kahkengwong/CEP-IP_Framework

CEP-IP: An Explainable Framework for Cell Subpopulation Identification in Single-cell Transcriptomics (by Kah Keng Wong, Sep 2025) (arXiv Preprint)

📊 Dataset Overview

This dataset contains processed single-cell RNA-seq data from prostate tissue samples, including both tumor and benign samples from seven prostate cancer (PCa) patients, of publicly-available dataset obtained from: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE185344

The data has been quality-controlled, normalized, and clustered using Seurat v5.1.0.

The aim of this repository is to host the processed Seurat object GSE185344_Seurat_processed.RData (9.52 GB), which exceeds GitHub's file size limits.

Contents:

• Processed Seurat objects for tumor and benign prostate samples: GSE185344_Seurat_processed.RData
• Quality-controlled expression data with batch correction
• UMAP embeddings and cluster annotations
• Cluster marker genes and metadata
• Ready for subsequent generalized additive model (GAM) analysis with mgcv package.

📚 Original Data Source

This processed dataset is based on publicly available data from: Wong HY, Sheng Q, Hesterberg AB, Croessmann S et al. Single cell analysis of cribriform prostate cancer reveals cell intrinsic and tumor microenvironmental pathways of aggressive disease. Nat Commun 2022;13(1):6036. https://doi.org/10.1038/s41467-022-33780-1

🛠️ Technical Details

Processing Information

• Seurat version: 5.1.0
• Key R packages used:
  • future (parallel processing)
  • parallel (core parallel processing)
  • scDblFinder (doublet detection)
  • Seurat (single-cell analysis)
  • SingleCellExperiment (data conversion)

Key Processing Steps

• Quality control filtering (>500 features per cell)
• Ribosomal gene filtering (cells above 90th percentile removed)
• Mitochondrial gene filtering (cells above 90th percentile removed)
• Cell cycle regression to remove phase effects
• Doublet removal using scDblFinder
• Batch effect correction via SCTransform integration
• Dimensionality reduction and clustering (resolution 0.5)
• UMAP visualization with cosine metric

🧮 Code Availability

The complete analysis pipeline of this GAM-REML-TPRS project is available on GitHub: https://github.com/kahkengwong/CEP-IP_Framework

After downloading the GSE185344_Seurat_processed.RData file, you can run the rest of the code available on GitHub starting from Part_2_UMAP_Heatmap_Spearman-Kendall's-matrix.r until Part_3.15_Monocle3_Pre-IP_vs_Post-IP_TREP.r.

🎯 Citation

If you use this processed dataset, please cite: Wong KK (2025). CEP-IP: An Explainable Framework for Cell Subpopulation Identification in Single-cell Transcriptomics. arXiv preprint arXiv:2509.12073. https://arxiv.org/abs/2509.12073

Please also cite the source dataset: Wong HY, Sheng Q, Hesterberg AB, Croessmann S et al (2022). Single cell analysis of cribriform prostate cancer reveals cell intrinsic and tumor microenvironmental pathways of aggressive disease. Nat Commun 13(1):6036. https://doi.org/10.1038/s41467-022-33780-1

📋 License

This dataset is licensed under the MIT License.

📝 Click to view complete MIT License

MIT License

Copyright (c) 2025 Kah Keng Wong

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

🔬 Detailed Analysis Pipeline

The complete processing pipeline includes: data loading → quality control → cell cycle regression → doublet removal → batch correction → clustering → UMAP visualization.

