Identification of high lead exposure locations in Ohio at the census tract scale using a generalizable geospatial hotspot approach.

BACKGROUND: Lead is a persistent, ubiquitous pollutant whose historical sources have been largely addressed through regulation and voluntary actions. The United States (U.S.) has achieved significant decreases in children's blood lead levels (BLL) over the past 40 years; however, there is no known safe level of Pb exposure. Some communities continue to be disproportionately impacted by exposure to Pb, including Black children and families living in older homes. OBJECTIVE: To identify Ohio (OH) census tracts with children exposed to Pb and evaluate potential exposure determinants. METHODS: We obtained individual children's blood Pb data from 2005-2018 in OH. The percent of children with elevated BLL (EBLL) was calculated for OH census tracts using three blood Pb reference values (3.5, 5, and 10 µg/dL). Getis-Ord Gi* geospatial hotspot or top 20th percentile methodologies were then applied to identify "hotspots." Findings across multiple time periods and blood Pb reference values were evaluated and compared with existing Pb exposure indices and models. RESULTS: Consistency was observed across different blood Pb reference values, with the main hotspots identified at 3.5 µg/dL, also identified at 5 and 10 µg/dL. Substantial gains in public health were demonstrated, with the biggest decreases in the number of census tracts with EBLL observed between 2008-2010 and 2011-2013. Across OH, 355 census tracts (of 2850) were identified as hotspots across 17 locations, with the majority in the most populated cites. Generally, old housing and sociodemographic factors were indicators of these EBLL hotspots. A smaller number of hotspots were not associated with these exposure determinants. Variables of race, income, and education level were all strong predictors of hotspots. IMPACT STATEMENT: The Getis-Ord Gi* geospatial hotspot analysis can inform local investigations into potential Pb exposures for children living in OH. The successful application of a generalizable childhood blood Pb methodology at the census tract scale provides results that are more readily actionable. The moderate agreement of the measured blood Pb results with public Pb indices provide confidence that these indices can be used in the absence of available blood Pb surveillance data. While not a replacement for universal blood Pb testing, a consistent approach can be applied to identify areas where Pb exposure may be problematic.

A U.S. Lead Exposure Hotspots Analysis.

To identify U.S. lead exposure risk hotspots, we expanded upon geospatial statistical methods from a published Michigan case study. The evaluation of identified hotspots using five lead indices, based on housing age and sociodemographic data, showed moderate-to-substantial agreement with state-identified higher-risk locations from nine public health department reports (45-78%) and with hotspots of children's blood lead data from Michigan and Ohio (e.g., Cohen's kappa scores of 0.49-0.63). Applying geospatial cluster analysis and 80th-100th percentile methods to the lead indices, the number of U.S. census tracts ranged from ∼8% (intersection of indices) to ∼41% (combination of indices). Analyses of the number of children <6 years old living in those census tracts revealed the states (e.g., Illinois, Michigan, New Jersey, New York, Ohio, Pennsylvania, Massachusetts, California, Texas) and counties with highest potential lead exposure risk. Results support use of available lead indices as surrogates to identify locations in the absence of consistent, complete blood lead level (BLL) data across the United States. Ground-truthing with local knowledge, additional BLL data, and environmental data is needed to improve identification and analysis of lead exposure and BLL hotspots for interventions. While the science evolves, these screening results can inform "deeper dive" analyses for targeting lead actions.

A Machine Learning Model to Estimate Toxicokinetic Half-Lives of Per- and Polyfluoro-Alkyl Substances (PFAS) in Multiple Species.

