Skip to contents

Find up-regulated marker genes for all clusters (quasi-NB GLM).

Source: R/getMarkers.R
getMarkers.Rd

Pseudo-bulks cells by cluster via spe2PB and, for each cluster, tests its genes against the average of all other clusters (1-vs-rest) using a quasi-likelihood approach. Because the full design is saturated (one pseudo-bulk per cluster, 0 residual df), each contrast is tested by fitting the corresponding null design (one fewer column, 1 residual df) with glmQLFit and using the quasi-deviance chi-square statistic. The test is one-sided: p-values reflect up-regulation in the cluster, so each table ranks that cluster's positive markers.

Usage

getMarkers(
  spe,
  cluster_name = "cluster",
  dispersion = 0.05,
  method = c("quasi", "lrt"),
  adjust.method = "BH",
  sort.by = "PValue",
  ...
)

Arguments

spe

A SpatialExperiment object.

cluster_name

Character. Column name in colData(spe) containing cluster labels. Default "cluster".

dispersion

Numeric. Fixed NB dispersion (BCV^2) used for the GLM fits. Required because the saturated full design (one pseudo-bulk per cluster) leaves no residual df for dispersion estimation. Default 0.05 (BCV \(\approx\) 0.224).

method

Character. Testing pipeline: "quasi" (default) fits each null design with glmQLFit and uses the quasi-deviance chi-square statistic; "lrt" uses the standard glmFit + glmLRT likelihood-ratio test. Both use the preset dispersion.

adjust.method

Character. Multiple-testing adjustment method passed to p.adjust (e.g. "BH", "BY", "holm", "bonferroni", "none"). Default "BH".

sort.by

Character. How to order each table: "PValue" (default), "logFC" (by absolute log fold change), "stat", or "none".

...

Additional arguments passed to glmQLFit for the null-model fits (used only when method = "quasi").

Value

A named list of data frames, one per cluster, each containing all genes. Columns:

logFC

log2 fold change of the cluster vs the average of the rest.

logCPM

average log2 counts-per-million across pseudo-bulk samples.

Prop

proportion of cells in the cluster with a non-zero count for the gene.

stat

signed quasi statistic (signed square root of the quasi chi-square statistic); positive means up-regulated in the cluster.

PValue

one-sided (up-regulation) p-value.

FDR

adjusted p-value (column named per adjust.method; "FDR" for BH/BY, "FWER" for family-wise methods). Absent when adjust.method = "none".

Details

Each per-cluster table holds all genes, sorted by sort.by with an adjusted p-value column added via adjust.method. Use topMarkers to extract and combine the top genes per cluster. For a direct comparison between two clusters (or two groups of clusters), see getDE.

Examples


data("xenium_bc_spe")
spe <- normalizeAssay(spe)
spe <- runPCA(spe)
#> Genes with 0 variance are excluded: ENSG00000135218 NegControlProbe_00002 NegControlCodeword_0504 NegControlCodeword_0509 NegControlCodeword_0510 NegControlCodeword_0511 NegControlCodeword_0512 NegControlCodeword_0516 NegControlCodeword_0517 NegControlCodeword_0518 NegControlCodeword_0519 NegControlCodeword_0520 NegControlCodeword_0522 NegControlCodeword_0526 NegControlCodeword_0527 NegControlCodeword_0530 NegControlCodeword_0536 NegControlCodeword_0537 BLANK_0030 BLANK_0163 BLANK_0165 BLANK_0212 BLANK_0221 BLANK_0230 BLANK_0237 BLANK_0311 BLANK_0361 BLANK_0365 BLANK_0382 BLANK_0384 BLANK_0387 BLANK_0388 BLANK_0391 BLANK_0393 BLANK_0397 BLANK_0399 BLANK_0404 BLANK_0406 BLANK_0410 BLANK_0411 BLANK_0418 BLANK_0425 BLANK_0432 BLANK_0447
spe <- findNbrsSNN(spe, dimred = "PCA")
#> [1] "Getting K-nearest neighbour"
#> [1] "Getting shared nearest neighbour"
spe <- getClusters(spe, resolution = 0.5)
#> Reassigned all 2 cells from clusters with <= 1 cells (0 by graph, 2 by distance).
markers <- getMarkers(spe)