Cell state activity inference from single cell data — SecAct.activity.inference.scRNAseq • SecAct

Calculate secreted protein signaling activity of cell states from single cell RNA-Sequencing data.

Usage

SecAct.activity.inference.scRNAseq(
  inputProfile,
  cellType_meta,
  is.singleCellLevel = FALSE,
  sigMatrix = "SecAct",
  is.filter.sig = FALSE,
  is.group.sig = TRUE,
  is.group.cor = 0.9,
  lambda = 5e+05,
  nrand = 1000,
  ncores = 1L,
  backend = "auto",
  rng_method = "mt19937",
  batch_size = NULL
)

Arguments

inputProfile: A Seurat object.
cellType_meta: Column name in meta data that includes cell-type annotations.
is.singleCellLevel: A logical flag indicating whether to calculate for each single cell (Default: FALSE).
sigMatrix: Secreted protein signature matrix. Could be "SecAct", "CytoSig", "SecAct-Breast", "SecAct-Colorectal", "SecAct-Glioblastoma", "SecAct-Kidney", "SecAct-Liver", "SecAct-Lung-Adeno", "SecAct-Ovarian", "SecAct-Pancreatic", "SecAct-Prostate". SecAct signatures were derived from all cancer ST samples; SecAct-XXX signatures were derived from XXX cancer ST samples.
is.filter.sig: A logical flag indicating whether to filter the secreted protein signatures based on the genes from inputProfile (Default: FALSE). Because some sequencing platforms (e.g., CosMx) cover only a subset of secreted proteins, setting this option to TRUE restricts the activity inference on those proteins.
is.group.sig: A logical flag indicating whether to group similar signatures (Default: TRUE). Many secreted proteins, such as cytokines with similar cell surface receptors and downstream pathways, have cellular effects that appear redundant within a cellular context. When enabled, this option clusters secreted proteins based on Pearson correlations among their composite signatures. The output still reports activity estimates for all secreted proteins prior to clustering. Secreted proteins assigned to the same non-redundant cluster share the same inferred activity.
is.group.cor: A numeric value specifying the correlation cutoff used to define similar signatures (Default: 0.90). When r > 0.90, 1,170 secreted protein signatures are grouped into 657 non-redundant signature groups.
lambda: Penalty factor in the ridge regression. If NULL, lambda will be assigned as 5e+05 or 10000 when sigMatrix = "SecAct" or "CytoSig", respectively.
nrand: Number of randomization in the permutation test, with a default value 1000.
ncores: Number of threads for accelerator backends (ignored by pure R). Default 1.
backend: One of "auto", "gpu", "cpu-fast", "cpu-pure". "auto" picks GPU (RidgeCuda) > CPU-fast (RidgeFast) > CPU-pure depending on what is installed. Default "auto".
rng_method: RNG for permutations. "mt19937" (default) is GSL-compatible MT19937 seed 0 — bit-identical across backends when ncores=1. Accelerators may support "srand" for faster, non-reproducible runs. Pure-R supports only "mt19937".
batch_size: Optional positive integer. When supplied, permutation inference is performed over column-batches of Y via the backend's ridge_batch (memory-efficient path for large sample counts). Default NULL (no batching).

Value

A Seurat object.