R/multivariateAnalysisHeatmap.R

In chapmandu2/CancerCellLineModelling: Modelling Drug Response In Cancer Cell Lines

#' multivariateAnalysisHeatmap
#'
#'  Make a heatmap of the results from a list of train objects generated by multivariateAnalysisXXX functions
#'
#' @param caret_res_list list of caret train objects
#' @param n number of top features to plot
#'
#' @return draws a heatmap to the current device
#' @export
#' @import pheatmap
#'
multivariateAnalysisHeatmap <- function(caret_res_list, n=10) {

    #apply VarImp function on all caret train objects in the list and combine output into data frame
    res_df <- lapply(caret_res_list, function(x) multivariateAnalysisVarImp(caret_res=x, glmnet_caret = TRUE, varImp_scale = TRUE)) %>%
        dplyr::bind_rows() %>%
        dplyr::mutate(col_id=paste(model_type, resp_var, sep='_'))

    #pivot the data using tidyr - feature id by endpoint
    res_ct <- res_df  %>%
        dplyr::select(feature_name, Imp, col_id) %>%
        tidyr::spread(col_id, Imp)

    #turn data frame into matrix
    res_mat <- res_ct %>%
        dplyr::select(-feature_name) %>%
        as.matrix()
    rownames(res_mat) <- res_ct$feature_name

    #determine the top features for each analysis and combine
    res_topfeatures <- res_df %>% dplyr::filter(idx <= n) %>% dplyr::select(feature_name) %>% dplyr::distinct()

    #generate a data frame of column information for plotting
    column_info <- res_df %>% dplyr::select(col_id, model_type, resp_var) %>% dplyr::distinct() %>% as.data.frame()
    rownames(column_info) <- column_info$col_id
    column_info <- column_info[,-1]

    #generate the heatmap
    pheatmap(res_mat[res_topfeatures$feature_name,],
             cluster_rows = TRUE, cluster_cols = FALSE,
             annotation_col = column_info,
             show_colnames = FALSE
    )

}