view mqppep_anova_script.Rmd @ 0:ba62d93a9ef5 draft

planemo upload for repository https://github.com/galaxyproteomics/tools-galaxyp/tree/master/tools/mqppep commit 3a7b3609d6e514c9e8f980ecb684960c6b2252fe

---
title: "MaxQuant Phosphoproteomic Enrichment Pipeline ANOVA/KSEA"
author:
- "Nick Graham^[ORCiD 0000-0002-6811-1941, University of Southern California: Los Angeles, CA, US]"
- "Larry Cheng^[ORCiD 0000-0002-6922-6433, Rutgers School of Graduate Studies: New Brunswick, NJ, US]"
- "Art Eschenlauer^[ORCiD 0000-0002-2882-0508, University of Minnesota: Minneapolis, Minnesota, US]"
date:
- "May 28, 2018"
- "; revised June 23, 2022"
output:
  pdf_document:
    toc: true
    toc_depth: 3
    keep_tex: true
header-includes:
  - \usepackage{longtable}
  - \newcommand\T{\rule{0pt}{2.6ex}}       % Top strut
  - \newcommand\B{\rule[-1.2ex]{0pt}{0pt}} % Bottom strut
params:
  alphaFile:            "test-data/alpha_levels.tabular"
  inputFile:            "test-data/test_input_for_anova.tabular"
  preprocDb:            "test-data/test_input_for_anova.sqlite"
  kseaAppPrepDb:        !r c(":memory:", "test-data/mqppep.sqlite")[2]
  show_toc:             true
  firstDataColumn:      "^Intensity[^_]"
  imputationMethod:     !r c("group-median", "median", "mean", "random")[1]
  meanPercentile:       1
  sdPercentile:         1.0
  regexSampleNames:     "\\.\\d+[A-Z]$"
  regexSampleGrouping:  "\\d+"
  imputedDataFilename:  "test-data/limbo/imputedDataFilename.txt"
  imputedQNLTDataFile:  "test-data/limbo/imputedQNLTDataFile.txt"
  anovaKseaMetadata:    "test-data/limbo/anovaKseaMetadata.txt"
  oneWayManyCategories: !r c("aov", "kruskal.test", "oneway.test")[1]
  oneWayTwoCategories:  !r c("aov", "kruskal.test", "oneway.test")[3]
  kseaCutoffStatistic:  !r c("p.value", "FDR")[2]
  kseaCutoffThreshold:  !r c( 0.1, 0.05)[2]
  kseaMinKinaseCount:   1
  intensityHeatmapRows: 75
---
<!--
  kseaCutoffStatistic:  !r c("p.value", "FDR")[2]
  kseaCutoffThreshold:  !r c(0.05, 0.1)[1]

  alphaFile:            "test-data/alpha_levels.tabular"
  inputFile:            "test-data/test_input_for_anova.tabular"
  preprocDb:            "test-data/test_input_for_anova.sqlite"
  kseaAppPrepDb:        !r c(":memory:", "test-data/mqppep.sqlite")[2]

  alphaFile:            "test-data/alpha_levels.tabular"
  inputFile:            "test-data/UT_phospho_ST_sites.preproc.tabular"
  preprocDb:            "test-data/UT_phospho_ST_sites.preproc.sqlite"
  kseaAppPrepDb:        !r c(":memory:", "test-data/UT_phospho_ST_sites.ksea.sqlite")[2]

  alphaFile:            "test-data/alpha_levels.tabular"
  inputFile:            "test-data/pY_Sites_NancyDu.txt.ppep_intensities.ppep_map.preproc.tabular"
  preprocDb:            "test-data/pY_Sites_NancyDu.txt.ppep_intensities.ppep_map.preproc.sqlite"
  kseaAppPrepDb:        !r c(":memory:", "test-data/pST_Sites_NancyDu.ksea.sqlite")[2]

  alphaFile:            "test-data/alpha_levels.tabular"
  inputFile:            "test-data/pST_Sites_NancyDu.txt.preproc.tabular"
  preprocDb:            "test-data/pST_Sites_NancyDu.txt.preproc.sqlite"
  kseaAppPrepDb:        !r c(":memory:", "test-data/pST_Sites_NancyDu.ksea.sqlite")[2]

