Run NEBULA on meta cells
nebula_mc.RdFits NEBULA's negative binomial gamma mixed model over aggregated counts.
The arithmetic is nebula_sc() verbatim, only the counts come out
of memory rather than the streamed store, and the fit is much cheaper
because there are far fewer rows.
What changes is the interpretation, not the numerics. The cell-level overdispersion becomes the spread between aggregates within a subject rather than between cells, so it is smaller and it absorbs whatever the aggregation smoothed away. The subject-level term keeps its meaning. Read the two as a variance decomposition over meta cells and do not compare them against a single cell run.
Usage
nebula_mc(
object,
subject_col,
design,
coef = NULL,
contrast = NULL,
genes_to_use = NULL,
offset = NULL,
nebula_params = params_nebula(),
.verbose = TRUE
)Arguments
- object
MetaCellsclass.- subject_col
String. The column in the obs table holding the subject (donor) identifier. This is what the random effect is over.
- design
Formula. The experimental design, evaluated against the obs table, e.g.
~ conditionor~ condition + age. Include the intercept.- coef
Optional integer or character. Which coefficient of the design the Wald test reports, as a 1-based column position or a column name. Defaults to the last column.
- contrast
Optional numeric vector. One weight per design column. Mutually exclusive with
coef.- genes_to_use
Optional character vector. The genes to fit. Defaults to every gene in the object.
- offset
Optional numeric vector. Strictly positive scaling factor per meta cell, aligned to the meta cells that survive the design. Defaults to
NULL, which uses the aggregated library sizes.- nebula_params
A list, see
params_nebula(). Seenebula_sc()for the individual elements.- .verbose
Boolean or integer. Controls verbosity and returns run times.
FALSE-> quiet,TRUEor1L-> normal verbosity,2L-> detailed verbosity.
Value
A ScNebula class, see new_sc_nebula_res().
Examples
# differential expression over meta cells with a donor random effect
sc <- demo_single_cells(
syn_data_params = params_sc_synthetic_data(
n_cells = 2000L,
n_genes = 50L,
n_samples = 6L,
sample_bias = "even"
)
)
mc <- generate_bt_meta_cells_sc(
sc,
sc_meta_cell_params = params_sc_bt_metacells(target_no_metacells = 100L),
.verbose = FALSE
)
# the meta cell obs carries no donor labels, so vote them over
sample_id <- sc[["sample_id"]]$sample_id
mc[["sample_id"]] <- vapply(
mc[[]]$original_cell_idx,
function(idx) names(which.max(table(sample_id[idx]))),
character(1)
)
mc[["condition"]] <- ifelse(
mc[["sample_id"]]$sample_id %in% c("sample_1", "sample_2", "sample_3"),
"ctr",
"trt"
)
res <- nebula_mc(
mc,
subject_col = "sample_id",
design = ~condition,
.verbose = FALSE
)
head(res$results)
#> gene_id log_fc effect_se z p_value fdr
#> <char> <num> <num> <num> <num> <num>
#> 1: gene_01 -0.4345582 0.2976426 -1.4599999 0.1442901 0.9274431
#> 2: gene_02 -0.1984760 0.2907377 -0.6826634 0.4948196 0.9274431
#> 3: gene_03 -0.1901702 0.2949573 -0.6447379 0.5190970 0.9274431
#> 4: gene_04 -0.2830729 0.3538626 -0.7999514 0.4237390 0.9274431
#> 5: gene_05 -0.3958592 0.4129879 -0.9585249 0.3377981 0.9274431
#> 6: gene_06 -0.2923441 0.2664164 -1.0973202 0.2725015 0.9274431
#> subject_overdispersion cell_overdispersion convergence sigma_at_bound
#> <num> <num> <int> <lgcl>
#> 1: 0.05448843 1.150486 1 FALSE
#> 2: 0.05224946 1.077529 1 FALSE
#> 3: 0.05974837 1.021993 -10 FALSE
#> 4: 0.10133330 1.170186 1 FALSE
#> 5: 0.14619244 1.391691 1 FALSE
#> 6: 0.03321586 1.091282 1 FALSE
#> cell_overdispersion_shrunk
#> <num>
#> 1: 1.119392
#> 2: 1.063358
#> 3: 1.001475
#> 4: 1.108070
#> 5: 1.310626
#> 6: 1.066635
unlink(sc@dir_data, recursive = TRUE, force = TRUE)