R/signature_potential_damage_analysis.R

In CuppenResearch/MutationalPatterns: Comprehensive genome-wide analysis of mutational processes

#' Potential damage analysis for the supplied mutational signatures
#'
#' The ratio of possible 'stop gain', 'mismatches', 'synonymous mutations' and
#' 'splice site mutations' is counted per signature. Normalized ratios are also
#' given. These were calculated by dividing the ratios in each signature, by the
#' ratios of a completely "flat" signature. A normalized ratio of 2 for "stop
#' gain" mutations, means that a signature is twice as likely to cause "stop
#' gain" mutations, compared to a completely random "flat" signature. N is the
#' number of possible mutations per context, multiplied by the signature
#' contribution per context, summed over all contexts. For mismatches the
#' blosum62 score is also calculated. A lower score means that the amino acids
#' in the mismatches are more dissimilar. More dissimilar amino acids are more
#' likely to have a detrimental effect. Normalized blosum62 scores are also
#' given. These are calculated by substracting the score of a completely "flat"
#' signature from the base blosum62 scores.
#' 
#' The function uses a tibble with the ratio of 'stop gain', 'mismatch',
#' 'synonymous' and 'splice site' mutations per mutation context as its input.
#' For each signature these ratios are linearly combined based on the signature
#' weights. They are then divided by a "flat" signature to get the normalized
#' ratios. The blosum62 scores are also linearly combined based on the signature
#' weights.
#'
#' Please take into account that this is a relatively basic analysis, that only
#' looks at mutational contexts. It does not take into account that signatures
#' can be influenced by open/closed chromatin, strand biases, hairpins or other
#' epigenetic features. This analysis is meant to give an indication, not a
#' definitive answer, of how damaging a signature might be. Further analyses
#' might be required, especially when using signatures in a clinical context.
#' 
#'
#' @param signatures Matrix containing signatures
#' @param contexts Vector of mutational contexts to use for the analysis.
#' @param context_mismatches A tibble with the ratio of 'stop gain', 'mismatch', 
#' 'synonymous' and 'splice site' mutations per mutation context.
#'
#' @return A tibble with the ratio of 'stop gain', 'mismatch', 'synonymous' 
#' and 'splice site' mutations per signature.
#' @export
#'
#' @examples
#'
#' ## Get the signatures
#' signatures <- get_known_signatures()
#'
#' ## See the 'mut_matrix()' example for how we obtained the
#' ## mutation matrix information:
#' mut_mat <- readRDS(system.file("states/mut_mat_data.rds",
#'   package = "MutationalPatterns"
#' ))
#'
#' contexts <- rownames(mut_mat)
#'
#' ## See the 'context_potential_damage_analysis()' example for how we obtained the
#' ## context_mismatches:
#' context_mismatches <- readRDS(system.file("states/context_mismatches.rds",
#'   package = "MutationalPatterns"
#' ))
#'
#' ## Determine the potential damage per signature
#' signature_potential_damage_analysis(signatures, contexts, context_mismatches)
signature_potential_damage_analysis <- function(signatures, contexts, context_mismatches) {

  # These variables use non standard evaluation.
  # To avoid R CMD check complaints we initialize them to NULL.
  context <- ratio <- contribution <- sig <- flat <- n <- NULL
  ratio_by_background <- flat_ratio <- flat_blosum62 <- NULL
  blosum62 <- blosum62_min_background <- NULL
  
  # Add context to signature matrix
  signatures <- as.data.frame(signatures)
  signatures$context <- contexts

  # Add flat background signature
  nr_features <- nrow(signatures)
  signatures$flat_background <- rep(1 / nr_features, nr_features)


  # Combine signatures and context_mismatches and make long.
  sig_context_mismatch <- signatures %>%
    dplyr::full_join(context_mismatches, by = "context") %>%
    tidyr::pivot_longer(cols = c(-context, -type, -ratio, -n, -blosum62), 
                        names_to = "sig", 
                        values_to = "contribution")

  # Combine weighted contexts per signature and mutation type
  sig_mismatch <- sig_context_mismatch %>%
    dplyr::mutate(ratio = contribution * ratio, 
                  n = contribution * n, 
                  blosum62 = contribution * blosum62) %>%
    dplyr::group_by(type, sig) %>%
    dplyr::summarise(n = sum(n), 
                     ratio = sum(ratio), 
                     blosum62 = sum(blosum62),
                     .groups = "drop_last") %>%
    dplyr::ungroup()

  # Get flat signature to serve as background
  flat_sig <- sig_mismatch %>%
    dplyr::filter(sig == "flat_background") %>%
    dplyr::select(type, flat_ratio = ratio, flat_blosum62 = blosum62)

  # Normalize against the flat background signature
  norm_sig_mismatch <- sig_mismatch %>%
    dplyr::full_join(flat_sig, by = "type") %>%
    dplyr::mutate(ratio_by_background = ratio / flat_ratio,
                  blosum62_min_background = blosum62 - flat_blosum62) %>%
    dplyr::select(type, 
                  sig, 
                  ratio, 
                  ratio_by_background, 
                  n, 
                  blosum62, 
                  blosum62_min_background) %>%
    dplyr::filter(sig != "flat_background") %>%
    dplyr::arrange(sig)

  return(norm_sig_mismatch)
}