R/predictAMR.R

In Cloris2000/MDRClassifier: Multi-Drug Resistance Classifier

#12 Nov 2021
#' Predict AMR from PCA.
#'
#' A function that predict multi-drug resistance by principle component analysis (PCA)
#' to show the new sample's in the original cluster.
#'
#' @param dataframe A dataframe with rownames representing the sample ID and each column
#'                  is one of sample's antimicrobial features. If the value equals to 1,
#'                  the sample has this antimicrobial feature. If the value equals to 0,
#'                  the sample does not have this antimicrobial feature.
#' @param new_data The new data of dataframe must contain columns (variables) with the same
#'                 names and in the same order as the data used to compute PCA.
#' @return Returns the PCA result of the new sample.
#'
#' @examples
#' # Example 3
#' # Using pca_data available with package
#'dim(pca_data)
#'data.active <- pca_data[c(1:17), c(2,9,16)]
#'new.active <- pca_data[c(18:20), c(2,9,16)]
#'resultsExample3 <- predictAMR(data.active, new.active)
#'resultsExample3
#'
#' @references
#'
#' @export
#' @import factoextra


predictAMR <- function(dataframe, new_data){

  #checking arguments

  #dataframe.active <- dataframe
  #calculation
  data.pca <- prcomp(dataframe, scale = TRUE)
  #visualize eigenvalues
  #fivz_eig(data.pca)
  new_ind_coord <- predict(data.pca, newdata = new_data)
  return(new_ind_coord)
}



#' Plot PCA.
#'
#' A function that generate PCA and dimension reduction plots.
#'
#' @param dataframe A dataframe with rownames representing the sample ID and each column
#'                  is one of sample's antimicrobial features. If the value equals to 1,
#'                  the sample has this antimicrobial feature. If the value equals to 0,
#'                  the sample does not have this antimicrobial feature.
#' @param new_data The new data of dataframe must contain columns (variables) with the same
#'                 names and in the same order as the data used to compute PCA.
#' @param plot_type The type of plot need to be generated. For individual PCA plot,
#'                  using "individual". For visualizing relationship between variables, using
#'                  "variables". For predict new data and its relationship with old data, using
#'                  "predict".
#' @return Returns the PCA plot.
#'
#' @examples
#' # Example 4
#' # Using pca_data available with package
#'dim(pca_data)
#'data.active <- pca_data[c(1:17), c(2,9,16)]
#'new.active <- pca_data[c(18:20), c(2,9,16)]
#'resultsExample4 <- plotPCA(data.active, new.active, "Predict")
#'resultsExample4
#'
#' @references
#'
#' @export
#' @import factoextra
#'
#'
#' @references
#'
#' @export
#' @import factoextra
#'
plotPCA <- function(dataframe, new_data=NULL, plot_type){
  #plot of active individuals
  data.pca <- prcomp(dataframe, scale = TRUE)
  p <- fviz_pca_ind(data.pca, repel = TRUE)
  if (plot_type == "individual"){
    #graph of individuals with similar profile are clustered
    return(fviz_pca_ind(data.pca,
                        col.ind = "cos2",
                        gradient.cols = c("#00AFBB", "#E7B800", "#FC4E07"),
                        repel = TRUE))
  }
  if (plot_type == "variables"){
    #graph of variables with positive correlated variables point to the same
    #direction of the plot. Vice versa.
    return(fviz_pca_var(data.pca,
                        col.var = "contrib",
                        gradient.cols = c("#00AFBB", "#E7B800", "#FC4E07"),
                        repel = TRUE))
  }
  if (plot_type == "predict" && (!is.null(new_data))) {
    # Add new supplementary individuals
    new_ind_coord <- predictAMR(dataframe, new_data)
    return(fviz_add(p, new_ind_coord, color = "blue"))
  }

}


#[END]


#devtools::load_all()
#devtools::check()
#devtools::document