R/normalize_random.R

In AlineTalhouk/nanostringr: Performs Quality Control, Data Normalization, and Batch Effect Correction for 'NanoString nCounter' Data

#' Normalize data using random reference samples
#'
#' Normalize nanostring gene expression using randomly chosen samples
#' for the reference-based approach for batch adjustment.
#'
#' The number of randomly chosen numbers can be selected, and optionally a
#' `strata` can be specified such that `n` reference samples are selected from
#' each level (like a stratified bootstrap). In relation to the reference
#' method, the random samples removed from `ref` form `R1`, the random samples
#' removed from `x` form `R2`, and the remaining samples from `x` form `Y`. See
#' `refMethod()` for details.
#'
#' In subsequent analyses, we refer to a method using `normalize_random(n)` as
#' the "Random n" method.
#'
#' @param x target data
#' @param ref reference data
#' @param n number of random reference samples to select for normalization
#' @param strata a grouping variable for stratified random sampling. If `strata`
#' has `k` levels, then `n * k` random samples are selected.
#' @param seed random seed for reproducibility
#' @return normalized gene expression
#'
#' @author Derek Chiu
#' @export
normalize_random <- function(x, ref, n = 1, strata = NULL, seed = NULL) {
  if (!is.null(seed)) {
    set.seed(seed)
  }

  df <- dplyr::bind_rows(ref, x)
  if (!is.null(strata)) {
    rand_df <- df %>%
      dplyr::bind_cols(strata = strata) %>%
      dplyr::distinct(.data$ottaID, strata) %>%
      dplyr::group_by(strata) %>%
      dplyr::slice_sample(n = n) %>%
      dplyr::ungroup()
  } else {
    rand_df <- df %>%
      dplyr::distinct(.data$ottaID) %>%
      dplyr::slice_sample(n = n)
  }

  ref_rand <- join_avg(ref, rand_df, "ottaID", "keep")
  x_rand <- join_avg(x, rand_df, "ottaID", "keep")
  x_counts <- join_avg(x, rand_df, "ottaID", "discard")
  x_norm <- as.data.frame(refMethod(x_counts, ref_rand, x_rand))
  return(x_norm)
}