Linakis et al. (2020): High Throughput Inhalation Model

Please send questions to

from “Development and evaluation of a high throughput inhalation model for organic chemicals”

Matthew W. Linakis, Risa R. Sayre, Robert G. Pearce, Mark A. Sfeir, Nisha S. Sipes, Heather A. Pangburn, Jeffery M. Gearhart, and John F. Wambaugh

Journal of Exposure Science & Environmental Epidemiology volume 30, pages 866–877 (2020)

https://doi.org/10.1038/s41370-020-0238-y

Abstract

Currently it is difficult to prospectively estimate human toxicokinetics (particularly for novel chemicals) in a high-throughput manner. The R software package httk has been developed, in part, to address this deficiency, and the aim of this investigation was to develop a generalized inhalation model for httk. The structure of the inhalation model was developed from two previously published physiologically-based models from Jongeneelen et al. (2011) and Clewell et al.  (2001) while calculated physicochemical data was obtained from EPA’s CompTox Chemicals Dashboard. In total, 142 exposure scenarios across 41 volatile organic chemicals were modeled and compared to published data. The slope of the regression line of best fit between log-transformed simulated and observed combined measured plasma and blood concentrations was 0.59 with an r2= 0.54 and a Root Mean Square Error (RMSE) of direct comparison between the log-transformed simulated and observed values of 0.87. Approximately 3.6% (n = 73) of the data points analyzed were > 2 orders of magnitude different than expected. The volatile organic chemicals examined in this investigation represent small, generally lipophilic molecules. Ultimately this paper details a generalized inhalation component that integrates with the httk physiologically-based toxicokinetic model to provide high-throughput estimates of inhalation chemical exposures.

HTTK Version

This vignette was created with httk v2.0.0. Although we attempt to maintain backward compatibility, if you encounter issues with the latest release of httk and cannot easily address the changes, historical versions of httk are available from: https://cran.r-project.org/src/contrib/Archive/httk/

Prepare for session

R package knitr generates html and PDF documents from this RMarkdown file, Each bit of code that follows is known as a “chunk”. We start by telling knitr how we want our chunks to look.

knitr::opts_chunk$set(echo = TRUE, fig.width=5, fig.height=4)

Clear the memory

It is a bad idea to let variables and other information from previous R sessions float around, so we first remove everything in the R memory.

rm(list=ls())

eval = execute.vignette

If you are using the RMarkdown version of this vignette (extension, .RMD) you will be able to see that several chunks of code in this vignette have the statement “eval = execute.vignette”. The next chunk of code, by default, sets execute.vignette = FALSE. This means that the code is included (and necessary) but was not run when the vignette was built. We do this because some steps require extensive computing time and the checks on CRAN limit how long we can spend building the package. If you want this vignette to work, you must run all code, either by cutting and pasting it into R. Or, if viewing the .RMD file, you can either change execute.vignette to TRUE or press “play” (the green arrow) on each chunk in RStudio.

# Set whether or not the following chunks will be executed (run):
execute.vignette <- FALSE

Load the relevant libraries

We use the command ‘library()’ to load various R packages for our analysis. If you get the message “Error in library(X) : there is no package called ‘X’” then you will need to install that package:

From the R command prompt:

install.packages(“X”)

Or, if using RStudio, look for ‘Install Packages’ under ‘Tools’ tab.

knitr::opts_chunk$set(echo = TRUE, fig.width=5, fig.height=4)
library(httk)
library(ggplot2)
library(gridExtra)
library(cowplot)
library(ggrepel)
library(dplyr)
library(stringr)
library(forcats)
library(smatr)

Get metabolism and concentration data

The table httk::concentration_data_Linakis2020 contains CvTdb data (Sayre, et al. 2020) for inhalation exposure. The units of all measurements have been converted to uM concentrations.

met_data <- metabolism_data_Linakis2020
conc_data <- concentration_data_Linakis2020
#conc_data <- subset(concentration_data_Linakis2020, !(SAMPLING_MATRIX %in%
#                      c("EEB","MEB","EB")))
#conc_data <- concentration_data_Linakis2020
#  subset(concentration_data_Linakis2020, !(SOURCE_CVT %in% c(
#  "11453305")))
conc_data[,"DOSE_U"] <- ifelse(conc_data[,"DOSE_U"] == "ppm",
                               yes = "ppmv",
                               conc_data[,"DOSE_U"]) 
conc_data[,"ORIG_CONC_U"] <- ifelse(conc_data[,"ORIG_CONC_U"] == "ppm",
                                    yes = "ppmv",
                                    conc_data[,"ORIG_CONC_U"]) 
# Not sure what to do with percent:
conc_data <- subset(conc_data,toupper(ORIG_CONC_U) != "PERCENT")
# Rename this column:
colnames(conc_data)[colnames(conc_data)=="ORIG_CONC_U"] <- "CONC_U"
conc_data$ORIGINAL_CONC_U <- conc_data$CONC_U
conc_data$ORIGINAL_CONC <- conc_data$CONCENTRATION
# Maybe Linakis et al translated all concentrations to uM?
conc_data$CONC_U <- "uM"

Sets the units based on the sampling matrix (gas/liquid): BL : blood EEB : end-exhaled breath MEB : mixed exhaled breath VBL : venous blood ABL : arterial blood EB : unspecified exhaled breath sample (assumed to be EEB) PL: plasma +W with work/exercise

Normalize units for gaseous samples to ppmv:

gas.media <- c("EB","MEB","EEB","EB (+W)")
gas.units <- unique(subset(conc_data,
  SAMPLING_MATRIX %in% gas.media)$CONC_U)
target.unit <- "ppmv"
for (this.unit in gas.units)
  if (this.unit != target.unit)
  {
    these.chems <- unique(subset(conc_data,
      SAMPLING_MATRIX %in% gas.media &
      CONC_U==this.unit)$DTXSID)    
    for (this.chem in these.chems)  
    {
      this.factor <- convert_units(
        input.units=this.unit, 
        output.units=target.unit, 
        dtxsid=this.chem, state="gas")
      print(paste("Use",this.factor,"to convert",this.unit,"to",target.unit))
      
      # Scale the observation    
      conc_data[conc_data$DTXSID==this.chem &
                  conc_data$SAMPLING_MATRIX %in% gas.media &
                  conc_data$CONC_U==this.unit,"CONCENTRATION"] <-
        this.factor * conc_data[
          conc_data$DTXSID==this.chem &
            conc_data$SAMPLING_MATRIX %in% gas.media &
            conc_data$CONC_U==this.unit,"CONCENTRATION"]
      # Change the reported unit
      conc_data[conc_data$DTXSID==this.chem &
                  conc_data$SAMPLING_MATRIX %in% gas.media &
                  conc_data$CONC_U==this.unit,"CONC_U"] <-
        target.unit

    }
  }

Normalize the units for tissue samples to uM:

tissue.media <- c("VBL","BL","ABL","PL","BL (+W)")
tissue.units <- unique(subset(conc_data,
  SAMPLING_MATRIX %in% tissue.media)$CONC_U)
target.unit <- "uM"
for (this.unit in tissue.units)
  if (this.unit != target.unit)
  {
    these.chems <- unique(subset(conc_data,
      SAMPLING_MATRIX %in% tissue.media &
      CONC_U==this.unit)$DTXSID)    
    for (this.chem in these.chems)  
    {
      this.factor <- try(convert_units(
        input.units=this.unit, 
        output.units=target.unit, 
        dtxsid=this.chem))
      print(paste("Use",this.factor,"to convert",this.unit,"to",target.unit))
      
      # Scale the observation    
      conc_data[conc_data$DTXSID==this.chem &
                  conc_data$SAMPLING_MATRIX %in% tissue.media &
                  conc_data$CONC_U==this.unit,"CONCENTRATION"] <-
        this.factor * conc_data[
          conc_data$DTXSID==this.chem &
            conc_data$SAMPLING_MATRIX %in% tissue.media &
            conc_data$CONC_U==this.unit,"CONCENTRATION"]
      # Change the reported unit
      conc_data[conc_data$DTXSID==this.chem &
                  conc_data$SAMPLING_MATRIX %in% tissue.media &
                  conc_data$CONC_U==this.unit,"CONC_U"] <-
        target.unit

