Fit the NEBULA negative binomial gamma mixed model over meta cells

The arithmetic is the single cell one verbatim, only the counts come from memory rather than the streamed store. What changes is the interpretation: the cell-level overdispersion becomes the spread between aggregates within a subject, not between cells, so it is smaller and absorbs whatever the aggregation smoothed away. The subject-level term keeps its meaning. Do not compare the two against a single cell run.

Usage

rs_nebula_mc(
  sparse_data,
  metacells_to_keep,
  gene_indices,
  subject_ids,
  design,
  offset,
  nebula_params,
  verbose
)

Arguments

sparse_data

A named list that needs to have data, indptr, indices, nrow, ncol and cs_type. Shape (metacells, genes), holding the raw counts.

metacells_to_keep

Integer vector. 0-indexed(!) positions of the meta cells to analyse, in any order. Must not hold duplicates.

gene_indices

Integer vector. 0-indexed(!) positions of the genes to fit.

subject_ids

Integer vector. 0-indexed(!) subject label per meta cell. One entry per row of sparse_data, not per element of metacells_to_keep.

design

Numeric matrix. Predictors of meta cells x coefficients, rows aligned to metacells_to_keep and including an intercept.

offset

Optional numeric vector. Strictly positive scaling factor per selected meta cell. NULL uses the aggregated library sizes.

nebula_params

Named list. The NEBULA parameters, see params_nebula(), plus either coef (a 0-indexed(!) coefficient) or contrast (one weight per coefficient).

verbose

Integer. 0L - quiet; 1L - normal verbosity; 2L - detailed verbosity.

Value

A list with the same elements rs_nebula_sc() returns.

References

He, et al., Commun Biol, 2021