📝 Click to view complete processing code

##########################################
# A. Dataset Description
##########################################
This dataset contains scRNA-seq data processed using Seurat v5.1.0, and the dataset was obtained from: 
https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE185344

The original data was published by Wong et al. Nat Commun 2022;13(1):6036 (doi: 10.1038/s41467-022-33780-1)
available from: https://pubmed.ncbi.nlm.nih.gov/36229464/ 


###############################################
# B. Processing Information with Seurat in R
###############################################
setwd("C:/...") # Set the local directory
library(dplyr)  
library(future)         
library(ggplot2)      
library(parallel)      
library(scales)        
library(scDblFinder) 
library(Seurat)          
library(SingleCellExperiment) 
library(viridis)  

# =========================================
# 1. scRNA-seq Dataset Pre-processing
# =========================================
# Load scRNA-seq dataset
loaded_df <- readRDS("C:/...directory.../GSE185344_PH_scRNA.final.rds")

# Extract Seurat object, the core data structure
seurat_obj <- loaded_df$obj

# Define sample names (tumor vs benign)
prostate_ca_samples <- c("HYW_4701_Tumor", "HYW_4847_Tumor", "HYW_4880_Tumor", 
                         "HYW_4881_Tumor", "HYW_5386_Tumor", "HYW_5742_Tumor", 
                         "HYW_5755_Tumor")
non_cancerous_samples <- c("HYW_4701_Benign", "HYW_4847_Benign", "HYW_4880_Benign", 
                           "HYW_4881_Benign", "HYW_5386_Benign", "HYW_5742_Benign", 
                           "HYW_5755_Benign")

# Subset and process Seurat object for normalization and feature selection
process_seurat <- function(seurat_obj, sample_names, project_name) {
    subset_obj <- subset(seurat_obj, subset = orig.ident %in% sample_names)  # Subset by sample ID
    subset_obj <- NormalizeData(subset_obj, normalization.method = "LogNormalize", scale.factor = 10000)  # Log-normalize (expression scaling)
    subset_obj <- FindVariableFeatures(subset_obj, selection.method = "vst", nfeatures = 2000)  # Top 2000 variable features (VST method)
    return(subset_obj)
}

# Process samples to split into tumor and benign
prostate_ca_seurat <- process_seurat(seurat_obj, prostate_ca_samples, "prostate-ca")
non_cancerous_seurat <- process_seurat(seurat_obj, non_cancerous_samples, "NonCancerous")

# Filter by feature and count thresholds to remove low-quality cells
prostate_ca_seurat <- subset(prostate_ca_seurat, subset = nFeature_RNA > 500 & nCount_RNA > 0)
non_cancerous_seurat <- subset(non_cancerous_seurat, subset = nFeature_RNA > 500 & nCount_RNA > 0)

# Filter high ribosomal content to mitigate bias from cells with highest expression of ribosomal genes (top 10th percentile)
filter_ribosomal <- function(seurat_obj, method = "fixed", cutoff = 10) {
    rp_genes <- grep("^RP[SL]|^MRP[SL]", rownames(seurat_obj), value = TRUE)  # Ribosomal genes (RP/MRP prefixes)
    seurat_obj[["percent.ribo"]] <- PercentageFeatureSet(seurat_obj, features = rp_genes)  # % ribosomal expression (per cell)
    threshold <- if (method == "percentile") quantile(seurat_obj$percent.ribo, probs = cutoff) else cutoff  # Threshold (percentile or fixed)
    plot <- ggplot(seurat_obj@meta.data, aes(x = percent.ribo)) +  # Plot distribution (with threshold line)
        geom_histogram(bins = 100) +
        geom_vline(xintercept = threshold, color = "red", linetype = "dashed") +
        ggtitle("Distribution of Ribosomal Gene Percentage")
    print(plot)
    genes_before <- nrow(seurat_obj); cells_before <- ncol(seurat_obj)  # Pre-filter counts (genes, cells)
    seurat_obj_filtered <- subset(seurat_obj, subset = percent.ribo < threshold)  # Apply filter (below threshold)
    genes_after <- nrow(seurat_obj_filtered); cells_after <- ncol(seurat_obj_filtered)  # Post-filter counts
    cat("Threshold:", threshold, "\nCells before:", cells_before, "\nCells after:", cells_after, 
        "\nRemoved:", round((cells_before - cells_after) / cells_before * 100, 2), "%\n",
        "Genes before:", genes_before, "\nGenes after:", genes_after, "\n")
    return(seurat_obj_filtered)
}