    }
  }

ANALYSIS

Identify chemicals currently in our metabolism data that we don’t have good concentration/time data for and remove them from our training dataset

Data summary for chemical properties

# Small molecule chemicals
summary(met_data$AVERAGE_MASS)
# Generally more lipophilic chemicals
summary(met_data$OCTANOL_WATER_PARTITION_LOGP_OPERA_PRED)
# Unsurprisingly then, the chemicals are generally less water-soluble
summary(met_data$WATER_SOLUBILITY_MOL.L_OPERA_PRED)
# ~60% of samples in humans
table(conc_data$CONC_SPECIES)/nrow(conc_data)*100
# ~72% of samples are from blood
table(conc_data$SAMPLING_MATRIX)/nrow(conc_data)*100

Exposure scenarios

# Create a dataframe with 1 row for each unique external exposure scenario
unique_scenarios <- conc_data[with(conc_data,                             
  order(PREFERRED_NAME,
        CONC_SPECIES,
        SAMPLING_MATRIX,
        as.numeric(as.character(DOSE)),EXP_LENGTH,-TIME)),] %>%
  distinct(DTXSID,DOSE,DOSE_U,EXP_LENGTH,CONC_SPECIES,SAMPLING_MATRIX, .keep_all = TRUE)

Observations and Predictions

Create a list of dataframes of observed and predicted concentrations for each unique external exposure scenario (corresponding to 142 studies)

# Store the output of each simulation:
simlist <- list()
# Store the Cvt data relevant to each simulation
obslist <- list()
# Conduct one simulation for each unique combination of chemical, species, dose:
for (i in 1:nrow(unique_scenarios))
  if (unique_scenarios$CASRN[i] %in% get_cheminfo(model="gas_pbtk",
                                                  suppress.messages = TRUE))
{
  # Identify relevant Cvt data:
    relconc <- subset(conc_data,conc_data$DTXSID == unique_scenarios$DTXSID[i] & 
      conc_data$DOSE == unique_scenarios$DOSE[i] & 
      conc_data$EXP_LENGTH == unique_scenarios$EXP_LENGTH[i] & 
      conc_data$CONC_SPECIES == unique_scenarios$CONC_SPECIES[i] & 
      conc_data$SAMPLING_MATRIX == unique_scenarios$SAMPLING_MATRIX[i])
    obslist[[i]] <- relconc
#
#
#
#
#
# THE FOLLOWING CODE RUNS solve_gas_pbtk FOR EACH SCENARIO 
# (UNIQUE COMBINATION OF CHEMICAL, SPECIES, DOSE, ETC.)
#
#
#
#
#
    solver.out <- try(suppressWarnings(as.data.frame(solve_gas_pbtk(
        chem.cas = unique_scenarios$CASRN[i], 
        days = (unique_scenarios$TIME[i]+unique_scenarios$EXP_LENGTH[i]), 
# Make sure we get predicted conc's at the observed times:
        times=unique(c(0,signif(obslist[[i]]$TIME,4))), # days
        tsteps = 500,
        exp.conc = as.numeric(unique_scenarios$DOSE[i]), # SED (06-22-2021) think this is ppmv for all scenarios
         input.units = unique_scenarios$DOSE_U[i], # specify the units for exp.conc (ppmv)
        exp.duration = unique_scenarios$EXP_LENGTH[i], # days 
        period = (unique_scenarios$TIME[i]+unique_scenarios$EXP_LENGTH[i]), # days 
        species = as.character(unique_scenarios$CONC_SPECIES[i]),
        monitor.vars = c(
          "Cven", 
          "Clung", 
          "Cart", 
          "Cplasma", 
          "Calvppmv", 
          "Cendexhppmv", 
          "Cmixexhppmv", 
          "Calv", 
          "Cendexh", 
          "Cmixexh", 
          "Cmuc", 
          "AUC"),
        vmax.km = FALSE,
        suppress.messages = TRUE))))
#
#
#
#
#
#
#
#
#
#
    if (class(solver.out) %in% "try-error") 
      solver.out <- data.frame(time=NA,Conc=NA)
    
    print(solver.out)
# Get the blood:plasma ratio:
    this.Rb2p <- suppressWarnings(available_rblood2plasma(
      chem.cas=unique_scenarios$CASRN[i],
      species=as.character(unique_scenarios$CONC_SPECIES[i])))
    solver.out$Rb2p <- this.Rb2p
    
    # The column simconc holds the appropriate prediction for the sampling matrix
    # BL : blood
    # EEB : end-exhaled breath
    # MEB : mixed exhaled breath
    # VBL : venous blood
    # ABL : arterial blood
    # EB : unspecified exhaled breath sample (assumed to be EEB)
    # PL: plasma
    # +W with work/exercise
    #
    # The model outputs are provided in the following units:
      # uM: Cven, Clung, Cart, Cplasma, Calv, Cendexh, Cmixexh, Cmuc
      # ppmv: Calvppmv, Cendexhppmv, Cmixexhppmv
      # uM*days: AUC
    if (unique_scenarios$SAMPLING_MATRIX[i] == "VBL" | 
      unique_scenarios$SAMPLING_MATRIX[i] == "BL" | 
      unique_scenarios$SAMPLING_MATRIX[i] == "BL (+W)")
    {
      solver.out$simconc <- solver.out$Cven*this.Rb2p
      solver.out$unit <- "uM"
    } else if (unique_scenarios$SAMPLING_MATRIX[i] == "ABL") {
      solver.out$simconc <- solver.out$Cart*this.Rb2p
      solver.out$unit <- "uM"
    } else if (unique_scenarios$SAMPLING_MATRIX[i] == "EB" |
      unique_scenarios$SAMPLING_MATRIX[i] == "EEB" | 
      unique_scenarios$SAMPLING_MATRIX[i] == "EB (+W)")
    {
      solver.out$simconc <- solver.out$Cendexh # uM
      solver.out$unit <- "uM"
    } else if (unique_scenarios$SAMPLING_MATRIX[i] == "MEB") {
        solver.out$simconc <- solver.out$Cmixexh # uM
        solver.out$unit <- "uM"
    } else if (unique_scenarios$SAMPLING_MATRIX[i] == "PL") {
      solver.out$simconc <- solver.out$Cplasma
      solver.out$unit <- "uM"
    } else {
      solver.out$simconc <- NA
      solver.out$unit <- NA
    }
    simlist[[i]] <- solver.out
}

Create a predicted vs. observed plot for each study:

cvtlist <- list()
for(i in 1:nrow(unique_scenarios))
{
    plot.data <- simlist[[i]]
    relconc <- obslist[[i]]

    if (!is.null(plot.data))
    {
#Right now this is only calculating real concentrations according to mg/L in blood
    cvtlist[[i]] <- ggplot(data=plot.data, aes(time*24, simconc)) + 
      geom_line() + 
      xlab("Time (h)") + 
      ylab(paste0("Simulated ", 
        unique_scenarios$SAMPLING_MATRIX[i], 
        "\nConcentration (" , 
        solver.out$unit, ")")) + 
      ggtitle(paste0(
        unique_scenarios$PREFERRED_NAME[i],
        " (", 
        unique_scenarios$CONC_SPECIES[i], 
        ", ",
        round(as.numeric(unique_scenarios$DOSE[i]), digits = 2),
        unique_scenarios$DOSE_U[i], 
        " for ",
        round(unique_scenarios$EXP_LENGTH[i]*24, digits = 2),
        "h in ", 
        unique_scenarios$SAMPLING_MATRIX[i], ")")) + 
      geom_point(data = relconc, aes(TIME*24,CONCENTRATION)) + 
      theme(plot.title = element_text(face = 'bold', size = 20),
        axis.title.x = element_text(face = 'bold', size = 20), 
        axis.text.x = element_text(size=16), 
        axis.title.y = element_text(face = 'bold', size = 20), 
        axis.text.y = element_text(size = 16),
        legend.title = element_text(face = 'bold', size = 16),
        legend.text = element_text(face = 'bold',size = 14))+
      theme_bw() 
    }
}

