Module Score

Purpose

Module scoring summarizes the expression of a gene set in each cell.

Typical questions:

• which cells show high interferon response?
• which clusters have high exhaustion signature?
• does a pathway-like signature differ between conditions?
• where is a custom marker signature enriched on UMAP?

In Seurat, module scoring is usually done with AddModuleScore().

Signature scoring is related to functional enrichment, but the output is different.

Analysis	Input	Output	Typical question
Signature scoring	expression matrix + predefined gene set	score per cell or sample	which cells have high activity for this signature?
Functional enrichment	marker genes, DE genes, or ranked gene list	enriched pathways or GO terms	what functions are over-represented in this gene list?

Methods such as AddModuleScore(), AUCell, and UCell assign scores to cells. GO, KEGG, Reactome, and GSEA-style analyses summarize gene lists or ranked gene lists.

Scoring Methods

Common single-cell signature scoring methods:

Method	Main idea	Notes
AddModuleScore()	average expression of a gene set compared with control genes	common in Seurat workflows
AUCell	AUC of gene-set genes in each cell’s expression ranking	rank-based, often used for regulon or signature activity
UCell	Mann-Whitney U statistic from per-cell gene rankings	rank-based and robust for sparse data
singscore	rank-based single-sample scoring	general signature scoring method
GSVA / ssGSEA	sample-wise gene set variation scoring	can be used on cell or pseudobulk-like matrices, but may be heavier

AddModuleScore() is a practical default in Seurat. Rank-based methods such as AUCell and UCell are useful when relative within-cell rankings are preferred over expression averages.

Input Gene Sets

AddModuleScore() expects a list of gene vectors.

One score:

ifn_genes <- list(
  c("ISG15", "IFIT1", "IFIT3", "MX1", "OAS1")
)

Multiple scores:

signature_genes <- list(
  IFN_response = c("ISG15", "IFIT1", "IFIT3", "MX1", "OAS1"),
  Cytotoxicity = c("NKG7", "GNLY", "GZMB", "PRF1")
)

Gene sets can come from gene-sets.qmd, marker signatures, pathway databases, or curated literature.

Run AddModuleScore

Basic example:

seu <- Seurat::AddModuleScore(
  object = seu,
  features = ifn_genes,
  assay = "RNA",
  name = "IFN_score",
  seed = 42
)

AddModuleScore() writes the result to metadata.

The output column usually has a number suffix:

colnames(seu@meta.data)

For the example above, the score column is usually:

seu$IFN_score1

The suffix exists because features is a list. If multiple gene sets are provided, Seurat creates one score column per list element.

Check Genes

Before scoring, check which genes are present in the object:

genes <- ifn_genes[[1]]

genes_found <- intersect(
  x = genes,
  y = rownames(seu)
)

genes_missing <- setdiff(
  x = genes,
  y = rownames(seu)
)

genes_found
genes_missing

If many genes are missing, the score may not be meaningful.

Visualize Score

Feature plot:

Seurat::FeaturePlot(
  object = seu,
  features = "IFN_score1",
  reduction = "umap",
  pt.size = 0.5,
  alpha = 0.7
)

Violin plot by cluster:

Seurat::VlnPlot(
  object = seu,
  features = "IFN_score1",
  group.by = "seurat_clusters",
  pt.size = 0
)

Violin plot by cell type:

Seurat::VlnPlot(
  object = seu,
  features = "IFN_score1",
  group.by = "cell_type",
  pt.size = 0
)

Compare By Condition

Compare score distributions by condition:

Seurat::VlnPlot(
  object = seu,
  features = "IFN_score1",
  group.by = "condition",
  pt.size = 0
)

Compare within one cell type:

t_cells <- subset(
  x = seu,
  subset = cell_type == "T cell"
)

Seurat::VlnPlot(
  object = t_cells,
  features = "IFN_score1",
  group.by = "condition",
  pt.size = 0
)

For formal condition comparison, use sample-level summaries rather than treating every cell as an independent biological replicate.

Summarize Scores

Summarize by cluster:

score_by_cluster <- seu@meta.data |>
  dplyr::group_by(seurat_clusters) |>
  dplyr::summarise(
    mean_score = mean(IFN_score1, na.rm = TRUE),
    median_score = median(IFN_score1, na.rm = TRUE),
    .groups = "drop"
  )

score_by_cluster

Summarize by sample and cell type:

score_by_sample_celltype <- seu@meta.data |>
  dplyr::group_by(sample, condition, cell_type) |>
  dplyr::summarise(
    mean_score = mean(IFN_score1, na.rm = TRUE),
    median_score = median(IFN_score1, na.rm = TRUE),
    n_cells = dplyr::n(),
    .groups = "drop"
  )

score_by_sample_celltype

This gives one row per sample-cell type combination.

AUCell

AUCell is another common method for single-cell signature scoring.

It scores whether genes from a gene set are enriched near the top of each cell’s expression ranking.

This makes AUCell a per-cell scoring method, even though it uses an enrichment-like idea internally.

Prepare Input

Use an expression matrix with genes as rows and cells as columns:

expr_mat <- Seurat::GetAssayData(
  object = seu,
  assay = "RNA",
  layer = "data"
)

Prepare gene sets:

gene_sets <- list(
  IFN_response = c("ISG15", "IFIT1", "IFIT3", "MX1", "OAS1"),
  Cytotoxicity = c("NKG7", "GNLY", "GZMB", "PRF1")
)

Check genes:

lapply(
  X = gene_sets,
  FUN = function(genes) {
    intersect(genes, rownames(expr_mat))
  }
)

Build Rankings

Build per-cell gene rankings:

rankings <- AUCell::AUCell_buildRankings(
  exprMat = expr_mat,
  nCores = 1,
  plotStats = FALSE
)

Each cell gets a ranking of genes based on expression.

Calculate AUC Scores

Calculate AUC scores:

auc <- AUCell::AUCell_calcAUC(
  geneSets = gene_sets,
  rankings = rankings
)

Extract the score matrix:

auc_mat <- SummarizedExperiment::assay(auc)

dim(auc_mat)
rownames(auc_mat)

Rows are gene sets and columns are cells.

Add AUCell Scores To Seurat

Add one AUCell score to metadata:

seu$IFN_AUCell <- as.numeric(
  auc_mat["IFN_response", colnames(seu)]
)

Add all AUCell scores:

auc_df <- as.data.frame(t(auc_mat))

colnames(auc_df) <- paste0(
  colnames(auc_df),
  "_AUCell"
)

seu <- Seurat::AddMetaData(
  object = seu,
  metadata = auc_df
)

Visualize AUCell score:

Seurat::FeaturePlot(
  object = seu,
  features = "IFN_response_AUCell",
  reduction = "umap",
  pt.size = 0.5,
  alpha = 0.7
)

AUCell is useful when the relative rank of genes within each cell is more important than the average expression level.

Notes

Module scores are relative scores, not direct pathway activity measurements.

Interpret scores carefully:

• gene sets should be biologically coherent
• very small gene sets can be unstable
• very broad gene sets can lose specificity
• missing genes weaken the score
• condition comparisons should respect sample-level replication

Use module scores as evidence alongside markers, annotation, metadata, and biological context.