Residual variance and the variable features it selects — rs_sc_residual

Computes the per-gene residual variance from a fitted model and selects the variable features from it. With more than one group the selection follows Seurat: rank within each group, take the top n_hvg of each, and union them, so a marker only one sample carries is not buried by a pooled ranking. The returned set can therefore be larger than n_hvg.

Each gene's residual row is regenerated, reduced and dropped, so memory is one row per worker rather than a genes-by-cells matrix.

Usage

rs_sc_residual_variance(
  f_path_gene,
  residual_fit,
  cell_indices,
  n_hvg,
  gene_batch_size,
  verbose
)

Arguments

f_path_gene: String. Path to the counts_genes.bin file.
residual_fit: List. A fit from rs_sc_fit_residuals().
cell_indices: Integer vector. The cell indices to use. (0-indexed!) Must be the selection the fit was fitted on.
n_hvg: Integer. Variable features to take from each group.
gene_batch_size: Integer or NULL. Genes held in memory per batch.
verbose: Integer. 0L - quiet; 1L - normal verbosity; 2L - detailed verbosity.

Value

A list with the following items

genes - The gene indices the variances are indexed by. (0-indexed!)
variance - Matrix of residual variance, genes by groups.
hvg - The selected gene indices, ascending. (0-indexed!)

References

Seurat v5, SCTransform.StdAssay