Create a list to hold the combined observations and predictions for each scenario:

# Creation of simulated vs. observed concentration dataset
unique_scenarios$RSQD <- 0
unique_scenarios$RMSE <- 0
unique_scenarios$AIC <- 0
simobslist <- list()
obvpredlist <- list()

Merge the simulations and observations on the basis of simulation time:

for(i in 1:length(simlist))
{
  obsdata <- as.data.frame(obslist[[i]])
  simdata <- as.data.frame(simlist[[i]])
# skips over anything for which there was no observed data or 
# insufficient information to run simulation:
  if (!is.null(simdata) & 
      !is.null(obsdata) &
      dim(simdata)[1]>1)
  { 
# Make sure we are looking at consistent time points:
    simobscomb <- simdata[simdata$time %in% signif(obsdata$TIME,4),]
    obsdata <- subset(obsdata,signif(TIME,4) %in% simobscomb$time)
# Merge with obsdata
    colnames(obsdata)[colnames(obsdata) ==
      "TIME"] <- 
      "obstime"
# Round to match sim time:
    obsdata$time <- signif(obsdata$obstime,4)
    colnames(obsdata)[colnames(obsdata) ==
      "CONCENTRATION"] <- 
      "obsconc"
    colnames(obsdata)[colnames(obsdata) ==
      "PREFERRED_NAME"] <- 
      "chem"
    colnames(obsdata)[colnames(obsdata) ==
      "DOSE"] <- 
      "dose"
    colnames(obsdata)[colnames(obsdata) ==
      "EXP_LENGTH"] <- 
      "explen"
    colnames(obsdata)[colnames(obsdata) ==
      "CONC_SPECIES"] <- 
      "species"
    colnames(obsdata)[colnames(obsdata) ==
      "SAMPLING_MATRIX"] <- 
      "matrix"
    colnames(obsdata)[colnames(obsdata) ==
      "AVERAGE_MASS"] <- 
      "mw"
    colnames(obsdata)[colnames(obsdata) ==
      "CONC_U"] <- 
      "CONC_U"
    simobscomb <- suppressWarnings(merge(obsdata[,c(
      "time",
      "obstime",
      "obsconc",
      "chem",
      "dose",
      "explen",
      "species",
      "matrix",
      "mw",
      "CONC_U",
      "ORIGINAL_CONC_U"
      )], simobscomb, by="time", all.x=TRUE))

# Merge with met_data
    this.met_data <- subset(met_data,
      PREFERRED_NAME == simobscomb[1,"chem"] &
      SPECIES == simobscomb[1,"species"])
    colnames(this.met_data)[colnames(this.met_data)=="CHEM_CLASS"] <-
      "chemclass"
    colnames(this.met_data)[colnames(this.met_data) ==
      "OCTANOL_WATER_PARTITION_LOGP_OPERA_PRED"] <-
      "logp"
    colnames(this.met_data)[colnames(this.met_data) ==
      "WATER_SOLUBILITY_MOL.L_OPERA_PRED"] <-
      "sol"
    colnames(this.met_data)[colnames(this.met_data) ==
      "HENRYS_LAW_ATM.M3.MOLE_OPERA_PRED"] <-
      "henry"
    colnames(this.met_data)[colnames(this.met_data) ==
      "VMAX"] <-
      "vmax"
    colnames(this.met_data)[colnames(this.met_data) ==
      "KM"] <-
      "km"
    simobscomb <- suppressWarnings(cbind(simobscomb,this.met_data[c(
      "chemclass",
      "logp",
      "sol",
      "henry",
      "vmax",
      "km")]))
    simobslist[[i]] <- simobscomb
  }
}

Identify which quartile each observation occurred in with respect to the latest (maximum) observed time

for(i in 1:length(simobslist))
  if (!is.null(simobslist[[i]]))
  {
    simobscomb <- simobslist[[i]]
    for (j in 1:nrow(simobscomb))
    { 
      max.time <- max(simobscomb$time,na.rm=TRUE)
      if (is.na(max.time)) simobscomb$tquart <- NA
      else if (max.time == 0) simobscomb$tquart <- "1"
      else if (!is.na(simobscomb$time[j])) 
      {
        simobscomb$tquart[j] <- as.character(1 +
          floor(simobscomb$time[j]/max.time/0.25))
        simobscomb$tquart[simobscomb$tquart=="5"] <-
          "4"
      } else simobscomb$tquart[j] >- NA
    }
    simobslist[[i]] <- simobscomb
  }

Calculate the area under the curve (AUC)

for(i in 1:length(simobslist))
  if (!is.null(simobslist[[i]]))
  {
    simobscomb <- simobslist[[i]]
# Calculat the AUC with the trapezoidal rule:    
    if (nrow(simobscomb)>1) 
    {
      for (k in 2:max(nrow(simobscomb)-1,2,na.rm=TRUE))
      {
        simobscomb$obsAUCtrap[1] <- 0
        simobscomb$simAUCtrap[1] <- 0
        if (min(simobscomb$time) <= (simobscomb$explen[1]*1.03) & 
            nrow(simobscomb) >=2)
        {
          simobscomb$obsAUCtrap[k] <- simobscomb$obsAUCtrap[k-1] + 
            0.5*(simobscomb$time[k] - simobscomb$time[k-1]) * 
            (simobscomb$obsconc[k] + simobscomb$obsconc[k-1])
          simobscomb$simAUCtrap[k] <- simobscomb$simAUCtrap[k-1] + 
            0.5*(simobscomb$time[k]-simobscomb$time[k-1]) * 
            (simobscomb$simconc[k] + simobscomb$simconc[k-1])
        } else {
          simobscomb$obsAUCtrap <- 0
          simobscomb$simAUCtrap <- 0
        }
      }
    } else {
      simobscomb$obsAUCtrap <- 0
      simobscomb$simAUCtrap <- 0
    }
    simobscomb$AUCobs <- max(simobscomb$obsAUCtrap)
    simobscomb$AUCsim <- max(simobscomb$simAUCtrap)
    simobscomb$calcAUC <- max(simobscomb$AUC)
    if (min(simobscomb$time) <= simobscomb$explen[1]*1.03)
    {
      simobscomb$Cmaxobs <- max(simobscomb$obsconc)
      simobscomb$Cmaxsim <- max(simobscomb$simconc)
    } else {
      simobscomb$Cmaxobs <- 0
      simobscomb$Cmaxsim <- 0
    }
    simobslist[[i]] <- simobscomb
  }

Calculate performance statistics

for(i in 1:length(simobslist))
  if (!is.null(simobslist[[i]]))
  {
    simobscomb <- simobslist[[i]]
    unique_scenarios$RSQD[i] <- 1 - (
      sum((simobscomb$obsconc - simobscomb$simconc)^2) / 
      sum((simobscomb$obsconc-mean(simobscomb$obsconc))^2)
      )
    unique_scenarios$RMSE[i] <- 
      sqrt(mean((simobscomb$simconc - simobscomb$obsconc)^2))
    unique_scenarios$AIC[i] <- 
      nrow(simobscomb)*(
        log(2*pi) + 1 +
        log((sum((simobscomb$obsconc-simobscomb$simconc)^2) /
          nrow(simobscomb)))
      ) + ((44+1)*2) #44 is the number of parameters from inhalation_inits.R
    simobslist[[i]] <- simobscomb
  }

