README

SlimR is an R package designed for annotating single-cell and spatial-transcriptomics (ST) datasets. It supports the creation of a unified marker list, Markers_list, using sources including: the package’s built-in curated species-specific cell type and marker reference databases (e.g., ‘Cellmarker2’, ‘PanglaoDB’, ‘scIBD’, ‘TCellSI’), Seurat objects containing cell label information, or user-provided Excel tables mapping cell types to markers.

SlimR can predict calculate parameters by machine learning algorithms (e.g., ‘Random Forest’, ‘Gradient Boosting’, ‘Support Vector Machine’, ‘Ensemble Learning’), and based on Markers_list, calculate gene expression of different cell types and predict annotation information and calculate corresponding AUC by Celltype_Calculate(), and annotate it by Celltype_Annotation(), then verify it by Celltype_Verification(). At the same time, it can calculate gene expression corresponding to the cell type to generate a reference map for manual annotation (e.g., ‘Heat Map’, ‘Feature Plots’, ‘Combined Plots’).

1. Preparation

1.1 Installation

Install SlimR directly from CRAN using: (Stable version, recommended when the version equivalent to GitHub package version)

install.packages("SlimR")

Note: Try adjusting the CRAN image to Global (CDN) or use BiocManager::install("SlimR") if you encounter a version mismatch during installation.

Install SlimR directly from GitHub using: (Development version, recommended when the version is higher than CRAN package version)

devtools::install_github("Zhaoqing-wang/SlimR")

Note: If the function doesn’t work, please run install.packages('devtools') first.

1.2 Loading SlimR

library(SlimR)

1.3 Prepare Seurat Object

For Seurat objects with multiple layers in the assay, please run SeuratObject::JoinLayers() first.

# For example, if you want to use the 'RNA' layer in the multilayered Seurat object assay.
sce@assays$RNA <- SeuratObject::JoinLayers(sce@assays$RNA)

Important: To ensure accuracy of the annotation, make sure that the entered Seurat object has run the standard process and removed batch effects.

Note: It is recommended to use the clustree package to determine the appropriate resolution for the input Seurat object.

1.4 Dependencies (Alternative)

SlimR requires R (≥ 3.5) and depends on the following packages: cowplot, dplyr, ggplot2, patchwork, pheatmap, readxl, scales, Seurat, tidyr, tools. If installation fails, please install missing dependencies using:

# Install dependencies if needed:
install.packages(c("cowplot", "dplyr", "ggplot2", "patchwork", 
                   "pheatmap", "readxl", "scales", "Seurat", 
                   "tidyr", "tools"))

2. Standardized Markers_list Input

SlimR requires a standardized list format for storing marker information, metrics (can be omitted), and corresponding cell types (list names = cell types (necessary), first column = markers (necessary), subsequent columns = metrics (can be omitted)).

2.1 From Cellmarker2 Database

Cellmarkers2: A database of cell types and markers covering different species and tissue types.

2.1.1 Load Database:

Cellmarker2 <- SlimR::Cellmarker2

2.1.2 Optional Metadata Exploration:

Cellmarker2_table <- SlimR::Cellmarker2_table
View(Cellmarker2_table)

2.1.3 Generate Markers_list:

Markers_list_Cellmarker2 <- Markers_filter_Cellmarker2(
  Cellmarker2,
  species = "Human",
  tissue_class = "Intestine",
  tissue_type = NULL,
  cancer_type = NULL,
  cell_type = NULL
)

Important: Select at least the species and tissue_class parameters to ensure the accuracy of the annotation.

2.2 From PanglaoDB Database

PanglaoDB: Database of cell types and markers covering different species and tissue types.

2.2.1 Load Database:

PanglaoDB <- SlimR::PanglaoDB

2.2.2 Optional Metadata Exploration:

PanglaoDB_table <- SlimR::PanglaoDB_table
View(PanglaoDB_table)

2.2.3 Generate Markers_list:

Markers_list_panglaoDB <- Markers_filter_PanglaoDB(
  PanglaoDB,
  species_input = 'Human',
  organ_input = 'GI tract'
)

Important: Select the species_input and organ_input parameters to ensure the accuracy of the annotation.

2.3 From scIBD Database

Markers_list_scIBD <- SlimR::Markers_list_scIBD

Important: This is for human intestinal annotation only. The input Seurat object was ensured to be a human intestinal type to ensure the accuracy of the labeling.