# Apply ribosomal filter (90th percentile cutoff)
cat("Filtering prostate cancer samples (ribosomal)\n")
prostate_ca_seurat <- filter_ribosomal(prostate_ca_seurat, method = "percentile", cutoff = 0.90)
cat("Filtering non-cancerous samples (ribosomal)\n")
non_cancerous_seurat <- filter_ribosomal(non_cancerous_seurat, method = "percentile", cutoff = 0.90)

# Filter high mitochondrial content to mitigate dying cells
filter_mitochondrial <- function(seurat_obj, method = "fixed", cutoff = 10) {
    mt_genes <- grep("^MT-", rownames(seurat_obj), value = TRUE)  # Mitochondrial genes (MT- prefix)
    seurat_obj[["percent.mt"]] <- PercentageFeatureSet(seurat_obj, features = mt_genes)  # % mitochondrial expression (per cell)
    threshold <- if (method == "percentile") quantile(seurat_obj$percent.mt, probs = cutoff) else cutoff  # Threshold (percentile or fixed)
    plot <- ggplot(seurat_obj@meta.data, aes(x = percent.mt)) +  # Plot distribution (with threshold line)
        geom_histogram(bins = 100) +
        geom_vline(xintercept = threshold, color = "red", linetype = "dashed") +
        ggtitle("Distribution of Mitochondrial Gene Percentage")
    print(plot)
    genes_before <- nrow(seurat_obj); cells_before <- ncol(seurat_obj)  # Pre-filter counts (genes, cells)
    seurat_obj_filtered <- subset(seurat_obj, subset = percent.mt < threshold)  # Apply filter (below threshold)
    genes_after <- nrow(seurat_obj_filtered); cells_after <- ncol(seurat_obj_filtered)  # Post-filter counts
    cat("Threshold:", threshold, "\nCells before:", cells_before, "\nCells after:", cells_after, 
        "\nRemoved:", round((cells_before - cells_after) / cells_before * 100, 2), "%\n",
        "Genes before:", genes_before, "\nGenes after:", genes_after, "\n")
    return(seurat_obj_filtered)
}

# Apply mitochondrial filter (90th percentile cutoff)
cat("Filtering prostate cancer samples (mitochondrial)\n")
prostate_ca_seurat <- filter_mitochondrial(prostate_ca_seurat, method = "percentile", cutoff = 0.90)
cat("Filtering non-cancerous samples (mitochondrial)\n")
non_cancerous_seurat <- filter_mitochondrial(non_cancerous_seurat, method = "percentile", cutoff = 0.90)

# =========================================
# 2. Cell Cycle Regression
# =========================================
# Check pre-regression cell count (after the previous steps)
cat("Cells before cell cycle regression (prostate cancer):", ncol(prostate_ca_seurat), "\n")

# Default cell cycle genes (common S and G2M phase markers)
s_genes_default <- c("MCM5", "PCNA", "TYMS", "FEN1", "MCM2", "MCM4", "RRM1", "UNG", "GINS2", "MCM6", "CDCA7", "DTL", "PRIM1", "UHRF1", "MLF1IP", "HELLS", "RFC2", "RPA2", "NASP", "RAD51AP1", "GMNN", "WDR76", "SLBP", "CCNE2", "UBR7", "POLD3", "MSH2", "ATAD2", "RAD51", "RRM2", "CDC45", "CDC6", "EXO1", "TIPIN", "DSCC1", "BLM", "CASP8AP2", "USP1", "CLSPN", "POLA1", "CHAF1B", "BRIP1", "E2F8")
g2m_genes_default <- c("HMGB2", "CDK1", "NUSAP1", "UBE2C", "BIRC5", "TPX2", "TOP2A", "NDC80", "CKS2", "NUF2", "CKS1B", "MKI67", "TMPO", "CENPF", "TACC3", "FAM64A", "SMC4", "CCNB2", "CKAP2L", "CKAP2", "AURKB", "BUB1", "KIF11", "ANP32E", "TUBB4B", "GTSE1", "KIF20B", "HJURP", "CDCA3", "HN1", "CDC20", "TTK", "CDC25C", "KIF2C", "RANGAP1", "NCAPD2", "DLGAP5", "CDCA2", "CDCA8", "ECT2", "KIF23", "HMMR", "AURKA", "PSRC1", "ANLN", "LBR", "CKAP5", "CENPE", "CTCF", "NEK2", "G2E3", "GAS2L3", "CBX5", "CENPA")

