selectionPlot: Plot Pair-wise Overlap or Selection Size Distribution of...

In ClassifyR: A framework for cross-validated classification problems, with applications to differential variability and differential distribution testing

Description

Pair-wise overlaps can be done for two types of analyses. Firstly, each cross-validation iteration can be considered within a single classification. This explores the feature selection stability. Secondly, the overlap may be considered between different classification results. This approach compares the feature selection commonality between different selection methods. Two types of commonality are possible to analyse. One summary is the average pair-wise overlap between a level of the comparison factor and the other summary is the pair-wise overlap of each level of the comparison factor that is not the reference level against the reference level. The overlaps are converted to percentages and plotted as lineplots.

Additionally, a heatmap of selection size frequencies can be made.

Usage

  ## S4 method for signature 'list'
selectionPlot(results,
              comparison = c("within", "size", "classificationName",
                             "validation", "datasetName", "selectionName"),
              referenceLevel = NULL,
              xVariable = c("classificationName", "datasetName", "validation", "selectionName"),
              boxFillColouring = c("classificationName", "size", "datasetName",
                                   "validation", "selectionName", "None"),
              boxFillColours = NULL,
              boxFillBinBoundaries = NULL, setSizeBinBoundaries = NULL,
              boxLineColouring = c("validation", "classificationName",
                                   "datasetName", "selectionName", "None"),
              boxLineColours = NULL,
              rowVariable = c("None", "validation", "datasetName",
                              "classificationName", "selectionName"),
              columnVariable = c("datasetName", "classificationName",
                                 "validation", "selectionName", "None"),
              yMax = 100, fontSizes = c(24, 16, 12, 16),
              title = if(comparison[1] == "within") "Feature Selection Stability"
                      else if(comparison == "size") "Feature Selection Size" else
                      "Feature Selection Commonality",
              xLabel = "Analysis",
              yLabel = if(is.null(referenceLevel) && comparison != "size") "Common Features (%)"
                       else if(comparison == "size") "Set Size" else
                       paste("Common Features with", referenceLevel, "(%)"),
              margin = grid::unit(c(1, 1, 1, 1), "lines"), rotate90 = FALSE,
              showLegend = TRUE, plot = TRUE, parallelParams = bpparam())

Arguments

results	A list of ClassifyResult or SelectResult objects.
comparison	The aspect of the experimental design to compare. See Details section for a detailed description.
referenceLevel	The level of the comparison factor to use as the reference to compare each non-reference level to. If NULL, then each level has the average pairwise overlap calculated to all other levels.
xVariable	The factor to make separate boxes in the boxplot for.
boxFillColouring	A factor to colour the boxes by.
boxFillColours	A vector of colours, one for each level of boxFillColouring. If NULL, a default palette is used.
boxFillBinBoundaries	Used only if comparison is "size". A vector of integers, specifying the bin boundaries of percentages of size bins observed. e.g. 0, 10, 20, 30, 40, 50.
setSizeBinBoundaries	Used only if comparison is "size". A vector of integers, specifying the bin boundaries of set size bins. e.g. 50, 100, 150, 200, 250.
boxLineColouring	A factor to colour the box lines by.
boxLineColours	A vector of colours, one for each level of boxLineColouring. If NULL, a default palette is used.
rowVariable	The slot name that different levels of are plotted as separate rows of boxplots.
columnVariable	The slot name that different levels of are plotted as separate columns of boxplots.
yMax	The maximum value of the percentage to plot.
fontSizes	A vector of length 4. The first number is the size of the title. The second number is the size of the axes titles. The third number is the size of the axes values. The fourth number is the font size of the titles of grouped plots, if any are produced. In other words, when rowVariable or columnVariable are not NULL.
title	An overall title for the plot.
xLabel	Label to be used for the x-axis.
yLabel	Label to be used for the y-axis of overlap percentages.
margin	The margin to have around the plot.
rotate90	Logical. If TRUE, the boxplot is horizontal.
showLegend	If TRUE, a legend is plotted next to the plot. If FALSE, it is hidden.
plot	Logical. If TRUE, a plot is produced on the current graphics device.
parallelParams	An object of class MulticoreParam or SnowParam.

Details

Possible values for characteristics are "datasetName", "classificationName", "size", "selectionName", and "validation". If "None", then that graphical element is not used.

