From big data to big insights: statistical and bioinformatic approaches for exploring the lipidome.

The goal of lipidomic studies is to provide a broad characterization of cellular lipids present and changing in a sample of interest. Recent lipidomic research has significantly contributed to revealing the multifaceted roles that lipids play in fundamental cellular processes, including signaling, energy storage, and structural support. Furthermore, these findings have shed light on how lipids dynamically respond to various perturbations. Continued advancement in analytical techniques has also led to improved abilities to detect and identify novel lipid species, resulting in increasingly large datasets. Statistical analysis of these datasets can be challenging not only because of their vast size, but also because of the highly correlated data structure that exists due to many lipids belonging to the same metabolic or regulatory pathways. Interpretation of these lipidomic datasets is also hindered by a lack of current biological knowledge for the individual lipids. These limitations can therefore make lipidomic data analysis a daunting task. To address these difficulties and shed light on opportunities and also weaknesses in current tools, we have assembled this review. Here, we illustrate common statistical approaches for finding patterns in lipidomic datasets, including univariate hypothesis testing, unsupervised clustering, supervised classification modeling, and deep learning approaches. We then describe various bioinformatic tools often used to biologically contextualize results of interest. Overall, this review provides a framework for guiding lipidomic data analysis to promote a greater assessment of lipidomic results, while understanding potential advantages and weaknesses along the way.

High-throughput screening and genome-wide analyses of 44 anticancer drugs in the 1000 Genomes cell lines reveals an association of the NQO1 gene with the response of multiple anticancer drugs.

Cancer patients exhibit a broad range of inter-individual variability in response and toxicity to widely used anticancer drugs, and genetic variation is a major contributor to this variability. To identify new genes that influence the response of 44 FDA-approved anticancer drug treatments widely used to treat various types of cancer, we conducted high-throughput screening and genome-wide association mapping using 680 lymphoblastoid cell lines from the 1000 Genomes Project. The drug treatments considered in this study represent nine drug classes widely used in the treatment of cancer in addition to the paclitaxel + epirubicin combination therapy commonly used for breast cancer patients. Our genome-wide association study (GWAS) found several significant and suggestive associations. We prioritized consistent associations for functional follow-up using gene-expression analyses. The NAD(P)H quinone dehydrogenase 1 (NQO1) gene was found to be associated with the dose-response of arsenic trioxide, erlotinib, trametinib, and a combination treatment of paclitaxel + epirubicin. NQO1 has previously been shown as a biomarker of epirubicin response, but our results reveal novel associations with these additional treatments. Baseline gene expression of NQO1 was positively correlated with response for 43 of the 44 treatments surveyed. By interrogating the functional mechanisms of this association, the results demonstrate differences in both baseline and drug-exposed induction.

Aggregated Molecular Phenotype Scores: Enhancing Assessment and Visualization of Mass Spectrometry Imaging Data for Tissue-Based Diagnostics.

Mass spectrometry imaging (MSI) has gained increasing popularity for tissue-based diagnostics due to its ability to identify and visualize molecular characteristics unique to different phenotypes within heterogeneous samples. Data from MSI experiments are often assessed and visualized using various supervised and unsupervised statistical approaches. However, these approaches tend to fall short in identifying and concisely visualizing subtle, phenotype-relevant molecular changes. To address these shortcomings, we developed aggregated molecular phenotype (AMP) scores. AMP scores are generated using an ensemble machine learning approach to first select features differentiating phenotypes, weight the features using logistic regression, and combine the weights and feature abundances. AMP scores are then scaled between 0 and 1, with lower values generally corresponding to class 1 phenotypes (typically control) and higher scores relating to class 2 phenotypes. AMP scores, therefore, allow the evaluation of multiple features simultaneously and showcase the degree to which these features correlate with various phenotypes. Due to the ensembled approach, AMP scores are able to overcome limitations associated with individual models, leading to high diagnostic accuracy and interpretability. Here, AMP score performance was evaluated using metabolomic data collected from desorption electrospray ionization MSI. Initial comparisons of cancerous human tissues to their normal or benign counterparts illustrated that AMP scores distinguished phenotypes with high accuracy, sensitivity, and specificity. Furthermore, when combined with spatial coordinates, AMP scores allow visualization of tissue sections in one map with distinguished phenotypic borders, highlighting their diagnostic utility.