# Score and regress out cell cycle to remove phase effects (prostate cancer)
prostate_ca_seurat <- CellCycleScoring(prostate_ca_seurat, s.features = s_genes_default, g2m.features = g2m_genes_default, set.ident = TRUE)
prostate_ca_seurat <- ScaleData(prostate_ca_seurat, vars.to.regress = c("S.Score", "G2M.Score"))

# Post-regression cell count to verify no cell loss
cat("Cells after cell cycle regression (prostate cancer):", ncol(prostate_ca_seurat), "\n") # 22796 (no changes)

# Score and regress out cell cycle (non-cancerous)
cat("Cells before cell cycle regression (non-cancerous):", ncol(non_cancerous_seurat), "\n")
non_cancerous_seurat <- CellCycleScoring(non_cancerous_seurat, s.features = s_genes_default, g2m.features = g2m_genes_default, set.ident = TRUE)
non_cancerous_seurat <- ScaleData(non_cancerous_seurat, vars.to.regress = c("S.Score", "G2M.Score"))

# =========================================
# 3. Doublets removal
# =========================================
# Set up parallel processing to speed up doublet detection
plan(multisession, workers = availableCores())

# Remove doublets using scDblFinder
remove_doublets <- function(seurat_obj) {
    sce_obj <- as.SingleCellExperiment(seurat_obj)  # Convert to SCE for scDblFinder
    samples <- seurat_obj@meta.data$orig.ident  # Batch info by sample IDs
    doublet_scores <- scDblFinder(sce_obj, samples = samples, k = 30, nfeatures = 2000)  # Run scDblFinder 
    batch_thresholds <- tapply(doublet_scores$scDblFinder.score, samples, function(x) quantile(x, probs = 0.95))  # Batch-specific thresholds (95th percentile)
    cat("Batch thresholds:\n"); print(batch_thresholds)
    doublet_cells <- colnames(sce_obj)[mapply(function(x, y) x > batch_thresholds[y], doublet_scores$scDblFinder.score, samples)]  # Identify doublets (above threshold)
    cat("Doublets:", length(doublet_cells), "\n")
    seurat_obj$doublet <- colnames(seurat_obj) %in% doublet_cells  # Mark doublets as TRUE/FALSE
    seurat_obj <- subset(seurat_obj, subset = doublet == FALSE)  # Remove doublets
    cat("Cells after removal:", ncol(seurat_obj), "\n")
    seurat_obj$filtered <- "filtered"  # Update metadata and flag filtered cells
    return(seurat_obj)
}

# Apply doublet removal to prostate ca and benign cases
cat("Processing prostate cancer samples\n")
prostate_ca_seurat <- remove_doublets(prostate_ca_seurat)
cat("Processing non-cancerous samples\n")
non_cancerous_seurat <- remove_doublets(non_cancerous_seurat)

# =========================================
# 4. Batch effects correction
# =========================================
# Disable parallel processing to avoid integration issues
plan(sequential)

