Consensus Modeling of Median Chemical Intake for the U.S. Population Based on Predictions of Exposure Pathways.

Prioritizing the potential risk posed to human health by chemicals requires tools that can estimate exposure from limited information. In this study, chemical structure and physicochemical properties were used to predict the probability that a chemical might be associated with any of four exposure pathways leading from sources-consumer (near-field), dietary, far-field industrial, and far-field pesticide-to the general population. The balanced accuracies of these source-based exposure pathway models range from 73 to 81%, with the error rate for identifying positive chemicals ranging from 17 to 36%. We then used exposure pathways to organize predictions from 13 different exposure models as well as other predictors of human intake rates. We created a consensus, meta-model using the Systematic Empirical Evaluation of Models framework in which the predictors of exposure were combined by pathway and weighted according to predictive ability for chemical intake rates inferred from human biomonitoring data for 114 chemicals. The consensus model yields an R2 of ∼0.8. We extrapolate to predict relevant pathway(s), median intake rate, and credible interval for 479 926 chemicals, mostly with minimal exposure information. This approach identifies 1880 chemicals for which the median population intake rates may exceed 0.1 mg/kg bodyweight/day, while there is 95% confidence that the median intake rate is below 1 µg/kg BW/day for 474572 compounds.

Development of a prioritization method for chemical-mediated effects on steroidogenesis using an integrated statistical analysis of high-throughput H295R data.

Synthesis of 11 steroid hormones in human adrenocortical carcinoma cells (H295R) was measured in a high-throughput steroidogenesis assay (HT-H295R) for 656 chemicals in concentration-response as part of the US Environmental Protection Agency's ToxCast program. This work extends previous analysis of the HT-H295R dataset and model by examining the utility of a novel prioritization metric based on the Mahalanobis distance that reduced these 11-dimensional data to 1-dimension via calculation of a mean Mahalanobis distance (mMd) at each chemical concentration screened for all hormone measures available. Herein, we evaluated the robustness of mMd values, and demonstrate that covariance and variance of the hormones measured appear independent of the chemicals screened and are inherent to the assay; the Type I error rate of the mMd method is less than 1%; and, absolute fold changes (up or down) of 1.5 to 2-fold have sufficient power for statistical significance. As a case study, we examined hormone responses for aromatase inhibitors in the HT-H295R assay and found high concordance with other ToxCast assays for known aromatase inhibitors. Finally, we used mMd and other ToxCast cytotoxicity data to demonstrate prioritization of the most selective and active chemicals as candidates for further in vitro or in silico screening.

tcpl: the ToxCast pipeline for high-throughput screening data.

Motivation: Large high-throughput screening (HTS) efforts are widely used in drug development and chemical toxicity screening. Wide use and integration of these data can benefit from an efficient, transparent and reproducible data pipeline. Summary: The tcpl R package and its associated MySQL database provide a generalized platform for efficiently storing, normalizing and dose-response modeling of large high-throughput and high-content chemical screening data. The novel dose-response modeling algorithm has been tested against millions of diverse dose-response series, and robustly fits data with outliers and cytotoxicity-related signal loss. Availability and Implementation: tcpl is freely available on the Comprehensive R Archive Network under the GPL-2 license. Contact: martin.matt@epa.gov.

Evaluation and calibration of high-throughput predictions of chemical distribution to tissues.

