R/preprocessing.R

In erah: Automated Spectral Deconvolution, Alignment, and Metabolite Identification in GC/MS-Based Untargeted Metabolomics

pre.process <- function(sampl, sigma.scans){
  sigma.model.filter <- round(sigma.scans)
  if(is.even(sigma.model.filter)) sigma.model.filter <- sigma.model.filter + 1 
  filt.order <- 3
  if(filt.order>=sigma.model.filter) filt.order <- sigma.model.filter - 1
  if(filt.order>=1) sampl <- soft.filter(sampl,filt.order,sigma.model.filter)
  sampl[sampl<0] <- 0
  
  k.filt <- trunc(sigma.scans*10)
  moving.maximas <- apply(sampl,2,function(x){max(x)})
  sampl[,moving.maximas<100] <- 0
  sampl <- apply(sampl,2,function(x) removeBaseline(x,k.filt))
  
  sampl[sampl<0] <- 0
  sampl
}

#' @importFrom stats sd

removeBaseline <- function (x, k){
  if (max(x) == 0) return(x)
  x.min <- runningmin(x, k)
  k.m <- k
  if(k.m>=length(x)) k.m <- length(x) - 1
  if (is.even(k.m) == T) k.m <- k.m - 1
  x.dupCopy <- c(x[length(x):1],x,x[length(x):1])
  x.med <- stats::runmed(x.dupCopy, k.m, endrule='keep')[(length(x)+1):(2*length(x))]
  
  base.sd <- sd(x.min)
  x.base <- x.med
  x.base[x.base > (x.min + base.sd)] <- x.min[x.base > (x.min + base.sd)] + base.sd
  x.base <- runningmean(x.base, k)
  x.clean <- x - x.base
  x.clean[x.clean < 0] <- 0
  x.clean
}


#' @importFrom signal sgolayfilt

soft.filter <- function (mat, p, n = NULL){
  if (is.null(n)) 
    n <- p + 3
  
  mat.f <- apply(mat, 2, function(x) sgolayfilt(x, p = p , n = n))
  mat.D <- normalize(mat.f)
  mat.D <- sweep(mat.D,2,apply(mat,2, max),"*")
  mat.D
}