Transparency in Modeling through Careful Application of OECD's QSAR/QSPR Principles via a Curated Water Solubility Data Set.

The need for careful assembly, training, and validation of quantitative structure-activity/property models (QSAR/QSPR) is more significant than ever as data sets become larger and sophisticated machine learning tools become increasingly ubiquitous and accessible to the scientific community. Regulatory agencies such as the United States Environmental Protection Agency must carefully scrutinize each aspect of a resulting QSAR/QSPR model to determine its potential use in environmental exposure and hazard assessment. Herein, we revisit the goals of the Organisation for Economic Cooperation and Development (OECD) in our application and discuss the validation principles for structure-activity models. We apply these principles to a model for predicting water solubility of organic compounds derived using random forest regression, a common machine learning approach in the QSA/PR literature. Using public sources, we carefully assembled and curated a data set consisting of 10,200 unique chemical structures with associated water solubility measurements. This data set was then used as a focal narrative to methodically consider the OECD's QSA/PR principles and how they can be applied to random forests. Despite some expert, mechanistically informed supervision of descriptor selection to enhance model interpretability, we achieved a model of water solubility with comparable performance to previously published models (5-fold cross validated performance 0.81 R2 and 0.98 RMSE). We hope this work will catalyze a necessary conversation around the importance of cautiously modernizing and explicitly leveraging OECD principles while pursuing state-of-the-art machine learning approaches to derive QSA/PR models suitable for regulatory consideration.

Prediction of aquatic toxicity mode of action using linear discriminant and random forest models.

The ability to determine the mode of action (MOA) for a diverse group of chemicals is a critical part of ecological risk assessment and chemical regulation. However, existing MOA assignment approaches in ecotoxicology have been limited to a relatively few MOAs, have high uncertainty, or rely on professional judgment. In this study, machine based learning algorithms (linear discriminant analysis and random forest) were used to develop models for assigning aquatic toxicity MOA. These methods were selected since they have been shown to be able to correlate diverse data sets and provide an indication of the most important descriptors. A data set of MOA assignments for 924 chemicals was developed using a combination of high confidence assignments, international consensus classifications, ASTER (ASessment Tools for the Evaluation of Risk) predictions, and weight of evidence professional judgment based an assessment of structure and literature information. The overall data set was randomly divided into a training set (75%) and a validation set (25%) and then used to develop linear discriminant analysis (LDA) and random forest (RF) MOA assignment models. The LDA and RF models had high internal concordance and specificity and were able to produce overall prediction accuracies ranging from 84.5 to 87.7% for the validation set. These results demonstrate that computational chemistry approaches can be used to determine the acute toxicity MOAs across a large range of structures and mechanisms.

Does rational selection of training and test sets improve the outcome of QSAR modeling?

Prior to using a quantitative structure activity relationship (QSAR) model for external predictions, its predictive power should be established and validated. In the absence of a true external data set, the best way to validate the predictive ability of a model is to perform its statistical external validation. In statistical external validation, the overall data set is divided into training and test sets. Commonly, this splitting is performed using random division. Rational splitting methods can divide data sets into training and test sets in an intelligent fashion. The purpose of this study was to determine whether rational division methods lead to more predictive models compared to random division. A special data splitting procedure was used to facilitate the comparison between random and rational division methods. For each toxicity end point, the overall data set was divided into a modeling set (80% of the overall set) and an external evaluation set (20% of the overall set) using random division. The modeling set was then subdivided into a training set (80% of the modeling set) and a test set (20% of the modeling set) using rational division methods and by using random division. The Kennard-Stone, minimal test set dissimilarity, and sphere exclusion algorithms were used as the rational division methods. The hierarchical clustering, random forest, and k-nearest neighbor (kNN) methods were used to develop QSAR models based on the training sets. For kNN QSAR, multiple training and test sets were generated, and multiple QSAR models were built. The results of this study indicate that models based on rational division methods generate better statistical results for the test sets than models based on random division, but the predictive power of both types of models are comparable.

Detection and toxicity modeling of anthraquinone dyes and chlorinated side products from a colored smoke pyrotechnic reaction.