Note: The Markers_list_scIBD was generated using section 2.5.2 and the parameters sort_by = "logFC" and gene_filter = 20 were set.

2.4 From TCellSI Database

Markers_list_TCellSI <- SlimR::Markers_list_TCellSI

Important: This is only for T cell subset annotation. Ensure that the input Seurat object is of T cell type to guarantee the accuracy of the annotation.

2.5 From Seurat Objects

2.5.1 Identify Markers and Generate Markers_list:

The standard Markers_list can be generated by the built-in read_seurat_markers() function after obtaining Markers through the Seurat::FindAllMarkers() function.

seurat_markers <- Seurat::FindAllMarkers(
    object = sce,
    group.by = "Cell_type",
    only.pos = TRUE)

Markers_list_Seurat <- Read_seurat_markers(seurat_markers,
    sources = "Seurat",
    sort_by = "FSS",
    gene_filter = 20
    )

Note: Recommend use the parameter sort_by = "FSS" to use the ‘Feature Significance Score’ (FSS, product value of log2FC and Expression ratio) or use the parameter sort_by = "avg_log2FC" as the ranking basis.

2.5.2 Use presto to Speed Up: (Alternative)

For large data sets, the presto::wilcoxauc() function can be used to speed up the operation. (Alternative, ~10x faster, sacrifice partial accuracy)

seurat_markers <- dplyr::filter(
    presto::wilcoxauc(
      X = sce,
      group_by = "Cell_type",
      seurat_assay = "RNA"
      ),
    padj < 0.05, logFC > 0.5
    )

Markers_list_Seurat <- Read_seurat_markers(seurat_markers,
    sources = "presto",
    sort_by = "FSS",
    gene_filter = 20
    )

Improtant: This feature depends on the presto packages, please run devtools::install_github('immunogenomics/presto') first.

Note: Recommend use the parameter sort_by = "logFC" or use the parameter sort_by = "FSS" to use the ‘Feature Significance Score’ (FSS, product value of log2FC and Expression ratio) as the ranking basis.

2.6 From Excel Tables

Markers_list_Excel <- Read_excel_markers("D:/Laboratory/Marker_load.xlsx")

3. Automated Annotation Workflow

3.1 Calculate Parameter (Alternative)

SlimR integrates multiple machine learning algorithms (e.g., Random Forest, Gradient Boosting, Support Vector Machine, Ensemble Learning) to automatically determine optimal min_expression and specificity_weight parameters in section 3.2 for cell types probability calculate.

# Basic usage uses default genes
SlimR_params <- Parameter_Calculate(
   seurat_obj = sce,
   features = c("CD3E", "CD4", "CD8A"),
   assay = "RNA",
   cluster_col = "seurat_clusters",
   method = "ensemble",
   n_models = 3,
   return_model = FALSE,
   verbose = TRUE
   )
 
 # Use with custom method: use the genes corresponding to a specific cell type in 'Markers_list' as input
SlimR_params <- Parameter_Calculate(
   seurat_obj = sce,
   features = unique(Markers_list_Cellmarker2$`B cell`$marker),
   assay = "RNA",
   cluster_col = "seurat_clusters",
   method = "rf",
   return_model = FALSE,
   verbose = TRUE
   )

Important: This scheme is optional and can be skipped to section 3.2 for cell type probability calculation using default parameters.

Note: Using the parameter method = "rf" in the function Parameter_Calculate () can modify the machine learning model used.Machine learning method: rf (Random Forest), gbm (Gradient Boosting), svm (Support Vector Machine), or ensemble (Ensemble Learning; default)

3.2 Calculate Cell Types

3.2.1 Calculate Cell Types (Core)

Uses markers_list to calculate probability, prediction results, calculate corresponding AUC (optional) and generate heat map and ROC graphs (optional) for cell annotation.

SlimR_anno_result <- Celltype_Calculate(seurat_obj = sce,
    gene_list = Markers_list,
    species = "Human",
    cluster_col = "seurat_clusters",
    assay = "RNA",
    min_expression = 0.1,
    specificity_weight = 3,
    threshold = 0.8,
    compute_AUC = TRUE,
    plot_AUC = TRUE,
    AUC_correction = TRUE,
    colour_low = "navy",
    colour_high = "firebrick3"
    )

You can use the min_expression = SlimR_params$min_expression and specificity_weight = SlimR_params$specificity_weight parameter in function Celltype_Calculate() if you have run the Parameter_Calculate () function in section 3.1 above.