Per- and polyfluoroalkyl substances (PFAS) are a diverse group of man-made chemicals that are commonly found in body tissues. The toxicokinetics of most PFAS are currently uncharacterized, but long half-lives (t½) have been observed in some cases. Knowledge of chemical-specific t½ is necessary for exposure reconstruction and extrapolation from toxicological studies. We used an ensemble machine learning method, random forest, to model the existing in vivo measured t½ across four species (human, monkey, rat, mouse) and eleven PFAS. Mechanistically motivated descriptors were examined, including two types of surrogates for renal transporters: (1) physiological descriptors, including kidney geometry, for renal transporter expression and (2) structural similarity of defluorinated PFAS to endogenous chemicals for transporter affinity. We developed a classification model for t½ (Bin 1: <12 h; Bin 2: <1 week; Bin 3: <2 months; Bin 4: >2 months). The model had an accuracy of 86.1% in contrast to 32.2% for a y-randomized null model. A total of 3890 compounds were within domain of the model, and t½ was predicted using the bin medians: 4.9 h, 2.2 days, 33 days, and 3.3 years. For human t½, 56% of PFAS were classified in Bin 4, 7% were classified in Bin 3, and 37% were classified in Bin 2. This model synthesizes the limited available data to allow tentative extrapolation and prioritization.

A Scoping Assessment of Implemented Toxicokinetic Models of Per- and Polyfluoro-Alkyl Substances, with a Focus on One-Compartment Models.

Toxicokinetic (TK) models have been used for decades to estimate concentrations of per-and polyfluoroalkyl substances (PFAS) in serum. However, model complexity has varied across studies depending on the application and the state of the science. This scoping effort seeks to systematically map the current landscape of PFAS TK models by categorizing different trends and similarities across model type, PFAS, and use scenario. A literature review using Web of Science and SWIFT-Review was used to identify TK models used for PFAS. The assessment covered publications from 2005-2020. PFOA, the PFAS for which most models were designed, was included in 69 of the 92 papers, followed by PFOS with 60, PFHxS with 22, and PFNA with 15. Only 4 of the 92 papers did not include analysis of PFOA, PFOS, PFNA, or PFHxS. Within the corpus, 50 papers contained a one-compartment model, 17 two-compartment models were found, and 33 used physiologically based pharmacokinetic (PBTK) models. The scoping assessment suggests that scientific interest has centered around two chemicals-PFOA and PFOS-and most analyses use one-compartment models in human exposure scenarios.

Lead Data Mapping to Prioritize US Locations for Whole-of-Government Exposure Prevention Efforts: State of the Science, Federal Collaborations, and Remaining Challenges.

For this state-of-science overview of geospatial approaches for identifying US communities with high lead-exposure risk, we compiled and summarized public data and national maps of lead indices and models, environmental lead indicators, and children's blood lead surveillance data. Currently available indices and models are primarily constructed from housing-age and sociodemographic data; differing methods, variables, data, weighting schemes, and geographic scales yield maps with different exposure risk profiles. Environmental lead indicators are available (e.g., air, drinking water, dust, soil) at different spatial scales, but key gaps remain. Blood lead level data have limitations as testing, reporting, and completeness vary across states. Mapping tools and approaches developed by federal agencies and other groups for different purposes present an opportunity for greater collaboration. Maps, data visualization tools, and analyses that synthesize available geospatial efforts can be evaluated and improved with local knowledge and blood lead data to refine identification of high-risk locations for prioritizing prevention efforts and targeting risk-reduction strategies. Remaining challenges are discussed along with a work-in-progress systematic approach for cross-agency data integration, toward advancing "whole-of-government" public health protection from lead exposures. (Am J Public Health. 2022;112(S7):S658-S669. https://doi.org/10.2105/AJPH.2022.307051).

A Generalizable Evaluated Approach, Applying Advanced Geospatial Statistical Methods, to Identify High Lead Exposure Locations at Census Tract Scale: Michigan Case Study.