# Correct batch effects by integrating across samples
correct_batch_effects <- function(seurat_obj) {
    cat("Metadata columns:\n"); print(colnames(seurat_obj@meta.data))
    cat("Unique orig.ident:\n"); print(unique(seurat_obj@meta.data$orig.ident))
    seurat_obj_before_integration <- FindVariableFeatures(seurat_obj, selection.method = "vst", nfeatures = 500)  # Pre-integration features (500 VST)
    seurat_obj_before_integration <- RunPCA(seurat_obj_before_integration, verbose = FALSE)  # PCA (dimensionality reduction)
    seurat_obj_before_integration <- RunUMAP(seurat_obj_before_integration, dims = 1:8, verbose = FALSE)  # UMAP (pre-integration visualization)
    plasma_colors <- viridis(n = length(unique(seurat_obj_before_integration$orig.ident)), option = "plasma")
    p1 <- DimPlot(seurat_obj_before_integration, group.by = "orig.ident", pt.size = 0.5, label = FALSE, repel = TRUE, cols = plasma_colors) + 
        ggtitle("UMAP Before Integration") + theme(legend.position = "right")  # Pre-integration UMAP (batch-colored)
    print(p1)
    sample_list <- SplitObject(seurat_obj_before_integration, split.by = "orig.ident")  # Split by batch (sample IDs)
    sample_list <- lapply(sample_list, function(x) {  # SCTransform and clean NAs (per sample)
        x <- SCTransform(x, verbose = FALSE, variable.features.n = 500, vst.flavor = "v2")
        x@meta.data <- x@meta.data[complete.cases(x@meta.data), ]
        x
    })
    anchors <- FindIntegrationAnchors(object.list = sample_list, dims = 1:5, verbose = FALSE)  # Find anchors (dims 1-5)
    seurat_obj_integrated <- IntegrateData(anchorset = anchors, dims = 1:5, verbose = FALSE)  # Integrate (batch-corrected)
    seurat_obj_integrated <- ScaleData(seurat_obj_integrated, verbose = FALSE)  # Scale (post-integration)
    seurat_obj_integrated <- RunPCA(seurat_obj_integrated, verbose = FALSE)  # PCA (integrated)
    seurat_obj_integrated <- FindNeighbors(seurat_obj_integrated, dims = 1:8)  # Neighbors (for clustering)
    seurat_obj_integrated <- FindClusters(seurat_obj_integrated, resolution = 0.5)  # Clusters (res 0.5)
    seurat_obj_integrated <- RunUMAP(seurat_obj_integrated, dims = 1:8, verbose = FALSE, umap.method = "uwot", metric = "cosine")  # UMAP (post-integration)
    plasma_colors <- viridis(n = length(unique(seurat_obj_integrated$orig.ident)), option = "plasma")
    p3 <- DimPlot(seurat_obj_integrated, group.by = "orig.ident", pt.size = 0.5, label = FALSE, repel = TRUE, cols = plasma_colors) + 
        ggtitle("UMAP After Integration") + theme(legend.position = "right")  # Post-integration UMAP (batch-colored)
    print(p3)
    p4 <- DimPlot(seurat_obj_integrated, group.by = "orig.ident", pt.size = 0.5, label = TRUE, repel = TRUE) + 
        ggtitle("UMAP After Integration") + theme(legend.position = "right")  # Labeled UMAP by batch IDs
    print(p4)
    return(seurat_obj_integrated)
}

# Remove parallelization limits to ensure stability
options(future.globals.maxSize = Inf)

# Apply batch correction for prostate ca and benign cases
cat("Processing prostate cancer samples (batch effects)\n")
prostate_ca_seurat_integrated <- correct_batch_effects(prostate_ca_seurat)
cat("Processing non-cancerous samples (batch effects)\n")
non_cancerous_seurat_integrated <- correct_batch_effects(non_cancerous_seurat)

