R/methods-barPlotModelComparison.R

In bhklab/PanCuRx: Pancreatic Ductal Adenocarcinoma Tool-Kit

#' Make A Bar Plot Comparing Performance Between Two `S4` Objects Representing
#'   Mathematical Models.
#'
#' @param model1 An`S4` object containing results of a mathematical model
#' @param model2 An `S4` object containing results of a different mathematical
#'   model, but with the same or overlapping samples.
#' @param ... Allow new parameters to be defined for this generic.
#'
#' @return A bar plot comparing some aspect of model1 and model2
#'
#' @examples
#' data(sampleCohortList)
#' data(sampleValPCOSPmodel)
#' data(sampleICGCmicro)
#'
#' # Set parallelization settings
#' BiocParallel::register(BiocParallel::SerialParam())
#'
#' # Setup the models
#' set.seed(1987)
#' clinicalModel <- ClinicalModel(sampleICGCmicro,
#'   formula='prognosis ~ sex + age + T + N + M + grade',
#'   randomSeed=1987)
#'
#' # Train the models
#' trainedClinicalModel <- trainModel(clinicalModel)
#'
#' # Make predctions
#' clinicalPredCohortList <- predictClasses(sampleCohortList[c('PCSI', 'TCGA')],
#'   model=trainedClinicalModel)
#'
#' # Validate the models
#' validatedClinicalModel <- validateModel(trainedClinicalModel,
#'   valData=clinicalPredCohortList)
#'
#' # Plot the comparison
#' modelCompBarPlot <- barPlotModelComparison(validatedClinicalModel,
#'  sampleValPCOSPmodel, stat='AUC')
#'
#' @md
#' @export
setGeneric("barPlotModelComparison", function(model1, model2, ...)
    standardGeneric('barPlotModelComparison'))
#'
#' Make a Bar Plot Comparison Model Performance Between a ClinicalModel
#'   and a PCOSP, RLSModel or RGAModel object.
#'
#' @param model1 A `ClinicalModel` object.
#' @param model2 A `PCOSP` or `RLSModel` or `RGAModel` object.
#' @param stat A `character` vector specifying which statistic to compare the
#'   models using. Options are 'AUC', 'log_D_index' or 'concordance_index'.
#' @param ... Not used.
#'
#' @return A `ggplot2` object showing a barplot coloured by the model and
#'   comparing the stat between all cohorts that both models were validated
#'   with.
#'
#' @examples
#' data(sampleValPCOSPmodel)
#' data(sampleCohortList)
#' data(sampleICGCmicro)
#'
#' # Set parallelization settings
#' BiocParallel::register(BiocParallel::SerialParam())
#'
#' # Setup the models
#' clinicalModel <- ClinicalModel(sampleICGCmicro,
#'   formula='prognosis ~ sex + age + T + N + M + grade',
#'   randomSeed=1987)
#'
#' # Train the models
#' trainedClinicalModel <- trainModel(clinicalModel)
#'
#' # Make predctions
#' clinicalPredCohortList <- predictClasses(sampleCohortList[c('PCSI', 'TCGA')],
#'   model=trainedClinicalModel)
#'
#' # Validate the models
#' validatedClinicalModel <- validateModel(trainedClinicalModel,
#'   valData=clinicalPredCohortList)
#'
#' # Plot the comparison
#' modelCompBarPlot <- barPlotModelComparison(validatedClinicalModel,
#'  sampleValPCOSPmodel, stat='AUC')
#'
#' @md
#' @include classUnions.R
#' @importFrom ggplot2 ggplot geom_col aes
#' @export
setMethod('barPlotModelComparison', signature(model1='ClinicalModel',
    model2='PCOSP_or_RLS_or_RGA'), function(model1, model2, stat, ...)
{

    valStats1 <- validationStats(model1)
    valStats1$model <- class(model1)
    valStats2 <- validationStats(model2)
    valStats2$model <- class(model2)

    valStats <- rbind(valStats1, valStats2[cohort %in% valStats1$cohort])
    valStats <- valStats[statistic == stat & !isSummary, ]

    plot <- ggplot(valStats, aes(x=cohort, y=estimate)) +
        geom_col(aes(fill=model), position='dodge')

    return(plot)

})