  inputFile:            "test-data/density_failure.preproc_tab.tabular"
  kseaAppPrepDb:        !r c(":memory:", "mqppep.sqlite")[2]
  latex_document: default
-->
```{r setup, include = FALSE}
#ref for debugging: https://yihui.org/tinytex/r/#debugging
options(tinytex.verbose = TRUE)

# ref for parameterizing Rmd document: https://stackoverflow.com/a/37940285
# ref for top and bottom struts: https://tex.stackexchange.com/a/50355
knitr::opts_chunk$set(echo = FALSE, fig.dim = c(9, 10))

# freeze the random number generator so the same results will be produced
#  from run to run
set.seed(28571)

### LIBRARIES
library(gplots)
library(DBI)
library(RSQLite)
# Suppress "Warning: no DISPLAY variable so Tk is not available"
suppressWarnings(suppressMessages(library(sqldf)))

# required but not added to search list:
# - DBI
# - RSQLite
# - ggplot2
# - knitr
# - latex2exp
# - preprocessCore
# - reshape2
# - vioplot

### CONSTANTS

const_parfin <- par("fin")
const_boxplot_fill <- "grey94"
const_stripchart_cex <- 0.5
const_stripsmall_cex <-
  sqrt(const_stripchart_cex * const_stripchart_cex / 2)
const_stripchart_jitter <- 0.3
const_write_debug_files <- FALSE
const_table_anchor_bp <- "bp"
const_table_anchor_ht <- "ht"
const_table_anchor_p <- "p"
const_table_anchor_tbp <- "tbp"


const_ksea_astrsk_kinases    <- 1
const_ksea_nonastrsk_kinases <- 2
const_ksea_all_kinases       <- 3

const_log10_e <- log10(exp(1))

### FUNCTIONS

# from `demo(error.catching)`
##' Catch *and* save both errors and warnings, and in the case of
##' a warning, also keep the computed result.
##'
##' @title tryCatch both warnings (with value) and errors
##' @param expr an \R expression to evaluate
##' @return a list with 'value' and 'warning', where
##'   'value' may be an error caught.
##' @author Martin Maechler;
##' Copyright (C) 2010-2012  The R Core Team
try_catch_w_e <- function(expr) {
  wrn <- NULL
  # warning handler
  w_handler <- function(w) {
    wrn <<- w
    invokeRestart("muffleWarning")
  }
  list(
    value = withCallingHandlers(
      tryCatch(
        expr,
        error = function(e) e
      ),
      warning = w_handler
    ),
    warning = wrn
  )
}


write_debug_file <- function(s) {
  if (const_write_debug_files) {
    s_path <- sprintf("test-data/%s.txt", deparse(substitute(s)))
    print(sprintf("DEBUG writing file %s", spath))
    write.table(
      s,
      file = s_path,
      sep = "\t",
      col.names = TRUE,
      row.names = TRUE,
      quote = FALSE
    )
  }
}

# ref: http://adv-r.had.co.nz/Environments.html
# "When creating your own environment, note that you should set its parent
#   environment to be the empty environment. This ensures you don't
#   accidentally inherit objects from somewhere else."
# Caution: this prevents `with(my_env, expr)` from working when `expr`
#   contains anything from the global environment, even operators!
#   Hence, `x <- 1; get("x", new_env())` fails by design.
new_env <- function() {
  new.env(parent = emptyenv())
}

### numerical/statistical helper functions

any_nan <- function(x) {
  !any(x == "NaN")
}

# determine standard deviation of quantile to impute
sd_finite <- function(x) {
  ok <- is.finite(x)
  sd(x[ok])
}

anova_func <- function(x, grouping_factor, one_way_f) {
  subject <- data.frame(
    intensity = x
  )
  x_aov <-
    one_way_f(
      formula = intensity ~ grouping_factor,
      data = subject
      )
  pvalue <-
    if (identical(one_way_f, aov))
      summary(x_aov)[[1]][["Pr(>F)"]][1]
    else
      pvalue <- x_aov$p.value
  pvalue
}


### LaTeX functions

latex_collapsed_vector <- function(collapse_string, v, underscore_whack = TRUE) {
  v_sub <- if (underscore_whack) gsub("_", "\\\\_", v) else v
  cat(
    paste0(
      v_sub,
      collapse = collapse_string
      )
    )
}

latex_itemized_collapsed <- function(collapse_string, v, underscore_whack = TRUE) {
  cat("\\begin{itemize}\n\\item ")
  latex_collapsed_vector(collapse_string, v, underscore_whack)
  cat("\n\\end{itemize}\n")
}

latex_itemized_list <- function(v, underscore_whack = TRUE) {
  latex_itemized_collapsed("\n\\item ", v, underscore_whack)
}

latex_enumerated_collapsed <- function(collapse_string, v, underscore_whack = TRUE) {
  cat("\\begin{enumerate}\n\\item ")
  latex_collapsed_vector(collapse_string, v, underscore_whack)
  cat("\n\\end{enumerate}\n")
}

latex_enumerated_list <- function(v) {
  latex_enumerated_collapsed("\n\\item ", v)
}

latex_table_row <- function(v, extra = "", underscore_whack = TRUE) {
  latex_collapsed_vector(" & ", v, underscore_whack)
  cat(extra)
  cat(" \\\\\n")
}

# Use this like print.data.frame, from which it is adapted:
data_frame_latex <-
  function(
    x,
    ...,
    # digits to pass to format.data.frame
    digits = NULL,
    # TRUE -> right-justify columns; FALSE -> left-justify
    right = TRUE,
    # maximumn number of rows to print
    max = NULL,
    # string with justification of each column
    justification = NULL,
    # TRUE to center on page
    centered = TRUE,
    # optional caption
    caption = NULL,
    # h(inline); b(bottom); t (top) or p (separate page)
    anchor = "h",
    # set underscore_whack to TRUE to escape underscores
    underscore_whack = TRUE
  ) {
    if (is.null(justification))
      justification <-
        Reduce(
          f = paste,
          x = rep_len(if (right) "r" else "l", length(colnames(x)))
          )
    n <- length(rownames(x))
    if (length(x) == 0L) {
      cat(
        sprintf(
          # if n is one, use singular 'row', else use plural 'rows'
          ngettext(
            n,
            "data frame with 0 columns and %d row",
            "data frame with 0 columns and %d rows"
            ),
          n
          ),
        "\n",
        sep = ""
        )
    } else if (n == 0L) {
      cat("0 rows for:\n")
      latex_itemized_list(
        v = names(x),
        underscore_whack = underscore_whack
        )
    } else {
      if (is.null(max))
        max <- getOption("max.print", 99999L)
      if (!is.finite(max))
        stop("invalid 'max' / getOption(\"max.print\"): ",
          max)
      omit <- (n0 <- max %/% length(x)) < n
      m <- as.matrix(
        format.data.frame(
          if (omit) x[seq_len(n0), , drop = FALSE] else x,
          digits = digits,
          na.encode = FALSE
          )
        )
      cat(
        # h(inline); b(bottom); t (top) or p (separate page)
        paste0("\\begin{table}[", anchor, "]\n")
        )
      if (!is.null(caption))
        cat(paste0(" \\caption{", caption, "}"))
      if (centered) cat("\\centering\n")
      cat(
        paste(
          " \\begin{tabular}{",
          justification,
          "}\n",
          sep = ""
          )
        )
      # ref: https://tex.stackexchange.com/a/50353
      #   Describes use of \rule{0pt}{3ex}
      if (!is.null(caption))
        cat("\\B \\\\ \\hline\\hline\n")
      # ref for top and bottom struts: https://tex.stackexchange.com/a/50355
      latex_table_row(
        v = colnames(m),
        extra = "\\T\\B",
        underscore_whack = underscore_whack
        )
      cat("\\hline\n")
      for (i in seq_len(length(m[, 1]))) {
        latex_table_row(
        v = m[i, ],
        underscore_whack = underscore_whack
        )
      }
      cat(
        paste(
          " \\end{tabular}",
          "\\end{table}",
          sep = "\n"
          )
        )
      if (omit)
        cat(" [ reached 'max' / getOption(\"max.print\") -- omitted",
          n - n0, "rows ]\n")
    }
    invisible(x)
  }

hypersub <-
  function(s) {
    hyper <- tolower(s)
    hyper <- gsub("[^a-z0-9]+", "-", hyper)
    hyper <- gsub("[-]+",       "-", hyper)
    hyper <- sub("^[-]",        "",  hyper)
    hyper <- sub("[-]$",        "",  hyper)
    return(hyper)
  }

subsection_header <-
  function(s) {
    hyper <- hypersub(s)
    cat(
      sprintf(
        "\\hypertarget{%s}\n{\\subsection{%s}\\label{%s}}\n",
        hyper, s, hyper
        )
      )
  }

subsubsection_header <-
  function(s) {
    hyper <- hypersub(s)
    cat(
      sprintf(
        "\\hypertarget{%s}\n{\\subsubsection{%s}\\label{%s}}\n",
        hyper, s, hyper
        )
      )
  }

### SQLite functions

ddl_exec <- function(db, sql) {
  discard <- DBI::dbExecute(conn = db, statement = sql)
  if (FALSE && discard != 0) {
    need_newpage <- TRUE
    if (need_newpage) {
      need_newpage <<- FALSE
      cat("\\newpage\n")
    }
    o_file <- stdout()
    cat("\n\\begin{verbatim}\n")
    cat(sql, file = o_file)
    cat(sprintf("\n%d rows affected by DDL\n", discard), file = o_file)
    cat("\n\\end{verbatim}\n")
  }
}

dml_no_rows_exec <- function(db, sql) {
  discard <- DBI::dbExecute(conn = db, statement = sql)
  if (discard != 0) {
    need_newpage <- TRUE
    if (need_newpage) {
      need_newpage <<- FALSE
      cat("\\newpage\n")
    }
    cat("\n\\begin{verbatim}\n")
    o_file <- stdout()
    cat(sql, file = o_file)
    cat(sprintf("\n%d rows affected by DML\n", discard), file = o_file)
    cat("\n\\end{verbatim}\n")
  }
}

### KSEA functions and helpers

# Adapted from KSEAapp::KSEA.Scores to allow retrieval of:
# - maximum log2(FC)
ksea_scores <- function(

  # For human data, typically, ksdata = KSEAapp::ksdata
  ksdata,

  # Input data file having columns:
  # - Protein     : abbreviated protein name
  # - Gene        : HUGO gene name
  # - Peptide     : peptide sequence without indications of phosphorylation
  # - Reside.Both : position(s) of phosphorylation within Gene sequence
  #                 - First letter designates AA that is modified
  #                 - Numbers indicate position within Gene
  #                 - Multiple values are separated by semicolons
  #   - p         : p-value
  #   - FC        : fold-change
  px,

  # A binary input of TRUE or FALSE, indicating whether or not to include
  #   NetworKIN predictions
  networkin,

  # A numeric value between 1 and infinity setting the minimum NetworKIN
  #   score (can be left out if networkin = FALSE)
  networkin_cutoff

) {
  if (length(grep(";", px$Residue.Both)) == 0) {
    # There are no Residue.Both entries having semicolons, so new is
    #   simply px except two columns are renamed and a column is added
    #   for log2(abs(fold-change))
    new <- px
    colnames(new)[c(2, 4)] <- c("SUB_GENE", "SUB_MOD_RSD")
    new$log2_fc <- log2(abs(as.numeric(as.character(new$FC))))
    new <- new[complete.cases(new$log2_fc), ]
  } else {
    # Split each row having semicolons in Residue.Both into rows that are
    #   duplicated in all respects except that each row has a single
    #   member of the set "split-on-semicolon-Residue.Both"
    px_double <- px[grep(";", px$Residue.Both), ]
    residues <- as.character(px_double$Residue.Both)
    residues <- as.matrix(residues, ncol = 1)
    split <- strsplit(residues, split = ";")
    # x gets count of residues in each row,
    #   i.e., 1 + count of semicolons
    x <- sapply(split, length)
    # Here is the set of split rows
    px_single <- data.frame(
      Protein      = rep(px_double$Protein, x),
      Gene         = rep(px_double$Gene,    x),
      Peptide      = rep(px_double$Peptide, x),
      Residue.Both = unlist(split),
      p            = rep(px_double$p,       x),
      FC           = rep(px_double$FC,      x)
      )
    # new first gets the split rows
    new <- px[-grep(";", px$Residue.Both), ]
    # to new, append the rows that didn't need splitting in the first place
    new <- rbind(new, px_single)
    # map Gene         to SUB_GENE
    # map Residue.Both to SUB_MOD_RSD
    colnames(new)[c(2, 4)] <- c("SUB_GENE", "SUB_MOD_RSD")
    # Eliminate any non-positive values to prevent introduction of
    #   infinite or NaN values
    new[(0 <= new$log2_fc), "log2_fc"] <- NA
    # Because of preceding step, there is no need for abs in the next line
    new$log2_fc <- log2(as.numeric(as.character(new$FC)))
    # Convert any illegal values from NaN to NA
    new[is.nan(new$log2_fc), "log2_fc"] <- NA
    # Eliminate rows having missing values (e.g., non-imputed data)
    new <- new[complete.cases(new$log2_fc), ]
  }
  if (networkin == TRUE) {
    # When NetworKIN is true, filter on NetworKIN.cutoff which includes
    #   PhosphoSitePlus data *because its networkin_score is set to Inf*
    ksdata_filtered <- ksdata[grep("[a-z]", ksdata$Source), ]
    ksdata_filtered <- ksdata_filtered[
      (ksdata_filtered$networkin_score >= networkin_cutoff), ]
  } else {
    # Otherwise, simply use PhosphSitePlus rows
    ksdata_filtered <- ksdata[
      grep("PhosphoSitePlus", ksdata$Source), ]
  }
  # Join the two data.frames on common columns SUB_GENE and SUB_MOD_RSD
  #   colnames of ksdata_filtered:
  #     "KINASE" "KIN_ACC_ID" "GENE" "KIN_ORGANISM" "SUBSTRATE" "SUB_GENE_ID"
  #     "SUB_ACC_ID" "SUB_GENE" "SUB_ORGANISM" "SUB_MOD_RSD" "SITE_GRP_ID"
  #     "SITE_...7_AA" "networkin_score" "Source"
  #   colnames of new:
  #     "Protein" "SUB_GENE" "Peptide" "SUB_MOD_RSD" "p" "FC" "log2_fc"
  # Equivalent to:
  #   SELECT a.*. b.Protein, b.Peptide, b.p, b.FC, b.log2_fc
  #     FROM ksdata_filtered a
  #       INNER JOIN new b
  #         ON a.SUB_GENE = b.SUB_GENE
  #           AND a.SUB_MOD_RSD = b.SUB_MOD_RSD
  ksdata_dataset <- base::merge(ksdata_filtered, new)
  #   colnames of ksdata_dataset:
  #     "KINASE"      "KIN_ACC_ID"   "GENE"       "KIN_ORGANISM" "SUBSTRATE"
  #     "SUB_GENE_ID" "SUB_ACC_ID"   "SUB_GENE"   "SUB_ORGANISM" "SUB_MOD_RSD"
  #     "SITE_GRP_ID" "SITE_...7_AA" "networkin_score"  "Source" "Protein"
  #     "Peptide"     "p"            "FC"         "log2_fc" (uniprot_no_isoform)
  # Re-order dataset; prior to accounting for isoforms
  ksdata_dataset <- ksdata_dataset[order(ksdata_dataset$GENE), ]
  # Extract non-isoform accession in UniProtKB
  ksdata_dataset$uniprot_no_isoform <- sapply(
    ksdata_dataset$KIN_ACC_ID,
    function(x) unlist(strsplit(as.character(x), split = "-"))[1]
    )
  # Discard previous results while selecting interesting columns ...
  ksdata_dataset_abbrev <- ksdata_dataset[, c(5, 1, 2, 16:19, 14)]
  # Column names are now:
  #   "GENE"       "SUB_GENE"        "SUB_MOD_RSD"    "Peptide" "p"
  #   "FC" "log2_fc" "Source"
  # Make column names human-readable
  colnames(ksdata_dataset_abbrev) <- c(
    "Kinase.Gene", "Substrate.Gene", "Substrate.Mod", "Peptide", "p",
    "FC", "log2FC", "Source"
    )
  # SELECT * FROM ksdata_dataset_abbrev
  #   ORDER BY Kinase.Gene, Substrate.Gene, Substrate.Mod, p
  ksdata_dataset_abbrev <-
    ksdata_dataset_abbrev[
      order(
        ksdata_dataset_abbrev$Kinase.Gene,
        ksdata_dataset_abbrev$Substrate.Gene,
        ksdata_dataset_abbrev$Substrate.Mod,
        ksdata_dataset_abbrev$p),
      ]
  # First aggregation step to account for multiply phosphorylated peptides
  #   and differing peptide sequences; the goal here is to combine results
  #   for all measurements of the same substrate.
  # SELECT  `Kinase.Gene`, `Substrate.Gene`, `Substrate.Mod`,
  #         `Source`, avg(log2FC) AS log2FC
  #   FROM  ksdata_dataset_abbrev
  #   GROUP BY `Kinase.Gene`, `Substrate.Gene`, `Substrate.Mod`,
  #         `Source`
  #   ORDER BY `Kinase.Gene`;
  # in two steps:
  # (1) compute average log_2(fold-change)
  ksdata_dataset_abbrev <- aggregate(
    log2FC ~ Kinase.Gene + Substrate.Gene + Substrate.Mod + Source,
    data = ksdata_dataset_abbrev,
    FUN = mean
    )
  # (2) order by Kinase.Gene
  ksdata_dataset_abbrev <-
    ksdata_dataset_abbrev[order(ksdata_dataset_abbrev$Kinase.Gene), ]
  # SELECT  `Kinase.Gene`, count(*)
  #   FROM  ksdata_dataset_abbrev
  #  GROUP BY `Kinase.Gene`;
  # in two steps:
  # (1) Extract the list of Kinase.Gene names
  kinase_list <- as.vector(ksdata_dataset_abbrev$Kinase.Gene)
  # (2) Convert to a named list of counts of kinases in ksdata_dataset_abrev,
  #   named by Kinase.Gene
  kinase_list <- as.matrix(table(kinase_list))
  # Second aggregation step to account for all substrates per kinase
  # CREATE TABLE mean_fc
  #   AS
  # SELECT  `Kinase.Gene`, avg(log2FC) AS log2FC
  #   FROM  ksdata_dataset_abbrev
  #   GROUP BY `Kinase.Gene`
  mean_fc <- aggregate(
    log2FC ~ Kinase.Gene,
    data = ksdata_dataset_abbrev,
    FUN = mean
    )
  # mean_fc columns: "Kinase.Gene", "log2FC"
  if (FALSE) {
    # I need to re-think this; I was trying to find the most-represented
    #   peptide, but that horse has already left the barn
    # SELECT  `Kinase.Gene`, max(abs(log2FC)) AS log2FC
    #   FROM  ksdata_dataset_abbrev
    #   GROUP BY `Kinase.Gene`
    max_fc <- aggregate(
      log2FC ~ Kinase.Gene,
      data = ksdata_dataset_abbrev,
      FUN = function(r) max(abs(r))
      )
  }

  # Create column 3: mS
  mean_fc$m_s <- mean_fc[, 2]
  # Create column 4: Enrichment
  mean_fc$enrichment <- mean_fc$m_s / abs(mean(new$log2_fc, na.rm = TRUE))
  # Create column 5: m, count of substrates
  mean_fc$m <- kinase_list
  # Create column 6: z-score
  mean_fc$z_score <- (
    (mean_fc$m_s - mean(new$log2_fc, na.rm = TRUE)) *
      sqrt(mean_fc$m)) / sd(new$log2_fc, na.rm = TRUE)
  # Create column 7: p-value, deduced from z-score
  mean_fc$p_value <- pnorm(-abs(mean_fc$z_score))
  # Create column 8: FDR, deduced by Benjamini-Hochberg adustment from p-value
  mean_fc$fdr <- p.adjust(mean_fc$p_value, method = "fdr")

  # Remove log2FC column, which is duplicated as mS
  mean_fc <- mean_fc[order(mean_fc$Kinase.Gene), -2]
  # Correct the column names which we had to hack because of the linter...
  colnames(mean_fc) <- c(
    "Kinase.Gene", "mS", "Enrichment", "m", "z.score", "p.value", "FDR"
    )
  return(mean_fc)
}

low_fdr_barplot <- function(
  rslt,
  i_cntrst,
  i,
  a_level,
  b_level,
  fold_change,
  caption
) {
  rslt_score_list_i <- rslt$score_list[[i]]
  if (!is.null(rslt_score_list_i)) {
    rslt_score_list_i_nrow <- nrow(rslt_score_list_i)
    k <- data.frame(
      contrast = as.integer(i_cntrst),
      a_level = rep.int(a_level, rslt_score_list_i_nrow),
      b_level = rep.int(b_level, rslt_score_list_i_nrow),
      kinase_gene = rslt_score_list_i$Kinase.Gene,
      mean_log2_fc = rslt_score_list_i$mS,
      enrichment = rslt_score_list_i$Enrichment,
      substrate_count = rslt_score_list_i$m,
      z_score = rslt_score_list_i$z.score,
      p_value = rslt_score_list_i$p.value,
      fdr = rslt_score_list_i$FDR
    )
    selector <- switch(
      ksea_cutoff_statistic,
      "FDR" = {
        k$fdr
        },
      "p.value" = {
        k$p_value
        },
      stop(
        sprintf(
          "Unexpected cutoff statistic %s rather than 'FDR' or 'p.value'",
          ksea_cutoff_statistic
          )
        )
      )

    k <- k[selector < ksea_cutoff_threshold, ]

    if (nrow(k) > 1) {
      op <- par(mai = c(1, 1.5, 0.4, 0.4))
      numeric_z_score <- as.numeric(k$z_score)
      z_score_order <- order(numeric_z_score)
      kinase_name <- k$kinase_gene
      long_caption <-
        sprintf(
          "Kinase z-score, %s < %s, %s",
          ksea_cutoff_statistic,
          ksea_cutoff_threshold,
          caption
          )
      my_cex_caption <- 65.0 / max(65.0, nchar(long_caption))
      cat("\n\\clearpage\n")
      barplot(
        height = numeric_z_score[z_score_order],
        border = NA,
        xpd = FALSE,
        cex.names = 1.0,
        cex.axis = 1.0,
        main = long_caption,
        cex.main = my_cex_caption,
        names.arg = kinase_name[z_score_order],
        horiz = TRUE,
        srt = 45,
        las = 1)
      par(op)
    }
  }
}

# note that this adds elements to the global variable `ksea_asterisk_hash`

low_fdr_print <- function(
  rslt,
  i_cntrst,
  i,
  a_level,
  b_level,
  fold_change,
  caption
) {
  rslt_score_list_i <- rslt$score_list[[i]]
  if (!is.null(rslt_score_list_i)) {
    rslt_score_list_i_nrow <- nrow(rslt_score_list_i)
    k <- contrast_ksea_scores <- data.frame(
      contrast = as.integer(i_cntrst),
      a_level = rep.int(a_level, rslt_score_list_i_nrow),
      b_level = rep.int(b_level, rslt_score_list_i_nrow),
      kinase_gene = rslt_score_list_i$Kinase.Gene,
      mean_log2_fc = rslt_score_list_i$mS,
      enrichment = rslt_score_list_i$Enrichment,
      substrate_count = rslt_score_list_i$m,
      z_score = rslt_score_list_i$z.score,
      p_value = rslt_score_list_i$p.value,
      fdr = rslt_score_list_i$FDR
    )

    selector <- switch(
      ksea_cutoff_statistic,
      "FDR" = {
        k$fdr
        },
      "p.value" = {
        k$p_value
        },
      stop(
        sprintf(
          "Unexpected cutoff statistic %s rather than 'FDR' or 'p.value'",
          ksea_cutoff_statistic
          )
        )
      )

    k <- k[selector < ksea_cutoff_threshold, ]
    # save kinase names to ksea_asterisk_hash
    for (kinase_name in k$kinase_gene) {
      ksea_asterisk_hash[[kinase_name]] <- 1
    }

    db_write_table_overwrite <- (i_cntrst < 2)
    db_write_table_append    <- !db_write_table_overwrite
    RSQLite::dbWriteTable(
      conn = db,
      name = "contrast_ksea_scores",
      value = contrast_ksea_scores,
      append = db_write_table_append
      )
    selector <- switch(
      ksea_cutoff_statistic,
      "FDR" = {
        contrast_ksea_scores$fdr
        },
      "p.value" = {
        contrast_ksea_scores$p_value
        },
      stop(
        sprintf(
          "Unexpected cutoff statistic %s rather than 'FDR' or 'p.value'",
          ksea_cutoff_statistic
          )
        )
      )
    output_df <- contrast_ksea_scores[
      selector < ksea_cutoff_threshold,
      c("kinase_gene", "mean_log2_fc", "enrichment", "substrate_count",
        "z_score", "p_value", "fdr")
      ]
    output_order <- with(output_df, order(mean_log2_fc, kinase_gene))
    output_df <- output_df[output_order, ]
    colnames(output_df) <-
      c(
        colnames(output_df)[1],
        colnames(output_df)[2],
        "enrichment",
        "m_s",
        "z_score",
        "p_value",
        "fdr"
      )
    output_df$fdr <- sprintf("%0.4f", output_df$fdr)
    output_df$p_value <- sprintf("%0.2e", output_df$p_value)
    output_df$z_score <- sprintf("%0.2f", output_df$z_score)
    output_df$m_s <- sprintf("%d", output_df$m_s)
    output_df$enrichment <- sprintf("%0.2f", output_df$enrichment)
    output_ncol <- ncol(output_df)
    colnames(output_df) <-
      c(
        "Kinase",
        "\\(\\overline{\\log_2 (|\\text{fold-change}|)}\\)",
        "Enrichment",
        "Substrates",
        "z-score",
        "p-value",
        "FDR"
      )
    selector <- switch(
      ksea_cutoff_statistic,
      "FDR" = {
        rslt$score_list[[i]]$FDR
        },
      "p.value" = {
        rslt$score_list[[i]]$p.value
        },
      stop(
        sprintf(
          "Unexpected cutoff statistic %s rather than 'FDR' or 'p.value'",
          ksea_cutoff_statistic
          )
        )
      )
    if (sum(selector < ksea_cutoff_threshold) > 0) {
      math_caption <- gsub("{", "\\{", caption,      fixed = TRUE)
      math_caption <- gsub("}", "\\}", math_caption, fixed = TRUE)
      data_frame_latex(
        x = output_df,
        justification = "l c c c c c c",
        centered = TRUE,
        caption = sprintf(
          "\\text{%s}, %s < %s",
          math_caption,
          ksea_cutoff_statistic,
          ksea_cutoff_threshold
          ),
        anchor = const_table_anchor_p
        )
    } else {
      cat(
        sprintf(
          "\\break
          No kinases had
          \\(\\text{%s}_\\text{enrichment} < %s\\)
          for contrast %s\\hfill\\break\n",
          ksea_cutoff_statistic,
          ksea_cutoff_threshold,
          caption
        )
      )
    }
  }
}

# create_breaks is a helper for ksea_heatmap
create_breaks <- function(merged_scores) {
  if (min(merged_scores, na.rm = TRUE) < -1.6) {
    breaks_neg <- seq(-1.6, 0, length.out = 30)
    breaks_neg <-
      append(
        seq(min(merged_scores, na.rm = TRUE), -1.6, length.out = 10),
        breaks_neg
        )
    breaks_neg <- sort(unique(breaks_neg))
  } else {
    breaks_neg <- seq(-1.6, 0, length.out = 30)
  }
  if (max(merged_scores, na.rm = TRUE) > 1.6) {
    breaks_pos <- seq(0, 1.6, length.out = 30)
    breaks_pos <-
      append(
        breaks_pos,
        seq(1.6, max(merged_scores, na.rm = TRUE),
        length.out = 10)
        )
    breaks_pos <- sort(unique(breaks_pos))
  } else {
    breaks_pos <- seq(0, 1.6, length.out = 30)
  }
  breaks_all <- unique(append(breaks_neg, breaks_pos))
  mycol_neg <-
    gplots::colorpanel(n = length(breaks_neg),
               low = "blue",
               high = "white")
  mycol_pos <-
    gplots::colorpanel(n = length(breaks_pos) - 1,
               low = "white",
               high = "red")
  mycol <- unique(append(mycol_neg, mycol_pos))
  color_breaks <- list(breaks_all, mycol)
  return(color_breaks)
}

# draw_kseaapp_summary_heatmap is a helper function for ksea_heatmap
draw_kseaapp_summary_heatmap <- function(
    x,
    sample_cluster,
    merged_asterisk,
    my_cex_row,
    color_breaks,
    margins,
    ...
) {
  merged_scores <- x
  if (!is.matrix(x)) {
    cat(
      paste0(
        "No plot because \\texttt{typeof(x)} is '",
        typeof(x),
        "' rather than 'matrix'.\n\n"
        )
      )
  } else if (nrow(x) < 2) {
    cat("No plot because matrix x has ", nrow(x), " rows.\n\n")
    cat("\\begin{verbatim}\n")
    str(x)
    cat("\\end{verbatim}\n")
  } else if (ncol(x) < 2) {
    cat("No plot because matrix x has ", ncol(x), " columns.\n\n")
    cat("\\begin{verbatim}\n")
    str(x)
    cat("\\end{verbatim}\n")
  } else {
    gplots::heatmap.2(
      x            = merged_scores,
      Colv         = sample_cluster,
      scale        = "none",
      cellnote     = merged_asterisk,
      notecol      = "white",
      cexCol       = 0.9,
      # Heuristically assign size of row labels
      cexRow       = min(1.0, ((3 * my_cex_row) ^ 1.7) / 2.25),
      srtCol       = 45,
      srtRow       = 45,
      notecex      = 3 * my_cex_row,
      col          = color_breaks[[2]],
      density.info = "none",
      trace        = "none",
      breaks       = color_breaks[[1]],
      lmat         = rbind(c(0, 3), c(2, 1), c(0, 4)),
      lhei         = c(0.4, 8.0, 1.1),
      lwid         = c(0.5, 3),
      key          = FALSE,
      margins      = margins,
      ...
    )
  }
}

# Adapted from KSEAapp::KSEA.Heatmap
ksea_heatmap <- function(
  # the data frame outputs from the KSEA.Scores() function, in list format
  score_list,
  # a character vector of all the sample names for heatmap annotation:
  # - the names must be in the same order as the data in score_list
  # - please avoid long names, as they may get cropped in the final image
  sample_labels,
  # character string of either "p.value" or "FDR" indicating the data column
  #   to use for marking statistically significant scores
  stats,
  # a numeric value between 0 and infinity indicating the min. number of
  #   substrates a kinase must have to be included in the heatmap
  m_cutoff,
  # a numeric value between 0 and 1 indicating the p-value/FDR cutoff
  #   for indicating significant kinases in the heatmap
  p_cutoff =
    stop("argument 'p_cutoff' is required for function 'ksea_heatmap'"),
  # a binary input of TRUE or FALSE, indicating whether or not to perform
  #   hierarchical clustering of the sample columns
  sample_cluster,
  # a binary input of TRUE or FALSE, indicating whether or not to export
  #   the heatmap as a .png image into the working directory
  export = FALSE,
  # bottom and right margins; adjust as needed if contrast names are too long
  margins = c(6, 20),
  # print which kinases?
  # - Mandatory argument, must be one of const_ksea_.*_kinases
  which_kinases,
  # additional arguments to gplots::heatmap.2, such as:
  # - main: main title of plot
  # - xlab: x-axis label
  # - ylab: y-axis label
  ...
) {
  filter_m <- function(dataset, m_cutoff) {
    filtered <- dataset[(dataset$m >= m_cutoff), ]
    return(filtered)
  }
  score_list_m <- lapply(score_list, function(...) filter_m(..., m_cutoff))
  for (i in seq_len(length(score_list_m))) {
    names <- colnames(score_list_m[[i]])[c(2:7)]
    colnames(score_list_m[[i]])[c(2:7)] <-
      paste(names, i, sep = ".")
  }
  master <-
    Reduce(
      f = function(...) {
        base::merge(..., by = "Kinase.Gene", all = FALSE)
      },
      x = score_list_m
      )

  row.names(master) <- master$Kinase.Gene
  columns <- as.character(colnames(master))
  merged_scores <- as.matrix(master[, grep("z.score", columns), drop = FALSE])
  colnames(merged_scores) <- sample_labels
  merged_stats <- as.matrix(master[, grep(stats, columns)])
  asterisk <- function(mtrx, p_cutoff) {
    new <- data.frame()
    for (i in seq_len(nrow(mtrx))) {
      for (j in seq_len(ncol(mtrx))) {
        my_value <- mtrx[i, j]
        if (!is.na(my_value) && my_value < p_cutoff) {
          new[i, j] <- "*"
        } else {
          new[i, j] <- ""
        }
      }
    }
    return(new)
  }
  merged_asterisk <- as.matrix(asterisk(merged_stats, p_cutoff))

  # begin hack to print only significant rows
  asterisk_rows <- rowSums(merged_asterisk == "*") > 0
  all_rows <- rownames(merged_stats)
  names(asterisk_rows) <- all_rows
  non_asterisk_rows <- names(asterisk_rows[asterisk_rows == FALSE])
  asterisk_rows <- names(asterisk_rows[asterisk_rows == TRUE])
  merged_scores_asterisk <- merged_scores[names(asterisk_rows), ]
  merged_scores_non_asterisk <- merged_scores[names(non_asterisk_rows), ]
  # end hack to print only significant rows

  row_list <- list()
  row_list[[const_ksea_astrsk_kinases]] <- asterisk_rows
  row_list[[const_ksea_all_kinases]] <- all_rows
  row_list[[const_ksea_nonastrsk_kinases]] <- non_asterisk_rows

  i <- which_kinases
  my_row_names <- row_list[[i]]
  scrs <- merged_scores[my_row_names, ]
  stts <- merged_stats[my_row_names, ]
  merged_asterisk <- as.matrix(asterisk(stts, p_cutoff))

  color_breaks <- create_breaks(scrs)
  plot_height <- nrow(scrs) ^ 0.55
  plot_width <- ncol(scrs) ^ 0.7
  my_cex_row <- 0.25 * 16 / plot_height
  if (export == "TRUE") {
    png(
      "KSEA.Merged.Heatmap.png",
      width = plot_width * 300,
      height = 2 * plot_height * 300,
      res = 300,
      pointsize = 14
    )
  }
  draw_kseaapp_summary_heatmap(
    x               = scrs,
    sample_cluster  = sample_cluster,
    merged_asterisk = merged_asterisk,
    my_cex_row      = my_cex_row,
    color_breaks    = color_breaks,
    margins         = margins
  )
  if (export == "TRUE") {
    dev.off()
  }
  return(my_row_names)
}

# helper for heatmaps of phosphopeptide intensities

draw_intensity_heatmap <-
  function(
    m,                              # matrix with rownames already formatted
    cutoff,                         # cutoff used by hm_heading_function
    hm_heading_function,            # construct and cat heading from m and cutoff
    hm_main_title,                  # main title for plot (drawn below heading)
    suppress_row_dendrogram = TRUE, # set to false to show dendrogram
    max_peptide_count               # experimental:
          = intensity_hm_rows,      #   values of 50 and 75 worked well
    ...                             # passthru parameters for heatmap
  ) {
    peptide_count <- 0
    # emit the heading for the heatmap
    if (hm_heading_function(m, cutoff)) {
      peptide_count <- min(max_peptide_count, nrow(m))
      if (nrow(m) > 1) {
        m_margin <- m[peptide_count:1, ]
        # Margin setting was heuristically derived
        margins <-
          c(0.5, # col
            max(80, sqrt(nchar(rownames(m_margin)))) * 5 / 16  # row
          )
        }
      if (nrow(m) > 1) {
        tryCatch(
          {
            old_oma <- par("oma")
            par(cex.main = 0.6)
            # Heuristically determined character size adjustment formula
            char_contractor <-
              250000 / (
                max(4500, (nchar(rownames(m_margin)))^2) * intensity_hm_rows
                )
            heatmap(
              m[peptide_count:1, ],
              Rowv = if (suppress_row_dendrogram) NA else NULL,
              Colv = NA,
              cexRow = char_contractor,
              cexCol = char_contractor * 50 / max_peptide_count,
              scale = "row",
              margins = margins,
              main =
                "Unimputed, unnormalized log(intensities)",
              xlab = "",
              las = 1,
              ...
              )
          },
          error = function(e) {
            cat(
              sprintf(
                "\nCould not draw heatmap, possibly because of too many missing values.  Internal message: %s\n",
                e$message
                )
              )
            },
          finally = par(old_oma)
        )
      }
    }
    return(peptide_count)
  }
```

```{r, echo = FALSE, fig.dim = c(9, 10), results = 'asis'}
cat("\\listoftables\n")
```
# Purpose