# =========================================
# 5. UMAP Clusters
# =========================================
# Generate elbow plots to assess PCA dimensionality reduction
generate_elbow_plot <- function(seurat_obj_integrated, output_prefix) {
    seurat_obj_integrated <- RunPCA(seurat_obj_integrated, verbose = FALSE)  # PCA (dimensionality reduction)
    elbow_plot <- ElbowPlot(seurat_obj_integrated, ndims = 50) +  # Elbow plot
        labs(title = paste("Elbow Plot for", output_prefix), x = "Principal Components", y = "Standard Deviation") +
        theme(plot.title = element_text(size = 14, face = "bold"), axis.title.x = element_text(size = 12), 
              axis.title.y = element_text(size = 12), axis.text.x = element_text(size = 10), axis.text.y = element_text(size = 10))
    ggsave(paste0(output_prefix, "_ElbowPlot.pdf"), elbow_plot, width = 6.83, height = 6.41)
    print(elbow_plot)
}

# Generate elbow plots (prostate ca and benign)
generate_elbow_plot(prostate_ca_seurat_integrated, "prostate_ca")
generate_elbow_plot(non_cancerous_seurat_integrated, "non_cancerous")

# Downstream analyses with UMAP (clustering and markers)
downstream_analyses <- function(seurat_obj_integrated, gene_of_interest, output_prefix, dims = 15) {
    set.seed(10)
    DefaultAssay(seurat_obj_integrated) <- "RNA"
    seurat_obj_integrated <- FindVariableFeatures(seurat_obj_integrated, selection.method = "vst", nfeatures = 2000)  # Variable features (2000 VST)
    seurat_obj_integrated <- ScaleData(seurat_obj_integrated, verbose = FALSE)  # Scale (center and normalize)
    seurat_obj_integrated <- RunPCA(seurat_obj_integrated, verbose = FALSE)  # PCA (dims reduction)
    seurat_obj_integrated <- RunUMAP(seurat_obj_integrated, dims = 1:dims, verbose = FALSE)  # UMAP (dims 1-15)
    set.seed(11); seurat_obj_integrated <- FindNeighbors(seurat_obj_integrated, dims = 1:dims)  # Neighbors (kNN graph)
    set.seed(12); seurat_obj_integrated <- FindClusters(seurat_obj_integrated, resolution = 0.5)  # Clusters (Louvain, res 0.5)
    set.seed(13); cluster_markers <- FindAllMarkers(seurat_obj_integrated, only.pos = TRUE, min.pct = 0.1, logfc.threshold = 0.25)  # Marker genes (positive, logFC > 0.25)
    print(paste("Cluster markers:", nrow(cluster_markers)))
    if (nrow(cluster_markers) == 0) {
        print("Cluster levels:"); print(levels(Idents(seurat_obj_integrated)))
        print("Cells per cluster:"); print(table(Idents(seurat_obj_integrated)))
    }
    umap_data <- as.data.frame(Embeddings(seurat_obj_integrated, "umap")); umap_data$cluster_id <- Idents(seurat_obj_integrated)  # UMAP data (coords + clusters)
    umap_data_mean <- aggregate(. ~ cluster_id, data = umap_data, FUN = mean)  # Mean coords (per cluster)
    plasma_func <- colorRampPalette(viridis::viridis(100, direction = -1, option = "plasma")); portion <- 0.8  # Colors (plasma palette)
    n_colors <- round(length(unique(umap_data$cluster_id)) / portion); plasma_colors <- plasma_func(n_colors)
    set.seed(14); umap_plot_with_labels <- ggplot(umap_data, aes(x = umap_1, y = umap_2, color = as.factor(cluster_id))) +  # Labeled UMAP (cluster IDs)
        geom_point(size = 0.3, alpha = 0.5) + scale_color_manual(values = plasma_colors) +
        geom_text(data = umap_data_mean, aes(label = cluster_id, x = umap_1, y = umap_2), color = "black", size = 3, fontface = "bold", check_overlap = TRUE) +
        theme(panel.border = element_rect(fill = NA, color = "black"), panel.grid.major = element_blank(), panel.grid.minor = element_blank(),
              axis.text.x = element_text(color = "black"), axis.ticks.x = element_line(color = "black"), axis.text.y = element_text(color = "black"),
