scdown

scdown is a minimal downstream package for annotated single-cell RNA-seq.

It assumes that QC, normalization, clustering, and annotation are already done. The package is intentionally narrow and focuses on five practical jobs inside one dataset:

• cluster and annotation maps
• marker discovery
• annotation checks with known marker panels
• cluster-level expression and signature summaries
• exploratory ligand-receptor interaction ranking

Website: https://dai540.github.io/scdown/

Installation

Install from GitHub with pak:

install.packages("pak")
pak::pak("dai540/scdown")

or remotes:

install.packages("remotes")
remotes::install_github("dai540/scdown")

Or install from a source tarball:

install.packages("path/to/scdown_<version>.tar.gz", repos = NULL, type = "source")

Then load the package:

library(scdown)

Input contract

scdown() accepts:

• a long-format table with cell, gene, value, and annotation columns
• a list with expr, cells, and optional embedding
• a SingleCellExperiment
• a Seurat object

The minimal metadata columns are:

• cell
• annotation
• sample if available

Minimal example

library(scdown)

obj <- scdown(
  scdown_example(),
  annotation_col = "annotation",
  sample_col = "sample"
)

res <- run_core(
  obj,
  signatures = c("cytotoxic", "antigen_presentation")
)

build_scdown_report(obj)

Core functions

• scdown()
• summarize_dataset()
• plot_map()
• find_markers()
• check_annotation()
• plot_average_expression()
• plot_cluster_similarity()
• plot_signature()
• infer_communication()
• run_core()
• build_scdown_report()

Built-in resources

• immune_signatures()
• immune_marker_panels()
• list_signatures()
• get_signature()

Main outputs

Typical outputs include:

• cluster map PNGs
• marker tables
• annotation check tables and heatmaps
• average-expression summaries
• cluster-similarity heatmaps
• signature score tables
• communication ranking tables
• one compact HTML report

What scdown does not do

scdown does not try to replace upstream or specialist tools.

• It does not do QC, ambient RNA correction, doublet detection, normalization, integration, or clustering.
• It does not do group-aware differential abundance or differential expression.
• It does not claim publication-grade communication inference.

Use Seurat, scater, scuttle, scran, SingleR, muscat, dreamlet, miloR, or CellChat for those stages.

Documentation

• Home: https://dai540.github.io/scdown/
• Getting Started: https://dai540.github.io/scdown/articles/getting-started.html
• Reference: https://dai540.github.io/scdown/reference/index.html
• News: https://dai540.github.io/scdown/news/index.html

Citation

citation("scdown")

Package layout

• R/object.R: input standardization
• R/downstream.R: minimal downstream workflow
• R/resources.R: built-in signatures, marker panels, and ligand-receptor pairs
• R/report.R: compact HTML reporting
• vignettes/: getting-started tutorial