To perform for phosphopeptides:

- imputation of missing values,
- quantile normalization,
- ANOVA (using the R stats::`r params$oneWayManyCategories` function), and
- KSEA (Kinase-Substrate Enrichment Analysis) using code adapted from the CRAN `KSEAapp` package to search for kinase substrates from the following databases:
  - PhosphoSitesPlus [https://www.phosphosite.org](https://www.phosphosite.org)
  - The Human Proteome Database [http://hprd.org](http://hprd.org)
  - NetworKIN [http://networkin.science/](http://networkin.science/)
  - Phosida [http://pegasus.biochem.mpg.de/phosida/help/motifs.aspx](http://pegasus.biochem.mpg.de/phosida/help/motifs.aspx)

```{r include = FALSE}

### GLOBAL VARIABLES

# parameters for KSEA

ksea_cutoff_statistic <- params$kseaCutoffStatistic
ksea_cutoff_threshold <- params$kseaCutoffThreshold
ksea_min_kinase_count <- params$kseaMinKinaseCount

ksea_heatmap_titles <- list()
ksea_heatmap_titles[[const_ksea_astrsk_kinases]] <-
  sprintf(
    "Summary for all kinases enriched in one or more contrasts at %s < %s",
    ksea_cutoff_statistic,
    ksea_cutoff_threshold
    )
ksea_heatmap_titles[[const_ksea_all_kinases]] <-
  "Summary figure for all contrasts and all kinases"
ksea_heatmap_titles[[const_ksea_nonastrsk_kinases]] <-
  sprintf(
    "Summary for all kinases not enriched at %s < %s in any contrast",
    ksea_cutoff_statistic,
    ksea_cutoff_threshold
    )
# hash to hold names of significantly enriched kinases
ksea_asterisk_hash <- new_env()

# READ PARAMETERS (mostly)

intensity_hm_rows <- params$intensityHeatmapRows
# Input Filename
input_file <- params$inputFile

# First data column - ideally, this could be detected via regexSampleNames,
#   but for now leave it as is.
first_data_column <- params$firstDataColumn
fdc_is_integer <- is.integer(first_data_column)
if (fdc_is_integer) {
  first_data_column <- as.integer(params$firstDataColumn)
}

# False discovery rate adjustment for ANOVA
#  Since pY abundance is low, set to 0.10 and 0.20 in addition to 0.05
val_fdr <-
  read.table(file = params$alphaFile, sep = "\t", header = FALSE, quote = "")

if (
  ncol(val_fdr) != 1 ||
  sum(!is.numeric(val_fdr[, 1])) ||
  sum(val_fdr[, 1] < 0) ||
  sum(val_fdr[, 1] > 1)
) {
  stop("alphaFile should be one column of numbers within the range [0.0,1.0]")
}
val_fdr <- val_fdr[, 1]

#Imputed Data filename
imputed_data_filename <- params$imputedDataFilename
imp_qn_lt_data_filenm <- params$imputedQNLTDataFile
anova_ksea_mtdt_file  <- params$anovaKseaMetadata

```

```{r echo = FALSE}
# Imputation method, should be one of
#   "random", "group-median", "median", or "mean"
imputation_method <- params$imputationMethod

# Selection of percentile of logvalue data to set the mean for random number
#   generation when using random imputation
mean_percentile <- params$meanPercentile / 100.0

# deviation adjustment-factor for random values; real number.
sd_percentile <- params$sdPercentile

# Regular expression of Sample Names, e.g., "\\.(\\d+)[A-Z]$"
regex_sample_names <- params$regexSampleNames

# Regular expression to extract Sample Grouping from Sample Name;
#   if error occurs, compare sample_treatment_levels vs. sample_name_matches
#   to see if groupings/pairs line up
#   e.g., "(\\d+)"
regex_sample_grouping <- params$regexSampleGrouping

one_way_all_categories_fname <- params$oneWayManyCategories
one_way_all_categories <- try_catch_w_e(
  match.fun(one_way_all_categories_fname))
if (!is.function(one_way_all_categories$value)) {
  write("fatal error for parameter oneWayManyCategories:", stderr())
  write(one_way_all_categories$value$message,             stderr())
  if (sys.nframe() > 0) quit(save = "no", status = 1)
  stop("Cannot continue. Goodbye.")
}
one_way_all_categories <- one_way_all_categories$value

one_way_two_categories_fname <- params$oneWayManyCategories
one_way_two_categories <- try_catch_w_e(
  match.fun(one_way_two_categories_fname))
if (!is.function(one_way_two_categories$value)) {
  cat("fatal error for parameter oneWayTwoCategories: \n")
  cat(one_way_two_categories$value$message, fill = TRUE)
  if (sys.nframe() > 0) quit(save = "no", status = 1)
  stop("Cannot continue. Goodbye.")
}
one_way_two_categories <- one_way_two_categories$value

preproc_db       <- params$preprocDb
ksea_app_prep_db <- params$kseaAppPrepDb
result <- file.copy(
  from      = preproc_db,
  to        = ksea_app_prep_db,
  overwrite = TRUE
  )
if (!result) {
  write(
    sprintf(
      "fatal error copying initial database '%s' to output '%s'",
      preproc_db,
      ksea_app_prep_db,
    ),
    stderr()
  )
  if (sys.nframe() > 0) quit(save = "no", status = 1)
  stop("Cannot continue. Goodbye.")
}
```

```{r echo = FALSE}
### READ DATA

# read.table reads a file in table format and creates a data frame from it.
#   - note that `quote = ""` means that quotation marks are treated literally.
full_data <- read.table(
  file = input_file,
  sep = "\t",
  header = TRUE,
  quote = "",
  check.names = FALSE
  )
```