Toxicokinetics (TK) provides critical information for integrating chemical toxicity and exposure assessments in order to determine potential chemical risk (i.e., the margin between toxic doses and plausible exposures). For thousands of chemicals that are present in our environment, in vivo TK data are lacking. The publicly available R package "httk" (version 1.8, named for "high throughput TK") draws from a database of in vitro data and physico-chemical properties in order to run physiologically-based TK (PBTK) models for 553 compounds. The PBTK model parameters include tissue:plasma partition coefficients (Kp) which the httk software predicts using the model of Schmitt (Toxicol In Vitro 22 (2):457-467, 2008). In this paper we evaluated and modified httk predictions, and quantified confidence using in vivo literature data. We used 964 rat Kp measured by in vivo experiments for 143 compounds. Initially, predicted Kp were significantly larger than measured Kp for many lipophilic compounds (log10 octanol:water partition coefficient > 3). Hence the approach for predicting Kp was revised to account for possible deficiencies in the in vitro protein binding assay, and the method for predicting membrane affinity was revised. These changes yielded improvements ranging from a factor of 10 to nearly a factor of 10,000 for 83 Kp across 23 compounds with only 3 Kp worsening by more than a factor of 10. The vast majority (92%) of Kp were predicted within a factor of 10 of the measured value (overall root mean squared error of 0.59 on log10-transformed scale). After applying the adjustments, regressions were performed to calibrate and evaluate the predictions for 12 tissues. Predictions for some tissues (e.g., spleen, bone, gut, lung) were observed to be better than predictions for other tissues (e.g., skin, brain, fat), indicating that confidence in the application of in silico tools to predict chemical partitioning varies depending upon the tissues involved. Our calibrated model was then evaluated using a second data set of human in vivo measurements of volume of distribution (Vss) for 498 compounds reviewed by Obach et al. (Drug Metab Dispos 36(7):1385-1405, 2008). We found that calibration of the model improved performance: a regression of the measured values as a function of the predictions has a slope of 1.03, intercept of - 0.04, and R2 of 0.43. Through careful evaluation of predictive methods for chemical partitioning into tissues, we have improved and calibrated these methods and quantified confidence for TK predictions in humans and rats.

httk: R Package for High-Throughput Toxicokinetics.

Thousands of chemicals have been profiled by high-throughput screening programs such as ToxCast and Tox21; these chemicals are tested in part because most of them have limited or no data on hazard, exposure, or toxicokinetics. Toxicokinetic models aid in predicting tissue concentrations resulting from chemical exposure, and a "reverse dosimetry" approach can be used to predict exposure doses sufficient to cause tissue concentrations that have been identified as bioactive by high-throughput screening. We have created four toxicokinetic models within the R software package httk. These models are designed to be parameterized using high-throughput in vitro data (plasma protein binding and hepatic clearance), as well as structure-derived physicochemical properties and species-specific physiological data. The package contains tools for Monte Carlo sampling and reverse dosimetry along with functions for the analysis of concentration vs. time simulations. The package can currently use human in vitro data to make predictions for 553 chemicals in humans, rats, mice, dogs, and rabbits, including 94 pharmaceuticals and 415 ToxCast chemicals. For 67 of these chemicals, the package includes rat-specific in vitro data. This package is structured to be augmented with additional chemical data as they become available. Package httk enables the inclusion of toxicokinetics in the statistical analysis of chemicals undergoing high-throughput screening.

Identifiability of PBPK models with applications to dimethylarsinic acid exposure.

Any statistical model should be identifiable in order for estimates and tests using it to be meaningful. We consider statistical analysis of physiologically-based pharmacokinetic (PBPK) models in which parameters cannot be estimated precisely from available data, and discuss different types of identifiability that occur in PBPK models and give reasons why they occur. We particularly focus on how the mathematical structure of a PBPK model and lack of appropriate data can lead to statistical models in which it is impossible to estimate at least some parameters precisely. Methods are reviewed which can determine whether a purely linear PBPK model is globally identifiable. We propose a theorem which determines when identifiability at a set of finite and specific values of the mathematical PBPK model (global discete identifiability) implies identifiability of the statistical model. However, we are unable to establish conditions that imply global discrete identifiability, and conclude that the only safe approach to analysis of PBPK models involves Bayesian analysis with truncated priors. Finally, computational issues regarding posterior simulations of PBPK models are discussed. The methodology is very general and can be applied to numerous PBPK models which can be expressed as linear time-invariant systems. A real data set of a PBPK model for exposure to dimethyl arsinic acid (DMA(V)) is presented to illustrate the proposed methodology.

Shape and steepness of toxicological dose-response relationships of continuous endpoints.

