t.distributed.error., palette=NULL, size=0.05) +
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="BayesSpace (t-distributed error)")

plot6 <- clusterPlot(dlpfc, label=clusters_df$BayesSpace..normal.error., palette=NULL, size=0.05) +
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="BayesSpace (normal error)")

plot7 <- clusterPlot(dlpfc, label=clusters_df$Giotto, palette=NULL, size=0.05) +
  scale_fill_viridis_d(option = "A", labels = 1:7) +
  labs(title="Giotto HMRF")
```

```{r}
grid.arrange(plot1, plot2, plot3, plot4, plot5, plot6, plot7, ncol = 4, nrow =2)
```

#### Evaluation

Finally, to evaluate each algorithm’s cluster assignments, we use the manual annotations provided with the original data by Maynard et al. We also compare the cluster assignments reported in this paper using the Walktrap clustering algorithm (a non-spatial hierarchical clustering algorithm.)

```{r}
clusters_df <- as.data.frame(clusters)
clusters_df
```

```{r}
clusters_df$Manual_annotation <- dlpfc@colData$layer_guess_reordered
clusters_df
```

```{r}
ARI <- purrr::map(clusters_df, function(x) mclust::adjustedRandIndex(x, clusters_df$Manual_annotation))
ARI
```

```{r}
library(ggplot2)
library(RColorBrewer)

colors <- c("#B2182B","#D6604D","orange","#F4A582","#FDDBC7","pink","#D1E5F0","lightblue","#92C5DE","#4393C3","#2166AC")

ari_data <- data.frame(
  Method = c(
    "BayesSpace", 
    "k-means", 
    "hclust",
    "Louvain", 
    "mclust", 
    "BASS", 
    "DR_SC (HVG)", 
    "DR_SC (SVG)", 
    "Giotto",
    "STitch3D"
  ),
  ARI = c(
    0.54647, 
    0.2734458, 
    0.2349343, 
    0.2860929, 
    0.4380811, 
    0.5987186, 
    0.322173, 
    0.486127, 
    0.3520176,
    0.5746382
  )
)

ari_data <- ari_data[order(ari_data$ARI, decreasing = TRUE), ]

ggplot(ari_data, aes(x = reorder(Method, -ARI), y = ARI, fill = Method)) +
  geom_bar(stat = "identity") +
  theme_minimal() +
  labs(title = "ARI Evaluation", x = "Method", y = "ARI") +
  coord_flip() +
  scale_fill_manual(values = colors)
```