"Green" pyrotechnics seek to remove known environmental pollutants and health hazards from their formulations. This chemical engineering approach often focuses on maintaining performance effects upon replacement of objectionable ingredients, yet neglects the chemical products formed by the exothermic reaction. In this work, milligram quantities of a lab-scale pyrotechnic red smoke composition were functioned within a thermal probe for product identification by pyrolysis-gas chromatography-mass spectrometry. Thermally decomposed ingredients and new side product derivatives were identified at lower relative abundances to the intact organic dye (as the engineered sublimation product). Side products included chlorination of the organic dye donated by the chlorate oxidizer. Machine learning quantitative structure-activity relationship models computed impacts to health and environmental hazards. High to very high toxicities were predicted for inhalation, mutagenicity, developmental, and endocrine disruption for common military pyrotechnic dyes and their analogous chlorinated side products. These results underscore the need to revise objectives of "green" pyrotechnic engineering.

Evolutionary genomics: codon volatility does not detect selection.

Plotkin et al. introduce a method to detect selection that is based on an index called codon volatility and that uses only the sequence of a single genome, claiming that this method is applicable to a large range of sequenced organisms. Volatility for a given codon is the ratio of non-synonymous codons to all sense codons accessible by one point mutation. The significance of each gene's volatility is assessed by comparison with a simulated distribution of 10(6) synonymous versions of each gene, with synonymous codons drawn randomly from average genome frequencies. Here we re-examine their method and data and find that codon volatility does not detect selection, and that, even if it did, the genomes of Mycobacterium tuberculosis and Plasmodium falciparum, as well as those of most sequenced organisms, do not meet the assumptions necessary for application of their method.

Integrating publicly available information to screen potential candidates for chemical prioritization under the Toxic Substances Control Act: A proof of concept case study using genotoxicity and carcinogenicity.

The Toxic Substances Control Act (TSCA) became law in the U.S. in 1976 and was amended in 2016. The amended law requires the U.S. EPA to perform risk-based evaluations of existing chemicals. Here, we developed a tiered approach to screen potential candidates based on their genotoxicity and carcinogenicity information to inform the selection of candidate chemicals for prioritization under TSCA. The approach was underpinned by a large database of carcinogenicity and genotoxicity information that had been compiled from various public sources. Carcinogenicity data included weight-of-evidence human carcinogenicity evaluations and animal cancer data. Genotoxicity data included bacterial gene mutation data from the Salmonella (Ames) and Escherichia coli WP2 assays and chromosomal mutation (clastogenicity) data. Additionally, Ames and clastogenicity outcomes were predicted using the alert schemes within the OECD QSAR Toolbox and the Toxicity Estimation Software Tool (TEST). The evaluation workflows for carcinogenicity and genotoxicity were developed along with associated scoring schemes to make an overall outcome determination. For this case study, two sets of chemicals, the TSCA Active Inventory non-confidential portion list available on the EPA CompTox Chemicals Dashboard (33,364 chemicals, 'TSCA Active List') and a representative proof-of-concept (POC) set of 238 chemicals were profiled through the two workflows to make determinations of carcinogenicity and genotoxicity potential. Of the 33,364 substances on the 'TSCA Active List', overall calls could be made for 20,371 substances. Here 46.67%% (9507) of substances were non-genotoxic, 0.5% (103) were scored as inconclusive, 43.93% (8949) were predicted genotoxic and 8.9% (1812) were genotoxic. Overall calls for genotoxicity could be made for 225 of the 238 POC chemicals. Of these, 40.44% (91) were non-genotoxic, 2.67% (6) were inconclusive, 6.22% (14) were predicted genotoxic, and 50.67% (114) genotoxic. The approach shows promise as a means to identify potential candidates for prioritization from a genotoxicity and carcinogenicity perspective.

Evaluation of Existing QSAR Models and Structural Alerts and Development of New Ensemble Models for Genotoxicity Using a Newly Compiled Experimental Dataset.