Leveraging high-throughput screening data, deep neural networks, and conditional generative adversarial networks to advance predictive toxicology.

There are currently 85,000 chemicals registered with the Environmental Protection Agency (EPA) under the Toxic Substances Control Act, but only a small fraction have measured toxicological data. To address this gap, high-throughput screening (HTS) and computational methods are vital. As part of one such HTS effort, embryonic zebrafish were used to examine a suite of morphological and mortality endpoints at six concentrations from over 1,000 unique chemicals found in the ToxCast library (phase 1 and 2). We hypothesized that by using a conditional generative adversarial network (cGAN) or deep neural networks (DNN), and leveraging this large set of toxicity data we could efficiently predict toxic outcomes of untested chemicals. Utilizing a novel method in this space, we converted the 3D structural information into a weighted set of points while retaining all information about the structure. In vivo toxicity and chemical data were used to train two neural network generators. The first was a DNN (Go-ZT) while the second utilized cGAN architecture (GAN-ZT) to train generators to produce toxicity data. Our results showed that Go-ZT significantly outperformed the cGAN, support vector machine, random forest and multilayer perceptron models in cross-validation, and when tested against an external test dataset. By combining both Go-ZT and GAN-ZT, our consensus model improved the SE, SP, PPV, and Kappa, to 71.4%, 95.9%, 71.4% and 0.673, respectively, resulting in an area under the receiver operating characteristic (AUROC) of 0.837. Considering their potential use as prescreening tools, these models could provide in vivo toxicity predictions and insight into the hundreds of thousands of untested chemicals to prioritize compounds for HT testing.

Utilizing Pine Needles to Temporally and Spatially Profile Per- and Polyfluoroalkyl Substances (PFAS).

As concerns over exposure to per- and polyfluoroalkyl substances (PFAS) are continually increasing, novel methods to monitor their presence and modifications are greatly needed, as some have known toxic and bioaccumulative characteristics while most have unknown effects. This task however is not simple, as the Environmental Protection Agency (EPA) CompTox PFAS list contains more than 9000 substances as of September 2020 with additional substances added continually. Nontargeted analyses are therefore crucial to investigating the presence of this immense list of possible PFAS. Here, we utilized archived and field-sampled pine needles as widely available passive samplers and a novel nontargeted, multidimensional analytical method coupling liquid chromatography, ion mobility spectrometry, and mass spectrometry (LC-IMS-MS) to evaluate the temporal and spatial presence of numerous PFAS. Over 70 PFAS were detected in the pine needles from this study, including both traditionally monitored legacy perfluoroalkyl acids (PFAAs) and their emerging replacements such as chlorinated derivatives, ultrashort chain PFAAs, perfluoroalkyl ether acids including hexafluoropropylene oxide dimer acid (HFPO-DA, "GenX") and Nafion byproduct 2, and a cyclic perfluorooctanesulfonic acid (PFOS) analog. Results from this study provide critical insight related to PFAS transport, contamination, and reduction efforts over the past six decades.

Demonstrating a systems approach for integrating disparate data streams to inform decisions on children's environmental health.

BACKGROUND: The use of systems science methodologies to understand complex environmental and human health relationships is increasing. Requirements for advanced datasets, models, and expertise limit current application of these approaches by many environmental and public health practitioners. METHODS: A conceptual system-of-systems model was applied for children in North Carolina counties that includes example indicators of children's physical environment (home age, Brownfield sites, Superfund sites), social environment (caregiver's income, education, insurance), and health (low birthweight, asthma, blood lead levels). The web-based Toxicological Prioritization Index (ToxPi) tool was used to normalize the data, rank the resulting vulnerability index, and visualize impacts from each indicator in a county. Hierarchical clustering was used to sort the 100 North Carolina counties into groups based on similar ToxPi model results. The ToxPi charts for each county were also superimposed over a map of percentage county population under age 5 to visualize spatial distribution of vulnerability clusters across the state. RESULTS: Data driven clustering for this systems model suggests 5 groups of counties. One group includes 6 counties with the highest vulnerability scores showing strong influences from all three categories of indicators (social environment, physical environment, and health). A second group contains 15 counties with high vulnerability scores driven by strong influences from home age in the physical environment and poverty in the social environment. A third group is driven by data on Superfund sites in the physical environment. CONCLUSIONS: This analysis demonstrated how systems science principles can be used to synthesize holistic insights for decision making using publicly available data and computational tools, focusing on a children's environmental health example. Where more traditional reductionist approaches can elucidate individual relationships between environmental variables and health, the study of collective, system-wide interactions can enable insights into the factors that contribute to regional vulnerabilities and interventions that better address complex real-world conditions.

