Fit marginal models for every feature — fitMarginalPop • scDesignPop

Fits a specified parametric model using various input parameters.

Usage

fitMarginalPop(
  data_list,
  mean_formula,
  model_family = "nb",
  interact_colnames = NULL,
  parallelization = c("pbmcapply", "future.apply", "parallel", "biocparallel"),
  n_cores = 2L,
  loc_colname = "POS",
  snp_colname = "snp_id",
  celltype_colname = "cell_type",
  indiv_colname = "indiv",
  filter_snps = TRUE,
  snpvar_thres = 0,
  force_formula = FALSE,
  keep_cellnames = FALSE,
  BPPARAM = NULL,
  future.seed = FALSE,
  data_maxsize = 1,
  ...
)

Arguments

data_list: a list of input data.
mean_formula: a string scalar to specify the mean formula, including random effects (if any) but without SNP genotypes or SNP genotype interaction effects.
model_family: a string scalar to specify model fitting used.
interact_colnames: a string scalar or vector for the variable names that have first-order interaction with SNP genotypes.
parallelization: a string scalar specifying the parallelization backend used during marginal fitting. Must be one of either "parallel", "future.apply", "biocparallel", or "pbmcapply". The default value is "pbmcapply". See details.
n_cores: positive integer value (greater or equal to 1) to specify the number of CPU cores used in parallelization. The default is 2.
loc_colname: a string scalar for column name of SNP position variable.
snp_colname: a string scalar for column name of SNP id variable.
celltype_colname: a string scalar for column name of cell type.
indiv_colname: a string scalar for column name of individuals (samples).
filter_snps: a logical scalar for whether to filter out SNP covariates with either low-variance or with only 1 distinct genotype (ie. all 1's) prior to fitting the model.
snpvar_thres: a numeric scalar (between 0 to 1) used to filter out SNPs whose variance of genotypes across samples are below this threshold. Used together when filter_snps = TRUE.
force_formula: a logical scalar for whether to bypass model parsimony check. If force_formula = TRUE, interaction terms whose covariates are not main effects in the model would be permitted. Results in error if force_formula = FALSE and length(geno_interact_names) > 0.
keep_cellnames: a logical scalar for whether to keep cell barcode names. If keep_cellnames = TRUE, the memory will be larger. The default is FALSE.
BPPARAM: a BiocParallelParam class object (from BiocParallel R package) that must be specified when using parallelization = "biocparallel". Either BiocParallel::SnowParam() or BiocParallel::MulticoreParam() can be used to initialize, depending on the operating system. BPPARAM is not used in other parallelization options. The default is NULL.
future.seed: a logical or an integer (of length one or seven), or a list of length(X) with pre-generated random seeds that can be specified when using parallelization = "future.apply". See future.apply::future_eapply documentation for more details on its usage. future.seed is not used in other parallelization options. The default is FALSE.
data_maxsize: a positive numeric value used to set max data_list size in GiB increments. Used only when parallelization = "future.apply". The default is 1.
...: additional arguments passed to internal functions.

Value

a list of named features, each containing a list with the following items:

fit: a glmmTMB fit object.
time: a numeric scalar for the elapsed time to fit the given feature.
snp_cov: a string scalar or vector of SNP ids used in the fit for given feature.
model_attr: a list of attributes extracted from each model. NOT currently implemented.
removed_cell: a string scalar or vector of the cell names removed due to low-variance (NOT currently implemented).

Details

Parallelization options

If "parallel" is used then mclapply is called from the parallel package; if "biocparallel" is used, then bplapply is called from the BiocParallel package; if "future.apply" is used, then future_lapply is called from the future.apply package; if "pbmcapply" is used, then pbmclapply is called from the pbmcapply package.

Examples

NULL
#> NULL