Regulatory agencies world-wide face the challenge of performing risk-based prioritization of thousands of substances in commerce. In this study, a major effort was undertaken to compile a large genotoxicity dataset (54,805 records for 9299 substances) from several public sources (e.g., TOXNET, COSMOS, eChemPortal). The names and outcomes of the different assays were harmonized, and assays were annotated by type: gene mutation in Salmonella bacteria (Ames assay) and chromosome mutation (clastogenicity) in vitro or in vivo (chromosome aberration, micronucleus, and mouse lymphoma Tk +/- assays). This dataset was then evaluated to assess genotoxic potential using a categorization scheme, whereby a substance was considered genotoxic if it was positive in at least one Ames or clastogen study. The categorization dataset comprised 8442 chemicals, of which 2728 chemicals were genotoxic, 5585 were not and 129 were inconclusive. QSAR models (TEST and VEGA) and the OECD Toolbox structural alerts/profilers (e.g., OASIS DNA alerts for Ames and chromosomal aberrations) were used to make in silico predictions of genotoxicity potential. The performance of the individual QSAR tools and structural alerts resulted in balanced accuracies of 57-73%. A Naïve Bayes consensus model was developed using combinations of QSAR models and structural alert predictions. The 'best' consensus model selected had a balanced accuracy of 81.2%, a sensitivity of 87.24% and a specificity of 75.20%. This in silico scheme offers promise as a first step in ranking thousands of substances as part of a prioritization approach for genotoxicity.

Applicability domains for classification problems: Benchmarking of distance to models for Ames mutagenicity set.

The estimation of accuracy and applicability of QSAR and QSPR models for biological and physicochemical properties represents a critical problem. The developed parameter of "distance to model" (DM) is defined as a metric of similarity between the training and test set compounds that have been subjected to QSAR/QSPR modeling. In our previous work, we demonstrated the utility and optimal performance of DM metrics that have been based on the standard deviation within an ensemble of QSAR models. The current study applies such analysis to 30 QSAR models for the Ames mutagenicity data set that were previously reported within the 2009 QSAR challenge. We demonstrate that the DMs based on an ensemble (consensus) model provide systematically better performance than other DMs. The presented approach identifies 30-60% of compounds having an accuracy of prediction similar to the interlaboratory accuracy of the Ames test, which is estimated to be 90%. Thus, the in silico predictions can be used to halve the cost of experimental measurements by providing a similar prediction accuracy. The developed model has been made publicly available at http://ochem.eu/models/1 .

An automated framework for compiling and integrating chemical hazard data.

Comparative chemical hazard assessment, which compares hazards for several endpoints across several chemicals, can be used for a variety of purposes including alternatives assessment and the prioritization of chemicals for further assessment. A new framework was developed to compile and integrate chemical hazard data for several human health and ecotoxicity endpoints from public online sources including hazardous chemical lists, Globally Harmonized System hazard codes (H-codes) or hazard categories from government health agencies, experimental quantitative toxicity values, and predicted values using Quantitative Structure-Activity Relationship (QSAR) models. QSAR model predictions were obtained using EPA's Toxicity Estimation Software Tool. Java programming was used to download hazard data, convert data from each source into a consistent score record format, and store the data in a database. Scoring criteria based on the EPA's Design for the Environment Program Alternatives Assessment Criteria for Hazard Evaluation were used to determine ordinal hazard scores (i.e., low, medium, high, or very high) for each score record. Different methodologies were assessed for integrating data from multiple sources into one score for each hazard endpoint for each chemical. The chemical hazard assessment (CHA) Database developed in this study currently contains more than 990,000 score records for more than 85,000 chemicals. The CHA Database and the methods used in its development may contribute to several cheminformatics, public health, and environmental activities.

Quantitative structure-activity relationship modeling of rat acute toxicity by oral exposure.

Few quantitative structure-activity relationship (QSAR) studies have successfully modeled large, diverse rodent toxicity end points. In this study, a comprehensive data set of 7385 compounds with their most conservative lethal dose (LD(50)) values has been compiled. A combinatorial QSAR approach has been employed to develop robust and predictive models of acute toxicity in rats caused by oral exposure to chemicals. To enable fair comparison between the predictive power of models generated in this study versus a commercial toxicity predictor, TOPKAT (Toxicity Prediction by Komputer Assisted Technology), a modeling subset of the entire data set was selected that included all 3472 compounds used in TOPKAT's training set. The remaining 3913 compounds, which were not present in the TOPKAT training set, were used as the external validation set. QSAR models of five different types were developed for the modeling set. The prediction accuracy for the external validation set was estimated by determination coefficient R(2) of linear regression between actual and predicted LD(50) values. The use of the applicability domain threshold implemented in most models generally improved the external prediction accuracy but expectedly led to the decrease in chemical space coverage; depending on the applicability domain threshold, R(2) ranged from 0.24 to 0.70. Ultimately, several consensus models were developed by averaging the predicted LD(50) for every compound using all five models. The consensus models afforded higher prediction accuracy for the external validation data set with the higher coverage as compared to individual constituent models. The validated consensus LD(50) models developed in this study can be used as reliable computational predictors of in vivo acute toxicity.