Multiomic Big Data Analysis Challenges: Increasing Confidence in the Interpretation of Artificial Intelligence Assessments.

The need for holistic molecular measurements to better understand disease initiation, development, diagnosis, and therapy has led to an increasing number of multiomic analyses. The wealth of information available from multiomic assessments, however, requires both the evaluation and interpretation of extremely large data sets, limiting analysis throughput and ease of adoption. Computational methods utilizing artificial intelligence (AI) provide the most promising way to address these challenges, yet despite the conceptual benefits of AI and its successful application in singular omic studies, the widespread use of AI in multiomic studies remains limited. Here, we discuss present and future capabilities of AI techniques in multiomic studies while introducing analytical checks and balances to validate the computational conclusions.

Structural-based connectivity and omic phenotype evaluations (SCOPE): a cheminformatics toolbox for investigating lipidomic changes in complex systems.

Since its inception, the main goal of the lipidomics field has been to characterize lipid species and their respective biological roles. However, difficulties in both full speciation and biological interpretation have rendered these objectives extremely challenging and as a result, limited our understanding of lipid mechanisms and dysregulation. While mass spectrometry-based advancements have significantly increased the ability to identify lipid species, less progress has been made surrounding biological interpretations. We have therefore developed a Structural-based Connectivity and Omic Phenotype Evaluations (SCOPE) cheminformatics toolbox to aid in these evaluations. SCOPE enables the assessment and visualization of two main lipidomic associations: structure/biological connections and metadata linkages either separately or in tandem. To assess structure and biological relationships, SCOPE utilizes key lipid structural moieties such as head group and fatty acyl composition and links them to their respective biological relationships through hierarchical clustering and grouped heatmaps. Metadata arising from phenotypic and environmental factors such as age and diet is then correlated with the lipid structures and/or biological relationships, utilizing Toxicological Prioritization Index (ToxPi) software. Here, SCOPE is demonstrated for various applications from environmental studies to clinical assessments to showcase new biological connections not previously observed with other techniques.

Concentration-response evaluation of ToxCast compounds for multivariate activity patterns of neural network function.

The US Environmental Protection Agency's ToxCast program has generated toxicity data for thousands of chemicals but does not adequately assess potential neurotoxicity. Networks of neurons grown on microelectrode arrays (MEAs) offer an efficient approach to screen compounds for neuroactivity and distinguish between compound effects on firing, bursting, and connectivity patterns. Previously, single concentrations of the ToxCast Phase II library were screened for effects on mean firing rate (MFR) in rat primary cortical networks. Here, we expand this approach by retesting 384 of those compounds (including 222 active in the previous screen) in concentration-response across 43 network activity parameters to evaluate neural network function. Using hierarchical clustering and machine learning methods on the full suite of chemical-parameter response data, we identified 15 network activity parameters crucial in characterizing activity of 237 compounds that were response actives ("hits"). Recognized neurotoxic compounds in this network function assay were often more potent compared to other ToxCast assays. Of these chemical-parameter responses, we identified three k-means clusters of chemical-parameter activity (i.e., multivariate MEA response patterns). Next, we evaluated the MEA clusters for enrichment of chemical features using a subset of ToxPrint chemotypes, revealing chemical structural features that distinguished the MEA clusters. Finally, we assessed distribution of neurotoxicants with known pharmacology within the clusters and found that compounds segregated differentially. Collectively, these results demonstrate that multivariate MEA activity patterns can efficiently screen for diverse chemical activities relevant to neurotoxicity, and that response patterns may have predictive value related to chemical structural features.

Determination of chemical-disease risk values to prioritize connections between environmental factors, genetic variants, and human diseases.

