Deconvolve bulk RNA-seq data using single-cell signatures

Usage

deconvolve_bulk(
  signatures,
  bulk_counts,
  gene_lengths,
  gene_weights = NULL,
  backend = "wgpu",
  insert_size = 500,
  init_log_exp = -10,
  learning_rate = 0.01,
  l1_lambda = 0,
  l2_lambda = 0,
  max_iter = 10000,
  poll_interval = 100,
  ll_tolerance = 1e-06,
  sparsity_tolerance = 1e-04,
  verbose = TRUE
)

Arguments

signatures: Matrix of cellular signatures (genes × cell_types). Each column should be normalized to sum to 1.
bulk_counts: Matrix of bulk RNA-seq counts (genes × samples). Gene names must match signatures.
gene_lengths: Numeric vector of gene lengths, same length as nrow(signatures).
gene_weights: Optional numeric vector of gene weights (0-1), same length as nrow(signatures). Default is all 1s.
backend: Which compute backend to use: "ndarray" (CPU, default), "wgpu" (GPU if compiled with wgpu feature), "cuda" (CUDA if available).
insert_size: Insert size for RNA-seq fragments. Default 500.
init_log_exp: Initial value for log exposures. Default -10.
learning_rate: Learning rate for optimization. Default 0.01.
l1_lambda: L1 regularization parameter. Default 0.0 (no regularization).
l2_lambda: L2 regularization parameter. Default 0.0 (no regularization).
max_iter: Maximum number of iterations. Default 10000.
poll_interval: Check convergence every N iterations. Default 100.
ll_tolerance: Log-likelihood convergence tolerance. Default 1e-6.
sparsity_tolerance: Sparsity convergence tolerance. Default 1e-4.
verbose: emit learning progress stats

Value

A list with:

exposures: Matrix of cell type proportions (cell_types+intercept × samples)
pred_counts: Matrix of predicted bulk counts (genes × samples)
proportions: Normalized cell type proportions excluding intercept

Examples

if (FALSE) { # \dontrun{
# CPU backend (default)
result <- deconvolve_bulk(
  signatures = signatures,
  bulk_counts = bulk_counts,
  gene_lengths = gene_lengths
)

# GPU backend (if available)
result <- deconvolve_bulk(
  signatures = signatures,
  bulk_counts = bulk_counts,
  gene_lengths = gene_lengths,
  backend = "wgpu"
)
} # }