# Extract Sample Names and Treatment Levels

Column names parsed from input file are shown in Table 1; sample names and treatment levels, in Table 2.

```{r echo = FALSE, results = 'asis'}

data_column_indices <- grep(first_data_column, names(full_data), perl = TRUE)

if (!fdc_is_integer) {
  if (length(data_column_indices) > 0) {
    first_data_column <- data_column_indices[1]
  } else {
    stop(paste("failed to convert firstDataColumn:", first_data_column))
  }
}

cat(
  sprintf(
    paste(
      "\n\nThe input data file has peptide-intensity data for each sample",
      "in one of columns %d through %d.\n\n"
      ),
    min(data_column_indices),
    max(data_column_indices)
    )
  )

# Write column names as a LaTeX enumerated list.
column_name_df <- data.frame(
  column = seq_len(length(colnames(full_data))),
  name = paste0("\\verb@", colnames(full_data), "@")
  )
data_frame_latex(
  x = column_name_df,
  justification = "l l",
  centered = TRUE,
  caption = "Input data column names",
  anchor = const_table_anchor_bp,
  underscore_whack = FALSE
  )

```

```{r echo = FALSE, results = 'asis'}
quant_data <- full_data[first_data_column:length(full_data)]
quant_data[quant_data == 0] <- NA
rownames(quant_data) <- rownames(full_data) <- full_data$Phosphopeptide
# Extract factors and trt-replicates using regular expressions.
# Typically:
#   regex_sample_names    is "\\.\\d+[A-Z]$"
#   regex_sample_grouping is "\\d+"
# This would distinguish trt-replicates by terminal letter [A-Z]
#   in sample names and group them into trts by the preceding digits.
#   e.g.:
#     group .1A .1B .1C into group 1;
#     group .2A .2B .2C, into group 2;
#     etc.
m <- regexpr(regex_sample_names, colnames(quant_data), perl = TRUE)
sample_name_matches <- regmatches(names(quant_data), m)
colnames(quant_data) <- sample_name_matches

write_debug_file(quant_data)

rx_match <- regexpr(regex_sample_grouping, sample_name_matches, perl = TRUE)
sample_treatment_levels <- as.factor(regmatches(sample_name_matches, rx_match))
number_of_samples <- length(sample_name_matches)
sample_treatment_df <- data.frame(
  level = sample_treatment_levels,
  sample = sample_name_matches
  )
data_frame_latex(
  x = sample_treatment_df,
  justification = "rp{0.2\\linewidth} lp{0.3\\linewidth}",
  centered = TRUE,
  caption = "Treatment levels",
  anchor = const_table_anchor_tbp,
  underscore_whack = FALSE
  )
```

```{r echo = FALSE, results = 'asis'}
cat("\\newpage\n")
```

## Are the log-transformed sample distributions similar?

```{r echo = FALSE, fig.dim = c(9, 5.5), results = 'asis'}

quant_data[quant_data == 0] <- NA  #replace 0 with NA
quant_data_log <- log10(quant_data)

rownames(quant_data_log) <- rownames(quant_data)
colnames(quant_data_log) <- sample_name_matches

write_debug_file(quant_data_log)

# data visualization
old_par <- par(
  mai = par("mai") + c(0.5, 0, 0, 0)
)
# ref: https://r-charts.com/distribution/add-points-boxplot/
# Vertical plot
boxplot(
  quant_data_log
, las = 1
, col = const_boxplot_fill
, ylab = latex2exp::TeX("$log_{10}$(peptide intensity)")
, xlab = "Sample"
)
par(old_par)



cat("\n\n\n")
cat("\n\n\n")

```

```{r echo = FALSE, fig.align = "left", fig.dim = c(9, 4), results = 'asis'}
if (nrow(quant_data_log) > 1) {
  quant_data_log_stack <- stack(quant_data_log)
  ggplot2::ggplot(quant_data_log_stack, ggplot2::aes(x = values)) +
    ggplot2::xlab(latex2exp::TeX("$log_{10}$(peptide intensity)")) +
    ggplot2::ylab("Probability density") +
    ggplot2::geom_density(ggplot2::aes(group = ind, colour = ind), na.rm = TRUE)
} else {
  cat("No density plot because there are too few peptides.\n\n")
}
```

## Globally, are peptide intensities are approximately unimodal?

<!--
# bquote could be used as an alternative to latex2exp::TeX below particularly
#   and when plotting math expressions generally, at the expense of mastering
#   another syntax, which hardly seems worthwhile when I need to use TeX
#   elsewhere; here's an introduction to bquote:
#   https://www.r-bloggers.com/2018/03/math-notation-for-r-plot-titles-expression-and-bquote/
-->
```{r echo = FALSE, fig.align = "left", fig.dim = c(9, 5), results = 'asis'}

# identify the location of missing values
fin <- is.finite(as.numeric(as.matrix(quant_data_log)))

logvalues <- as.numeric(as.matrix(quant_data_log))[fin]
logvalues_density <- density(logvalues)
plot(
  x = logvalues_density,
  main = latex2exp::TeX(
    "Smoothed estimated probability density vs. $log_{10}$(peptide intensity)"
    ),
  xlab = latex2exp::TeX("$log_{10}$(peptide intensity)"),
  ylab = "Probability density"
  )
hist(
  x = as.numeric(as.matrix(quant_data_log)),
  xlim = c(min(logvalues_density$x), max(logvalues_density$x)),
  breaks = 100,
  main = latex2exp::TeX("Frequency vs. $log_{10}$(peptide intensity)"),
  xlab = latex2exp::TeX("$log_{10}$(peptide intensity)")
)
```

## Distribution of standard deviations of $log_{10}(\text{intensity})$, ignoring missing values

```{r echo = FALSE, fig.align = "left", fig.dim = c(9, 5), results = 'asis'}
# determine quantile
q1 <- quantile(logvalues, probs = mean_percentile)[1]

# 1 = row of matrix (ie, phosphopeptide)
sds <- apply(quant_data_log, 1, sd_finite)
if (sum(!is.na(sds)) > 2) {
  plot(
    density(sds, na.rm = TRUE)
  , main = "Smoothed estimated probability density vs. std. deviation"
  , sub = "(probability estimation made with Gaussian smoothing)"
  , ylab = "Probability density"
  )
} else {
  cat(
    "At least two non-missing values are required to plot",
    "probability density.\n\n"
    )
}

```

```{r echo = FALSE}
# Determine number of cells to impute
temp <- quant_data[is.na(quant_data)]

# Determine number of values to impute
number_to_impute <- length(temp)

# Determine percent of missing values
pct_missing_values <-
  round(length(temp) / (length(logvalues) + length(temp)) * 100)
```

```{r echo = FALSE}

# prep for trt-median based imputation

```
# Impute Missing Values

```{r echo = FALSE}

imp_smry_pot_peptides_before <- nrow(quant_data_log)
imp_smry_missing_values_before   <- number_to_impute
imp_smry_pct_missing      <- pct_missing_values

```

```{r echo = FALSE}
#Determine number of cells to impute

```
```{r echo = FALSE}

# Identify which values are missing and need to be imputed
ind <- which(is.na(quant_data), arr.ind = TRUE)

```
```{r echo = FALSE, results = 'asis'}

# Apply imputation
switch(
  imputation_method
, "group-median" = {
    quant_data_imp <- quant_data
    imputation_method_description <-
      paste("Substitute missing value with",
        "median peptide-intensity for sample group.\n"
        )
    sample_level_integers <- as.integer(sample_treatment_levels)
    # Take the accurate ln(x+1) because the data are log-normally distributed
    #   and because median can involve an average of two measurements.
    quant_data_imp <- log1p(quant_data_imp)
    for (i in seq_len(length(levels(sample_treatment_levels)))) {
      # Determine the columns for this factor-level
      level_cols <- i == sample_level_integers
      # Extract those columns
      lvlsbst <- quant_data_imp[, level_cols, drop = FALSE]
      # assign to ind the row-column pairs corresponding to each NA
      ind <- which(is.na(lvlsbst), arr.ind = TRUE)
      # No group-median exists if there is only one sample
      #   a given ppep has no measurement; otherwise, proceed.
      if (ncol(lvlsbst) > 1) {
        the_centers <-
          apply(lvlsbst, 1, median, na.rm = TRUE)
        for (j in seq_len(nrow(lvlsbst))) {
          for (k in seq_len(ncol(lvlsbst))) {
            if (is.na(lvlsbst[j, k])) {
              lvlsbst[j, k] <- the_centers[j]
            }
          }
        }
        quant_data_imp[, level_cols] <- lvlsbst
      }
    }
    # Take the accurate e^x - 1 to match scaling of original input.
    quant_data_imp <- round(expm1(quant_data_imp_ln <- quant_data_imp))
    good_rows <- !is.na(rowMeans(quant_data_imp))
  }
, "median" = {
    quant_data_imp <- quant_data
    imputation_method_description <-
      paste("Substitute missing value with",
        "median peptide-intensity across all sample classes.\n"
        )
    # Take the accurate ln(x+1) because the data are log-normally distributed
    #   and because median can involve an average of two measurements.
    quant_data_imp <- log1p(quant_data_imp)
    quant_data_imp[ind] <- apply(quant_data_imp, 1, median, na.rm = TRUE)[ind[, 1]]
    # Take the accurate e^x - 1 to match scaling of original input.
    quant_data_imp <- round(expm1(quant_data_imp_ln <- quant_data_imp))
    good_rows <- !is.nan(rowMeans(quant_data_imp))
  }
, "mean" = {
    quant_data_imp <- quant_data
    imputation_method_description <-
      paste("Substitute missing value with",
        "geometric-mean peptide intensity across all sample classes.\n"
        )
    # Take the accurate ln(x+1) because the data are log-normally distributed,
    #   so arguments to mean should be previously transformed.
    #   this will have to be
    quant_data_imp <- log1p(quant_data_imp)
    # Assign to NA cells the mean for the row
    quant_data_imp[ind] <- apply(quant_data_imp, 1, mean, na.rm = TRUE)[ind[, 1]]
    # Take the accurate e^x - 1 to match scaling of original input.
    quant_data_imp <- round(expm1(quant_data_imp_ln <- quant_data_imp))
    good_rows <- !is.nan(rowMeans(quant_data_imp))
  }
, "random" = {
    quant_data_imp <- quant_data
    m1 <- median(sds, na.rm = TRUE) * sd_percentile #sd to be used is the median sd
    # If you want results to be reproducible, you will want to call
    #   base::set.seed before calling stats::rnorm
    imputation_method_description <-
      paste("Substitute each missing value with random intensity",
        sprintf(
          "random intensity $N \\sim (%0.2f, %0.2f)$.\n",
          q1, m1
          )
        )
    cat(sprintf("mean_percentile (from input parameter) is %2.0f\n\n",
      100 * mean_percentile))
    cat(sprintf("sd_percentile (from input parameter) is %0.2f\n\n",
      sd_percentile))
    quant_data_imp[ind] <-
      10 ^ rnorm(number_to_impute, mean = q1, sd = m1)
    quant_data_imp_ln <- log1p(quant_data_imp)
    good_rows <- !is.nan(rowMeans(quant_data_imp))
  }
)
quant_data_imp_log10 <- quant_data_imp_ln * const_log10_e

if (length(good_rows) < 1) {
  print("ERROR: Cannot impute data; there are no good rows!")
  return(-1)
  }
```

```{r echo = FALSE, results = 'asis'}
cat("\\quad\n\nImputation method:\n\n\n", imputation_method_description)
```

```{r echo = FALSE}

imp_smry_pot_peptides_after <- sum(good_rows)
imp_smry_rejected_after  <- sum(!good_rows)
imp_smry_missing_values_after   <- sum(is.na(quant_data_imp[good_rows, ]))
```
```{r echo = FALSE, results = 'asis'}
# ref: http://www1.maths.leeds.ac.uk/latex/TableHelp1.pdf
tabular_lines_fmt <- paste(
  "\\begin{table}[hb]", # h(inline); b(bottom); t (top) or p (separate page)
  " \\caption{Imputation Results}",
  " \\centering", # \centering centers the table on the page
  " \\begin{tabular}{l c c c}",
  "  \\hline\\hline",
  "  \\  & potential peptides   & missing values & rejected",
  "    peptides \\\\ [0.5ex]",
  "  \\hline",
  "  before imputation & %d     & %d (%d\\%s)    &    \\\\",
  "  after imputation  & %d     & %d             & %d \\\\ [1ex]",
  "  \\hline",
  " \\end{tabular}",
  #" \\label{table:nonlin}", # may be used to refer this table in the text
  "\\end{table}",
  sep = "\n"
  )
tabular_lines <-
  sprintf(
    tabular_lines_fmt,
    imp_smry_pot_peptides_before,
    imp_smry_missing_values_before,
    imp_smry_pct_missing,
    "%",
    imp_smry_pot_peptides_after,
    imp_smry_missing_values_after,
    imp_smry_rejected_after
    )
cat(tabular_lines)
```
```{r echo = FALSE}


# Zap rows where imputation was ineffective
full_data         <- full_data        [good_rows, ]
quant_data        <- quant_data       [good_rows, ]

quant_data_imp <- quant_data_imp[good_rows, ]
write_debug_file(quant_data_imp)
quant_data_imp_good_rows <- quant_data_imp

write_debug_file(quant_data_imp_good_rows)
```

```{r echo = FALSE, results = 'asis'}

can_plot_before_after_imp <- TRUE
d_combined <-
  as.numeric(
    as.matrix(
      log10(quant_data_imp)
      )
    )
d_original <-
  as.numeric(
    as.matrix(
      log10(quant_data_imp[!is.na(quant_data)])
      )
    )

if (sum(!is.na(d_original)) > 2) {
  d_original <- density(d_original)
} else {
  can_plot_before_after_imp <- FALSE
}
if (can_plot_before_after_imp && sum(is.na(d_combined)) < 1) {
  d_combined <- density(d_combined)
} else {
  can_plot_before_after_imp <- FALSE
}

if (sum(is.na(quant_data)) > 0) {
  # There ARE missing values
  d_imputed <-
    as.numeric(
      as.matrix(
        log10(quant_data_imp[is.na(quant_data)])
        )
      )
  if (can_plot_before_after_imp && sum(is.na(d_imputed)) < 1) {
    d_imputed <- (density(d_imputed))
  } else {
    can_plot_before_after_imp <- FALSE
  }
} else {
  # There are NO missing values
  d_imputed <- d_combined
}