Traditional methods for chemical risk assessment are too time-consuming and resource-intensive to characterize either the diversity of chemicals to which humans are exposed or how that diversity may manifest in population susceptibility differences. The advent of novel toxicological data sources and their integration with bioinformatic databases affords opportunities for modern approaches that consider gene-environment (GxE) interactions in population risk assessment. Here, we present an approach that systematically links multiple data sources to relate chemical risk values to diseases and gene-disease variants. These data sources include high-throughput screening (HTS) results from Tox21/ToxCast, chemical-disease relationships from the Comparative Toxicogenomics Database (CTD), hazard data from resources like the Integrated Risk Information System, exposure data from the ExpoCast initiative, and gene-variant-disease information from the DisGeNET database. We use these integrated data to identify variants implicated in chemical-disease enrichments and develop a new value that estimates the risk of these associations toward differential population responses. Finally, we use this value to prioritize chemical-disease associations by exploring the genomic distribution of variants implicated in high-risk diseases. We offer this modular approach, termed DisQGOS (Disease Quotient Genetic Overview Score), for relating overall chemical-disease risk to potential for population variable responses, as a complement to methods aiming to modernize aspects of risk assessment.

Population-based toxicity screening in human induced pluripotent stem cell-derived cardiomyocytes.

The potential for cardiotoxicity is carefully evaluated for pharmaceuticals, as it is a major safety liability. However, environmental chemicals are seldom tested for their cardiotoxic potential. Moreover, there is a large variability in both baseline and drug-induced cardiovascular risk in humans, but data are lacking on the degree to which susceptibility to chemically-induced cardiotoxicity may also vary. Human induced pluripotent stem cell (iPSC)-derived cardiomyocytes have become an important in vitro model for drug screening. Thus, we hypothesized that a population-based model of iPSC-derived cardiomyocytes from a diverse set of individuals can be used to assess potential hazard and inter-individual variability in chemical effects on these cells. We conducted concentration-response screening of 134 chemicals (pharmaceuticals, industrial and environmental chemicals and food constituents) in iPSC-derived cardiomyocytes from 43 individuals, comprising both sexes and diverse ancestry. We measured kinetic calcium flux and conducted high-content imaging following chemical exposure, and utilized a panel of functional and cytotoxicity parameters in concentration-response for each chemical and donor. We show reproducible inter-individual variability in both baseline and chemical-induced effects on iPSC-derived cardiomyocytes. Further, chemical-specific variability in potency and degree of population variability were quantified. This study shows the feasibility of using an organotypic population-based human in vitro model to quantitatively assess chemicals for which little cardiotoxicity information is available. Ultimately, these results advance in vitro toxicity testing methodologies by providing an innovative tool for population-based cardiotoxicity screening, contributing to the paradigm shift from traditional animal models of toxicity to in vitro toxicity testing methods.

ToxPi Graphical User Interface 2.0: Dynamic exploration, visualization, and sharing of integrated data models.

BACKGROUND: Drawing integrated conclusions from diverse source data requires synthesis across multiple types of information. The ToxPi (Toxicological Prioritization Index) is an analytical framework that was developed to enable integration of multiple sources of evidence by transforming data into integrated, visual profiles. Methodological improvements have advanced ToxPi and expanded its applicability, necessitating a new, consolidated software platform to provide functionality, while preserving flexibility for future updates. RESULTS: We detail the implementation of a new graphical user interface for ToxPi (Toxicological Prioritization Index) that provides interactive visualization, analysis, reporting, and portability. The interface is deployed as a stand-alone, platform-independent Java application, with a modular design to accommodate inclusion of future analytics. The new ToxPi interface introduces several features, from flexible data import formats (including legacy formats that permit backward compatibility) to similarity-based clustering to options for high-resolution graphical output. CONCLUSIONS: We present the new ToxPi interface for dynamic exploration, visualization, and sharing of integrated data models. The ToxPi interface is freely-available as a single compressed download that includes the main Java executable, all libraries, example data files, and a complete user manual from http://toxpi.org .

Population genetic diversity in zebrafish lines.

