Skip to contents

3. Calculate Clustering Metrics For Tabula Muris

3_clustering_metrics.Rmd

First we get a list of Seurat object files containing the clustering results for callback, sc-SHC, and CHOIR.


tissue_seurat_files <- list.files(pattern = "cluster_results_seurat.rds")

Then, we loop over all of the Seurat objects and calculate ARI, V-measure, homogeneity, completeness, FMI, and Jaccard similarity for the clusters determined by each method.

cluster_metrics_df <- data.frame()

for (tissue_rds_file in tissue_seurat_files) {
  tissue_seurat_obj <- readRDS(tissue_rds_file)
  
  
  tissue_name <- sub('cluster_results_seurat.rds', "", tissue_rds_file)
  
  # callback metrics
  ari <- pdfCluster::adj.rand.index(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                    tissue_seurat_obj@meta.data$callback_idents)
  
  v_measure_results <- clevr::v_measure(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                        tissue_seurat_obj@meta.data$callback_idents)
  
  homogeneity_results <- clevr::homogeneity(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                            tissue_seurat_obj@meta.data$callback_idents)
  
  completeness_results <- clevr::completeness(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                              tissue_seurat_obj@meta.data$callback_idents)
  
  fowlkes_mallows_results <- clevr::fowlkes_mallows(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                                    tissue_seurat_obj@meta.data$callback_idents)
  
  jaccard_results <- clusteval::jaccard(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                        tissue_seurat_obj@meta.data$callback_idents)
  
  method <- "callback"
  
  callback_metrics_df_row <- data.frame(tissue_name,
                                     method,
                                     ari,
                                     v_measure_results,
                                     fowlkes_mallows_results,
                                     homogeneity_results,
                                     completeness_results,
                                     jaccard_results)
  
  cluster_metrics_df <- rbind(cluster_metrics_df, callback_metrics_df_row)
  
  
  # sc-SHC metrics
  ari <- pdfCluster::adj.rand.index(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                    tissue_seurat_obj@meta.data$scSHC_clusters)
  
  v_measure_results <- clevr::v_measure(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                        tissue_seurat_obj@meta.data$scSHC_clusters)
  
  homogeneity_results <- clevr::homogeneity(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                            tissue_seurat_obj@meta.data$scSHC_clusters)
  
  completeness_results <- clevr::completeness(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                              tissue_seurat_obj@meta.data$scSHC_clusters)
  
  fowlkes_mallows_results <- clevr::fowlkes_mallows(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                                    tissue_seurat_obj@meta.data$scSHC_clusters)
  
  jaccard_results <- clusteval::jaccard(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                        tissue_seurat_obj@meta.data$scSHC_clusters)
  
  method <- "sc-SHC"
  
  scSHC_metrics_df_row <- data.frame(tissue_name,
                                     method,
                                     ari,
                                     v_measure_results,
                                     fowlkes_mallows_results,
                                     homogeneity_results,
                                     completeness_results,
                                     jaccard_results)
  
  cluster_metrics_df <- rbind(cluster_metrics_df, scSHC_metrics_df_row)
  
  # choir metrics
  ari <- pdfCluster::adj.rand.index(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                    tissue_seurat_obj@meta.data$CHOIR_clusters_0.05)
  
  v_measure_results <- clevr::v_measure(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                        tissue_seurat_obj@meta.data$CHOIR_clusters_0.05)
  
  homogeneity_results <- clevr::homogeneity(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                            tissue_seurat_obj@meta.data$CHOIR_clusters_0.05)
  
  completeness_results <- clevr::completeness(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                              tissue_seurat_obj@meta.data$CHOIR_clusters_0.05)
  
  fowlkes_mallows_results <- clevr::fowlkes_mallows(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                                    tissue_seurat_obj@meta.data$CHOIR_clusters_0.05)
  
  jaccard_results <- clusteval::jaccard(tissue_seurat_obj@meta.data$cell_ontology_class, 
                                        tissue_seurat_obj@meta.data$CHOIR_clusters_0.05)
  
  method <- "CHOIR"

  
  choir_metrics_df_row <- data.frame(tissue_name,
                                   method,
                                   ari,
                                   v_measure_results,
                                   fowlkes_mallows_results,
                                   homogeneity_results,
                                   completeness_results,
                                   jaccard_results)
  
  cluster_metrics_df <- rbind(cluster_metrics_df, choir_metrics_df_row)
  
}

Finally, we save the metrics to a CSV file.

write.csv(cluster_metrics_df, file="cluster_metrics_df.csv")