If comparison is "within", then the feature selection overlaps are compared within a particular analysis. The result will inform how stable the selections are between different iterations of cross-validation for a particular analysis. If comparison is "classificationName", then the feature selections are compared across different classification algorithm types, for each level of "datasetName", "selectionName" and "validation". The result will inform how stable the feature selections are between different classification algorithms, for every cross-validation scheme, selection algorithm and data set. If comparison is "selectionName", then the feature selections are compared across different feature selection algorithms, for each level of "datasetName", "classificationName" and "validation". The result will inform how stable the feature selections are between feature selection algorithms, for every data set, classification algorithm, and cross-validation scheme. If comparison is "validation", then the feature selections are compared across different cross-validation schemes, for each level of "classificationName", "selectionName" and "datasetName". The result will inform how stable the feature selections are between different cross-validation schemes, for every selection algorithm, classification algorithm and every data set. If comparison is "datasetName", then the feature selections are compared across different data sets, for each level of "classificationName", "selectionName", and "validation". The result will inform how stable the feature selections are between different data sets, for every classification algorithm and every data set. This could be used to consider if different experimental studies have a highly overlapping feature selection pattern.

Calculating all pair-wise set overlaps can be time-consuming. This stage can be done on multiple CPUs by providing the relevant options to parallelParams. The percentage is calculated as the intersection of two sets of features divided by the union of the sets, multiplied by 100.

For the selection size mode, boxFillBins is used to create bins which include the lowest value for the first bin, and the highest value for the last bin using cut.

Value

An object of class ggplot and a plot on the current graphics device, if plot is TRUE.

Author(s)

Dario Strbenac

Examples

  predicted <- data.frame(sample = sample(10, 100, replace = TRUE),
                          class = rep(c("Healthy", "Cancer"), each = 50))
  actual <- factor(rep(c("Healthy", "Cancer"), each = 5))
  features <- sapply(1:100, function(index) paste(sample(LETTERS, 3), collapse = ''))
  rankList <- list(list(features[1:100], features[c(5:1, 6:100)]),
                   list(features[c(1:9, 11, 10, 12:100)], features[c(1:50, 60:51, 61:100)]))
  result1 <- ClassifyResult("Example", "Differential Expression",
                            "Example Selection", LETTERS[1:10], features,
                            100, rankList,
                            list(list(rankList[[1]][[1]][1:15], rankList[[1]][[2]][1:15]),
                                 list(rankList[[2]][[1]][1:10], rankList[[2]][[2]][1:10])),
                            list(function(oracle){}),
                            list(predicted), actual, list("permuteFold", 2, 2))
  
  predicted[, "class"] <- sample(predicted[, "class"])
  rankList <- list(list(features[1:100], features[c(sample(20), 21:100)]),
                   list(features[c(1:9, 11, 10, 12:100)], features[c(1:50, 60:51, 61:100)]))
  result2 <- ClassifyResult("Example", "Differential Variability",
                            "Example Selection",
                            LETTERS[1:10], features, 100, rankList,
                            list(list(rankList[[1]][[1]][1:15], rankList[[1]][[2]][1:15]),
                                 list(rankList[[2]][[1]][1:10], rankList[[2]][[2]][1:10])),
                            list(function(oracle){}),
                            list(predicted), actual, validation = list("permuteFold", 2, 2))
                            
  selectionPlot(list(result1, result2), xVariable = "classificationName",
                xLabel = "Analysis", columnVariable = "None", rowVariable = "None",
                boxFillColouring = "classificationName")
  
  selectionPlot(list(result1, result2), comparison = "size",
                xVariable = "classificationName", xLabel = "Analysis",
                columnVariable = "None", rowVariable = "None",
                boxFillColouring = "size", boxFillBinBoundaries = seq(0, 100, 10),
                setSizeBinBoundaries = seq(0, 25, 5), boxLineColouring = "None")
  
  oneRanking <- features[c(10, 8, 1, 2, 3, 4, 7, 9, 5, 6)]
  otherRanking <- features[c(8, 2, 3, 4, 1, 10, 6, 9, 7, 5)]
  oneResult <- SelectResult("Example", "One Method", 10, list(oneRanking), list(oneRanking[1:5]))
  otherResult <- SelectResult("Example", "Another Method", 10, list(otherRanking), list(otherRanking[1:2]))
  
  selectionPlot(list(oneResult, otherResult), comparison = "selectionName",
                xVariable = "selectionName", xLabel = "Selection Method",
                columnVariable = "None", rowVariable = "None",
                boxFillColouring = "selectionName", boxLineColouring = "None")

ClassifyR documentation built on Nov. 8, 2020, 6:53 p.m.