```

```{r echo = FALSE, fig.dim = c(9, 5.5), results = 'asis'}
zero_sd_rownames <-
  rownames(quant_data_imp)[
    is.na((apply(quant_data_imp, 1, sd, na.rm = TRUE)) == 0)
  ]

if (length(zero_sd_rownames) >= nrow(quant_data_imp)) {
  stop("All peptides have zero standard deviation.  Cannot continue.")
}
if (length(zero_sd_rownames) > 0) {
  cat(
    sprintf("%d peptides with zero variance were removed from statistical consideration",
      length(zero_sd_rownames)
    )
  )
  zap_named_rows <- function(df, nms) {
    return(df[!(row.names(df) %in% nms), ])
  }
  quant_data_imp <- zap_named_rows(quant_data_imp, zero_sd_rownames)
  quant_data     <- zap_named_rows(quant_data,     zero_sd_rownames)
  full_data      <- zap_named_rows(full_data,      zero_sd_rownames)
}

if (sum(is.na(quant_data)) > 0) {
  cat("\\leavevmode\\newpage\n")
  # data visualization
  old_par <- par(
    mai = par("mai") + c(0.5, 0, 0, 0)
  )
  # Copy quant data to x
  x <- quant_data
  # x gets to have values of:
  #  - NA for observed values
  #  - 1 for missing values
  x[is.na(x)] <- 0
  x[x > 1] <- NA
  x[x == 0] <- 1
  # Log-transform imputed data
  # update variable because rows may have been eliminated from quant_data_imp
  quant_data_imp_log10 <- log10(quant_data_imp)

  write_debug_file(quant_data_imp_log10)

  # Set blue_dots to log of quant data or NA for NA quant data
  blue_dots <- log10(quant_data)
  # Set red_dots to log of imputed data or NA for observed quant data
  red_dots <- quant_data_imp_log10 * x

  count_red <- sum(!is.na(red_dots))
  count_blue <- sum(!is.na(blue_dots))
  ylim_save <- ylim <- c(
    min(red_dots, blue_dots, na.rm = TRUE),
    max(red_dots, blue_dots, na.rm = TRUE)
    )
  show_stripchart <-
    50 > (count_red + count_blue) / length(sample_name_matches)
  if (show_stripchart) {
    boxplot_sub <- "Light blue = data before imputation; Red = imputed data"
  } else {
    boxplot_sub <- ""
  }

  # Vertical plot
  colnames(blue_dots) <- sample_name_matches
  boxplot(
      blue_dots
    , las = 1 # "always horizontal"
    , col = const_boxplot_fill
    , ylim = ylim
    , main = "Peptide intensities after eliminating unusable peptides"
    , sub = boxplot_sub
    , xlab = "Sample"
    , ylab = latex2exp::TeX("$log_{10}$(peptide intensity)")
    )

  if (show_stripchart) {
    # Points
    # ref: https://r-charts.com/distribution/add-points-boxplot/
    # NA values are not plotted
    stripchart(
      blue_dots,                 # Data
      method = "jitter",          # Random noise
      jitter = const_stripchart_jitter,
      pch = 19,                   # Pch symbols
      cex = const_stripsmall_cex, # Size of symbols reduced
      col = "lightblue",          # Color of the symbol
      vertical = TRUE,            # Vertical mode
      add = TRUE                  # Add it over
      )
    stripchart(
      red_dots,                   # Data
      method = "jitter",          # Random noise
      jitter = const_stripchart_jitter,
      pch = 19,                   # Pch symbols
      cex = const_stripsmall_cex, # Size of symbols reduced
      col = "red",                # Color of the symbol
      vertical = TRUE,            # Vertical mode
      add = TRUE                  # Add it over
      )

  }
  if (TRUE) {
    # show measured values in blue on left half-violin plot
    cat("\\leavevmode\n\\quad\n\n\\quad\n\n")
    vioplot::vioplot(
      x = lapply(blue_dots, function(x) x[!is.na(x)]),
      col = "lightblue1",
      side = "left",
      plotCentre = "line",
      ylim = ylim_save,
      main = "Distributions of observed and imputed data",
      sub = "Light blue = observed data; Pink = imputed data",
      xlab = "Sample",
      ylab = latex2exp::TeX("$log_{10}$(peptide intensity)")
      )
    red_violins <- lapply(red_dots, function(x) x[!is.na(x)])
    cols_to_delete <- c()
    for (ix in seq_len(length(red_violins))) {
      if (length(red_violins[[ix]]) < 1) {
        cols_to_delete <- c(cols_to_delete, ix)
      }
    }
    # destroy any unimputable columns
    if (!is.null(cols_to_delete)) {
      red_violins <- red_violins[-cols_to_delete]
    }
    # plot imputed values in red on right half-violin plot
    vioplot::vioplot(
      x = red_violins,
      col = "lightpink1",
      side = "right",
      plotCentre = "line",
      add = TRUE
      )
  }

  par(old_par)

  # density plot
  cat("\\leavevmode\n\n\n\n\n\n\n")
  if (can_plot_before_after_imp) {
    ylim <- c(
      0,
      max(
        if (is.list(d_combined)) d_combined$y else d_combined,
        if (is.list(d_original)) d_original$y else d_original,
        if (is.list(d_imputed)) d_imputed$y else d_imputed,
        na.rm = TRUE
      )
    )
    plot(
      d_combined,
      ylim = ylim,
      sub =
        paste(
          "Blue = data before imputation; Red = imputed data;",
          "Black = combined"
          ),
      main = "Density of peptide intensity before and after imputation",
      xlab = latex2exp::TeX("$log_{10}$(peptide intensity)"),
      ylab = "Probability density"
    )
    lines(d_original, col = "blue")
    lines(d_imputed, col = "red")
  } else {
    cat(
      "There are too few points to plot the density of peptide intensity",
      "before and after imputation."
      )
  }
  cat("\\leavevmode\\newpage\n")
}
```

# Perform Quantile Normalization

The excellent `normalize.quantiles` function from
*[the `preprocessCore` Bioconductor package](http://bioconductor.org/packages/release/bioc/html/preprocessCore.html)*
performs "quantile normalization" as described Bolstad *et al.* (2003),
DOI *[10.1093/bioinformatics/19.2.185](https://doi.org/10.1093%2Fbioinformatics%2F19.2.185)*
and *its supplementary material [http://bmbolstad.com/misc/normalize/normalize.html](http://bmbolstad.com/misc/normalize/normalize.html)*,
i.e., it assumes that the goal is to detect
subtle differences among grossly similar samples (having similar distributions)
by equailzing intra-quantile quantitations.
Unfortunately, one software library upon which it depends
*[suffers from a concurrency defect](https://support.bioconductor.org/p/122925/#9135989)*
that requires that a specific, non-concurrent version of the library be
installed.  The installation command equivalent to what was used to install the library to produce the results presented here is:
```
    conda install bioconductor-preprocesscore openblas=0.3.3
```


<!--
# Apply quantile normalization using preprocessCore::normalize.quantiles
# ---
# tool repository: http://bioconductor.org/packages/release/bioc/html/preprocessCore.html
#   except this: https://support.bioconductor.org/p/122925/#9135989
#   says to install it like this:
#     ```
#     BiocManager::install("preprocessCore", configure.args="--disable-threading", force = TRUE, lib=.libPaths()[1])
#     ```
# conda installation (necessary because of a bug in recent openblas):
#   conda install bioconductor-preprocesscore openblas=0.3.3
# ...
# ---
# normalize.quantiles {preprocessCore} --  Quantile Normalization
#
# Description:
#   Using a normalization based upon quantiles, this function normalizes a
#     matrix of probe level intensities.
#
#   THIS FUNCTIONS WILL HANDLE MISSING DATA (ie NA values), based on the
#     assumption that the data is missing at random.
#
# Usage:
#   normalize.quantiles(x, copy = TRUE, keep.names = FALSE)
#
# Arguments:
#
#   - x: A matrix of intensities where each column corresponds to a chip and each row is a probe.
#
#   - copy: Make a copy of matrix before normalizing. Usually safer to work with a copy,
#       but in certain situations not making a copy of the matrix, but instead normalizing
#       it in place will be more memory friendly.
#
#   - keep.names: Boolean option to preserve matrix row and column names in output.
#
# Details:
#   This method is based upon the concept of a quantile-quantile plot extended to n dimensions.
#     No special allowances are made for outliers. If you make use of quantile normalization
#     please cite Bolstad et al, Bioinformatics (2003).
#
#   This functions will handle missing data (ie NA values), based on
#     the assumption that the data is missing at random.
#
#   Note that the current implementation optimizes for better memory usage
#     at the cost of some additional run-time.
#
# Value: A normalized matrix.
#
# Author: Ben Bolstad, bmbolstad.com
#
# References
#
#   - Bolstad, B (2001) Probe Level Quantile Normalization of High Density Oligonucleotide
#       Array Data. Unpublished manuscript http://bmbolstad.com/stuff/qnorm.pdf
#
#   - Bolstad, B. M., Irizarry R. A., Astrand, M, and Speed, T. P. (2003) A Comparison of
#       Normalization Methods for High Density Oligonucleotide Array Data Based on Bias
#       and Variance. Bioinformatics 19(2), pp 185-193. DOI 10.1093/bioinformatics/19.2.185
#       http://bmbolstad.com/misc/normalize/normalize.html
# ...
-->
```{r echo = FALSE, results = 'asis'}

if (nrow(quant_data_imp) > 0) {
  quant_data_imp_qn <- preprocessCore::normalize.quantiles(
     as.matrix(quant_data_imp), keep.names = TRUE
   )
} else {
  quant_data_imp_qn <- as.matrix(quant_data_imp)
}

quant_data_imp_qn <- as.data.frame(quant_data_imp_qn)

write_debug_file(quant_data_imp_qn)

quant_data_imp_qn_log <- log10(quant_data_imp_qn)

write_debug_file(quant_data_imp_qn_log)

quant_data_imp_qn_ls <- t(scale(t(log10(quant_data_imp_qn))))

sel <- apply(quant_data_imp_qn_ls, 1, any_nan)
quant_data_imp_qn_ls2 <- quant_data_imp_qn_ls

quant_data_imp_qn_ls2 <- quant_data_imp_qn_ls2[which(sel), ]
quant_data_imp_qn_ls2 <- as.data.frame(quant_data_imp_qn_ls2)

quant_data_imp_qn_ls <- as.data.frame(quant_data_imp_qn_ls)

write_debug_file(quant_data_imp_qn_ls)
write_debug_file(quant_data_imp_qn_ls2)

# Create data.frame used by ANOVA analysis
data_table_imp_qn_lt <- cbind(full_data[1:9], quant_data_imp_qn_log)
```

<!-- ACE insertion begin -->
## Are normalized, imputed, log-transformed sample distributions similar?

```{r echo = FALSE, fig.dim = c(9, 5.5), results = 'asis'}

# Save unimputed quant_data_log for plotting below
unimputed_quant_data_log <- quant_data_log

# log10 transform (after preparing for zero values,
#   which should never happen...)
quant_data_imp_qn[quant_data_imp_qn == 0] <- .000000001
quant_data_log <- log10(quant_data_imp_qn)

how_many_peptides <- nrow(quant_data_log)

if ((how_many_peptides) > 0) {
  cat(
    sprintf(
      "Intensities for %d peptides:\n\n\n",
      how_many_peptides
      )
    )
  cat("\n\n\n")


  # data visualization
  old_par <- par(
    mai = par("mai") + c(0.5, 0, 0, 0)
  , oma = par("oma") + c(0.5, 0, 0, 0)
  )
  # ref: https://r-charts.com/distribution/add-points-boxplot/
  # Vertical plot
  colnames(quant_data_log) <- sample_name_matches
  boxplot(
    quant_data_log
  , las = 1
  , col = const_boxplot_fill
  , ylab = latex2exp::TeX("$log_{10}$(peptide intensity)")
  , xlab = "Sample"
  )
  par(old_par)
} else {
  cat("There are no peptides to plot\n")
}

cat("\n\n\n")

```

```{r echo = FALSE, fig.align = "left", fig.dim = c(9, 4), results = 'asis'}
if (nrow(quant_data_log) > 1) {
  quant_data_log_stack <- stack(quant_data_log)
  ggplot2::ggplot(quant_data_log_stack, ggplot2::aes(x = values)) +
    ggplot2::xlab(latex2exp::TeX("$log_{10}$(peptide intensity)")) +
    ggplot2::ylab("Probability density") +
    ggplot2::geom_density(ggplot2::aes(group = ind, colour = ind), na.rm = TRUE)
} else {
  cat("No density plot because there are fewer than two peptides to plot.\n\n")
}
```
```{r echo = FALSE, results = 'asis'}
cat("\\leavevmode\\newpage\n")
```