BACKGROUND: Despite great progress in reducing environmental lead (Pb) levels, many children in the United States are still being exposed. OBJECTIVE: Our aim was to develop a generalizable approach for systematically identifying, verifying, and analyzing locations with high prevalence of children's elevated blood Pb levels (EBLLs) and to assess available Pb models/indices as surrogates, using a Michigan case study. METHODS: We obtained ∼1.9 million BLL test results of children <6 years of age in Michigan from 2006-2016; we then evaluated them for data representativeness by comparing two percentage EBLL (%EBLL) rates (number of children tested with EBLL divided by both number of children tested and total population). We analyzed %EBLLs across census tracts over three time periods and between two EBLL reference values (≥5 vs. ≥10µg/dL) to evaluate consistency. Locations with high %EBLLs were identified by a top 20 percentile method and a Getis-Ord Gi* geospatial cluster "hotspot" analysis. For the locations identified, we analyzed convergences with three available Pb exposure models/indices based on old housing and sociodemographics. RESULTS: Analyses of 2014-2016 %EBLL data identified 11 Michigan locations via cluster analysis and 80 additional locations via the top 20 percentile method and their associated census tracts. Data representativeness and consistency were supported by a 0.93 correlation coefficient between the two EBLL rates over 11 y, and a Kappa score of ∼0.8 of %EBLL hotspots across the time periods (2014-2016) and reference values. Many EBLL hotspot locations converge with current Pb exposure models/indices; others diverge, suggesting additional Pb sources for targeted interventions. DISCUSSION: This analysis confirmed known Pb hotspot locations and revealed new ones at a finer geographic resolution than previously available, using advanced geospatial statistical methods and mapping/visualization. It also assessed the utility of surrogates in the absence of blood Pb data. This approach could be applied to other states to inform Pb mitigation and prevention efforts. https://doi.org/10.1289/EHP9705.

Assessment of Non-Occupational 1,4-Dioxane Exposure Pathways from Drinking Water and Product Use.

1,4-Dioxane is a persistent and mobile organic chemical that has been found by the United States Environmental Protection Agency (USEPA) to be an unreasonable risk to human health in some occupational contexts. 1,4-Dioxane is released into the environment as industrial waste and occurs in some personal-care products as an unintended byproduct. However, limited exposure assessments have been conducted outside of an occupational context. In this study, the USEPA simulation modeling tool, Stochastic Human Exposure and Dose Simulator-High Throughput (SHEDS-HT), was adapted to estimate the exposure and chemical mass released down the drain (DTD) from drinking water consumption and product use. 1,4-Dioxane concentrations measured in drinking water and consumer products were used by SHEDS-HT to evaluate and compare the contributions of these sources to exposure and mass released DTD. Modeling results showed that compared to people whose daily per capita exposure came from only products (2.29 × 10-7 to 2.92 × 10-7 mg/kg/day), people exposed to both contaminated water and product use had higher per capita median exposures (1.90 × 10-6 to 4.27 × 10-6 mg/kg/day), with exposure mass primarily attributable to water consumption (75-91%). Last, we demonstrate through simulation that while a potential regulatory action could broadly reduce DTD release, the proportional reduction in exposure would be most significant for people with no or low water contamination.

Correction to Designing QSARs for Parameters of High Throughput Toxicokinetic Models Using Open-Source Descriptors.

The intrinsic metabolic clearance rate (Clint) and fraction of chemical unbound in plasma (fup) serve as important parameters for high throughput toxicokinetic models, but experimental data are limited for many chemicals. Open-source quantitative structure-activity relationship (QSAR) models for both parameters were developed to offer reliable in silico predictions for a diverse set of chemicals regulated under U.S. law, including pharmaceuticals, pesticides, and industrial chemicals. As a case study to demonstrate their utility, model predictions served as inputs to the TK component of a risk-based prioritization approach based on Bioactivity: Exposure Ratios (BER), in which a BER < 1 indicates exposures are predicted to exceed a biological activity threshold. When applied to a subset of the Tox21 screening library (6631 chemicals) we found that the proportion of chemicals with BER < 1 was similar using either in silico (1337/6631; 20.16%) or in vitro (151/850; 17.76%) parameters. Further, when considering only the chemicals in the Tox21 set with in vitro data, there was a high concordance of chemicals classified with either BER < 1 or >1 using either in silico or in vitro parameters (776/850, 91.30%). Thus, the presented QSARs may be suitable for prioritizing the risk posed by many chemicals for which measured in vitro TK data are lacking.

Mining of Consumer Product Ingredient and Purchasing Data to Identify Potential Chemical Coexposures.