A re-analysis of a large number of historical dose-response data for continuous endpoints indicates that an exponential or a Hill model with four parameters adequately describes toxicological dose-responses. No exceptions were found for the datasets considered, which related to a wide variety of endpoints and to both in vivo and in vitro studies of various types. For a given endpoint/study type dose-response shapes were found to be homogenous among chemicals in the in vitro studies considered, while a mild among-chemical variation in the steepness parameter seemed to be present in the in vivo studies. Our findings have various practical consequences. For continuous endpoints, model selection in the BMD approach is not a crucial issue. The often applied approach of using constraints on the model parameters to prevent "infinite" slopes at dose zero in fitting a model is not in line with our findings, and appears to be unjustified. Instead, more realistic ranges of parameter values could be derived from re-analyses of large numbers of historical dose-response datasets in the same endpoint and study type, which could be used as parameter constraints in future individual datasets. This approach will be particularly useful for weak datasets (e.g. few doses, much scatter). In addition, this approach may open the way to use fewer animals in future studies. In the discussion, we argue that distinctions between linear, sub/supralinear or thresholded dose-response shapes, based on visual inspection of plots, are not biologically meaningful nor useful for risk assessment.

High throughput heuristics for prioritizing human exposure to environmental chemicals.

The risk posed to human health by any of the thousands of untested anthropogenic chemicals in our environment is a function of both the hazard presented by the chemical and the extent of exposure. However, many chemicals lack estimates of exposure intake, limiting the understanding of health risks. We aim to develop a rapid heuristic method to determine potential human exposure to chemicals for application to the thousands of chemicals with little or no exposure data. We used Bayesian methodology to infer ranges of exposure consistent with biomarkers identified in urine samples from the U.S. population by the National Health and Nutrition Examination Survey (NHANES). We performed linear regression on inferred exposure for demographic subsets of NHANES demarked by age, gender, and weight using chemical descriptors and use information from multiple databases and structure-based calculators. Five descriptors are capable of explaining roughly 50% of the variability in geometric means across 106 NHANES chemicals for all the demographic groups, including children aged 6-11. We use these descriptors to estimate human exposure to 7968 chemicals, the majority of which have no other quantitative exposure prediction. For thousands of chemicals with no other information, this approach allows forecasting of average exposure intake of environmental chemicals.

Characterizing Chemical Exposure Trends from NHANES Urinary Biomonitoring Data.

BACKGROUND: Xenobiotic metabolites are widely present in human urine and can indicate recent exposure to environmental chemicals. Proper inference of which chemicals contribute to these metabolites can inform human exposure and risk. Furthermore, longitudinal biomonitoring studies provide insight into how chemical exposures change over time. OBJECTIVES: We constructed an exposure landscape for as many human-exposure relevant chemicals over as large a time span as possible to characterize exposure trends across demographic groups and chemical types. METHODS: We analyzed urine data of nine 2-y cohorts (1999-2016) from the National Health and Nutrition Examination Survey (NHANES). Chemical daily intake rates (in milligrams per kilogram bodyweight per day) were inferred, using the R package bayesmarker, from metabolite concentrations in each cohort individually to identify exposure trends. Trends for metabolites and parents were clustered to find chemicals with similar exposure patterns. Exposure variation by age, gender, and body mass index were also assessed. RESULTS: Intake rates for 179 parent chemicals were inferred from 151 metabolites (96 measured in five or more cohorts). Seventeen metabolites and 44 parent chemicals exhibited fold-changes ≥10 between any two cohorts (deltamethrin, di-n-octyl phthalate, and di-isononyl phthalate had the greatest exposure increases). Di-2-ethylhexyl phthalate intake began decreasing in 2007, whereas both di-isobutyl and di-isononyl phthalate began increasing shortly before. Intake for four parabens was markedly higher in females, especially reproductive-age females, compared with males and children. Cadmium and arsenobetaine exhibited higher exposure for individuals >65 years of age and lower for individuals <20 years of age. DISCUSSION: With appropriate analysis, NHANES indicates trends in chemical exposures over the past two decades. Decreases in exposure are observable as the result of regulatory action, with some being accompanied by increases in replacement chemicals. Age- and gender-specific variations in exposure were observed for multiple chemicals. Continued estimation of demographic-specific exposures is needed to both monitor and identify potential vulnerable populations. https://doi.org/10.1289/EHP12188.

Is the assumption of normality or log-normality for continuous response data critical for benchmark dose estimation?