Linking Molecular Structure via Functional Group to Chemical Literature for Establishing a Reaction Lineage for Application to Alternatives Assessment.

The evaluation of potential alternatives for chemicals of concern (CoC) requires an understanding of their potential human health and environmental impacts during the manufacture, use, recycle and disposal life stages. During the manufacturing phase, the processes used to produce a desired chemical are defined based on the sequence of chemical reactions and unit operations required to produce the molecule and separate it from other materials used or produced during its manufacture. This paper introduces and demonstrates a tool that links a chemical's structure to information about its synthesis route and the manufacturing process for that chemical. The structure of the chemical is entered using either a SMILES string or the molecule MOL file, and the molecule is searched to identify functional groups present. Based on those functional groups present, the respective named reactions that can be used in its synthesis routes are identified. This information can be used to identify input and output materials for each named reaction, along with reaction conditions, solvents, and catalysts that participate in the reaction. Additionally, the reaction database contains links to internet references and appropriate reaction-specific keywords, further increasing its comprehensiveness. The tool is designed to facilitate the cataloging and use of the chemical literature in a way that allows user to identify and evaluate information about the reactions, such as alternative solvents, catalysts, reaction conditions and other reaction products which enable the comparison of various reaction pathways for the manufacture of the subject chemical. The chemical manufacturing processing steps can be linked to a chemical process ontology to estimate releases and exposures occurring during the manufacturing phase of a chemical.

The Next Generation Blueprint of Computational Toxicology at the U.S. Environmental Protection Agency.

The U.S. Environmental Protection Agency (EPA) is faced with the challenge of efficiently and credibly evaluating chemical safety often with limited or no available toxicity data. The expanding number of chemicals found in commerce and the environment, coupled with time and resource requirements for traditional toxicity testing and exposure characterization, continue to underscore the need for new approaches. In 2005, EPA charted a new course to address this challenge by embracing computational toxicology (CompTox) and investing in the technologies and capabilities to push the field forward. The return on this investment has been demonstrated through results and applications across a range of human and environmental health problems, as well as initial application to regulatory decision-making within programs such as the EPA's Endocrine Disruptor Screening Program. The CompTox initiative at EPA is more than a decade old. This manuscript presents a blueprint to guide the strategic and operational direction over the next 5 years. The primary goal is to obtain broader acceptance of the CompTox approaches for application to higher tier regulatory decisions, such as chemical assessments. To achieve this goal, the blueprint expands and refines the use of high-throughput and computational modeling approaches to transform the components in chemical risk assessment, while systematically addressing key challenges that have hindered progress. In addition, the blueprint outlines additional investments in cross-cutting efforts to characterize uncertainty and variability, develop software and information technology tools, provide outreach and training, and establish scientific confidence for application to different public health and environmental regulatory decisions.

A hierarchical clustering methodology for the estimation of toxicity.

ABSTRACT A quantitative structure-activity relationship (QSAR) methodology based on hierarchical clustering was developed to predict toxicological endpoints. This methodology utilizes Ward's method to divide a training set into a series of structurally similar clusters. The structural similarity is defined in terms of 2-D physicochemical descriptors (such as connectivity and E-state indices). A genetic algorithm-based technique is used to generate statistically valid QSAR models for each cluster (using the pool of descriptors described above). The toxicity for a given query compound is estimated using the weighted average of the predictions from the closest cluster from each step in the hierarchical clustering assuming that the compound is within the domain of applicability of the cluster. The hierarchical clustering methodology was tested using a Tetrahymena pyriformis acute toxicity data set containing 644 chemicals in the training set and with two prediction sets containing 339 and 110 chemicals. The results from the hierarchical clustering methodology were compared to the results from several different QSAR methodologies.

A framework for an alternatives assessment dashboard for evaluating chemical alternatives applied to flame retardants for electronic applications.