BACKGROUND: Chemicals in consumer products are a major contributor to human chemical coexposures. Consumers purchase and use a wide variety of products containing potentially thousands of chemicals. There is a need to identify potential real-world chemical coexposures to prioritize in vitro toxicity screening. However, due to the vast number of potential chemical combinations, this identification has been a major challenge. OBJECTIVES: We aimed to develop and implement a data-driven procedure for identifying prevalent chemical combinations to which humans are exposed through purchase and use of consumer products. METHODS: We applied frequent itemset mining to an integrated data set linking consumer product chemical ingredient data with product purchasing data from 60,000 households to identify chemical combinations resulting from co-use of consumer products. RESULTS: We identified co-occurrence patterns of chemicals over all households as well as those specific to demographic groups based on race/ethnicity, income, education, and family composition. We also identified chemicals with the highest potential for aggregate exposure by identifying chemicals occurring in multiple products used by the same household. Last, a case study of chemicals active in estrogen and androgen receptor in silico models revealed priority chemical combinations co-targeting receptors involved in important biological signaling pathways. DISCUSSION: Integration and comprehensive analysis of household purchasing data and product-chemical information provided a means to assess human near-field exposure and inform selection of chemical combinations for high-throughput screening in in vitro assays. https://doi.org/10.1289/EHP8610.

Evaluation of Quantitative Structure Property Relationship Algorithms for Predicting Plasma Protein Binding in Humans.

The extent of plasma protein binding is an important compound-specific property that influences a compound's pharmacokinetic behavior and is a critical input parameter for predicting exposure in physiologically based pharmacokinetic (PBPK) modeling. When experimentally determined fraction unbound in plasma (fup) data are not available, quantitative structure-property relationship (QSPR) models can be used for prediction. Because available QSPR models were developed based on training sets containing pharmaceutical-like compounds, we compared their prediction accuracy for environmentally relevant and pharmaceutical compounds. Fup values were calculated using Ingle et al., Watanabe et al. and ADMET Predictor (Simulation Plus). The test set included 818 pharmaceutical and environmentally relevant compounds with fup values ranging from 0.01 to 1. Overall, the three QSPR models resulted in over-prediction of fup for highly binding compounds and under-prediction for low or moderately binding compounds. For highly binding compounds (0.01≤ fup ≤ 0.25), Watanabe et al. performed better with a lower mean absolute error (MAE) of 6.7% and a lower mean absolute relative prediction error (RPE) of 171.7 % than other methods. For low to moderately binding compounds, both Ingle et al. and ADMET Predictor performed better than Watanabe et al. with superior MAE and RPE values. The positive polar surface area, the number of basic functional groups and lipophilicity were the most important chemical descriptors for predicting fup. This study demonstrated that the prediction of fup was the most uncertain for highly binding compounds. This suggested that QSPR-predicted fup values should be used with caution in PBPK modeling.

Designing QSARs for Parameters of High-Throughput Toxicokinetic Models Using Open-Source Descriptors.

The intrinsic metabolic clearance rate (Clint) and the fraction of the chemical unbound in plasma (fup) serve as important parameters for high-throughput toxicokinetic (TK) models, but experimental data are limited for many chemicals. Open-source quantitative structure-activity relationship (QSAR) models for both parameters were developed to offer reliable in silico predictions for a diverse set of chemicals regulated under the U.S. law, including pharmaceuticals, pesticides, and industrial chemicals. As a case study to demonstrate their utility, model predictions served as inputs to the TK component of a risk-based prioritization approach based on bioactivity/exposure ratios (BERs), in which a BER < 1 indicates that exposures are predicted to exceed a biological activity threshold. When applied to a subset of the Tox21 screening library (6484 chemicals), we found that the proportion of chemicals with BER <1 was similar using either in silico (1133/6484; 17.5%) or in vitro (148/848; 17.5%) parameters. Further, when considering only the chemicals in the Tox21 set with in vitro data, there was a high concordance of chemicals classified with either BER <1 or >1 using either in silico or in vitro parameters (767/848, 90.4%). Thus, the presented QSARs may be suitable for prioritizing the risk posed by many chemicals for which measured in vitro TK data are lacking.

