Interpret cluster markers for cell-type annotation

Usage

sn_interpret_annotation(
  object,
  de_name,
  cluster_col = "seurat_clusters",
  n_markers = 10,
  background = NULL,
  output_format = c("llm", "human"),
  provider = NULL,
  model = NULL,
  return_prompt = FALSE,
  store_name = "default",
  return_object = TRUE,
  ...
)

Arguments

object: A Seurat object.
de_name: Name of a stored marker result.
cluster_col: Metadata column containing cluster labels.
n_markers: Number of top markers per cluster.
background: Optional study-specific background information to provide additional interpretation context.
output_format: One of "llm" for a model-ready prompt bundle or "human" for a human-readable summary.
provider: Optional model provider function.
model: Optional model identifier.
return_prompt: If TRUE, return the prompt bundle without calling the provider.
store_name: Name used under object@misc$interpretation_results.
return_object: If TRUE, return the updated Seurat object.
...: Additional arguments forwarded to provider.

Value

A prompt bundle, response, or updated Seurat object.

Examples

if (requireNamespace("Seurat", quietly = TRUE)) {
  counts <- matrix(rpois(20 * 24, lambda = 1), nrow = 20, ncol = 24)
  rownames(counts) <- c(
    paste0("GENE", 1:14),
    "CD3D", "CD3E", "TRAC", "MS4A1", "CD79A", "HLA-DRA"
  )
  colnames(counts) <- paste0("cell", 1:24)
  counts[c("CD3D", "CD3E", "TRAC"), 1:12] <-
    counts[c("CD3D", "CD3E", "TRAC"), 1:12] + 20
  counts[c("MS4A1", "CD79A", "HLA-DRA"), 13:24] <-
    counts[c("MS4A1", "CD79A", "HLA-DRA"), 13:24] + 20
  obj <- sn_initialize_seurat_object(counts, species = "human")
  obj$cell_type <- rep(c("Tcell", "Bcell"), each = 12)
  Seurat::Idents(obj) <- obj$cell_type
  obj <- Seurat::NormalizeData(obj, verbose = FALSE)
  obj <- sn_find_de(obj, analysis = "markers", group_by = "cell_type",
    layer = "data", min_pct = 0, logfc_threshold = 0,
    store_name = "celltype_markers", return_object = TRUE, verbose = FALSE
  )
  prompt <- sn_interpret_annotation(
    obj,
    de_name = "celltype_markers",
    cluster_col = "cell_type",
    return_prompt = TRUE
  )
  prompt$task
}
#> INFO [2026-03-26 18:52:22] Initializing Seurat object for project: Shennong.
#> INFO [2026-03-26 18:52:22] Running QC metrics for human.
#> INFO [2026-03-26 18:52:23] Seurat object initialization complete.
#> [1] "annotation"