# ANOVA Analysis

```{r, echo = FALSE}
# Make new data frame containing only Phosphopeptides
#   to connect preANOVA to ANOVA (connect_df)
connect_df <- data.frame(
    data_table_imp_qn_lt$Phosphopeptide
  , data_table_imp_qn_lt[, first_data_column]
  )
colnames(connect_df) <- c("Phosphopeptide", "Intensity")
```

```{r echo = FALSE, fig.dim = c(9, 10), results = 'asis'}
count_of_treatment_levels <- length(levels(sample_treatment_levels))
if (count_of_treatment_levels < 2) {
  nuke_control_sequences <-
    function(s) {
      s <- gsub("[\\]", "xyzzy_plugh", s)
      s <- gsub("[$]", "\\\\$", s)
      s <- gsub("xyzzy_plugh", "$\\\\backslash$", s)
      return(s)
    }
  cat(
    "ERROR!!!! Cannot perform ANOVA analysis",
    "(see next page)\\newpage\n"
  )
  cat(
    "ERROR: ANOVA analysis",
    "requires two or more factor levels!\n\n\n"
  )

  cat("\n\n\n\n\n")
  cat("Unparsed sample names are:\n\n\n",
    "\\begin{quote}\n",
    paste(names(quant_data_imp_qn_log), collapse = "\n\n\n"),
    "\n\\end{quote}\n\n")

  regex_sample_names <- nuke_control_sequences(regex_sample_names)

  cat("\\leavevmode\n\n\n")
  cat("Parsing rule for SampleNames is",
    "\n\n\n",
    "\\text{'",
    regex_sample_names,
    "'}\n\n\n",
    sep = ""
    )

  cat("\nParsed sample names are:\n",
    "\\begin{quote}\n",
    paste(sample_name_matches, collapse = "\n\n\n"),
    "\n\\end{quote}\n\n")

  regex_sample_grouping <- nuke_control_sequences(regex_sample_grouping)

  cat("\\leavevmode\n\n\n")
  cat("Parsing rule for SampleGrouping is",
    "\n\n\n",
    "\\text{'",
    regex_sample_grouping,
    "'}\n\n\n",
    sep = ""
    )

  cat("\n\n\n")
  cat("Sample group assignments are:\n",
    "\\begin{quote}\n",
    paste(regmatches(sample_name_matches, rx_match), collapse = "\n\n\n"),
    "\n\\end{quote}\n\n")

} else {

  p_value_data_anova_ps <-
    apply(
      quant_data_imp_qn_log,
      1,
      anova_func,
      grouping_factor = sample_treatment_levels,
      one_way_f = one_way_all_categories
      )

  p_value_data_anova_ps_fdr <-
    p.adjust(p_value_data_anova_ps, method = "fdr")
  p_value_data <- data.frame(
    phosphopeptide = full_data[, 1]
    ,
    raw_anova_p = p_value_data_anova_ps
    ,
    fdr_adjusted_anova_p = p_value_data_anova_ps_fdr
  )

  # output ANOVA file to constructed filename,
  #   e.g.    "Outputfile_pST_ANOVA_STEP5.txt"
  #   becomes "Outpufile_pST_ANOVA_STEP5_FDR0.05.txt"

  # Re-output datasets to include p-values
  metadata_plus_p <- cbind(full_data[1:9], p_value_data[, 2:3])
  write.table(
    cbind(metadata_plus_p, quant_data_imp),
    file = imputed_data_filename,
    sep = "\t",
    col.names = TRUE,
    row.names = FALSE,
    quote = FALSE
    )

  write.table(
    cbind(metadata_plus_p, quant_data_imp_qn_log),
    file = imp_qn_lt_data_filenm,
    sep = "\t",
    col.names = TRUE,
    row.names = FALSE,
    quote = FALSE
    )


  p_value_data <-
    p_value_data[order(p_value_data$fdr_adjusted_anova_p), ]

  first_page_suppress <- 1
  number_of_peptides_found <- 0
  cutoff <- val_fdr[1]
  for (cutoff in val_fdr) {
    if (number_of_peptides_found > 49) {
      cat("\\leavevmode\n\n\n")
      break
      }

    #loop through FDR cutoffs

    filtered_p <-
      p_value_data[
        which(p_value_data$fdr_adjusted_anova_p < cutoff),
        , drop = FALSE
        ]
    filtered_data_filtered <-
      quant_data_imp_qn_log[
        rownames(filtered_p),
        , drop = FALSE
        ]
    filtered_data_filtered <-
      filtered_data_filtered[
        order(filtered_p$fdr_adjusted_anova_p),
        , drop = FALSE
        ]

    # <!-- ACE insertion start -->

    if (nrow(filtered_p) && nrow(filtered_data_filtered) > 0) {
      if (first_page_suppress == 1) {
        first_page_suppress <- 0
        } else {
          cat("\\newpage\n")
        }
      if (nrow(filtered_data_filtered) > 1) {
        subsection_header(sprintf(
          "Intensity distributions for %d phosphopeptides whose adjusted p-value < %0.2f\n",
          nrow(filtered_data_filtered),
          cutoff
        ))
      } else {
        subsection_header(sprintf(
          "Intensity distribution for one phosphopeptide (%s) whose adjusted p-value < %0.2f\n",
          rownames(filtered_data_filtered)[1],
          cutoff
        ))
      }
      cat("\n\n\n")
      cat("\n\n\n")

      old_oma <- par("oma")
      old_par <- par(
        mai = (par("mai") + c(0.7, 0, 0, 0)) * c(1, 1, 0.3, 1),
        oma = old_oma * c(1, 1, 0.3, 1),
        cex.main = 0.9,
        cex.axis = 0.7,
        fin = c(9, 7.25)
        )
      # ref: https://r-charts.com/distribution/add-points-boxplot/
      # Vertical plot
      colnames(filtered_data_filtered) <- sample_name_matches
      tryCatch(
        boxplot(
          filtered_data_filtered,
          main = "Imputed, normalized intensities", # no line plot
          las = 1,
          col = const_boxplot_fill,
          ylab = latex2exp::TeX("$log_{10}$(peptide intensity)")
        ),
        error = function(e) print(e)
      )
      par(old_par)
    } else {
      cat(sprintf(
        "%s < %0.2f\n\n\n\n\n",
        "No peptides were found to have cutoff adjusted p-value",
        cutoff
      ))
    }

    if (nrow(filtered_data_filtered) > 0) {
      # Add Phosphopeptide column to anova_filtered table
      # The assumption here is that the first intensity is unique;
      #   this is a hokey assumption but almost definitely will
      #   be true in the real world unless there is a computation
      #   error upstream.
      anova_filtered_merge <- base::merge(
        x = connect_df,
        y = filtered_data_filtered,
        by.x = "Intensity",
        by.y = 1
      )
      anova_filtered_merge_order <- rownames(filtered_p)

      anova_filtered <- data.frame(
        ppep    = anova_filtered_merge$Phosphopeptide,
        intense = anova_filtered_merge$Intensity,
        data    = anova_filtered_merge[, 2:number_of_samples + 1]
      )
      colnames(anova_filtered) <-
        c("Phosphopeptide", colnames(filtered_data_filtered))

      # Merge qualitative columns into the ANOVA data
      output_table <- data.frame(anova_filtered$Phosphopeptide)
      output_table <- base::merge(
        x = output_table,
        y = data_table_imp_qn_lt,
        by.x = "anova_filtered.Phosphopeptide",
        by.y = "Phosphopeptide"
      )

      # Produce heatmap to visualize significance and the effect of imputation

      anova_filtered_merge_format <- sapply(
        X = filtered_p$fdr_adjusted_anova_p
        ,
        FUN = function(x) {
          if (x > 0.0001)
            paste0("(%0.", 1 + ceiling(-log10(x)), "f) %s")
          else
            paste0("(%0.4e) %s")
        }
      )

      cat_hm_heading <- function(m, cutoff) {
        cat("\\newpage\n")
        if (nrow(m) > intensity_hm_rows) {
          subsection_header(
            paste(
              sprintf("Heatmap for the %d most-significant peptides",
                intensity_hm_rows),
              sprintf("whose adjusted p-value < %0.2f\n", cutoff)
            )
          )
        } else {
          if (nrow(m) == 1) {
            return(FALSE)
          } else {
            subsection_header(
              paste(
                sprintf("Heatmap for %d usable peptides whose", nrow(m)),
                sprintf("adjusted p-value < %0.2f\n", cutoff)
              )
            )
          }
        }
        cat("\n\n\n")
        cat("\n\n\n")
        return(TRUE)
      }

      # construct matrix with appropriate rownames
      m <-
        as.matrix(unimputed_quant_data_log[anova_filtered_merge_order, ])
      if (nrow(m) > 0) {
        rownames_m <- rownames(m)
        rownames(m) <- sapply(
          X = seq_len(nrow(m))
          ,
          FUN = function(i) {
            sprintf(
              anova_filtered_merge_format[i],
              filtered_p$fdr_adjusted_anova_p[i],
              rownames_m[i]
            )
          }
        )
      }
      # draw the heading and heatmap
      if (nrow(m) > 0) {
          number_of_peptides_found <-
            draw_intensity_heatmap(
              m                       = m,
              cutoff                  = cutoff,
              hm_heading_function     = cat_hm_heading,
              hm_main_title           = "Unimputed, unnormalized log(intensities)",
              suppress_row_dendrogram = FALSE
            )
      }
    }
  }
}
cat("\\leavevmode\n\n\n")
```

```{r sqlite, echo = FALSE, fig.dim = c(9, 10), results = 'asis'}

if (count_of_treatment_levels > 1) {
  # Prepare two-way contrasts with adjusted p-values
  # Strategy:
  # - use imputed, log-transformed data:
  #   - remember this when computing log2(fold-change)
  # - each contrast is between a combination of trt levels
  # - for each contrast, compute samples that are members
  # - compute one-way test:
  #   - use `oneway.test` (Welch test) if numbers of samples
  #     are not equivalent between trt levels
  #   - otherwise, aov is fine but offers no advantage
  # - adjust p-value, assuming that
  #   (# of pppeps)*(# of contrasts) tests were performed

  # Each contrast is between a combination of trt levels
  m2 <- combn(
    x = seq_len(length(levels(sample_treatment_levels))),
    m = 2,
    simplify = TRUE
  )
  contrast_count <- ncol(m2)

  # For each contrast, compute samples that are members
  # - local function to construct a data.frame for each contrast
  #   - the contrast in the first "column"
  f_m2 <-
    function(cntrst, lvl1, lvl2) {
      return(
        data.frame(
          contrast = cntrst,
          level = sample_treatment_levels[
              sample_treatment_levels %in%
                levels(sample_treatment_levels)[c(lvl1, lvl2)]
            ],
          label = sample_name_matches[
              sample_treatment_levels %in%
                levels(sample_treatment_levels)[c(lvl1, lvl2)]
            ]
        )
      )
    }
  # - compute a df for each contrast
  sample_level_dfs <- lapply(
    X = 1:contrast_count,
    FUN = function(i) f_m2(i, m2[1, i], m2[2, i])
  )
  # - compute a single df for all contrasts
  combined_contrast_df <- Reduce(f = rbind, x = sample_level_dfs)

  # - dispose objects to free resources
  rm(sample_level_dfs)

  # - write the df to a DB for later join-per-contrast
  db <- RSQLite::dbConnect(RSQLite::SQLite(), ksea_app_prep_db)

  RSQLite::dbWriteTable(
    conn = db,
    name = "contrast",
    value = combined_contrast_df,
    overwrite = TRUE
  )

  # Create UK for insert
  ddl_exec(db, "
    CREATE UNIQUE INDEX IF NOT EXISTS contrast__uk__idx
      ON contrast(contrast, label);
    "
  )
  # Create indexes for join
  ddl_exec(db, "
    -- index for join in contrast_ppep_smpl_qnlt on a.label < b.label
    CREATE INDEX IF NOT EXISTS contrast__label__idx
      ON contrast(label);
    "
  )
  ddl_exec(db, "
    -- index for joining two contrast_lvl_ppep_avg_quant on contrast
    CREATE INDEX IF NOT EXISTS contrast__contrast__idx
      ON contrast(contrast);
    "
  )
  ddl_exec(db, "
    -- index for joining two contrast_lvl_ppep_avg_quant on phophospep
    CREATE INDEX IF NOT EXISTS contrast__level__idx
      ON contrast(level);
    "
  )
  # - dispose objects to free resources
  rm(combined_contrast_df)

  # Use imputed, log-transformed data
  # - remember that this was donoe when computing log2(fold-change)
  # - melt data matrix for use in later join-per-contrast
  casted <- cbind(
    data.frame(vrbl = rownames(quant_data_imp_qn_log)),
    quant_data_imp_qn_log
  )
  quant_data_imp_qn_log_melted <- reshape2::melt(
    casted,
    id.vars = "vrbl"
  )
  colnames(quant_data_imp_qn_log_melted) <-
    c("phosphopep", "sample", "quant")
  # - dispose objects to free resources
  rm(casted)

  # - write the df to a DB for use in later join-per-contrast
  RSQLite::dbWriteTable(
    conn = db,
    name = "ppep_smpl_qnlt",
    value = quant_data_imp_qn_log_melted,
    overwrite = TRUE
  )
  # Create UK for insert
  ddl_exec(db, "
    CREATE UNIQUE INDEX IF NOT EXISTS ppep_smpl_qnlt__uk__idx
      ON ppep_smpl_qnlt(phosphopep, sample);
    "
  )
  # Create index for join
  ddl_exec(db, "
    -- index for join in contrast_ppep_smpl_qnlt
    CREATE INDEX IF NOT EXISTS ppep_smpl_qnlt__sample__idx
      ON ppep_smpl_qnlt(sample);
    "
  )
  ddl_exec(db, "
    -- index for joining two contrast_lvl_ppep_avg_quant on phopho.pep
    CREATE INDEX IF NOT EXISTS ppep_smpl_qnlt__phosphopep__idx
      ON ppep_smpl_qnlt(phosphopep);
    "
  )
  # - dispose objects to free resources
  rm(quant_data_imp_qn_log_melted)

  # - drop views if exist
  ddl_exec(db, "
    -- drop view dependent on contrast_lvl_ppep_avg_quant
    DROP VIEW IF EXISTS v_contrast_log2_fc;
    "
  )
  ddl_exec(db, "
    -- drop table dependent on contrast_ppep_smpl_qnlt
    DROP TABLE IF EXISTS contrast_lvl_ppep_avg_quant;
    "
  )
  ddl_exec(db, "
    DROP TABLE IF EXISTS contrast_lvl_lvl_metadata;
    "
  )
  ddl_exec(db, "
    DROP VIEW IF EXISTS v_contrast_lvl_metadata;
    "
  )
  ddl_exec(db, "
    -- drop view dependent on contrast_ppep_smpl_qnlt
    DROP VIEW IF EXISTS v_contrast_lvl_ppep_avg_quant;
    "
  )
  ddl_exec(db, "
    DROP VIEW IF EXISTS v_contrast_lvl_lvl;
    "
  )
  ddl_exec(db, "
    -- drop view upon which other views depend
    DROP VIEW IF EXISTS contrast_ppep_smpl_qnlt;
    "
  )
  # - create view
  dml_no_rows_exec(db, "
      -- view contrast_ppep_smpl_qnlt is used for each phopshopep to
      --   compute p-value for test of trt effect for two trt levels
      CREATE VIEW contrast_ppep_smpl_qnlt
        AS
      SELECT  contrast,
              level,
              phosphopep,
              sample,
              quant
        FROM  contrast AS c,
              ppep_smpl_qnlt AS q
        WHERE q.sample = c.label
        ORDER BY contrast, level, phosphopep
      ;
    "
  )
  # - create simplification views
  dml_no_rows_exec(db, "
      CREATE VIEW v_contrast_lvl_metadata
        AS
      SELECT  contrast,
              level,
              group_concat(label, ';') AS samples
        FROM  contrast
        GROUP BY contrast, level
        /* view v_contrast_lvl_metadata is used
           to simplify creation of table contrast_lvl_lvl_metadata */
      ;
    "
  )
  dml_no_rows_exec(db, "
      CREATE VIEW v_contrast_lvl_ppep_avg_quant
        AS
      SELECT  contrast,
              level,
              phosphopep,
              avg(quant)                AS avg_quant
        FROM  contrast_ppep_smpl_qnlt
        GROUP BY contrast, level, phosphopep
        /* view v_contrast_lvl_ppep_avg_quant is used
           to simplify view v_contrast_log2_fc */
      ;
    "
  )

  # - create contrast-metadata table
  dml_no_rows_exec(db, "
      CREATE TABLE contrast_lvl_lvl_metadata
        AS
      SELECT DISTINCT
              a.contrast              AS ab_contrast,
              a.level                 AS a_level,
              b.level                 AS b_level,
              a.samples               AS a_samples,
              b.samples               AS b_samples,
              'log2(level_'||a.level||'/level_'||b.level||')'
                                      AS fc_description
        FROM  v_contrast_lvl_metadata AS a,
              v_contrast_lvl_metadata AS b
        WHERE a.contrast = b.contrast
          AND a.level > b.level
        /* view v_contrast_lvl_lvl is used
           to simplify view v_contrast_log2_fc */
      ;
    "
  )
  # - create pseudo-materialized view table
  dml_no_rows_exec(db, "
      CREATE VIEW v_contrast_lvl_lvl
        AS
      SELECT DISTINCT
              a.contrast AS ab_contrast,
              a.level    AS a_level,
              b.level    AS b_level
        FROM  contrast   AS a,
              contrast   AS b
        WHERE a.contrast = b.contrast
          AND a.level > b.level
        /* view v_contrast_lvl_lvl is used
           to simplify view v_contrast_log2_fc */
      ;
    "
  )

  # - create view to compute log2(fold-change)
  dml_no_rows_exec(db, "
      CREATE VIEW v_contrast_log2_fc
        AS
      SELECT  ab.ab_contrast                         AS contrast,
              m.a_level                              AS a_level,
              c.avg_quant                            AS a_quant,
              m.a_samples                            AS a_samples,
              ab.b_level                             AS b_level,
              d.avg_quant                            AS b_quant,
              m.b_samples                            AS b_samples,
              m.fc_description                       AS fc_description,
              3.32193 * ( d.avg_quant - c.avg_quant) AS log2_fc,
              d.phosphopep                           AS phosphopep
        FROM  contrast_lvl_lvl_metadata                  AS m,
              v_contrast_lvl_ppep_avg_quant              AS d,
              v_contrast_lvl_lvl AS ab
                INNER JOIN v_contrast_lvl_ppep_avg_quant AS c
                  ON c.contrast = ab.ab_contrast
                  AND c.level   = ab.a_level
        WHERE d.contrast    = ab.ab_contrast
          AND m.ab_contrast = ab.ab_contrast
          AND d.level       = ab.b_level
          AND d.phosphopep  = c.phosphopep
        /* view to compute log2(fold-change) */
        ;
    "
  )

  # For each contrast, compute samples that are members
  # compute one-way test:
  # - use `oneway.test` (Welch test) if numbers of samples
  #   are not equivalent between trt levels
  # - otherwise, aov is fine but offers no advantage
  for (contrast in contrast_count:2) {
    invisible(contrast)
  }
  for (contrast in 1:contrast_count) {
    contrast_df <- sqldf::sqldf(
      x = paste0("
        SELECT level, phosphopep, sample, quant
          FROM contrast_ppep_smpl_qnlt
          WHERE contrast = ", contrast, "
          ORDER BY phosphopep, level, sample
      "),
      connection = db
    )
    contrast_cast <- reshape2::dcast(
      data = contrast_df,
      formula = phosphopep ~ sample,
      value.var = "quant"
    )
    contrast_cast_ncol <- ncol(contrast_cast)
    contrast_cast_data <- contrast_cast[, 2:contrast_cast_ncol]
    contrast_cast_samples <- colnames(contrast_cast_data)

    # - order grouping_factor by order of sample columns of contrast_cast_data
    grouping_factor <- sqldf::sqldf(
      x = paste0("
        SELECT sample, level
          FROM contrast_ppep_smpl_qnlt
          WHERE contrast = ", contrast, "
          ORDER BY phosphopep, level, sample
          LIMIT ", contrast_cast_ncol - 1
      ),
      connection = db
    )
    rownames(grouping_factor) <- grouping_factor$sample
    grouping_factor <- grouping_factor[, "level", drop = FALSE]

    # - run the two-level (one-way) test
    p_value_data_contrast_ps <-
      apply(
        X = contrast_cast_data,
        MARGIN = 1, # apply to rows
        FUN = anova_func,
        grouping_factor =
          as.factor(as.numeric(grouping_factor$level)), # anova_func arg2
        one_way_f = one_way_two_categories, # anova_func arg3
        simplify = TRUE # TRUE is the default for simplify
        )
    contrast_data_adj_p_values <- p.adjust(
        p = p_value_data_contrast_ps,
        method = "fdr",
        n = length(p_value_data_contrast_ps) # this is the default, length(p)
      )
    # - compute the fold-change
    contrast_p_df <-
      data.frame(
        contrast = contrast,
        phosphopep = contrast_cast$phosphopep,
        p_value_raw = p_value_data_contrast_ps,
        p_value_adj = contrast_data_adj_p_values
      )
    db_write_table_overwrite <- (contrast < 2)
    db_write_table_append <- !db_write_table_overwrite
    RSQLite::dbWriteTable(
      conn = db,
      name = "contrast_ppep_p_val",
      value = contrast_p_df,
      append = db_write_table_append
      )
    # Create UK for insert
    ddl_exec(db, "
      CREATE UNIQUE INDEX IF NOT EXISTS contrast_ppep_p_val__uk__idx
        ON contrast_ppep_p_val(phosphopep, contrast);
      "
    )
  }
  # Perhaps this could be done more elegantly using unique keys
  #   or creating the tables before saving data to them, but this
  #   is fast and, if the database exists on disk rather than in
  #   memory, it doesn't stress memory.
  dml_no_rows_exec(db, "
    CREATE TEMP table contrast_log2_fc
      AS
    SELECT *
      FROM  v_contrast_log2_fc
      ORDER BY contrast, phosphopep
    ;
    "
  )
  dml_no_rows_exec(db, "
    CREATE TEMP table ppep_p_val
      AS
    SELECT  p_value_raw,
            p_value_adj,
            contrast   AS p_val_contrast,
            phosphopep AS p_val_ppep
      FROM  contrast_ppep_p_val
      ORDER BY contrast, phosphopep
    ;
    "
  )
  dml_no_rows_exec(db, "
    DROP TABLE IF EXISTS contrast_log2_fc_p_val
    ;
    "
  )
  dml_no_rows_exec(db, "
    CREATE TABLE contrast_log2_fc_p_val
      AS
    SELECT  a.*,
            b.p_value_raw,
            b.p_value_adj,
            b.p_val_contrast,
            b.p_val_ppep
      FROM  contrast_log2_fc a, ppep_p_val b
      WHERE a.rowid = b.rowid
        AND a.phosphopep = b.p_val_ppep
    ;
    "
  )
  # Create UK
  ddl_exec(db, "
    CREATE UNIQUE INDEX IF NOT EXISTS contrast_log2_fc_p_val__uk__idx
      ON contrast_log2_fc_p_val(phosphopep, contrast);
    "
  )
  # Create indices for future queries
  ddl_exec(db, "
    CREATE INDEX IF NOT EXISTS contrast_log2_fc_p_val__contrast__idx
      ON contrast_log2_fc_p_val(contrast);
    "
  )
  ddl_exec(db, "
    CREATE INDEX IF NOT EXISTS contrast_log2_fc_p_val__phosphopep__idx
      ON contrast_log2_fc_p_val(phosphopep);
    "
  )
  ddl_exec(db, "
    CREATE INDEX IF NOT EXISTS contrast_log2_fc_p_val__p_value_raw__idx
      ON contrast_log2_fc_p_val(p_value_raw);
    "
  )
  ddl_exec(db, "
    CREATE INDEX IF NOT EXISTS contrast_log2_fc_p_val__p_value_adj__idx
      ON contrast_log2_fc_p_val(p_value_adj);
    "
  )
  dml_no_rows_exec(db, "
    DROP VIEW IF EXISTS v_contrast_log2_fc_p_val
    ;
    "
  )
  dml_no_rows_exec(db, "
      CREATE VIEW v_contrast_log2_fc_p_val
        AS
      SELECT  contrast,
              a_level,
              a_samples,
              b_level,
              b_samples,
              a_quant,
              b_quant,
              fc_description,
              log2_fc,
              p_value_raw,
              p_value_adj,
              phosphopep
        FROM  contrast_log2_fc_p_val
        ORDER BY contrast, phosphopep
        ;
    "
  )
  ddl_exec(db, "
    DROP TABLE IF EXISTS kseaapp_metadata
    ;
    "
  )
  dml_no_rows_exec(db, "
    CREATE TABLE kseaapp_metadata
      AS
    WITH extended(deppep, ppep, gene_name, uniprot_id, phosphoresidue) AS (
              SELECT DISTINCT
                  deppep.seq,
                  ppep.seq,
                  GeneName||';',
                  UniProtID||';',
                  PhosphoResidue||';'
              FROM
                ppep, deppep, mrgfltr_metadata
              WHERE
                  mrgfltr_metadata.ppep_id = ppep.id
                AND
                  ppep.deppep_id = deppep.id
        )
    SELECT
        ppep                                                  AS `ppep`,
        SUBSTR(uniprot_id,     1, INSTR(uniprot_id,';') - 1 ) AS `Protein`,
        SUBSTR(gene_name,      1, INSTR(gene_name,';')  - 1 ) AS `Gene`,
        deppep                                                AS `Peptide`,
        REPLACE(
          REPLACE(
            SUBSTR(phosphoresidue, 1, INSTR(phosphoresidue,';') - 1 ),
            'p',
            ''
            ),
          ', ',
          ';'
          )                                                   AS `Residue.Both`
      FROM extended
      ;
    "
  )
  # Create indexes for join
  ddl_exec(db, "
    CREATE INDEX IF NOT EXISTS kseaapp_metadata__ppep__idx
      ON kseaapp_metadata(ppep);
    "
  )
  ddl_exec(db, "
    DROP VIEW IF EXISTS v_kseaapp_contrast
    ;
    "
  )
  dml_no_rows_exec(db, "
    CREATE VIEW v_kseaapp_contrast
      AS
    SELECT  a.*, b.Protein, b.Gene, b.Peptide, b.`Residue.Both`
      FROM  v_contrast_log2_fc_p_val a, kseaapp_metadata b
      WHERE b.ppep = a.phosphopep
      ;
      "
  )
  ddl_exec(db, "
    DROP VIEW IF EXISTS v_kseaapp_input
    ;
    "
  )
  dml_no_rows_exec(db, "
    CREATE VIEW v_kseaapp_input
      AS
    SELECT  v.contrast,
            v.phosphopep,
            m.`Protein`,
            m.`Gene`,
            m.`Peptide`,
            m.`Residue.Both`,
            v.p_value_raw AS `p`,
            v.log2_fc AS `FC`
      FROM  kseaapp_metadata AS m,
            v_contrast_log2_fc_p_val AS v
      WHERE m.ppep = v.phosphopep
        AND NOT m.`Gene` = 'No_Gene_Name'
        AND NOT v.log2_fc = 0
      ;
      "
  )
}
```

```{r echo = FALSE, results = 'asis'}
cat("\\newpage\n")
```

