Initialize eSVD

For each gene, this function fits one ridge-penalized Poisson GLM of the counts on covariates (with offset_variables as offsets; stats::glm when at most one covariate is left to estimate) to obtain z_mat, and then takes a rank-k SVD of log1p(dat) minus the fitted covariate effects to obtain x_mat and y_mat. The result is stored as fit_Init.

Usage

initialize_esvd(
  dat,
  covariates,
  metadata_individual,
  bool_intercept = FALSE,
  case_control_variable = NULL,
  k = 30,
  lambda = 0.01,
  library_size_variable = "Log_UMI",
  offset_variables = "Log_UMI",
  metadata_case_control = NULL,
  verbose = 0
)

Arguments

dat

Dataset (either matrix or dgCMatrix) where the \(n\) rows represent cells and \(p\) columns represent genes. The rows and columns of the matrix should be named.

covariates

matrix object with \(n\) rows with the same rownames as dat where the columns represent the different covariates. Notably, this should contain only numerical columns (i.e., all categorical variables should have already been split into numerous indicator variables), and all the columns in covariates will (strictly speaking) be included in the eSVD matrix factorization model.

metadata_individual

factor vector of length \(n\) that denotes which cell originates from which individual.

bool_intercept

Boolean on whether or not an intercept will be included as a covariate.

case_control_variable

A string of the column name of covariates which depicts the case-control status of each cell. Notably, this should be a binary variable where a 1 is hard-coded to describe case, and a 0 to describe control.

k

Number of latent dimensions.

lambda

Ridge penalty used by glmnet::glmnet on every covariate that is neither the intercept nor an offset; must lie in \([10^{-4}, 10^4]\).

library_size_variable

A string of the variable name (which must be in covariates) of which variable denotes the sequenced (i.e., observed) library size.

offset_variables

A vector of strings depicting which column names in covariate will be set to have a coefficient of 1 automatically (i.e., there will be no estimation of their coefficient).

metadata_case_control

(Optional) vector of length \(n\) with values strictly 0 or 1 that denotes if a cell is from cases or controls. By default, this is set to NULL since the code will extract this information from covariates.

verbose

Integer

Value

eSVD object with elements dat, covariates, param, fit_Init (an eSVD_Fit with x_mat, y_mat, z_mat), latest_Fit ("fit_Init"), case_control and individual

Examples

set.seed(10)
sim <- generate_null(cell_per_person = 15, num_genes = 40,
                     num_individuals = 8)
esvd_obj <- initialize_esvd(dat = sim$obs_mat,
                            covariates = sim$covariates,
                            metadata_individual = sim$metadata_individual,
                            case_control_variable = "CC",
                            bool_intercept = TRUE,
                            k = 2,
                            lambda = 0.1)
esvd_obj <- reparameterization_esvd_covariates(input_obj = esvd_obj,
                                               fit_name = "fit_Init",
                                               omitted_variables = "Log_UMI")
names(esvd_obj)
#> [1] "dat"          "covariates"   "param"        "fit_Init"     "latest_Fit"  
#> [6] "case_control" "individual"  
dim(esvd_obj$fit_Init$x_mat)
#> [1] 120   2