Continuous responses (e.g. body weight) are widely used in risk assessment for determining the benchmark dose (BMD) which is used to derive a U.S. EPA reference dose. One critical question that is not often addressed in dose-response assessments is whether to model the continuous data as normally or log-normally distributed. Additionally, if lognormality is assumed, and only summarized response data (i.e., mean±standard deviation) are available as is usual in the peer-reviewed literature, the BMD can only be approximated. In this study, using the "hybrid" method and relative deviation approach, we first evaluate six representative continuous dose-response datasets reporting individual animal responses to investigate the impact on BMD/BMDL estimates of (1) the distribution assumption and (2) the use of summarized versus individual animal data when a log-normal distribution is assumed. We also conduct simulation studies evaluating model fits to various known distributions to investigate whether the distribution assumption has influence on BMD/BMDL estimates. Our results indicate that BMDs estimated using the hybrid method are more sensitive to the distribution assumption than counterpart BMDs estimated using the relative deviation approach. The choice of distribution assumption has limited impact on the BMD/BMDL estimates when the within dose-group variance is small, while the lognormality assumption is a better choice for relative deviation method when data are more skewed because of its appropriateness in describing the relationship between mean and standard deviation. Additionally, the results suggest that the use of summarized data versus individual response data to characterize log-normal distributions has minimal impact on BMD estimates.

High-throughput models for exposure-based chemical prioritization in the ExpoCast project.

The United States Environmental Protection Agency (U.S. EPA) must characterize potential risks to human health and the environment associated with manufacture and use of thousands of chemicals. High-throughput screening (HTS) for biological activity allows the ToxCast research program to prioritize chemical inventories for potential hazard. Similar capabilities for estimating exposure potential would support rapid risk-based prioritization for chemicals with limited information; here, we propose a framework for high-throughput exposure assessment. To demonstrate application, an analysis was conducted that predicts human exposure potential for chemicals and estimates uncertainty in these predictions by comparison to biomonitoring data. We evaluated 1936 chemicals using far-field mass balance human exposure models (USEtox and RAIDAR) and an indicator for indoor and/or consumer use. These predictions were compared to exposures inferred by Bayesian analysis from urine concentrations for 82 chemicals reported in the National Health and Nutrition Examination Survey (NHANES). Joint regression on all factors provided a calibrated consensus prediction, the variance of which serves as an empirical determination of uncertainty for prioritization on absolute exposure potential. Information on use was found to be most predictive; generally, chemicals above the limit of detection in NHANES had consumer/indoor use. Coupled with hazard HTS, exposure HTS can place risk earlier in decision processes. High-priority chemicals become targets for further data collection.

Characterizing surface water concentrations of hundreds of organic chemicals in United States for environmental risk prioritization.

BACKGROUND: Thousands of chemicals are observed in freshwater, typically at trace levels. Measurements are collected for different purposes, so sample characteristics vary. Due to inconsistent data availability for exposure and hazard, it is complex to prioritize which chemicals may pose risks. OBJECTIVE: We evaluated the influence of data curation and statistical practices aggregating surface water measurements of organic chemicals into exposure distributions intended for prioritizing based on nation-scale potential risk. METHODS: The Water Quality Portal includes millions of observations describing over 1700 chemicals in 93% of hydrologic subbasins across the United States. After filtering to maintain quality and applicability while including all possible samples, we compared concentrations across sample types. We evaluated statistical methods to estimate per-chemical distributions for chosen samples. Overlaps between resulting exposure ranges and distributions representing no-effect concentrations for multiple freshwater species were used to rank estimated chemical risks for further assessment. RESULTS: When we apply explicit data quality and statistical assumptions, we find that there are 186 organic chemicals for which we can make screening-level estimates of surface water chemical concentration. Of the original 1700 observed chemicals, this number decreased primarily due to a predominance of censored values (that is, observations indicating concentrations too low to be measured). We further identify 423 chemicals where all measurements were censored but, through consideration of detection limits, risk might still be prioritized based on the detection limits themselves. In the final set of 1.5 million samples, the median environmental concentration of one chemical (acetic acid) exceeded the 5th percentile of no-effect concentrations for the most delicate freshwater species (the highest priority risk condition identified here), and a further 29 chemicals were identified for possible further evaluation based on a small margin between occurrence and toxicity values. SIGNIFICANCE: This method shows the broad range of chemical concentrations seen for organic chemicals across the country and identifies methods of determining their central tendency, allowing for researchers to characterize higher-than-normal or lower-than-normal surface water conditions as well as providing an overall indication of the presence of organic chemicals in the United States. The highest chemical concentrations did not always indicate the highest-risk conditions. Even when accounting for the high level of uncertainty in these data due to differences in data collection and reporting across the set, some chemicals may still be categorized as higher environmental risk than others using this method, providing value to chemical safety decision makers and researchers by suggesting avenues for more focused investigation.