Combine individual studies into single table

obsvpredlist <- list()
for(i in 1:length(simobslist))
  if (!is.null(simobslist[[i]]))
  {
    simobscomb <- simobslist[[i]]
    obsvpredlist[[i]] <- ggplot(simobscomb, aes(x = simconc, y = obsconc)) + 
      geom_point() + 
      geom_abline() + 
      xlab("Simulated Concentrations (uM)") + 
      ylab("Observed Concentrations (uM)") + 
      ggtitle(paste0(
        unique_scenarios$PREFERRED_NAME[i],
        " (", 
        unique_scenarios$CONC_SPECIES[i],
        ", ",
        round(as.numeric(unique_scenarios$DOSE[i]), digits = 2),
        unique_scenarios$DOSE_U[i], 
        " for ",
        round(unique_scenarios$EXP_LENGTH[i]*24, digits = 2),
        "h in ", 
        unique_scenarios$SAMPLING_MATRIX[i], ")")) + 
      theme_bw() + 
      theme(plot.title = element_text(face = 'bold', size = 20),
        axis.title.x = element_text(face = 'bold', size = 20), 
        axis.text.x = element_text(size=16), 
        axis.title.y = element_text(face = 'bold', size = 20), 
        axis.text.y = element_text(size = 16),
        legend.title = element_text(face = 'bold', size = 16),
        legend.text = element_text(face = 'bold',size = 14))
  }

Write out the study-level figures:

# Create pdfs of observed vs. predicted concentration plots
dir.create("Linakis2020")
pdf("Linakis2020/obvpredplots.pdf", width = 10, height = 10)
for (i in 1:length(obsvpredlist)) {
  print(obsvpredlist[[i]])
}
dev.off()

Finish simulated concentration/time plots

for (i in 1:length(cvtlist))
  if (!is.null(cvtlist[[i]]))
{
  cvtlist[[i]] <- cvtlist[[i]] + 
    geom_text(
      x = Inf, 
      y = Inf, 
      hjust = 1.3, 
      vjust = 1.3, 
#      size = 6, 
      label = paste0(
        "RMSE: ", 
        round(unique_scenarios$RMSE[i],digits = 2),
        "\nAIC: ", 
        round(unique_scenarios$AIC[i],digits = 2)))# + 
#    theme(
#      plot.title = element_text(face = 'bold', size = 15),
#      axis.title.x = element_text(face = 'bold', size = 20), 
#      axis.text.x = element_text(size=16), 
#      axis.title.y = element_text(face = 'bold', size = 20), 
#      axis.text.y = element_text(size = 16),
#      legend.title = element_text(face = 'bold', size = 16),
#      legend.text = element_text(face = 'bold',size = 14))
}
# Create pdfs of simulated concentration-time plots against observed c-t values
pdf("Linakis2020/simdataplots.pdf")
for (i in 1:length(cvtlist)) {
  print(cvtlist[[i]])
}
dev.off()

Combine obs. vs. pred. into single table:

simobsfull <- do.call("rbind",simobslist)
simobsfullrat <- subset(simobsfull, simobsfull$species == "Rat")
simobsfullhum <- subset(simobsfull, simobsfull$species == "Human")
unique_scenarios <- subset(unique_scenarios,!is.na(unique_scenarios$RSQD))

The observations in simobsfull are in both uM and ppmv – standardize to uM

these.chems <- unique(subset(simobsfull,unit=="ppmv")$chem)
for (this.chem in these.chems)
{
  # Use HTTK unit conversion:
  this.factor <- convert_units(
    input.units="ppmv", output.units="um", chem.name=this.chem, state="gas")
      
  # Scale the observation    
  simobsfull[simobsfull$chem==this.chem & 
              simobsfull$unit=="ppmv","obsconc"] <-
    this.factor * simobsfull[
      simobsfull$chem==this.chem & 
        simobsfull$unit=="ppmv","obsconc"]
  # Scale the prediction    
  simobsfull[simobsfull$chem==this.chem & 
              simobsfull$unit=="ppmv","simconc"] <-
    this.factor * simobsfull[
      simobsfull$chem==this.chem & 
        simobsfull$unit=="ppmv","simconc"]
  # Change the reported unit
  simobsfull[simobsfull$chem==this.chem & 
              simobsfull$unit=="ppmv","unit"] <-
    "uM"
}

Regressions

Other analytics including linear regression on overall concentration vs. time observed vs. predicted

table(unique_scenarios$CONC_SPECIES)
nrow(simobsfull) - nrow(simobsfull[
  !is.na(simobsfull$simconc) & 
  simobsfull$simconc > 0 & 
  simobsfull$obsconc > 0,])
pmiss <- (nrow(simobsfull) - 
  nrow(simobsfull[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc > 0,])) /
  nrow(simobsfull) * 100
missdata <- (simobsfull[
  is.na(simobsfull$simconc) | 
  simobsfull$simconc <= 0 | 
  simobsfull$obsconc <= 0,])
t0df <- simobsfull[simobsfull$obstime == 0,]
lmall <- lm(
#log transforms:
  log10(simobsfull$obsconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc > 0]) ~ 
#log transforms:
  log10(simobsfull$simconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc > 0])) 
#Linear binned 1
lmsub1 <- lm(
  simobsfull$obsconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc < 0.1] ~ 
  simobsfull$simconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc < 0.1])
#Linear binned 2
lmsub2 <- lm(
  simobsfull$obsconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc >= 0.1 & 
    simobsfull$obsconc < 10] ~ 
  simobsfull$simconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc >= 0.1 & 
    simobsfull$obsconc < 10]) 
#Linear binned 3
lmsub3 <- lm(
  simobsfull$obsconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc >= 10] ~ 
  simobsfull$simconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc >= 10]) 
lmrat <- lm(
  log10(simobsfullrat$obsconc[
    !is.na(simobsfullrat$simconc) & 
    simobsfullrat$simconc > 0 & 
    simobsfullrat$obsconc > 0]) ~ 
  log10(simobsfullrat$simconc[
    !is.na(simobsfullrat$simconc) & 
    simobsfullrat$simconc > 0 & 
    simobsfullrat$obsconc > 0]))
unique(simobsfullrat$chem)
lmhum <- lm(
  log10(simobsfullhum$obsconc[
    !is.na(simobsfullhum$simconc) & 
    simobsfullhum$simconc > 0 & 
    simobsfullhum$obsconc > 0]) ~ 
  log10(simobsfullhum$simconc[
    !is.na(simobsfullhum$simconc) & 
    simobsfullhum$simconc > 0 & 
    simobsfullhum$obsconc > 0]))
unique(simobsfullhum$chem)
concregslope <- summary(lmall)$coef[2,1]
concregr2 <- summary(lmall)$r.squared
concregrmse <- sqrt(mean(lmall$residuals^2))
totalrmse <- sqrt(mean((
  log10(simobsfull$simconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc > 0]) - 
  log10(simobsfull$obsconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc > 0]))^2, 
   na.rm = TRUE))
totalmae <- mean(abs(
  log10(simobsfull$simconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc > 0]) - 
  log10(simobsfull$obsconc[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc > 0 & 
    simobsfull$obsconc > 0])), 
  na.rm = TRUE)
totalaic <- nrow(
  simobsfull[
    !is.na(simobsfull$simconc) & 
    simobsfull$simconc >0 & 
    simobsfull$obsconc > 
    0,]) *
  (log(2*pi) + 
     1 +
     log((sum(
       (simobsfull$obsconc[
         !is.na(simobsfull$simconc) & 
         simobsfull$simconc > 0 & 
         simobsfull$obsconc > 0] - 
       simobsfull$simconc[
         !is.na(simobsfull$simconc) & 
         simobsfull$simconc > 0 & 
         simobsfull$obsconc > 0])^2,
       na.rm=TRUE) / 
     nrow(simobsfull[
       !is.na(simobsfull$simconc) & 
       simobsfull$simconc > 0 & 
       simobsfull$obsconc > 0,])))) + 
  ((44+1)*2) #44 is the number of parameters from inhalation_inits.R
mispred <- table(abs(
  log10(simobsfull$simconc) -
  log10(simobsfull$obsconc))>2 & 
  simobsfull$simconc>0)
mispred[2]
mispred[2] / nrow(simobsfull[
  !is.na(simobsfull$simconc) & 
    simobsfull$simconc >0 & 
    simobsfull$obsconc > 0,])*100
overpred <- table(
  log10(simobsfull$simconc) -
  log10(simobsfull$obsconc)>2 & 
  simobsfull$simconc>0)
overpred[2]
overpred[2] / nrow(simobsfull[
  !is.na(simobsfull$simconc) & 
  simobsfull$simconc >0 & 
  simobsfull$obsconc > 0,])*100
underpred <- table(
  log10(simobsfull$obsconc) - 
  log10(simobsfull$simconc)>2 & 
  simobsfull$simconc>0)
underpred[2]
underpred[2] / nrow(simobsfull[
  !is.na(simobsfull$simconc) & 
  simobsfull$simconc >0 & 
  simobsfull$obsconc > 0,])*100
mispredhalf <- table(abs(
  log10(simobsfull$simconc) -
  log10(simobsfull$obsconc))>0.5 & 
  simobsfull$simconc>0)
mispredhalf[2]
mispredhalf[2] / nrow(simobsfull[
  !is.na(simobsfull$simconc) & 
  simobsfull$simconc >0 & 
  simobsfull$obsconc > 0,])*100
overpredhalf <- table(
  log10(simobsfull$simconc) - 
  log10(simobsfull$obsconc)>0.5 & 
  simobsfull$simconc>0)
overpredhalf[2]
overpredhalf[2] / nrow(simobsfull[
  !is.na(simobsfull$simconc) & 
  simobsfull$simconc >0 & 
  simobsfull$obsconc > 0,])*100
underpredhalf <- table(
  log10(simobsfull$obsconc) - 
  log10(simobsfull$simconc)>0.5 & 
  simobsfull$simconc>0)
underpredhalf[2]
underpredhalf[2] / nrow(simobsfull[
  !is.na(simobsfull$simconc) & 
  simobsfull$simconc > 0 & 
  simobsfull$obsconc > 0,])*100
chemunderpred <- subset(simobsfull,
  log10(simobsfull$simconc) -
  log10(simobsfull$obsconc) < 0 & 
  simobsfull$simconc > 0)
table(chemunderpred$chemclass) / table(simobsfull$chemclass)*100