              axis.ticks.y = element_line(color = "black"), panel.background = element_rect(fill = "white")) +
        labs(title = "UMAP plot colored by cluster (with labels)", x = "umap_1", y = "umap_2", color = "Cluster") +
        guides(color = guide_legend(override.aes = list(size = 3)))
    print(umap_plot_with_labels)
    set.seed(15); umap_plot_no_labels <- ggplot(umap_data, aes(x = umap_1, y = umap_2, color = as.factor(cluster_id))) +  # Unlabeled UMAP (clusters only)
        geom_point(size = 0.3, alpha = 0.5) + scale_color_manual(values = plasma_colors) +
        theme(panel.border = element_rect(fill = NA, color = "black"), panel.grid.major = element_blank(), panel.grid.minor = element_blank(),
              axis.text.x = element_text(color = "black"), axis.ticks.x = element_line(color = "black"), axis.text.y = element_text(color = "black"),
              axis.ticks.y = element_line(color = "black"), panel.background = element_rect(fill = "white")) +
        labs(title = "UMAP plot colored by cluster (without labels)", x = "umap_1", y = "umap_2", color = "Cluster") +
        guides(color = guide_legend(override.aes = list(size = 3)))
    print(umap_plot_no_labels)
    if (gene_of_interest %in% rownames(seurat_obj_integrated)) {  # Gene expression UMAP (if gene exists)
        gene_colors_alpha <- c(scales::alpha("lightgray", 0.85), scales::alpha("lightpink", 0.85), scales::alpha("#FF6666", 0.85), 
                               scales::alpha("#BC2727", 0.85), scales::alpha("#660000", 0.85))
        set.seed(16); feature_plot <- FeaturePlot(seurat_obj_integrated, features = gene_of_interest, min.cutoff = 'q10', max.cutoff = 'q90',
                                                  pt.size = 0.2, cols = gene_colors_alpha) +
            theme(panel.border = element_rect(fill = NA, color = "black"), panel.grid.major = element_blank(), panel.grid.minor = element_blank(),
                  axis.text.x = element_text(color = "black"), axis.ticks.x = element_line(color = "black"), axis.text.y = element_text(color = "black"),
                  axis.ticks.y = element_line(color = "black"), panel.background = element_rect(fill = "white")) +
            labs(title = paste("UMAP plot colored by", gene_of_interest, "expression"), x = "umap_1", y = "umap_2")
        print(feature_plot)
    } else {
        cat(paste("Warning: Gene", gene_of_interest, "not found.\nAvailable genes:\n"))
        print(head(rownames(seurat_obj_integrated), 20))
    }
    if (nrow(cluster_markers) > 0) {  # Top markers (50 per cluster)
        top_markers <- cluster_markers %>% group_by(cluster) %>% top_n(n = 50, wt = avg_log2FC)
    } else {
        top_markers <- data.frame()
        warning("No cluster markers found.")
    }
    write.table(top_markers, file = paste0(output_prefix, "_top_markers_for_each_cluster_vRibo.tsv"), sep = "\t", col.names = TRUE, row.names = TRUE, quote = FALSE)  # Save markers (TSV)
    return(list(seurat_obj = seurat_obj_integrated, cluster_markers = cluster_markers, top_markers = top_markers))
}

# Apply downstream analyses (TRPM4 focus)
set.seed(42)
prostate_results <- downstream_analyses(prostate_ca_seurat_integrated, "TRPM4", "prostate_ca", dims = 15)
non_cancerous_results <- downstream_analyses(non_cancerous_seurat_integrated, "TRPM4", "non_cancerous", dims = 15)

# Save workspace
save.image(file = "GSE185344_Seurat_processed.RData")

# For subsequent analysis, load the saved file
load("GSE185344_Seurat_processed.RData")


#########################################################
# C. Subsequent GAM-REML-TPRS Analysis Code Availability
#########################################################
The subsequent analysis pipeline of this project is available on GitHub: https://github.com/kahkengwong/CEP-IP_Framework