Data Mining Approaches for Assessing Chemical Coexposures Using Consumer Product Purchase Data.

The use of consumer products presents a potential for chemical exposures to humans. Toxicity testing and exposure models are routinely employed to estimate risks from their use; however, a key challenge is the sparseness of information concerning who uses products and which products are used contemporaneously. Our goal was to demonstrate a method to infer use patterns by way of purchase data. We examined purchase patterns for three types of personal care products (cosmetics, hair care, and skin care) and two household care products (household cleaners and laundry supplies) using data from 60,000 households collected over a one-year period in 2012. The market basket analysis methodology frequent itemset mining (FIM) was used to identify co-occurring sets of product purchases for all households and demographic groups based on income, education, race/ethnicity, and family composition. Our methodology captured robust co-occurrence patterns for personal and household products, globally and for different demographic groups. FIM identified cosmetic co-occurrence patterns captured in prior surveys of cosmetic use, as well as a trend of increased diversity of cosmetic purchases as children mature to teenage years. We propose that consumer product purchase data can be mined to inform person-oriented use patterns for high-throughput chemical screening applications, for aggregate and combined chemical risk evaluations.

Advancing systematic-review methodology in exposure science for environmental health decision making.

Systematic review (SR) is a rigorous methodology applied to synthesize and evaluate a body of scientific evidence to answer a research or policy question. Effective use of systematic-review methodology enables use of research evidence by decision makers. In addition, as reliance on systematic reviews increases, the required standards for quality of evidence enhances the policy relevance of research. Authoritative guidance has been developed for use of SR to evaluate evidence in the fields of medicine, social science, environmental epidemiology, toxicology, as well as ecology and evolutionary biology. In these fields, SR is typically used to evaluate a cause-effect relationship, such as the effect of an intervention, procedure, therapy, or exposure on an outcome. However, SR is emerging to be a useful methodology to transparently review and integrate evidence for a wider range of scientifically informed decisions and actions across disciplines. As SR is being used more broadly, there is growing consensus for developing resources, guidelines, ontologies, and technology to make SR more efficient and transparent, especially for handling large amounts of diverse data being generated across multiple scientific disciplines. In this article, we advocate for advancing SR methodology as a best practice in the field of exposure science to synthesize exposure evidence and enhance the value of exposure studies. We discuss available standards and tools that can be applied and extended by exposure scientists and highlight early examples of SRs being developed to address exposure research questions. Finally, we invite the exposure science community to engage in further development of standards and guidance to grow application of SR in this field and expand the opportunities for exposure science to inform environment and public health decision making.

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.

Systematic review and meta-analyses of lead (Pb) concentrations in environmental media (soil, dust, water, food, and air) reported in the United States from 1996 to 2016.

Environmental lead (Pb) contamination is a persistent public health issue that prominently impacts communities across the United States. Multimedia Pb exposure assessments are utilized to provide a holistic evaluation of Pb exposure and inform the development of programs and regulations that are protective of human health. To conduct multimedia exposure assessments, robust, media-specific environmental Pb concentration data are necessary. To support this effort, systematic review and meta-analysis methods were used to conduct a comprehensive synthesis of research measuring Pb in multiple environmental media (soil, dust, water, food, and air) over a 20-year period within the United States. The breadth of the resulting database allowed for the evaluation of sample characteristics that can serve as indicators of environmental Pb contamination. Random effects models run on literature and national survey datasets generated overall mean estimates of Pb concentrations that can be used for multimedia Pb exposure modeling for general and high-exposure-risk populations. Results from our study highlighted several important trends: 1) The mean estimate of Pb in residential soils is three times higher for urbanized areas than non-urbanized areas; 2) The mean estimate of Pb in produce reported in the literature is approximately three orders of magnitude greater than commercially-sourced raw produce monitored in national surveys; 3) The mean estimate of Pb in soils from shooting ranges is two times greater than non-residential Pb contaminated Superfund sites reported in the literature; 4) Research reporting environmental Pb concentrations for school and daycare sites is very limited; 5) Inconsistent sample collection and reporting of results limited synthesis efforts; and 6) The U.S. EPA's Air Quality System was the most robust, publicly available national survey resource. Results from these analyses will inform future multimedia Pb exposure assessments and be useful in prioritizing future research and program development.