Toxicological and pharmacological researchers have seized upon the many benefits of zebrafish, including the short generation time, well-characterized development, and early maturation as clear embryos. A major difference from many model organisms is that standard husbandry practices in zebrafish are designed to maintain population diversity. While this diversity is attractive for translational applications in human and ecological health, it raises critical questions on how interindividual genetic variation might contribute to chemical exposure or disease susceptibility differences. Findings from pooled samples of zebrafish support this supposition of diversity yet cannot directly measure allele frequencies for reference versus alternate alleles. Using the Tanguay lab Tropical 5D zebrafish line (T5D), we performed whole genome sequencing on a large group (n = 276) of individual zebrafish embryos. Paired-end reads were collected on an Illumina 3000HT, then aligned to the most recent zebrafish reference genome (GRCz10). These data were used to compare observed population genetic variation across species (humans, mice, zebrafish), then across lines within zebrafish. We found more single nucleotide polymorphisms (SNPs) in T5D than have been reported in SNP databases for any of the WIK, TU, TL, or AB lines. We theorize that some subset of the novel SNPs may be shared with other zebrafish lines but have not been identified in other studies due to the limitations of capturing population diversity in pooled sequencing strategies. We establish T5D as a model that is representative of diversity levels within laboratory zebrafish lines and demonstrate that experimental design and analysis can exert major effects when characterizing genetic diversity in heterogeneous populations.

A data-driven weighting scheme for multivariate phenotypic endpoints recapitulates zebrafish developmental cascades.

Zebrafish have become a key alternative model for studying health effects of environmental stressors, partly due to their genetic similarity to humans, fast generation time, and the efficiency of generating high-dimensional systematic data. Studies aiming to characterize adverse health effects in zebrafish typically include several phenotypic measurements (endpoints). While there is a solid biomedical basis for capturing a comprehensive set of endpoints, making summary judgments regarding health effects requires thoughtful integration across endpoints. Here, we introduce a Bayesian method to quantify the informativeness of 17 distinct zebrafish endpoints as a data-driven weighting scheme for a multi-endpoint summary measure, called weighted Aggregate Entropy (wAggE). We implement wAggE using high-throughput screening (HTS) data from zebrafish exposed to five concentrations of all 1060 ToxCast chemicals. Our results show that our empirical weighting scheme provides better performance in terms of the Receiver Operating Characteristic (ROC) curve for identifying significant morphological effects and improves robustness over traditional curve-fitting approaches. From a biological perspective, our results suggest that developmental cascade effects triggered by chemical exposure can be recapitulated by analyzing the relationships among endpoints. Thus, wAggE offers a powerful approach for analysis of multivariate phenotypes that can reveal underlying etiological processes.

Transgenerational inheritance of neurobehavioral and physiological deficits from developmental exposure to benzo[a]pyrene in zebrafish.

Benzo[a]pyrene (B[a]P) is a well-known genotoxic polycylic aromatic compound whose toxicity is dependent on signaling via the aryl hydrocarbon receptor (AHR). It is unclear to what extent detrimental effects of B[a]P exposures might impact future generations and whether transgenerational effects might be AHR-dependent. This study examined the effects of developmental B[a]P exposure on 3 generations of zebrafish. Zebrafish embryos were exposed from 6 to 120h post fertilization (hpf) to 5 and 10µM B[a]P and raised in chemical-free water until adulthood (F0). Two generations were raised from F0 fish to evaluate transgenerational inheritance. Morphological, physiological and neurobehavioral parameters were measured at two life stages. Juveniles of the F0 and F2 exhibited hyper locomotor activity, decreased heartbeat and mitochondrial function. B[a]P exposure during development resulted in decreased global DNA methylation levels and generally reduced expression of DNA methyltransferases in wild type zebrafish, with the latter effect largely reversed in an AHR2-null background. Adults from the F0 B[a]P exposed lineage displayed social anxiety-like behavior. Adults in the F2 transgeneration manifested gender-specific increased body mass index (BMI), increased oxygen consumption and hyper-avoidance behavior. Exposure to benzo[a]pyrene during development resulted in transgenerational inheritance of neurobehavioral and physiological deficiencies. Indirect evidence suggested the potential for an AHR2-dependent epigenetic route.

High-throughput characterization of chemical-associated embryonic behavioral changes predicts teratogenic outcomes.

New strategies are needed to address the data gap between the bioactivity of chemicals in the environment versus existing hazard information. We address whether a high-throughput screening (HTS) system using a vertebrate organism (embryonic zebrafish) can characterize chemical-elicited behavioral responses at an early, 24 hours post-fertilization (hpf) stage that predict teratogenic consequences at a later developmental stage. The system was used to generate full concentration-response behavioral profiles at 24 hpf across 1060 ToxCast™ chemicals. Detailed, morphological evaluation of all individuals was performed as experimental follow-up at 5 days post-fertilization (dpf). Chemicals eliciting behavioral responses were also mapped against external HTS in vitro results to identify specific molecular targets and neurosignalling pathways. We found that, as an integrative measure of normal development, significant alterations in movement highlighted active chemicals representing several modes of action. These early behavioral responses were predictive for 17 specific developmental abnormalities and mortality measured at 5 dpf, often at lower (i.e., more potent) concentrations than those at which morphological effects were observed. Therefore, this system can provide rapid characterization of chemical-elicited behavioral responses at an early developmental stage that are predictive of observable adverse effects later in life.