TABLE AND PLOT GENERATION

Figure 2: overall observed vs. predicted plot

fig2 <- ggplot(
  data = simobsfull[
    simobsfull$simconc > 0 & 
    simobsfull$obsconc > 0,], 
  aes(x = log10(simconc), y = log10(obsconc))) + 
  geom_point(
    color = ifelse(
      abs(
        log10(simobsfull[
          simobsfull$simconc > 0 & 
          simobsfull$obsconc > 0,]$simconc) -
        log10(simobsfull[
          simobsfull$simconc > 0 & 
          simobsfull$obsconc > 0,]$obsconc)) >2,
      'red',
      'black')) + 
  geom_abline() + 
  xlab("Log(Simulated Concentrations)") + 
  ylab("Log(Observed Concentrations)") + 
  theme_bw() + 
  geom_smooth(method = 'lm',se = FALSE, aes(color = 'Overall')) + 
  geom_smooth(method = 'lm', se = FALSE, aes(color = species)) + 
  geom_text(
    x = Inf, 
    y = -Inf, 
    hjust = 1.05, 
    vjust = -0.15, 
    size = 8, 
    label = paste0(
  #    "Regression slope: ", 
#      round(summary(lmall)$coef[2,1],digits = 2),
      "\nRegression R^2: ", 
      round(summary(lmall)$r.squared,digits = 2),
      "\nRegression RMSE: ", 
      round(sqrt(mean(lmall$residuals^2)),digits = 2),
      "\nRMSE (Identity): ", 
      round(totalrmse,digits = 2)
 #     "\n% Missing:", 
#      round(pmiss, digits = 2), "%")
    )) + 
  #geom_text(
  #  data = simobsfull[
  #    abs(log10(simobsfull$simconc) - log10(simobsfull$obsconc))>7 & 
  #    simobsfull$simconc>0 & simobsfull$obsconc > 0,], 
  #  aes(label = paste(chem,species,matrix)),
   ## fontface = 'bold',
  #  check_overlap = TRUE,
#    size = 3.5, 
  #  hjust = 0.5, 
  #  vjust = -0.8) + 
  scale_color_discrete(name = 'Species', breaks = c("Overall","Human","Rat")) +
  theme(
    plot.title = element_text(face = 'bold', size = 15),
    axis.title.x = element_text(face = 'bold', size = 30), 
    axis.text.x = element_text(size=16), 
    axis.title.y = element_text(face = 'bold', size = 30), 
    axis.text.y = element_text(size = 16),
    legend.title = element_text(face = 'bold', size = 24),
    legend.text = element_text(face = 'bold',size = 24))
fig2 #Display plot in R
library(scales)
# Function for formatting tick labels:
scientific_10 <- function(x) {                                  
  out <- gsub("1e", "10^", scientific_format()(x))              
  out <- gsub("\\+","",out)                                     
  out <- gsub("10\\^01","10",out)                               
  out <- parse(text=gsub("10\\^00","1",out))                    
}  

font.size.large <- 10
font.size.small <- 8

figaddmodels <- ggplot(
  data = simobsfull[
    simobsfull$simconc > 0 & 
    simobsfull$obsconc > 0,], 
  aes(x = simconc, y = obsconc)) + 
  geom_point(
    color = ifelse(
      abs(
        log10(simobsfull[
          simobsfull$simconc > 0 & 
          simobsfull$obsconc > 0,]$simconc) -
        log10(simobsfull[
          simobsfull$simconc > 0 & 
          simobsfull$obsconc > 0,]$obsconc)) >2,
      'red',
      'black'),alpha=0.15) + 
  geom_abline() + 
  scale_y_log10(label=scientific_10,limits=c(1e-4,1e4))+
  scale_x_log10(label=scientific_10,limits=c(1e-4,1e4))+
  xlab("Simulated Concentrations") + 
  ylab("Observed Concentrations") + 
  theme_bw() + 
  geom_smooth(method = 'lm',se = FALSE, aes(color = 'Overall', linetype="Overall")) + 
  geom_smooth(method = 'lm', se = FALSE, aes(color = species, linetype = species)) + 
  geom_text(
    x = 2, 
    y = -1, 
    size = 4, 
    label = paste0("RMSLE: ",
            round(totalrmse,digits = 2)
            )) + 
  scale_color_discrete(name = 'Species', breaks = c("Overall","Human","Rat")) +
  scale_linetype_discrete(name = 'Species', breaks = c("Overall","Human","Rat")) +
  theme(
    plot.title = element_text(face = 'bold', size = font.size.small),
    axis.title.x = element_text(face = 'bold', size = font.size.large), 
    axis.text.x = element_text(size=font.size.small), 
    axis.title.y = element_text(face = 'bold', size = font.size.large), 
    axis.text.y = element_text(size = font.size.small),
    legend.title = element_text(face = 'bold', size = font.size.large),
    legend.text = element_text(face = 'bold',size = font.size.large)) 
print(figaddmodels)
ggsave("c:/users/jwambaug/AddModelsFig1.tiff", width=6, height=4, dpi=300)

counts <- simobsfull[,c("chem","dose","explen","species")]
counts <- subset(counts,!duplicated(counts))
paste(length(unique(counts$chem)),
      "chemicals across",dim(counts)[1],
      "experimental conditions in",
      length(unique(counts$species)),"species.")
pdf("Linakis2020/Figure2.pdf", width = 10, height = 10)
print(fig2)
dev.off()