Children's Lead Exposure: A Multimedia Modeling Analysis to Guide Public Health Decision-Making.

BACKGROUND: Drinking water and other sources for lead are the subject of public health concerns around the Flint, Michigan, drinking water and East Chicago, Indiana, lead in soil crises. In 2015, the U.S. Environmental Protection Agency (EPA)'s National Drinking Water Advisory Council (NDWAC) recommended establishment of a "health-based, household action level" for lead in drinking water based on children's exposure. OBJECTIVES: The primary objective was to develop a coupled exposure-dose modeling approach that can be used to determine what drinking water lead concentrations keep children's blood lead levels (BLLs) below specified values, considering exposures from water, soil, dust, food, and air. Related objectives were to evaluate the coupled model estimates using real-world blood lead data, to quantify relative contributions by the various media, and to identify key model inputs. METHODS: A modeling approach using the EPA's Stochastic Human Exposure and Dose Simulation (SHEDS)-Multimedia and Integrated Exposure Uptake and Biokinetic (IEUBK) models was developed using available data. This analysis for the U.S. population of young children probabilistically simulated multimedia exposures and estimated relative contributions of media to BLLs across all population percentiles for several age groups. RESULTS: Modeled BLLs compared well with nationally representative BLLs (0-23% relative error). Analyses revealed relative importance of soil and dust ingestion exposure pathways and associated Pb intake rates; water ingestion was also a main pathway, especially for infants. CONCLUSIONS: This methodology advances scientific understanding of the relationship between lead concentrations in drinking water and BLLs in children. It can guide national health-based benchmarks for lead and related community public health decisions. https://doi.org/10.1289/EHP1605.

Associations between socio-demographic characteristics and chemical concentrations contributing to cumulative exposures in the United States.

Association rule mining (ARM) has been widely used to identify associations between various entities in many fields. Although some studies have utilized it to analyze the relationship between chemicals and human health effects, fewer have used this technique to identify and quantify associations between environmental and social stressors. Socio-demographic variables were generated based on U.S. Census tract-level income, race/ethnicity population percentage, education level, and age information from the 2010-2014, 5-Year Summary files in the American Community Survey (ACS) database, and chemical variables were generated by utilizing the 2011 National-Scale Air Toxics Assessment (NATA) census tract-level air pollutant exposure concentration data. Six mobile- and industrial-source pollutants were chosen for analysis, including acetaldehyde, benzene, cyanide, particulate matter components of diesel engine emissions (namely, diesel PM), toluene, and 1,3-butadiene. ARM was then applied to quantify and visualize the associations between the chemical and socio-demographic variables. Census tracts with a high percentage of racial/ethnic minorities and populations with low income tended to have higher estimated chemical exposure concentrations (fourth quartile), especially for diesel PM, 1,3-butadiene, and toluene. In contrast, census tracts with an average population age of 40-50 years, a low percentage of racial/ethnic minorities, and moderate-income levels were more likely to have lower estimated chemical exposure concentrations (first quartile). Unsupervised data mining methods can be used to evaluate potential associations between environmental inequalities and social disparities, while providing support in public health decision-making contexts.

A Method for Identifying Prevalent Chemical Combinations in the U.S. Population.