# KSEA Analysis

Results of Kinase-Substrate Enrichment Analysis are presented here, if the substrates for any kinases are relatively enriched.   Enrichments are found by the CRAN `KSEAapp` package:

- The package is available on CRAN, at https:/cran.r-project.org/package=KSEAapp
- The method used is described in Casado et al. (2013) [doi:10.1126/scisignal.2003573](https:/doi.org/10.1126/scisignal.2003573) and Wiredja et al (2017) [doi:10.1093/bioinformatics/btx415](https:/doi.org/10.1093/bioinformatics/btx415).
- An online alternative (supporting only analysis of human data) is available at [https:/casecpb.shinyapps.io/ksea/](https:/casecpb.shinyapps.io/ksea/).

For each kinase, $i$, and each two-way contrast of treatments, $j$, an enrichment $z$-score is computed as:

$$
\text{kinase enrichment score}_{j,i} = \frac{(\overline{s}_{j,i} - \overline{p}_j)\sqrt{m_{j,i}}}{\delta_j}
$$

and fold-enrichment is computed as:

$$
\text{Enrichment}_{j,i} = \frac{\overline{s}_{j,i}}{\overline{p}_j}
$$

where:

- $\overline{s}_{j,i}$ is the mean $\log_2 (|\text{fold-change|})$ in intensities (for contrast $j$) of known substrates of the kinase $i$,
- $\overline{p}_j$ is the mean $\log_2 (|\text{fold-change}|)$ of all phosphosites identified in contrast $j$, and
- $m_{j,i}$ is the total number of phosphosite substrates of kinase $i$ identified in contrast $j$,
- $\delta_j$ is the standard deviation of the $\log_2 (|\text{fold-change}|)$ for contrast $j$ across all phosphosites in the dataset.
- Note that the absolute value of fold-change is used so that both increased and decreased substrates of a kinase will contribute to its enrichment score.

$\text{FDR}_{j,i}$ is computed from the $p$-value for the z-score using the R `stats::p.adjust` function, applying the False Discovery Rate correction from Benjamini and Hochberg (1995) [doi:10.1111/j.2517-6161.1995.tb02031.x](https:/doi.org/10.1111/j.2517-6161.1995.tb02031.x)

Color intensity in heatmaps reflects magnitude of $z$-score for enrichment of respective kinase in respective contrast; hue reflects the sign of the $z$-score (blue, negative; red, positive).

Asterisks in heatmaps reflect enrichments that are significant at `r ksea_cutoff_statistic` < `r ksea_cutoff_threshold`.