Create and read out plots of overall cvt, cmax, and auc observed vs. pred

# Create data and run calculations for populating plots
cmaxfull <- subset(simobsfull, !duplicated(simobsfull$AUCobs) & simobsfull$Cmaxobs != 0)
cmaxobs <- cmaxfull$Cmaxobs
cmaxsim <- cmaxfull$Cmaxsim
cmaxobs <- cmaxobs[!is.nan(cmaxsim)]
cmaxsim <- cmaxsim[!is.nan(cmaxsim)]
cmaxsim[!is.finite(log10(cmaxsim))] <- NA
cmaxlm <- lm(log10(cmaxobs)~log10(cmaxsim), na.action = na.exclude)
cmaxvcbkg <- subset(cmaxfull, 
  paste(
    cmaxfull$chem, 
    cmaxfull$dose, 
    cmaxfull$explen, 
    cmaxfull$species, 
    cmaxfull$matrix) %in% 
  paste(
    t0df$chem, 
    t0df$dose, 
    t0df$explen, 
    t0df$species,
    t0df$matrix))
cmaxvcbkg$cmaxcbkgratio <- cmaxvcbkg$Cmaxobs / cmaxvcbkg$obsconc
cmaxvcbkg$adjustedCmaxsim <- cmaxvcbkg$Cmaxsim - cmaxvcbkg$obsconc
aucfull <-subset(simobsfull, 
  !duplicated(simobsfull$AUCobs) & 
  simobsfull$AUCobs != 0)
aucobs <- aucfull$AUCobs
aucsim <- aucfull$AUCsim
aucobs <- aucobs[!is.nan(aucsim)]
aucsim <- aucsim[!is.nan(aucsim)]
aucsim[!is.finite(log10(aucsim))] <- NA
auclm <- lm(log10(aucobs)~log10(aucsim), na.action = na.exclude)
cmaxslope <- summary(cmaxlm)$coef[2,1]
cmaxrsq <- summary(cmaxlm)$r.squared
totalrmsecmax <- sqrt(mean((log10(cmaxfull$Cmaxsim) - 
  log10(cmaxfull$Cmaxobs))^2, na.rm = TRUE))
cmaxmiss <- nrow(cmaxfull[
  abs(log10(cmaxfull$Cmaxsim) - 
  log10(cmaxfull$Cmaxobs)) > 1,])
cmaxmissp <- nrow(cmaxfull[
  abs(log10(cmaxfull$Cmaxsim) - 
  log10(cmaxfull$Cmaxobs)) > 1,]) /
  nrow(cmaxfull) * 100
cmaxmisschem <- table(cmaxfull[
  abs(log10(cmaxfull$Cmaxsim) - 
  log10(cmaxfull$Cmaxobs)) > 1,]$chem)
aucslope <- summary(auclm)$coef[2,1]
aucrsq <- summary(auclm)$r.squared
totalrmseauc <- sqrt(mean((
  log10(aucfull$AUCsim) - 
  log10(aucfull$AUCobs))^2, na.rm = TRUE))
aucmiss <- nrow(aucfull[
  abs(log10(aucfull$AUCsim) - 
  log10(aucfull$AUCobs)) > 1,])
aucmissp <- nrow(aucfull[
  abs(log10(aucfull$AUCsim) - 
  log10(aucfull$AUCobs)) > 1,]) / 
  nrow(aucfull) * 100
aucmisschem <- table(aucfull[
  abs(log10(aucfull$AUCsim) - 
  log10(aucfull$AUCobs)) > 1,]$chem)

Figure 4: Cmax and AUC observed vs. Predicted Values

cmaxp <- ggplot(data = cmaxfull, aes(x = log10(Cmaxsim), y = log10(Cmaxobs))) +
  geom_point(color = 
    ifelse(abs(log10(cmaxfull$Cmaxsim)  -log10(cmaxfull$Cmaxobs))>=2, "red","black"))  +
  geom_abline()  + 
  xlab("Log(Simulated Max Concentration)") + 
  ylab("Log(Observed Max Concentration)") + 
  theme_bw() + 
  geom_smooth(method = 'lm', se = FALSE, aes(color = 'Overall')) + 
  geom_smooth(method = 'lm', se = FALSE, aes(color = species)) +
  geom_text(x = Inf, 
    y = -Inf, 
    hjust = 1.05, 
    vjust = -0.15, 
#    size = 6, 
    label = paste0("Regression slope: ", 
      round(summary(cmaxlm)$coef[2,1],digits = 2),
      "\nRegression R^2: ", 
      round(summary(cmaxlm)$r.squared,digits = 2))) +
  geom_text_repel(
    data = cmaxfull[
      (log10(cmaxfull$Cmaxsim)-log10(cmaxfull$Cmaxobs))>=2 & 
      log10(cmaxfull$Cmaxsim) > 2,], 
    aes(label = paste(chem,species,matrix)), 
    force = 2, 
 #   size = 5.3, 
    fontface = 'bold', 
    color = 'black', 
    hjust = -0.05, 
    vjust = 0.5) + 
  scale_y_continuous(lim = c (-1,5)) + 
  scale_x_continuous(lim = c(-1,5)) + 
  geom_text_repel(
    data = cmaxfull[
      (log10(cmaxfull$Cmaxsim)-log10(cmaxfull$Cmaxobs))>=2 &
      log10(cmaxfull$Cmaxsim) <= 2,], 
    aes(label = paste(chem,species,matrix)), 
    nudge_x = 0.0,
    nudge_y = -0.2, 
    force = 2, 
#    size = 5.3, 
    fontface = 'bold', 
    color = 'black', 
    hjust = -0.05, 
    vjust = 0.5) + 
  geom_text(
    data = cmaxfull[
      (log10(cmaxfull$Cmaxsim)-log10(cmaxfull$Cmaxobs))<=-2,], 
    aes(label = paste(chem,species,matrix)), 
#    size = 5.3, 
    fontface = 'bold', 
    color = 'black', 
    hjust = 0.5, 
    vjust = -0.7) + 
  scale_color_discrete(
    name = 'Species', 
    breaks = c("Overall","Human","Rat")) #+ 
#  theme(plot.title = element_text(face = 'bold', size = 10),
#    axis.title.x = element_text(face = 'bold', size = 10), 
#    axis.text.x = element_text(size=8), 
#    axis.title.y = element_text(face = 'bold', size = 10), 
#    axis.text.y = element_text(size = 8),
#    legend.title = element_text(face = 'bold', size = 8),
#    legend.text = element_text(face = 'bold',size = 8))
cmaxp
aucp <- ggplot(
  data = aucfull, 
  aes(x = log10(AUCsim), y = log10(AUCobs))) + 
  geom_point(color = 
    ifelse(abs(log10(aucfull$AUCsim)-log10(aucfull$AUCobs))>=2, "red","black"))  +
  geom_abline()  + 
  xlab("Log(Simulated AUC)") + 
  ylab("Log(Observed AUC)") + 
  theme_bw() + 
  geom_smooth(method = 'lm', se = FALSE, aes(color = "Overall")) + 
  geom_smooth(method = 'lm', se = FALSE, aes(color = species)) + 
  geom_text(
    x = Inf, 
    y = -Inf, 
    hjust = 1.05, 
    vjust = -0.15, 
#    size = 6, 
    label = paste0(
      "Regression slope: ", 
      round(summary(auclm)$coef[2,1],digits = 2),
      "\nRegression R^2: ", 
      round(summary(auclm)$r.squared,digits = 2))) + 
  geom_text_repel(
    data = aucfull[(log10(aucfull$AUCsim)-log10(aucfull$AUCobs))>=2,], 
    aes(label = paste(chem,species,matrix)), 
#    size = 5.3, 
    fontface = 'bold', 
    color = 'black', 
    hjust = -0.05, 
    vjust = 0.5) + 
  scale_y_continuous(lim = c (-2,4)) + 
  scale_x_continuous(lim = c(-2,4)) + 
  geom_text(
    data = aucfull[(log10(aucfull$AUCsim)-log10(aucfull$AUCobs))<=-2,], 
    aes(label = paste(chem,species,matrix)), 
#    size = 5.3, 
    fontface = 'bold', 
    color = 'black', 
    hjust = 0.5, 
    vjust = -0.8) + 
  scale_color_discrete(name = 'Species', breaks = c("Overall","Human","Rat")) #+
#  theme(
#    plot.title = element_text(face = 'bold', size = 15),
#    axis.title.x = element_text(face = 'bold', size = 20), 
#    axis.text.x = element_text(size=16), 
#    axis.title.y = element_text(face = 'bold', size = 20), 
#    axis.text.y = element_text(size = 16),
#    legend.title = element_text(face = 'bold', size = 16),
#    legend.text = element_text(face = 'bold',size = 14))
aucp
pdf("Linakis2020/Figure4.pdf", width = 20, height = 10)
plot_grid(cmaxp,aucp,nrow = 1, labels = c('A','B'), label_size = 30)
dev.off()