BACKGROUND: Through the food and water they ingest, the air they breathe, and the consumer products with which they interact at home and at work, humans are exposed to tens of thousands of chemicals, many of which have not been evaluated to determine their potential toxicities. Furthermore, while current chemical testing tends to focus on individual chemicals, the exposures that people actually experience involve mixtures of chemicals. Unfortunately, the number of mixtures that can be formed from the thousands of environmental chemicals is enormous, and testing all of them would be impossible. OBJECTIVES: We seek to develop and demonstrate a method for identifying those mixtures that are most prevalent in humans. METHODS: We applied frequent itemset mining, a technique traditionally used for market basket analysis, to biomonitoring data from the 2009-2010 cycle of the continuous National Health and Nutrition Examination Survey (NHANES) to identify combinations of chemicals that frequently co-occur in people. RESULTS: We identified 90 chemical combinations consisting of relatively few chemicals that occur in at least 30% of the U.S. population, as well as three supercombinations consisting of relatively many chemicals that occur in a small but nonnegligible proportion of the U.S. population. CONCLUSIONS: We demonstrated how FIM can be used in conjunction with biomonitoring data to narrow a large number of possible chemical combinations down to a smaller set of prevalent chemical combinations. https://doi.org/10.1289/EHP1265.

Informing the Human Plasma Protein Binding of Environmental Chemicals by Machine Learning in the Pharmaceutical Space: Applicability Domain and Limits of Predictability.

The free fraction of a xenobiotic in plasma (Fub) is an important determinant of chemical adsorption, distribution, metabolism, elimination, and toxicity, yet experimental plasma protein binding data are scarce for environmentally relevant chemicals. The presented work explores the merit of utilizing available pharmaceutical data to predict Fub for environmentally relevant chemicals via machine learning techniques. Quantitative structure-activity relationship (QSAR) models were constructed with k nearest neighbors (kNN), support vector machines (SVM), and random forest (RF) machine learning algorithms from a training set of 1045 pharmaceuticals. The models were then evaluated with independent test sets of pharmaceuticals (200 compounds) and environmentally relevant ToxCast chemicals (406 total, in two groups of 238 and 168 compounds). The selection of a minimal feature set of 10-15 2D molecular descriptors allowed for both informative feature interpretation and practical applicability domain assessment via a bounded box of descriptor ranges and principal component analysis. The diverse pharmaceutical and environmental chemical sets exhibit similarities in terms of chemical space (99-82% overlap), as well as comparable bias and variance in constructed learning curves. All the models exhibit significant predictability with mean absolute errors (MAE) in the range of 0.10-0.18Fub. The models performed best for highly bound chemicals (MAE 0.07-0.12), neutrals (MAE 0.11-0.14), and acids (MAE 0.14-0.17). A consensus model had the highest accuracy across both pharmaceuticals (MAE 0.151-0.155) and environmentally relevant chemicals (MAE 0.110-0.131). The inclusion of the majority of the ToxCast test sets within the AD of the consensus model, coupled with high prediction accuracy for these chemicals, indicates the model provides a QSAR for Fub that is broadly applicable to both pharmaceuticals and environmentally relevant chemicals.

Developing a Physiologically-Based Pharmacokinetic Model Knowledgebase in Support of Provisional Model Construction.

Developing physiologically-based pharmacokinetic (PBPK) models for chemicals can be resource-intensive, as neither chemical-specific parameters nor in vivo pharmacokinetic data are easily available for model construction. Previously developed, well-parameterized, and thoroughly-vetted models can be a great resource for the construction of models pertaining to new chemicals. A PBPK knowledgebase was compiled and developed from existing PBPK-related articles and used to develop new models. From 2,039 PBPK-related articles published between 1977 and 2013, 307 unique chemicals were identified for use as the basis of our knowledgebase. Keywords related to species, gender, developmental stages, and organs were analyzed from the articles within the PBPK knowledgebase. A correlation matrix of the 307 chemicals in the PBPK knowledgebase was calculated based on pharmacokinetic-relevant molecular descriptors. Chemicals in the PBPK knowledgebase were ranked based on their correlation toward ethylbenzene and gefitinib. Next, multiple chemicals were selected to represent exact matches, close analogues, or non-analogues of the target case study chemicals. Parameters, equations, or experimental data relevant to existing models for these chemicals and their analogues were used to construct new models, and model predictions were compared to observed values. This compiled knowledgebase provides a chemical structure-based approach for identifying PBPK models relevant to other chemical entities. Using suitable correlation metrics, we demonstrated that models of chemical analogues in the PBPK knowledgebase can guide the construction of PBPK models for other chemicals.