Important: The parameter cluster_col in the function Celltype_Calculate() and the function Celltype_Annotation() must be strictly the same to avoid false matches.

Note: Using the parameter AUC_correction = TRUE takes a little longer to compute (~20% longer than only setting parameter plot_AUC = TRUE; ~40% longer than only setting parameter compute_AUC = TRUE), but it is recommended to correct the predicted cell type this way in order to obtain more accurate cell type prediction results. The lower the parameter threshold, the more alternative cell types will be checked by AUC, and the longer the run time will be.

3.2.2 Plot Heat Map (Optional)

Check the annotation probability of the cell type to be annotated in the input cluster_col column and cell types in Markers_list with the following code.

print(SlimR_anno_result$Heatmap_plot)

Note: If the heat map is not generated properly, please run the function library(pheatmap) first.

3.2.3 View Prediction Results (Optional)

Cell type information results predicted by SlimR can be viewed with the following code.

View(SlimR_anno_result$Prediction_results)

3.2.4 Plot ROC Curve and AUC Value (Optional)

Furthermore, the ROC curve and AUC value of the corresponding cluster_col and predicted cell types can be viewed by the following code.

print(SlimR_anno_result$AUC_plot)

Note: If the heat map is not generated properly, please run the function library(ggplot2) first.

3.2.5 Correction for Predicted Cell Types (Alternative)

After viewing the list of predicted cell types and the corresponding AUC values, the predicted cell types can be corrected with the following code.

# For example, cluster '15' in 'cluster_col' corresponds to cell type 'Intestinal stem cell'.
SlimR_anno_result$Prediction_results$Predicted_cell_type[
  SlimR_anno_result$Prediction_results$cluster_col == 15
] <- "Intestinal stem cell"

# For example, a predicted cell type with an AUC of 0.5 or less should be labeled 'Unknown'.
SlimR_anno_result$Prediction_results$Predicted_cell_type[
  SlimR_anno_result$Prediction_results$AUC <= 0.5
] <- "Unknown"

After modifying the corresponding predicted cell type, the following code is used to view the updated predicted cell type table.

View(SlimR_anno_result$Prediction_results)

Improtant: It is strongly recommended that if you need to correct the cell type, use cell types in SlimR_anno_result$Prediction_results$Alternative_cell_type.

3.3 Annotate Cell Types

Assigns SlimR predicted cell types information in SlimR_anno_result$Prediction_results$Predicted_cell_type to the Seurat object based on cluster annotations, and stores the results into seurat_obj@meta.data$annotation_col.

sce <- Celltype_Annotation(seurat_obj = sce,
    cluster_col = "seurat_clusters",
    SlimR_anno_result = SlimR_anno_result,
    plot_UMAP = TRUE,
    annotation_col = "Cell_type_SlimR"
    )

Important: The parameter cluster_col in the function Celltype_Calculate() and the function Celltype_Annotation() must be strictly the same to avoid false matches. And the parameter annotation_col in the function Celltype_Annotation() and the function Celltype_Verification() must be strictly the same to avoid false matches.

3.4 Verify Cell Types

Use the cell group identity information in seurat_obj@meta.data$annotation_col and use the ‘Feature Significance Score’ (FSS, product value of log2FC and Expression ratio) as the ranking basis.

Celltype_Verification(seurat_obj = sce,
    SlimR_anno_result = SlimR_anno_result,
    gene_number = 5,
    assay = "RNA",
    colour_low = "white",
    colour_high = "navy",
    annotation_col = "Cell_type_SlimR"
    )

Important: The parameter annotation_col in the function Celltype_Annotation() and the function Celltype_Verification() must be strictly the same to avoid false matches.

Note: Cell types located in SlimR_anno_result$Prediction_results were verified using the markers information from SlimR_anno_result$Expression_list; cell types that are not in the above list are validated using the markers information from the function FindMarkers().

4. Semi-Automated Annotation Workflow

4.1 Annotation Heat Map

Generate a heat map to estimate the likelihood that various cell clusters exhibited similarity to control cell types:

Celltype_Annotation_Heatmap(
  seurat_obj = sce,
  gene_list = Markers_list,
  species = "Human",
  cluster_col = "seurat_cluster",
  min_expression = 0.1,
  specificity_weight = 3,
  colour_low = "navy",
  colour_high = "firebrick3"
)

Note: Now this function has been incorporated into Celltype_Calculate(), and it is recommended to use Celltype_Calculate() instead.