Figure 3: Separation by chemical class

simobsfull$aggscen <- as.factor(paste(simobsfull$chem, 
  simobsfull$species, 
  simobsfull$matrix))
chem.lm <- lm(
  log10(simconc) - log10(obsconc) ~ 
  aggscen, 
  data = simobsfull[simobsfull$simconc >0 & simobsfull$obsconc > 0,])
chem.res <- resid(chem.lm)
# Number of observations per chemical class
numpt <- simobsfull[simobsfull$simconc >0 & simobsfull$obsconc > 0,] %>%
  group_by(chemclass) %>% summarize(n = paste("n =", length(simconc))) 
fig3 <- ggplot(
  data = simobsfull[simobsfull$simconc >0 & simobsfull$obsconc > 0,], 
  aes(x = aggscen, y = log10(simconc)-log10(obsconc), fill = chemclass)) +
  geom_boxplot() +
  geom_hline(yintercept = 0) +
  geom_hline(yintercept = 2, linetype = 2)+
  geom_hline(yintercept = -2, linetype = 2)+
  xlab("Exposure Scenario") +
  ylab("Log(Simulated Concentration)-\nLog(Observed Concentration)\n") +
  facet_wrap(vars(chemclass), scales = 'free_x', nrow = 1) + #35 by 12 for poster
  theme_bw() +
  geom_text(
    data = numpt, 
    aes(x = Inf, y = -Inf, hjust = 1.05, vjust = -0.5, label = n), 
    size = 10, 
    colour = 'black', 
    inherit.aes = FALSE, 
    parse = FALSE) +
  theme(
    axis.text.x = element_text(angle = 90, hjust = 1,vjust=0.5,size = 20, face = 'bold'),
    strip.text.x = element_text(face = 'bold', size = 24), 
    legend.position = 'none', 
    axis.title.x = element_text(face = 'bold', size = 30), 
    axis.title.y = element_text(face = 'bold', size = 30), 
    axis.text.y = element_text(face = 'bold',size = 25, color = 'black'))
fig3
pdf("Linakis2020/Figure3.pdf", width = 40, height = 13)
print(fig3)
dev.off()

Figures S1A-S1D: Separation by time quartile and physicochemical properties

figs1a <- ggplot(
    data = simobsfull[simobsfull$simconc >0 & simobsfull$obsconc > 0,], 
    aes(x = tquart, y = log10(simconc)-log10(obsconc))) +
  geom_boxplot()+
  geom_hline(yintercept = 0)+
  geom_hline(yintercept = 2, linetype = 2)+
  geom_hline(yintercept = -2, linetype = 2)+
  xlab("\nTime Quartile\n") +
  ylab("Log(Simulated Concentration)-\nLog(Observed Concentration)\n") +
  theme_bw()+
  theme(
    axis.text.x = element_text(size = 20, face = 'bold'), 
    strip.text.x = element_text(face = 'bold', size = 20), 
    legend.position = 'none', 
    axis.title.x = element_text(face = 'bold', size = 20), 
    axis.title.y = element_text(face = 'bold', size = 20), 
    axis.text.y = element_text(size = 20, face = 'bold'))
figs1a
figs1b <- ggplot(
    data = simobsfull[simobsfull$simconc >0 & simobsfull$obsconc > 0,], 
    aes(x = mw, y = log10(simconc)-log10(obsconc))) +
  geom_point()+
  geom_hline(yintercept = 0)+
  geom_hline(yintercept = 2, linetype = 2)+
  geom_hline(yintercept = -2, linetype = 2)+
  xlab("\nMolecular Weight (g/mol)\n") +
  ylab("\nLog(Simulated Concentration)-\nLog(Observed Concentration)\n") +
  theme_bw()+
  theme(
    axis.text.x = element_text(size = 20, face = 'bold'), 
    strip.text.x = element_text(face = 'bold', size = 20), 
    legend.position = 'none', 
    axis.title.x = element_text(face = 'bold', size = 20), 
    axis.title.y = element_text(face = 'bold', size = 20), 
    axis.text.y = element_text(size = 20, face = 'bold'))
figs1b
figs1c <- ggplot(
    data = simobsfull[simobsfull$simconc >0 & simobsfull$obsconc > 0,], 
    aes(x = logp, y = log10(simconc)-log10(obsconc))) +
  geom_point()+
  geom_hline(yintercept = 0)+
  geom_hline(yintercept = 2, linetype = 2)+
  geom_hline(yintercept = -2, linetype = 2)+
  xlab("\nLog P") +
  ylab("Log(Simulated Concentration)-\nLog(Observed Concentration)\n") +
  theme_bw() +
  theme(
    axis.text.x = element_text(size = 20, face = 'bold'), 
    strip.text.x = element_text(face = 'bold', size = 20), 
    legend.position = 'none', 
    axis.title.x = element_text(face = 'bold', size = 20), 
    axis.title.y = element_text(face = 'bold', size = 20), 
    axis.text.y = element_text(size = 20, face = 'bold'))
figs1c
figs1d <- ggplot(
    data = simobsfull[simobsfull$simconc >0 & simobsfull$obsconc > 0,], 
    aes(x = sol, y = log10(simconc)-log10(obsconc))) +
  geom_point()+
  geom_hline(yintercept = 0)+
  geom_hline(yintercept = 2, linetype = 2)+
  geom_hline(yintercept = -2, linetype = 2)+
  xlab("\nSolubility (mol/L)") +
  ylab("\nLog(Simulated Concentration)-\nLog(Observed Concentration)\n") +
  scale_x_log10()+
  theme_bw() +
  theme(
    axis.text.x = element_text(size = 20, face = 'bold'), 
    strip.text.x = element_text(face = 'bold', size = 20), 
    legend.position = 'none', 
    axis.title.x = element_text(face = 'bold', size = 20), 
    axis.title.y = element_text(face = 'bold', size = 20), 
    axis.text.y = element_text(size = 20, face = 'bold'))
figs1d
pdf("Linakis2020/FigureS1.pdf", width = 20, height = 20)
plot_grid(figs1a,figs1b,figs1c,figs1d,nrow = 2, labels = c('A','B','C','D'), label_size = 30)
dev.off()

