Apply In-Silico Perturbation — ApplyPerturbation • compact

This function applies an in-silico perturbation (knock-out, knock-down, or knock-in) to selected features in a Seurat object.

Usage

ApplyPerturbation(
  seurat_obj,
  exp,
  features,
  perturb_dir,
  cells_use = NULL,
  group.by = NULL,
  layer = "counts",
  slot = "counts",
  assay = "RNA"
)

Arguments

seurat_obj: A Seurat object containing the dataset.
exp: A features-by-cells matrix (typically a sparse matrix) containing the observed expression data.
features: Character vector. The selected features to apply the perturbation on.
perturb_dir: Numeric. Determines the type of perturbation to apply. Negative values for knock-down, positive for knock-in, and 0 for knock-out.
cells_use: Character vector. Specific cells to apply the perturbation to. If NULL, defaults to handling internally.
group.by: Character. Column in seurat_obj@meta.data used to group cells for modeling.
layer: Character. The Seurat v5 layer to extract data from. Default is 'counts'.
slot: Character. The Seurat v4 slot to extract data from. Default is 'counts'.
assay: Character. The assay in seurat_obj containing expression information. Default is 'RNA'.

Value

A dgCMatrix object containing the updated expression matrix with the applied perturbations.

Details

The function models the baseline expression of target features using a Zero-Inflated Negative Binomial (ZINB) distribution to capture dropout and overdispersion characteristics typical of single-cell RNA-seq data . It samples from this distribution, scales the values by perturb_dir, and updates the original expression matrix. Any perturbed counts that fall below zero are strictly bounded to zero.