Reproducibility of organ-level effects in repeat dose animal studies.

This work estimates benchmarks for new approach method (NAM) performance in predicting organ-level effects in repeat dose studies of adult animals based on variability in replicate animal studies. Treatment-related effect values from the Toxicity Reference database (v2.1) for weight, gross, or histopathological changes in the adrenal gland, liver, kidney, spleen, stomach, and thyroid were used. Rates of chemical concordance among organ-level findings in replicate studies, defined by repeated chemical only, chemical and species, or chemical and study type, were calculated. Concordance was 39 - 88%, depending on organ, and was highest within species. Variance in treatment-related effect values, including lowest effect level (LEL) values and benchmark dose (BMD) values when available, was calculated by organ. Multilinear regression modeling, using study descriptors of organ-level effect values as covariates, was used to estimate total variance, mean square error (MSE), and root residual mean square error (RMSE). MSE values, interpreted as estimates of unexplained variance, suggest study descriptors accounted for 52-69% of total variance in organ-level LELs. RMSE ranged from 0.41 - 0.68 log10-mg/kg/day. Differences between organ-level effects from chronic (CHR) and subchronic (SUB) dosing regimens were also quantified. Odds ratios indicated CHR organ effects were unlikely if the SUB study was negative. Mean differences of CHR - SUB organ-level LELs ranged from -0.38 to -0.19 log10 mg/kg/day; the magnitudes of these mean differences were less than RMSE for replicate studies. Finally, in vitro to in vivo extrapolation (IVIVE) was employed to compare bioactive concentrations from in vitro NAMs for kidney and liver to LELs. The observed mean difference between LELs and mean IVIVE dose predictions approached 0.5 log10-mg/kg/day, but differences by chemical ranged widely. Overall, variability in repeat dose organ-level effects suggests expectations for quantitative accuracy of NAM prediction of LELs should be at least ± 1 log10-mg/kg/day, with qualitative accuracy not exceeding 70%.

Bayesian inference of chemical exposures from NHANES urine biomonitoring data.

BACKGROUND: Knowing which environmental chemicals contribute to metabolites observed in humans is necessary for meaningful estimates of exposure and risk from biomonitoring data. OBJECTIVE: Employ a modeling approach that combines biomonitoring data with chemical metabolism information to produce chemical exposure intake rate estimates with well-quantified uncertainty. METHODS: Bayesian methodology was used to infer ranges of exposure for parent chemicals of biomarkers measured in urine samples from the U.S population by the National Health and Nutrition Examination Survey (NHANES). Metabolites were probabilistically linked to parent chemicals using the NHANES reports and text mining of PubMed abstracts. RESULTS: Chemical exposures were estimated for various population groups and translated to risk-based prioritization using toxicokinetic (TK) modeling and experimental data. Exposure estimates were investigated more closely for children aged 3 to 5 years, a population group that debuted with the 2015-2016 NHANES cohort. SIGNIFICANCE: The methods described here have been compiled into an R package, bayesmarker, and made publicly available on GitHub. These inferred exposures, when coupled with predicted toxic doses via high throughput TK, can help aid in the identification of public health priority chemicals via risk-based bioactivity-to-exposure ratios.

Influence of living in the same home on biomonitored levels of consumer product chemicals.

