Skip to contents

Format covariates

Source: R/format_covariates.R
format_covariates.Rd

Mainly, this method splits the categorical variables (which should be `factor` variables) into indicator variables (i.e., one-hot encoding), dropping the first level of each factor as the reference, then rescales the numerical variables named in rescale_numeric_variables (only those; the others are left as they are), and computes the "Log_UMI" (i.e., log total counts) for each cell. "Log_UMI" is added as its own column.

Usage

format_covariates(
  dat,
  covariate_df,
  bool_center = FALSE,
  rescale_numeric_variables = NULL,
  variables_enumerate_all = NULL
)

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.

covariate_df

data.frame where each row represents a cell, and the columns are the different categorical or numerical variables that you wish to adjust for

bool_center

Boolean if the numerical variables should be centered around zero, default is FALSE

rescale_numeric_variables

A vector of strings denoting the column names in covariate_df that are numerical and you wish to rescale

variables_enumerate_all

Names of factor variables in covariate_df that get an indicator for every level, with no reference level dropped. Every factor is split into indicators either way; by default (NULL) each drops its first level. A full set of indicators is collinear with the intercept, which initialize_esvd refuses.

Value

a matrix with the same number of rows as dat, whose first two columns are "Intercept" and "Log_UMI", followed by the numerical variables and then one indicator column "<variable>_<level>" per retained factor level.

Details

The rescaling divides each named column by its root-mean-square (scale(x, center = FALSE, scale = TRUE)), not by its standard deviation, so the result does not have unit variance unless the column is also centered (bool_center = TRUE). What it does deliver is invariance to the unit the covariate was recorded in: age in years and age in months give identical columns.

Examples

set.seed(10)
dat <- matrix(stats::rpois(60 * 5, lambda = 3), nrow = 60, ncol = 5)
dimnames(dat) <- list(paste0("cell", 1:60), paste0("gene", 1:5))
covariate_df <- data.frame(
  CC = factor(rep(c("control", "case"), each = 30),
              levels = c("control", "case")),
  Sex = factor(rep(c("F", "M"), times = 30)),
  Age = stats::rnorm(60, mean = 40, sd = 10)
)
covariates <- format_covariates(dat = dat,
                                covariate_df = covariate_df,
                                rescale_numeric_variables = "Age")
colnames(covariates)
#> [1] "Intercept" "Log_UMI"   "Age"       "CC_case"   "Sex_M"    
utils::head(covariates, 3)
#>       Intercept  Log_UMI       Age CC_case Sex_M
#> cell1         1 2.079442 0.9020923       0     0
#> cell2         1 2.197225 0.9373963       0     1
#> cell3         1 2.639057 1.2885881       0     0