Redo the test at another cap on the nuisance rate, without fitting again

The factorization (x_mat, y_mat, z_mat) is the expensive part of an analysis and does not depend on the cap on the nuisance rate; see estimate_nuisance. This function takes a fitted and tested eSVD object, applies another cap_multiplier to the maximum-likelihood rates stored on it, and repeats what comes after the rates: the posterior, the test statistic, the log2 fold change and the p-values. Nothing is estimated again, and every other setting is the one recorded in input_obj$param when the object was built.

Usage

recompute_pvalue(input_obj, cap_multiplier, seurat_obj = NULL, verbose = 0)

Arguments

input_obj

eSVD object that has been tested, by eSVD, eSVD_helper, compute_test_per_gene or compute_pvalue, with version 1.2.0 or later.

cap_multiplier

One positive number: no gene's nuisance rate may exceed this many times the median of its library size. Inf means no cap. See estimate_nuisance.

seurat_obj

NULL, or the Seurat object the analysis was run on. Required when input_obj has no counts (input_obj$dat) or covariates, and ignored otherwise.

verbose

Integer; 0 is silent.

Value

The eSVD object with teststat_vec, case_mean, control_mean, case_var, control_var, log2fc_vec, log2fc_se_vec and pvalue_list replaced, and, in the fit named by latest_Fit, nuisance_vec, nuisance_status and the posterior matrices (if the object had them) replaced. param$nuisance_cap_multiplier, param$nuisance_num_capped and param$nuisance_num_boundary describe the new cap. The factorization and nuisance_mle_vec are unchanged, so the function can be called again at any other cap. An object that had no counts is returned without them.

Details

The posterior needs the counts and the covariates. An object built with eSVD(bool_diet = FALSE), or stage by stage, carries them. An object built with bool_diet = TRUE (the default of eSVD and eSVD_helper) does not, so pass seurat_obj, the Seurat object the analysis was run on. The cells and genes of the fit are looked up in it by name, so it may hold more cells and genes than were analyzed (the individuals and the all-zero genes eSVD_helper removed), and the covariates are rebuilt from its meta.data with the variable names eSVD recorded. Summaries of the rebuilt counts and covariates are compared with those recorded at the fit, and the function stops if they differ, for example because the meta.data was edited since.

The result is what eSVD would have returned had it been called with this cap_multiplier.