BACKGROUND: Individuals living in the same home may share exposures from direct contact with sources or indirectly through contamination of the home environment. OBJECTIVE: We investigated the influence of sharing a home on urine levels of ten phenolic chemicals present in some consumer products. METHODS: We used data from Silent Spring Institute's Detox Me Action Kit (DMAK), a crowdsourced biomonitoring program in the US. Of the 726 DMAK participants, 185 lived in the same home with one or more other DMAK participants (n = 137 pairs, up to six participants in a home). The concentration distributions included values below the detection limit so we used statistical methods that account for left-censored data, including non-parametric correlation estimation and hierarchical Bayesian regression models. RESULTS: Concentrations were significantly positively correlated between pair-members sharing a home for nine of the ten chemicals. Concentrations of 2,5-dichlorophenol were the most strongly correlated between pair-members (tau = 0.46), followed by benzophenone-3 (tau = 0.31) and bisphenol A (tau = 0.21). The relative contribution of personal product use reported product use of other household members (up to 5 others), and the residual contribution from a shared household, including exposures not asked about, varied by chemical. Paraben concentrations were largely influenced by personal behaviors whereas dichlorophenol and bisphenol concentrations were largely influenced by shared home exposures not related to reported behaviors. SIGNIFICANCE: Measuring the influence of personal and household practices on biomonitoring exposures helps pinpoint major sources of exposure and highlights chemical-specific intervention strategies to reduce them.

Estimating toxicity-related biological pathway altering doses for high-throughput chemical risk assessment.

We describe a framework for estimating the human dose at which a chemical significantly alters a biological pathway in vivo, making use of in vitro assay data and an in vitro-derived pharmacokinetic model, coupled with estimates of population variability and uncertainty. The quantity we calculate, the biological pathway altering dose (BPAD), is analogous to current risk assessment metrics in that it combines dose-response data with analysis of uncertainty and population variability to arrive at conservative exposure limits. The analogy is closest when perturbation of a pathway is a key event in the mode of action (MOA) leading to a specified adverse outcome. Because BPADs are derived from relatively inexpensive, high-throughput screening (HTS) in vitro data, this approach can be applied to high-throughput risk assessments (HTRA) for thousands of data-poor environmental chemicals. We envisage the first step of HTRA to be an assessment of in vitro concentration-response relationships across biologically important pathways to derive biological pathway altering concentrations (BPAC). Pharmacokinetic (PK) modeling is then used to estimate the in vivo doses required to achieve the BPACs in the blood at steady state. Uncertainty and variability are incorporated in both the BPAC and the PK parameters and then combined to yield a probability distribution for the dose required to perturb the critical pathway. We finally define the BPADL as the lower confidence bound of this pathway-altering dose. This perspective outlines a framework for using HTRA to estimate BPAD values; provides examples of the use of this approach, including a comparison of BPAD values with published dose-response data from in vivo studies; and discusses challenges and alternative formulations.

Comparison of Approaches for Determining Bioactivity Hits from High-Dimensional Profiling Data.

Phenotypic profiling assays are untargeted screening assays that measure a large number (hundreds to thousands) of cellular features in response to a stimulus and often yield diverse and unanticipated profiles of phenotypic effects, leading to challenges in distinguishing active from inactive treatments. Here, we compare a variety of different strategies for hit identification in imaging-based phenotypic profiling assays using a previously published Cell Painting data set. Hit identification strategies based on multiconcentration analysis involve curve fitting at several levels of data aggregation (e.g., individual feature level, aggregation of similarly derived features into categories, and global modeling of all features) and on computed metrics (e.g., Euclidean and Mahalanobis distance metrics and eigenfeatures). Hit identification strategies based on single-concentration analysis included measurement of signal strength (e.g., total effect magnitude) and correlation of profiles among biological replicates. Modeling parameters for each approach were optimized to retain the ability to detect a reference chemical with subtle phenotypic effects while limiting the false-positive rate to 10%. The percentage of test chemicals identified as hits was highest for feature-level and category-based approaches, followed by global fitting, whereas signal strength and profile correlation approaches detected the fewest number of active hits at the fixed false-positive rate. Approaches involving fitting of distance metrics had the lowest likelihood for identifying high-potency false-positive hits that may be associated with assay noise. Most of the methods achieved a 100% hit rate for the reference chemical and high concordance for 82% of test chemicals, indicating that hit calls are robust across different analysis approaches.