4.2 Annotation Feature Plots

Generates per-cell-type expression dot plot with metric heat map (when the metric information exists):

Celltype_Annotation_Features(
  seurat_obj = sce,
  gene_list = Markers_list,
  gene_list_type = "Cellmarker2",
  species = "Human",
  save_path = "./SlimR/Celltype_Annotation_Features/",
  colour_low = "white",
  colour_high = "navy",
  colour_low_mertic = "white",
  colour_high_mertic = "navy"
  )

Each resulting combined image consists of a dot plot above and a heat map below (if metric information present). Dot plot show the expression level and expression ratio relationship between the cell type and corresponding markers. Below it, there is a metric heat map for the corresponding markers (if the metric information exists).

4.3 Annotation Combined Plots

Celltype_Annotation_Combined(
  seurat_obj = sce,
  gene_list = Markers_list, 
  species = "Human",
  cluster_col = "seurat_cluster",
  assay = "RNA",
  save_path = "./SlimR/Celltype_Annotation_Combined/",
  colour_low = "white",
  colour_high = "navy"
)

Each generated combined plot shows the box plot of the expression levels of the corresponding markers for that cell type, with the colors corresponding to the average expression levels of the markers.

5. Other Functions Provided by SlimR

Functions in section 5.1, 5.2, 5.3 and 5.4 has been incorporated into Celltype_Annotation_Features(), and it is recommended to use Celltype_Annotation_Features() and set corresponding parameters (for example, gene_list_type = "Cellmarker2") instead. For more information, please refer to section 4.2.

5.1 Annotation Feature Plots with Cellmarker2 Database

Celltype_annotation_Cellmarker2(
  seurat_obj = sce,
  gene_list = Markers_list_Cellmarker2,
  species = "Human",
  cluster_col = "seurat_cluster",
  assay = "RNA",
  save_path = "./SlimR/Celltype_annotation_Cellmarkers2/",
  colour_low = "white",
  colour_high = "navy",
  colour_low_mertic = "white",
  colour_high_mertic = "navy"
)

Note: To call this function, set the parameter gene_list_type = "Cellmarker2" in the function Celltype_Annotation_Features().

5.2 Annotation Feature Plots with PanglaoDB Database

Celltype_annotation_PanglaoDB(
  seurat_obj = sce,
  gene_list = Markers_list_panglaoDB,
  species = "Human",
  cluster_col = "seurat_cluster",
  assay = "RNA",
  save_path = "./SlimR/Celltype_annotation_PanglaoDB/",
  colour_low = "white",
  colour_high = "navy",
  colour_low_mertic = "white",
  colour_high_mertic = "navy"
)

Note: To call this function, set the parameter gene_list_type = "PanglaoDB" in the function Celltype_Annotation_Features().

5.3 Annotation Feature Plots with Seurat-Based Markers List

Celltype_annotation_Seurat(
  seurat_obj = sce,
  gene_list = Markers_list_Seurat,
  species = "Human",
  cluster_col = "seurat_cluster",
  assay = "RNA",
  save_path = "./SlimR/Celltype_annotation_Seurat/",
  colour_low = "white",
  colour_high = "navy",
  colour_low_mertic = "white",
  colour_high_mertic = "navy"
)

Note: To call this function, set the parameter gene_list_type = "Seurat" in the function Celltype_Annotation_Features().

5.4 Annotation Feature Plots with Excel-Based Markers List

Celltype_annotation_Excel(
  seurat_obj = sce,
  gene_list = Markers_list_Excel,
  species = "Human",
  cluster_col = "seurat_cluster",
  assay = "RNA",
  save_path = "./SlimR/Celltype_annotation_Excel/",
  colour_low = "white",
  colour_high = "navy",
  colour_low_mertic = "white",
  colour_high_mertic = "navy"
)

Note: To call this function, set the parameter gene_list_type = "Excel" in the function Celltype_Annotation_Features. This function also works with Markers_list without metric information or with metric information generated in other ways.

6. Conclusion

Thank you for using SlimR. For questions, issues, or suggestions, please submit them in the issue section or discussion section on GitHub (suggested) or send an email (alternative):

SlimR: Machine Learning-Assisted, Marker-Based Tool for Single-Cell and Spatial Transcriptomics Annotation

Overview

Table of Contents