Flag low-quality cells

Usage

flagLowQuality(
  metrics,
  method = c("threshold", "gmm", "hdbscan"),
  score_cutoff = 2,
  auto_k = 3,
  obj = NULL,
  celltype_col = NULL,
  sample_col = NULL,
  rescue_mode = c("moderate", "lenient", "strict", "none"),
  cancer_bypass = FALSE,
  min_cluster_size = NULL,
  doublet_action = c("remove", "borderline", "none"),
  doublet_col = "is_doublet",
  qc_strength = c("auto", "default", "lenient", "strict"),
  save_dir = NULL
)

Arguments

metrics: data.frame of QC metrics (rows=cells). If it contains a logical column 'is_doublet' (as added by calcQCmetrics), it will be honored.
method: "threshold", "gmm", or "hdbscan"
score_cutoff: integer for threshold mode (default 2)
auto_k: kept for API compat (unused here)
obj: optional Seurat object for rescue step
celltype_col: optional meta column for rescue label coherence
sample_col: optional sample column (passed to detector; may be NULL)
rescue_mode: "moderate","lenient","strict", or "none"
cancer_bypass: Logical. If TRUE, clusters with healthy splicing profiles but high removal rates are exempt from the removal-fraction penalty, allowing potential cancer cell populations to be rescued. Default FALSE.
min_cluster_size: minimum cluster size to consider for rescue; NULL = adaptive (~0.5% or >=10)
doublet_action: what to do with predicted doublets in metrics$is_doublet: "remove" (default), "borderline", or "none"
doublet_col: name of the doublet flag column in metrics (default "is_doublet")
qc_strength: One of c("auto","default","lenient","strict")
save_dir: Character path or NULL. If non-NULL, write intermediate CSVs here.

Value

data.frame with rownames=cells and columns: qc_status, qc_score, cluster, reason,