Variability in in vivo studies: Defining the upper limit of performance for predictions of systemic effect levels.

New approach methodologies (NAMs) for chemical hazard assessment are often evaluated via comparison to animal studies; however, variability in animal study data limits NAM accuracy. The US EPA Toxicity Reference Database (ToxRefDB) enables consideration of variability in effect levels, including the lowest effect level (LEL) for a treatment-related effect and the lowest observable adverse effect level (LOAEL) defined by expert review, from subacute, subchronic, chronic, multi-generation reproductive, and developmental toxicity studies. The objectives of this work were to quantify the variance within systemic LEL and LOAEL values, defined as potency values for effects in adult or parental animals only, and to estimate the upper limit of NAM prediction accuracy. Multiple linear regression (MLR) and augmented cell means (ACM) models were used to quantify the total variance, and the fraction of variance in systemic LEL and LOAEL values explained by available study descriptors (e.g., administration route, study type). The MLR approach considered each study descriptor as an independent contributor to variance, whereas the ACM approach combined categorical descriptors into cells to define replicates. Using these approaches, total variance in systemic LEL and LOAEL values (in log10-mg/kg/day units) ranged from 0.74 to 0.92. Unexplained variance in LEL and LOAEL values, approximated by the residual mean square error (MSE), ranged from 0.20-0.39. Considering subchronic, chronic, or developmental study designs separately resulted in similar values. Based on the relationship between MSE and R-squared for goodness-of-fit, the maximal R-squared may approach 55 to 73% for a NAM-based predictive model of systemic toxicity using these data as reference. The root mean square error (RMSE) ranged from 0.47 to 0.63 log10-mg/kg/day, depending on dataset and regression approach, suggesting that a two-sided minimum prediction interval for systemic effect levels may have a width of 58 to 284-fold. These findings suggest quantitative considerations for building scientific confidence in NAM-based systemic toxicity predictions.

Simulating exposure-related behaviors using agent-based models embedded with needs-based artificial intelligence.

Exposure to a chemical is a critical consideration in the assessment of risk, as it adds real-world context to toxicological information. Descriptions of where and how individuals spend their time are important for characterizing exposures to chemicals in consumer products and in indoor environments. Herein we create an agent-based model (ABM) that simulates longitudinal patterns in human behavior. By basing the ABM upon an artificial intelligence (AI) system, we create agents that mimic human decisions on performing behaviors relevant for determining exposures to chemicals and other stressors. We implement the ABM in a computer program called the Agent-Based Model of Human Activity Patterns (ABMHAP) that predicts the longitudinal patterns for sleeping, eating, commuting, and working. We then show that ABMHAP is capable of simulating behavior over extended periods of time. We propose that this framework, and models based on it, can generate longitudinal human behavior data for use in exposure assessments.

Empirical models for anatomical and physiological changes in a human mother and fetus during pregnancy and gestation.

Many parameters treated as constants in traditional physiologically based pharmacokinetic models must be formulated as time-varying quantities when modeling pregnancy and gestation due to the dramatic physiological and anatomical changes that occur during this period. While several collections of empirical models for such parameters have been published, each has shortcomings. We sought to create a repository of empirical models for tissue volumes, blood flow rates, and other quantities that undergo substantial changes in a human mother and her fetus during the time between conception and birth, and to address deficiencies with similar, previously published repositories. We used maximum likelihood estimation to calibrate various models for the time-varying quantities of interest, and then used the Akaike information criterion to select an optimal model for each quantity. For quantities of interest for which time-course data were not available, we constructed composite models using percentages and/or models describing related quantities. In this way, we developed a comprehensive collection of formulae describing parameters essential for constructing a PBPK model of a human mother and her fetus throughout the approximately 40 weeks of pregnancy and gestation. We included models describing blood flow rates through various fetal blood routes that have no counterparts in adults. Our repository of mathematical models for anatomical and physiological quantities of interest provides a basis for PBPK models of human pregnancy and gestation, and as such, it can ultimately be used to support decision-making with respect to optimal pharmacological dosing and risk assessment for pregnant women and their developing fetuses. The views expressed in this article are those of the authors and do not necessarily represent the views or policies of the U.S. Environmental Protection Agency.