The goal of alternatives assessment (AA) is to facilitate a comparison of alternatives to a chemical of concern, resulting in the identification of safer alternatives. A two stage methodology for comparing chemical alternatives was developed. In the first stage, alternatives are compared using a variety of human health effects, ecotoxicity, and physicochemical properties. Hazard profiles are completed using a variety of online sources and quantitative structure activity relationship models. In the second stage, alternatives are evaluated utilizing an exposure/risk assessment over the entire life cycle. Exposure values are calculated using screening-level near-field and far-field exposure models. The second stage allows one to more accurately compare potential exposure to each alternative and consider additional factors that may not be obvious from separate binned persistence, bioaccumulation, and toxicity scores. The methodology was utilized to compare phosphate-based alternatives for decabromodiphenyl ether (decaBDE) in electronics applications.

A Bayesian network model for predicting aquatic toxicity mode of action using two dimensional theoretical molecular descriptors.

The mode of toxic action (MoA) has been recognized as a key determinant of chemical toxicity, but development of predictive MoA classification models in aquatic toxicology has been limited. We developed a Bayesian network model to classify aquatic toxicity MoA using a recently published dataset containing over one thousand chemicals with MoA assignments for aquatic animal toxicity. Two dimensional theoretical chemical descriptors were generated for each chemical using the Toxicity Estimation Software Tool. The model was developed through augmented Markov blanket discovery from the dataset of 1098 chemicals with the MoA broad classifications as a target node. From cross validation, the overall precision for the model was 80.2%. The best precision was for the AChEI MoA (93.5%) where 257 chemicals out of 275 were correctly classified. Model precision was poorest for the reactivity MoA (48.5%) where 48 out of 99 reactive chemicals were correctly classified. Narcosis represented the largest class within the MoA dataset and had a precision and reliability of 80.0%, reflecting the global precision across all of the MoAs. False negatives for narcosis most often fell into electron transport inhibition, neurotoxicity or reactivity MoAs. False negatives for all other MoAs were most often narcosis. A probabilistic sensitivity analysis was undertaken for each MoA to examine the sensitivity to individual and multiple descriptor findings. The results show that the Markov blanket of a structurally complex dataset can simplify analysis and interpretation by identifying a subset of the key chemical descriptors associated with broad aquatic toxicity MoAs, and by providing a computational chemistry-based network classification model with reasonable prediction accuracy.

Preparation of budesonide and budesonide-PLA microparticles using supercritical fluid precipitation technology.

The objective of this study was to prepare and characterize microparticles of budesonide alone and budesonide and polylactic acid (PLA) using supercritical fluid (SCF) technology. A precipitation with a compressed antisolvent (PCA) technique employing supercritical CO2 and a nozzle with 100- microm internal diameter was used to prepare microparticles of budesonide and budesonide-PLA. The effect of various operating variables (temperature and pressure of CO2 and flow rates of drug-polymer solution and/or CO2) and formulation variables (0.25%, 0.5%, and 1% budesonide in methylene chloride) on the morphology and size distribution of the microparticles was determined using scanning electron microscopy. In addition, budesonide-PLA particles were characterized for their surface charge and drug-polymer interactions using a zeta meter and differential scanning calorimetry (DSC), respectively. Furthermore, in vitro budesonide release from budesonide-PLA microparticles was determined at 37 degrees C. Using the PCA process, budesonide and budesonide-PLA microparticles with mean diameters of 1 to 2 microm were prepared. An increase in budesonide concentration (0.25%-1% wt/vol) resulted in budesonide microparticles that were fairly spherical and less agglomerated. In addition, the size of the microparticles increased with an increase in the drug-polymer solution flow rate (1.4-4.7 mL/min) or with a decrease in the CO2 flow rate (50-10 mL/min). Budesonide-PLA microparticles had a drug loading of 7.94%, equivalent to approximately 80% encapsulation efficiency. Budesonide-PLA microparticles had a zeta potential of -37 +/- 4 mV, and DSC studies indicated that SCF processing of budesonide-PLA microparticles resulted in the loss of budesonide crystallinity. Finally, in vitro drug release studies at 37 degrees C indicated 50% budesonide release from the budesonide-PLA microparticles at the end of 28 days. Thus, the PCA process was successful in producing budesonide and budesonide-PLA microparticles. In addition, budesonide-PLA microparticles sustained budesonide release for 4 weeks.