- Kinase names are generally as presented at Phospho.ELM [http://phospho.elm.eu.org/kinases.html](http://phospho.elm.eu.org/kinases.html) (when available), although Phospho.ELM data are not yet incorporated into this analysis.
- Kinase names having the suffix '(HPRD)' are as presented at [http://hprd.org/serine_motifs](http://hprd.org/serine_motifs) and [http://hprd.org/tyrosine_motifs](http://hprd.org/tyrosine_motifs) and are as originally reported in the Amanchy et al., 2007 (doi: [10.1038/nbt0307-285](https://doi.org/10.1038/nbt0307-285)).
- Kinase-strate deata were also taken from [http://networkin.science/download.shtml](http://networkin.science/download.shtml) and from PhosphoSitePlus [https://www.phosphosite.org/staticDownloads](https://www.phosphosite.org/staticDownloads).

```{r ksea, echo = FALSE, fig.dim = c(9, 10), results = 'asis'}

db <- RSQLite::dbConnect(RSQLite::SQLite(), ksea_app_prep_db)

# -- eliminate the table that's about to be defined
ddl_exec(db, "
DROP TABLE IF EXISTS site_metadata;
")

# -- define the site_metadata table
ddl_exec(db, "
CREATE TABLE site_metadata(
  id            INTEGER PRIMARY KEY
, site_type_id  INTEGER REFERENCES site_type(id)
, full          TEXT    UNIQUE ON CONFLICT IGNORE
, abbrev        TEXT
, pattern       TEXT
, motif         TEXT
);
")

# -- populate the table with initial values
ddl_exec(db, "
INSERT INTO site_metadata(full, abbrev, site_type_id)
  SELECT  DISTINCT kinase_map, kinase_map, site_type_id
    FROM  ppep_gene_site
    ORDER BY kinase_map;
")

# -- drop bogus KSData view if exists
ddl_exec(db, "
DROP VIEW IF EXISTS ks_data_v;
")

# -- create view to serve as an impostor for  KSEAapp::KSData
ddl_exec(db, "
CREATE VIEW IF NOT EXISTS ks_data_v
AS
SELECT
  'NA' AS KINASE,
  'NA' AS KIN_ACC_ID,
  kinase_map AS GENE,
  'NA' AS KIN_ORGANISM,
  'NA' AS SUBSTRATE,
  0 AS SUB_GENE_ID,
  'NA' AS SUB_ACC_ID,
  gene_names AS SUB_GENE,
  'NA' AS SUB_ORGANISM,
  phospho_peptide AS SUB_MOD_RSD,
  0 AS SITE_GROUP_ID,
  'NA' AS 'SITE_7AA',
  2 AS networkin_score,
  type_name AS Source
FROM ppep_gene_site_view;
")

contrast_metadata_df <-
  sqldf::sqldf("select * from contrast_lvl_lvl_metadata", connection = db)
rslt <- new_env()
rslt$score_list <- list()
rslt$name_list  <- list()
rslt$longname_list  <- list()

ddl_exec(db, "
  DROP TABLE IF EXISTS contrast_ksea_scores;
  "
)

next_index <- 0
err_na_subscr_df_const <-
 "missing values are not allowed in subscripted assignments of data frames"

for (i_cntrst in seq_len(nrow(contrast_metadata_df))) {
  cntrst_a_level <- contrast_metadata_df[i_cntrst, "a_level"]
  cntrst_b_level <- contrast_metadata_df[i_cntrst, "b_level"]
  cntrst_fold_change <- contrast_metadata_df[i_cntrst, 6]
  contrast_label <- sprintf("%s -> %s", cntrst_b_level, cntrst_a_level)
  contrast_longlabel <- (
    sprintf(
      "Trt %s {%s} -> Trt %s {%s}",
      contrast_metadata_df[i_cntrst, "b_level"],
      gsub(
        pattern = ";",
        replacement = ", ",
        x = contrast_metadata_df[i_cntrst, "b_samples"],
        fixed = TRUE
      ),
      contrast_metadata_df[i_cntrst, "a_level"],
      gsub(
        pattern = ";",
        replacement = ", ",
        x = contrast_metadata_df[i_cntrst, "a_samples"],
        fixed = TRUE
      )
    )
  )
  kseaapp_input <-
    sqldf::sqldf(
      x = sprintf("
        SELECT `Protein`, `Gene`, `Peptide`, phosphopep AS `Residue.Both`, `p`, `FC`
          FROM v_kseaapp_input
          WHERE contrast = %d
        ",
        i_cntrst
      ),
    connection = db
    )

  pseudo_ksdata <- dbReadTable(db, "ks_data_v")

  # This hack is because SQL table has the log2-transformed values
  kseaapp_input[, "FC"] <- 2 ** kseaapp_input[, "FC", drop = TRUE]
  main_title <- (
      sprintf(
      "Change from treatment %s to treatment %s",
      contrast_metadata_df[i_cntrst, "b_level"],
      contrast_metadata_df[i_cntrst, "a_level"]
    )
  )
  sub_title <- contrast_longlabel
  tryCatch(
    expr = {
        ksea_scores_rslt <-
          ksea_scores(
            ksdata           = pseudo_ksdata, # KSEAapp::KSData,
            px               = kseaapp_input,
            networkin        = TRUE,
            networkin_cutoff = 2
            )

        if (0 < sum(!is.nan(ksea_scores_rslt$FDR))) {
          next_index <- 1 + next_index
          rslt$score_list[[next_index]] <- ksea_scores_rslt
          rslt$name_list[[next_index]] <- contrast_label
          rslt$longname_list[[next_index]] <- contrast_longlabel
          low_fdr_print(
            rslt = rslt,
            i_cntrst = i_cntrst,
            i = next_index,
            a_level = cntrst_a_level,
            b_level = cntrst_b_level,
            fold_change = cntrst_fold_change,
            caption = contrast_longlabel
            )
        }
      },
    error = function(e) str(e)
  )
}

plotted_kinases <- NULL
if (length(rslt$score_list) > 1) {
  for (i in seq_len(length(ksea_heatmap_titles))) {
    hdr <- ksea_heatmap_titles[[i]]
    which_kinases <- i

    cat("\\clearpage\n\\begin{center}\n")
    if (i == const_ksea_astrsk_kinases) {
      subsection_header(hdr)
    } else {
      subsection_header(hdr)
    }
    cat("\\end{center}\n")

    plotted_kinases <- ksea_heatmap(
      # the data frame outputs from the KSEA.Scores() function, in list format
      score_list = rslt$score_list,
      # a character vector of all the sample names for heatmap annotation:
      # - the names must be in the same order as the data in score_list
      # - please avoid long names, as they may get cropped in the final image
      sample_labels = rslt$name_list,
      # character string of either "p.value" or "FDR" indicating the data column
      #   to use for marking statistically significant scores
      stats = c("p.value", "FDR")[2],
      # a numeric value between 0 and infinity indicating the min. number of
      #   substrates a kinase must have to be included in the heatmap
      m_cutoff = 1,
      # a numeric value between 0 and 1 indicating the p-value/FDR cutoff
      #   for indicating significant kinases in the heatmap
      p_cutoff = 0.05,
      # a binary input of TRUE or FALSE, indicating whether or not to perform
      #   hierarchical clustering of the sample columns
      sample_cluster = TRUE,
      # a binary input of TRUE or FALSE, indicating whether or not to export
      #   the heatmap as a .png image into the working directory
      export = FALSE,
      # additional arguments to gplots::heatmap.2, such as:
      # - main: main title of plot
      # - xlab: x-axis label
      # - ylab: y-axis label
      xlab = "Contrast",
      ylab = "Kinase",
      # print which kinases:
      # - 1 : all kinases
      # - 2 : significant kinases
      # - 3 : non-significant kinases
      which_kinases = which_kinases
    )
    cat("\\begin{center}\n")
    cat("Color intensities reflects $z$-score magnitudes; hue reflects $z$-score sign.  Asterisks reflect significance.\n")
    cat("\\end{center}\n")
  } # end for (i in ...
} # end if (length ...

for (i_cntrst in seq_len(length(rslt$score_list))) {
  next_index <- i_cntrst
  cntrst_a_level <- contrast_metadata_df[i_cntrst, "a_level"]
  cntrst_b_level <- contrast_metadata_df[i_cntrst, "b_level"]
  cntrst_fold_change <- contrast_metadata_df[i_cntrst, 6]
  contrast_label <- sprintf("%s -> %s", cntrst_b_level, cntrst_a_level)
  contrast_longlabel <- (
    sprintf(
      "Trt %s {%s} -> Trt %s {%s}",
      contrast_metadata_df[i_cntrst, "b_level"],
      gsub(
        pattern = ";",
        replacement = ", ",
        x = contrast_metadata_df[i_cntrst, "b_samples"],
        fixed = TRUE
      ),
      contrast_metadata_df[i_cntrst, "a_level"],
      gsub(
        pattern = ";",
        replacement = ", ",
        x = contrast_metadata_df[i_cntrst, "a_samples"],
        fixed = TRUE
      )
    )
  )
  main_title <- (
      sprintf(
      "Change from treatment %s to treatment %s",
      contrast_metadata_df[i_cntrst, "b_level"],
      contrast_metadata_df[i_cntrst, "a_level"]
    )
  )
  sub_title <- contrast_longlabel
  tryCatch(
    expr = {
        ksea_scores_rslt <- rslt$score_list[[next_index]]

        if (0 < sum(!is.nan(ksea_scores_rslt$FDR))) {
          low_fdr_barplot(
            rslt = rslt,
            i_cntrst = i_cntrst,
            i = next_index,
            a_level = cntrst_a_level,
            b_level = cntrst_b_level,
            fold_change = cntrst_fold_change,
            caption = contrast_longlabel
            )
        }
      },
    error = function(e) str(e)
  )
}
```

```{r enriched, echo = FALSE, fig.dim = c(9, 10), results = 'asis'}

# Use enriched kinases to find enriched kinase-substrate pairs
enriched_kinases <- data.frame(kinase = ls(ksea_asterisk_hash))
all_enriched_substrates <- sqldf("
  SELECT
    gene AS kinase,
    ppep,
    '('||group_concat(gene||'-'||sub_gene)||') '||ppep AS label
  FROM (
    SELECT DISTINCT gene, sub_gene, SUB_MOD_RSD AS ppep
      FROM pseudo_ksdata
      WHERE GENE IN (SELECT kinase FROM enriched_kinases)
    )
  GROUP BY ppep
  ")

# helper used to label per-kinase substrate enrichment figure
cat_enriched_heading <- function(m, cut_args) {
  cutoff <- cut_args$cutoff
  kinase <- cut_args$kinase
  statistic <- cut_args$statistic
  threshold <- cut_args$threshold
  cat("\\newpage\n")
  if (nrow(m) > intensity_hm_rows) {
    subsection_header(
      paste(
        sprintf(
          "Lowest p-valued %d (of %d) enriched %s-substrates,",
          intensity_hm_rows,
          nrow(m),
          kinase
        ),
        sprintf(" KSEA %s < %0.2f\n", statistic, threshold)
      )
    )
  } else {
    if (nrow(m) == 1) {
      return(FALSE)
    } else {
      subsection_header(
        paste(
          sprintf(
            "%d enriched %s-substrates,",
            nrow(m),
            kinase
            ),
          sprintf(
            " KSEA %s < %0.2f\n",
            statistic,
            threshold
            )
        )
      )
    }
  }
  cat("\n\n\n")
  cat("\n\n\n")
  return(TRUE)
}

# Disabling heatmaps for substrates pending decision whether to eliminate them altogether
if (FALSE)
  for (kinase_name in sort(enriched_kinases$kinase)) {
    enriched_substrates <-
      all_enriched_substrates[
        all_enriched_substrates$kinase == kinase_name,
        ,
        drop = FALSE
        ]
    # Get the intensity values for the heatmap
    enriched_intensities <-
      as.matrix(unimputed_quant_data_log[enriched_substrates$ppep, , drop = FALSE])
    # Remove rows having too many NA values to be relevant
    na_counter <- is.na(enriched_intensities)
    na_counts <- apply(na_counter, 1, sum)
    enriched_intensities <-
      enriched_intensities[na_counts < ncol(enriched_intensities) / 2, , drop = FALSE]
    # Rename the rows with the display-name for the heatmap
    rownames(enriched_intensities) <-
      sapply(
        X = rownames(enriched_intensities),
        FUN = function(rn) {
          enriched_substrates[enriched_substrates$ppep == rn, "label"]
        }
      )
    # Format as matrix for heatmap
    m <- as.matrix(enriched_intensities)
    # Draw the heading and heatmap
    if (nrow(m) > 0) {
      cut_args <- new_env()
      cut_args$cutoff <- cutoff
      cut_args$kinase <- kinase_name
      cut_args$statistic <- ksea_cutoff_statistic
      cut_args$threshold <- ksea_cutoff_threshold
      number_of_peptides_found <-
        draw_intensity_heatmap(
          m                       = m,
          cutoff                  = cut_args,
          hm_heading_function     = cat_enriched_heading,
          hm_main_title
            = "Unnormalized (zero-imputed) intensities of enriched kinase-substrates",
          suppress_row_dendrogram = FALSE
        )
    }
  }

# Write output tabular files

# get kinase, ppep, concat(kinase) tuples for enriched kinases

kinase_ppep_label <- sqldf("
  WITH
    t(ppep, label) AS
      (
        SELECT DISTINCT
            SUB_MOD_RSD AS ppep,
            group_concat(gene, '; ') AS label
          FROM pseudo_ksdata
          WHERE GENE IN (SELECT kinase FROM enriched_kinases)
          GROUP BY ppep
      ),
    k(kinase, ppep_join) AS
      (
      SELECT DISTINCT gene AS kinase, SUB_MOD_RSD AS ppep_join
        FROM pseudo_ksdata
        WHERE GENE IN (SELECT kinase FROM enriched_kinases)
      )
  SELECT k.kinase, t.ppep, t.label
    FROM  t,  k
    WHERE t.ppep = k.ppep_join
    ORDER BY k.kinase, t.ppep
  ")

# extract what we need from full_data
impish <- cbind(rownames(quant_data_imp), quant_data_imp)
colnames(impish)[1] <- "Phosphopeptide"
data_table_imputed_sql <- "
  SELECT
    f.*,
    k.label AS KSEA_enrichments,
    q.*
  FROM
    metadata_plus_p f
      LEFT JOIN kinase_ppep_label k
        ON f.Phosphopeptide = k.ppep,
    impish q
  WHERE
    f.Phosphopeptide = q.Phosphopeptide
  "
data_table_imputed <- sqldf(data_table_imputed_sql)
# Zap the duplicated 'Phosphopeptide' column named 'ppep'
data_table_imputed <-
    data_table_imputed[, c(1:12, 14:ncol(data_table_imputed))]

# Output with imputed, un-normalized data

write.table(
    data_table_imputed
  , file = imputed_data_filename
  , sep = "\t"
  , col.names = TRUE
  , row.names = FALSE
  , quote = FALSE
  )


#output quantile normalized data
impish <- cbind(rownames(quant_data_imp_qn_log), quant_data_imp_qn_log)
colnames(impish)[1] <- "Phosphopeptide"
data_table_imputed <- sqldf(data_table_imputed_sql)
# Zap the duplicated 'Phosphopeptide' column named 'ppep'
data_table_imputed <-
    data_table_imputed[, c(1:12, 14:ncol(data_table_imputed))]
write.table(
  data_table_imputed,
  file = imp_qn_lt_data_filenm,
  sep = "\t",
  col.names = TRUE,
  row.names = FALSE,
  quote = FALSE
)

ppep_kinase <- sqldf("
  SELECT DISTINCT k.ppep, k.kinase
    FROM (
      SELECT DISTINCT gene AS kinase, SUB_MOD_RSD AS ppep
        FROM pseudo_ksdata
        WHERE GENE IN (SELECT kinase FROM enriched_kinases)
      ) k
    ORDER BY k.ppep, k.kinase
  ")

RSQLite::dbWriteTable(
  conn = db,
  name = "ksea_enriched_ks",
  value = ppep_kinase,
  append = FALSE
  )

RSQLite::dbWriteTable(
  conn = db,
  name = "anova_signif",
  value = p_value_data,
  append = FALSE
  )

  ddl_exec(db, "
    DROP VIEW IF EXISTS stats_metadata_v;
    "
  )
  dml_no_rows_exec(db, "
      CREATE VIEW stats_metadata_v
        AS
      SELECT DISTINCT  m.*,
          p.raw_anova_p,
          p.fdr_adjusted_anova_p,
          kek.kinase AS ksea_enrichments
        FROM
          mrgfltr_metadata_view m
            LEFT JOIN anova_signif p
              ON m.phospho_peptide = p.phosphopeptide
            LEFT JOIN ksea_enriched_ks kek
              ON m.phospho_peptide = kek.ppep
      ;
    "
  )

write.table(
  dbReadTable(db, "stats_metadata_v"),
  file = anova_ksea_mtdt_file,
  sep = "\t",
  col.names = TRUE,
  row.names = FALSE,
  quote = FALSE
  )


```

```{r parmlist, echo = FALSE, fig.dim = c(9, 10), results = 'asis'}
cat("\\leavevmode\n\n\n")

# write parameters to report

param_unlist <- unlist(as.list(params))
param_df <- data.frame(
  parameter = paste0("\\verb@", names(param_unlist), "@"),
  value = paste0("\\verb@", gsub("$", "\\$", param_unlist, fixed = TRUE), "@")
  )

data_frame_latex(
  x = param_df,
  justification = "p{0.35\\linewidth} p{0.6\\linewidth}",
  centered = TRUE,
  caption = "Input parameters",
  anchor = const_table_anchor_bp,
  underscore_whack = FALSE
  )

# write parameters to SQLite output

mqppep_anova_script_param_df <- data.frame(
  script    = "mqppep_anova_script.Rmd",
  parameter = names(param_unlist),
  value     = param_unlist
  )
ddl_exec(db, "
  DROP TABLE IF EXISTS script_parameter;
  "
)
ddl_exec(db, "
  CREATE TABLE IF NOT EXISTS script_parameter(
    script    TEXT,
    parameter TEXT,
    value     ANY,
    UNIQUE (script, parameter) ON CONFLICT REPLACE
    )
    ;
  "
)
RSQLite::dbWriteTable(
  conn = db,
  name = "script_parameter",
  value = mqppep_anova_script_param_df,
  append = TRUE
)

# We are done with output
RSQLite::dbDisconnect(db)
```
<!--
There's gotta be a better way...

loaded_packages_df <-  sessioninfo::package_info("loaded")
loaded_packages_df[, "library"] <- as.character(loaded_packages_df$library)
loaded_packages_df <- data.frame(
  package = loaded_packages_df$package,
  version = loaded_packages_df$loadedversion,
  date    = loaded_packages_df$date
  )
data_frame_latex(
  x = loaded_packages_df,
  justification = "l | l l",
  centered = FALSE,
  caption = "Loaded R packages",
  anchor = const_table_anchor_bp
  )
-->