MAGIC: Markov Affinity-based Graph Imputation of Cells

Main function for MAGIC imputation of single-cell data.

Performs MAGIC (Markov Affinity-based Graph Imputation of Cells) for denoising and imputation of single-cell data.

Run MAGIC on a Seurat object and store results in a new assay.

Run MAGIC on a SingleCellExperiment object.

Usage

magic(data, ...)

# Default S3 method
magic(
  data,
  genes = NULL,
  knn = 5,
  knn_max = NULL,
  decay = 1,
  t = 3,
  npca = 100,
  solver = c("exact", "approximate"),
  init = NULL,
  t_max = 20,
  knn_dist = c("euclidean", "cosine", "manhattan", "correlation"),
  n_jobs = 1,
  seed = NULL,
  verbose = TRUE,
  ...
)

# S3 method for class 'matrix'
magic(
  data,
  genes = NULL,
  knn = 5,
  knn_max = NULL,
  decay = 1,
  t = 3,
  npca = 100,
  solver = c("exact", "approximate"),
  init = NULL,
  t_max = 20,
  knn_dist = c("euclidean", "cosine", "manhattan", "correlation"),
  n_jobs = 1,
  seed = NULL,
  verbose = TRUE,
  ...
)

# S3 method for class 'dgCMatrix'
magic(data, ...)

# S3 method for class 'data.frame'
magic(data, ...)

# S3 method for class 'Seurat'
magic(
  data,
  assay = NULL,
  slot = "data",
  genes = "all_genes",
  new_assay_name = NULL,
  knn = 5,
  knn_max = NULL,
  decay = 1,
  t = 3,
  npca = 100,
  solver = c("exact", "approximate"),
  init = NULL,
  t_max = 20,
  knn_dist = c("euclidean", "cosine", "manhattan", "correlation"),
  n_jobs = 1,
  seed = NULL,
  verbose = TRUE,
  ...
)

# S3 method for class 'SingleCellExperiment'
magic(
  data,
  assay_name = "logcounts",
  genes = "all_genes",
  new_assay_name = "MAGIC",
  knn = 5,
  knn_max = NULL,
  decay = 1,
  t = 3,
  npca = 100,
  solver = c("exact", "approximate"),
  init = NULL,
  t_max = 20,
  knn_dist = c("euclidean", "cosine", "manhattan", "correlation"),
  n_jobs = 1,
  seed = NULL,
  verbose = TRUE,
  ...
)

Arguments

data: A SingleCellExperiment object.
...: Additional arguments.
genes: Gene selection. Default is "all_genes".
knn: Number of neighbors. Default is 5.
knn_max: Maximum neighbors. Default is NULL.
decay: Alpha decay. Default is 1.
t: Diffusion steps. Default is 3.
npca: PCA components. Default is 100.
solver: Solver type. Default is "exact".
init: Previous result. Default is NULL.
t_max: Maximum t. Default is 20.
knn_dist: Distance metric. Default is "euclidean".
n_jobs: Parallel workers. Default is 1.
seed: Random seed. Default is NULL.
verbose: Verbosity. Default is TRUE.
assay: Character. Name of the assay to use as input. Default is the default assay of the object.
slot: Character. Slot to use for input data. Default is "data" (normalized data). For Seurat v5, this corresponds to the layer name.
new_assay_name: Name for the new assay. Default is "MAGIC".
assay_name: Character. Name of the assay to use. Default is "logcounts".

Value

A "magic" object (for matrices) or modified input object (for Seurat/SCE).

The Seurat object with a new assay containing MAGIC-imputed values. The new assay is named "MAGIC_<original_assay>" by default.

SingleCellExperiment object with MAGIC results in a new assay.

Details

MAGIC denoises single-cell data using diffusion on a cell similarity graph:

Optional PCA dimension reduction
Build k-nearest neighbor graph
Construct alpha-decaying kernel
Create row-stochastic diffusion operator
Apply diffusion operator t times

The imputation formula is: \(X_{imputed} = P^t X\)

This function:

Extracts normalized data from the specified assay
Transposes to cells x genes format
Runs MAGIC imputation
Creates a new assay with the imputed data
Stores MAGIC parameters in the object's misc slot

The function is compatible with both Seurat v4 and v5. For v4, it uses GetAssayData/SetAssayData. For v5, it uses LayerData where available.

References

van Dijk, D., et al. (2018). Recovering gene interactions from single-cell data using data diffusion. Cell, 174(3), 716-729. doi:10.1016/j.cell.2018.05.061

Examples

if (FALSE) { # \dontrun{
# Basic usage
set.seed(42)
data <- matrix(rpois(5000, 2), nrow = 500, ncol = 10)
result <- magic(data, t = 3)
imputed <- as.matrix(result)

# Automatic t selection
result <- magic(data, t = "auto")

# Specific genes
result <- magic(data, genes = c(1, 2, 3))
} # }
if (FALSE) { # \dontrun{
library(Seurat)

# Create example Seurat object
counts <- matrix(rpois(3000, lambda = 5), nrow = 30, ncol = 100)
rownames(counts) <- paste0("Gene", 1:30)
colnames(counts) <- paste0("Cell", 1:100)

seurat_obj <- CreateSeuratObject(counts = counts)
seurat_obj <- NormalizeData(seurat_obj)

# Run MAGIC
seurat_obj <- magic(seurat_obj, t = 3)

# Access MAGIC results
GetAssayData(seurat_obj, assay = "MAGIC_RNA", slot = "data")

# Make MAGIC assay the default
DefaultAssay(seurat_obj) <- "MAGIC_RNA"
} # }