Build a Rectangle signature model — build_model

Builds a hierarchical signature model from single-cell RNA-seq data using DESeq2-based marker gene selection and optional hierarchical clustering. The resulting model can be passed to deconvolute() to avoid rebuilding the signature for every bulk dataset.

build_model_rectanglepy(
  single_cell_object,
  cell_type_annotations,
  bulk_gene_expression = NULL,
  model_path = NULL,
  optimize_cutoffs = TRUE,
  p = 0.015,
  lfc = 1.5,
  n_cpus = NULL,
  gene_expression_threshold = 0.5,
  verbose = FALSE,
  ...
)

Arguments

single_cell_object: A matrix with the single-cell data. Rows are genes, columns are cells. Row and column names must be set. Raw counts expected.
cell_type_annotations: A vector of cell type labels, one per column of single_cell_object.
bulk_gene_expression: Optional. A matrix of bulk data (rows = genes, columns = samples). When provided, the signature is restricted to genes present in the bulk data.
model_path: Optional path where the signature pickle file should be saved. If NULL (default), a temporary file is used. Providing a permanent path allows the signature to be reused across R sessions by passing the same path to deconvolute() or extract_signature_rectanglepy().
optimize_cutoffs: Whether to optimize p-value and log-fold-change cutoffs via grid search. Recommended but slow. Default: TRUE.
p: P-value cutoff for DE analysis. Only used if optimize_cutoffs = FALSE.
lfc: Log fold-change cutoff for DE analysis. Only used if optimize_cutoffs = FALSE.
n_cpus: Number of CPUs for DE analysis. NULL uses all available.
gene_expression_threshold: Minimum fraction of cells that must express a gene for it to be included in DE analysis. Default: 0.5.
verbose: Whether to print progress to the console. Default: FALSE.
...: Additional arguments, silently ignored. Allows extra parameters to pass through the deconvolute() dispatcher without error.

Value

Path to a pickle file containing the RectangleSignatureResult. Pass this path as the signature argument to deconvolute() or extract_signature_rectanglepy().