ToxPi GUI: an interactive visualization tool for transparent integration of data from diverse sources of evidence.

MOTIVATION: Scientists and regulators are often faced with complex decisions, where use of scarce resources must be prioritized using collections of diverse information. The Toxicological Prioritization Index (ToxPi™) was developed to enable integration of multiple sources of evidence on exposure and/or safety, transformed into transparent visual rankings to facilitate decision making. The rankings and associated graphical profiles can be used to prioritize resources in various decision contexts, such as testing chemical toxicity or assessing similarity of predicted compound bioactivity profiles. The amount and types of information available to decision makers are increasing exponentially, while the complex decisions must rely on specialized domain knowledge across multiple criteria of varying importance. Thus, the ToxPi bridges a gap, combining rigorous aggregation of evidence with ease of communication to stakeholders. RESULTS: An interactive ToxPi graphical user interface (GUI) application has been implemented to allow straightforward decision support across a variety of decision-making contexts in environmental health. The GUI allows users to easily import and recombine data, then analyze, visualize, highlight, export and communicate ToxPi results. It also provides a statistical metric of stability for both individual ToxPi scores and relative prioritized ranks. AVAILABILITY: The ToxPi GUI application, complete user manual and example data files are freely available from http://comptox.unc.edu/toxpi.php.

Hierarchical dose-response modeling for high-throughput toxicity screening of environmental chemicals.

High-throughput screening (HTS) of environmental chemicals is used to identify chemicals with high potential for adverse human health and environmental effects from among the thousands of untested chemicals. Predicting physiologically relevant activity with HTS data requires estimating the response of a large number of chemicals across a battery of screening assays based on sparse dose-response data for each chemical-assay combination. Many standard dose-response methods are inadequate because they treat each curve separately and under-perform when there are as few as 6-10 observations per curve. We propose a semiparametric Bayesian model that borrows strength across chemicals and assays. Our method directly parametrizes the efficacy and potency of the chemicals as well as the probability of response. We use the ToxCast data from the U.S. Environmental Protection Agency (EPA) as motivation. We demonstrate that our hierarchical method provides more accurate estimates of the probability of response, efficacy, and potency than separate curve estimation in a simulation study. We use our semiparametric method to compare the efficacy of chemicals in the ToxCast data to well-characterized reference chemicals on estrogen receptor α (ERα) and peroxisome proliferator-activated receptor γ (PPARγ) assays, then estimate the probability that other chemicals are active at lower concentrations than the reference chemicals.

Predictive endocrine testing in the 21st century using in vitro assays of estrogen receptor signaling responses.

Thousands of environmental chemicals are subject to regulatory review for their potential to be endocrine disruptors (ED). In vitro high-throughput screening (HTS) assays have emerged as a potential tool for prioritizing chemicals for ED-related whole-animal tests. In this study, 1814 chemicals including pesticide active and inert ingredients, industrial chemicals, food additives, and pharmaceuticals were evaluated in a panel of 13 in vitro HTS assays. The panel of in vitro assays interrogated multiple end points related to estrogen receptor (ER) signaling, namely binding, agonist, antagonist, and cell growth responses. The results from the in vitro assays were used to create an ER Interaction Score. For 36 reference chemicals, an ER Interaction Score >0 showed 100% sensitivity and 87.5% specificity for classifying potential ER activity. The magnitude of the ER Interaction Score was significantly related to the potency classification of the reference chemicals (p < 0.0001). ERα/ERß selectivity was also evaluated, but relatively few chemicals showed significant selectivity for a specific isoform. When applied to a broader set of chemicals with in vivo uterotrophic data, the ER Interaction Scores showed 91% sensitivity and 65% specificity. Overall, this study provides a novel method for combining in vitro concentration response data from multiple assays and, when applied to a large set of ER data, accurately predicted estrogenic responses and demonstrated its utility for chemical prioritization.