Supplemental Table 2: Leave-one-out Chemical Sensitivity Analysis

senschem <- list()
sensslope <- list()
sensrsq <- list()
sensregrmse <- list()
senstotalrmse <- list()
senspmiss <- list()
senscmaxslope <- list()
senscmaxrsq <- list()
senstotalrmsecmax <- list()
sensaucslope <- list()
sensaucrsq <- list()
senstotalrmseauc <- list()
for (i in 1:nrow(simobsfull))
{
  simobsfullsens <- subset(simobsfull,simobsfull$chem != simobsfull$chem[i])
  senschem[i] <- as.character(simobsfull$chem[i])
  senslm <- lm(
    log10(simobsfullsens$obsconc[
      !is.na(simobsfullsens$simconc) & 
      simobsfullsens$simconc > 0 & 
      simobsfullsens$obsconc > 0]) ~ 
    log10(simobsfullsens$simconc[
      !is.na(simobsfullsens$simconc) & 
      simobsfullsens$simconc >0 & 
      simobsfullsens$obsconc > 0]))
  sensslope[i] <- round(summary(senslm)$coef[2,1],digits = 2)
  sensrsq[i] <- round(summary(senslm)$r.squared,digits = 2)
  sensregrmse[i] <- round(sqrt(mean(senslm$residuals^2)),digits = 2)
  senstotalrmse[i] <- round(sqrt(mean((
    log10(simobsfullsens$simconc[
      !is.na(simobsfullsens$simconc) & 
      simobsfullsens$simconc >0 & 
      simobsfullsens$obsconc > 0]) -   
    log10(simobsfullsens$obsconc[
      !is.na(simobsfullsens$simconc) & 
        simobsfullsens$simconc >0 & 
        simobsfullsens$obsconc > 0]))^2, 
    na.rm = TRUE)), digits = 2)
  senspmiss[i] <- round((nrow(simobsfullsens) - 
    nrow(simobsfullsens[
      !is.na(simobsfullsens$simconc) & 
      simobsfullsens$simconc >0 & 
      simobsfullsens$obsconc > 0,])) / nrow(simobsfullsens) * 100, 
    digits = 2)
  senscmaxfull <- subset(simobsfullsens, !duplicated(simobsfullsens$Cmaxobs))
  senscmaxlm <- lm(
    log10(senscmaxfull$Cmaxobs[senscmaxfull$Cmaxobs>0]) ~
    log10(senscmaxfull$Cmaxsim[senscmaxfull$Cmaxobs>0]), 
    na.action = na.exclude)
  sensaucfull <-subset(simobsfullsens, !duplicated(simobsfullsens$AUCobs))
  sensauclm <- lm(
    log10(aucfull$AUCobs[aucfull$AUCobs>0]) ~
    log10(aucfull$AUCsim[aucfull$AUCobs>0]), 
    na.action = na.exclude)
  senscmaxslope[i] <- round(summary(senscmaxlm)$coef[2,1],digits = 2)
  senscmaxrsq[i] <- round(summary(senscmaxlm)$r.squared,digits = 2)
  senstotalrmsecmax[i] <- sqrt(mean((log10(senscmaxfull$Cmaxsim[senscmaxfull$Cmaxobs>0]) - log10(senscmaxfull$Cmaxobs[senscmaxfull$Cmaxobs>0]))^2, na.rm = TRUE))
  sensaucslope[i] <- round(summary(sensauclm)$coef[2,1],digits = 2)
  sensaucrsq[i] <- round(summary(sensauclm)$r.squared,digits = 2)
  senstotalrmseauc[i] <- sqrt(mean((log10(sensaucfull$AUCsim[sensaucfull$AUCobs>0]) - log10(sensaucfull$AUCobs[sensaucfull$AUCobs>0]))^2, na.rm = TRUE))
}
sensitivitydf <- data.frame(Chemical <- as.character(senschem),
                            sensSlope <- as.numeric(sensslope),
                            sensRsq <- as.numeric(sensrsq),
                            sensRegrmse <- as.numeric(sensregrmse),
                            sensTotrmse <- as.numeric(senstotalrmse),
                            sensPmiss <- as.numeric(senspmiss),
                            sensCmaxslope <- as.numeric(senscmaxslope),
                            sensCmaxrsq <- as.numeric(senscmaxrsq),
                            sensCmaxrmse <- as.numeric(senstotalrmsecmax),
                            sensAUCslope <- as.numeric(sensaucslope),
                            sensAUCrsq <- as.numeric(sensaucrsq),
                            sensAUCrmse <- as.numeric(senstotalrmseauc),
                            stringsAsFactors=FALSE)
sensitivitydf <- subset(sensitivitydf,!duplicated(sensitivitydf$Chemical....as.character.senschem.))
names(sensitivitydf) <- c('Chemical Dropped','Regression Slope','Regression R^2','Regression RMSE','Orthogonal RMSE', 'Percent Data Censored','Cmax Regression Slope','Cmax Regression R^2','Cmax Orthogonal RMSE','AUC Regression Slope','AUC Regression R^2','AUC Orthogonal RMSE')
notdropped <- c('None',concregslope,concregr2,concregrmse,totalrmse,pmiss,cmaxslope,cmaxrsq,totalrmsecmax,aucslope,aucrsq,totalrmseauc)
sensitivitydf <- rbind(notdropped, sensitivitydf)
sensitivitydf[,-1] <- sapply(sensitivitydf[,-1],as.numeric)
sensitivitydf[,-1] <- round(sensitivitydf[,-1],2)
  # Clean up and write file
rm(chem.lm, obvpredplot, p, pdata1, plot.data, plots, relconc, sensaucfull, sensauclm, sensaucrsq, sensaucslope, senschem, senscmaxfull, senscmaxlm, senscmaxrsq, senscmaxslope, senslm, senspmiss, sensregrmse, sensrsq, sensslope, senstotalrmse, senstotalrmseauc, senstotalrmsecmax, solve, AUCrmse, AUCrsq, AUCslope, chem.res, Chemical, Cmaxrmse, Cmaxrsq, Cmaxslope, colors, count, i, j, k, met_col, name, name1, Pmiss, Regrmse, Rsq, Slope, rem, Totrmse)

write.csv(sensitivitydf, 'supptab2.csv',row.names = FALSE)

Supplemental Table 1

supptab1 <- subset(unique_scenarios, !duplicated(unique_scenarios$PREFERRED_NAME) | !duplicated(unique_scenarios$SOURCE_CVT) | !duplicated(unique_scenarios$CONC_SPECIES))
for(i in 1:nrow(supptab1)){
  tryCatch({
    supptab1$Metabolism_Source[i] <- met_data$SOURCE_MET[met_data$DTXSID %in% supptab1$DTXSID[i] & met_data$SPECIES %in% supptab1$CONC_SPECIES[i]]
    supptab1$Log_P[i] <- met_data$OCTANOL_WATER_PARTITION_LOGP_OPERA_PRED[met_data$DTXSID %in% supptab1$DTXSID[i]& met_data$SPECIES %in% supptab1$CONC_SPECIES[i]]
    supptab1$Solubility[i] <- met_data$WATER_SOLUBILITY_MOL.L_OPERA_PRED[met_data$DTXSID %in% supptab1$DTXSID[i]& met_data$SPECIES %in% supptab1$CONC_SPECIES[i]]
    supptab1$Blood_Air_Partition_Coefficient[i] <- met_data$CALC_PBA[met_data$DTXSID %in% supptab1$DTXSID[i]& met_data$SPECIES %in% supptab1$CONC_SPECIES[i]]
    supptab1$Chem_Class[i] <- met_data$CHEM_CLASS[met_data$DTXSID %in% supptab1$DTXSID[i] & met_data$SPECIES %in% supptab1$CONC_SPECIES[i]]
    supptab1$Species[i] <- met_data$SPECIES[met_data$DTXSID %in% supptab1$DTXSID[i] & met_data$SPECIES %in% supptab1$CONC_SPECIES[i]]
    supptab1$Vmax[i] <- met_data$VMAX[met_data$DTXSID %in% supptab1$DTXSID[i] & met_data$SPECIES %in% supptab1$CONC_SPECIES[i]]
    supptab1$Km[i] <- met_data$KM[met_data$DTXSID %in% supptab1$DTXSID[i] & met_data$SPECIES %in% supptab1$CONC_SPECIES[i]]
  }, error = function(e){})
}
supptab1 <- supptab1[c('PREFERRED_NAME','DTXSID','CASRN','Chem_Class','AVERAGE_MASS','Log_P','Solubility','Blood_Air_Partition_Coefficient','Species','Vmax','Km','Metabolism_Source','SOURCE_CVT')]
names(supptab1) <- c('Chemical','DTXSID','CASRN','CAMEO Chemical Class','Molecular Weight (g/mol)','Log P','Solubility (mol/L)','Blood Air Partition Coefficient','Available Species Data','Vmax (pmol/min/10^6 cells)','KM (uM)','Metabolism Data Source','Concentration-Time Data Source')
write.csv(supptab1, "supptab1.csv", row.names = FALSE)