Review of Existing QSAR/QSPR Models Developed for Properties Used in Hazardous Chemicals Classification System

Potential for Machine Learning to Address Data Gaps in Human Toxicity and Ecotoxicity Characterization

Machine Learning (ML) is increasingly applied to fill data gaps in assessments to quantify impacts associated with chemical emissions and chemicals in products. However, the systematic application of ML-based approaches to fill chemical data gaps is still limited, and their potential for addressing a wide range of chemicals is unknown. We prioritized chemical-related parameters for chemical toxicity characterization to inform ML model development based on two criteria: (1) each parameter's relevance to robustly characterize chemical toxicity described by the uncertainty in characterization results attributable to each parameter and (2) the potential for ML-based approaches to predict parameter values for a wide range of chemicals described by the availability of chemicals with measured parameter data. We prioritized 13 out of 38 parameters for developing ML-based approaches, while flagging another nine with critical data gaps. For all prioritized parameters, we performed a chemical space analysis to assess further the potential for ML-based approaches to predict data for diverse chemicals considering the structural diversity of available measured data, showing that ML-based approaches can potentially predict 8-46% of marketed chemicals based on 1-10% with available measured data. Our results can systematically inform future ML model development efforts to address data gaps in chemical toxicity characterization.

QSPR models to predict the physical hazards of mixtures: a state of art

Physical hazards of chemical mixtures, associated for example with their fire or explosion risks, are generally characterized using experimental tools. These tests can be expensive, complex, long to perform and even dangerous for operators. Therefore, for several years and especially with the implementation of the REACH regulation, predictive methods like quantitative structure-property relationships have been encouraged as alternatives tests to determine (eco)toxicological but also physical hazards of chemical substances. Initially, these approaches were intended for pure products, by considering a molecular similarity principle. However, additional to those for pure products, QSPR models for mixtures recently appeared and represent an increasing field of research. This study proposes a state of the art of existing QSPR models specifically dedicated to the prediction of the physical hazards of mixtures. Identified models have been analysed on the key elements of model development (experimental data and fields of application, descriptors used, development and validation methods). It draws up an overview of the potential and limitations of current models as well as areas of progress towards enlarged deployment as a complement to experimental characterizations, for example in the search for safer substances (according to safety-by-design concepts).

Toward a general and interpretable umami taste predictor using a multi-objective machine learning approach

The umami taste is one of the five basic taste modalities normally linked to the protein content in food. The implementation of fast and cost-effective tools for the prediction of the umami taste of a molecule remains extremely interesting to understand the molecular basis of this taste and to effectively rationalise the production and consumption of specific foods and ingredients. However, the only examples of umami predictors available in the literature rely on the amino acid sequence of the analysed peptides, limiting the applicability of the models. In the present study, we developed a novel ML-based algorithm, named VirtuousUmami, able to predict the umami taste of a query compound starting from its SMILES representation, thus opening up the possibility of potentially using such a model on any database through a standard and more general molecular description. Herein, we have tested our model on five databases related to foods or natural compounds. The proposed tool will pave the way toward the rationalisation of the molecular features underlying the umami taste and toward the design of specific peptide-inspired compounds with specific taste properties.

Building Chemical Property Models for Energetic Materials from Small Datasets Using a Transfer Learning Approach

For many experimentally measured chemical properties that cannot be directly computed from first-principles, the existing physics-based models do not extrapolate well to out-of-sample molecules, and experimental datasets themselves are too small for traditional machine learning (ML) approaches. To overcome these limitations, we apply a transfer learning approach, whereby we simultaneously train a multi-target regression model on a small number of molecules with experimentally measured values and a large number of molecules with related computed properties. We demonstrate this methodology on predicting the experimentally measured impact sensitivity of energetic crystals, finding that both characteristics of the computed dataset and model architecture are important to prediction accuracy of the small experimental dataset. Our directed-message passing neural network (D-MPNN) ML model using transfer learning outperforms direct-ML and physics-based models on a diverse test set, and the new methods described here are widely applicable to modeling many other structure-property relationships.

Classifying the toxicity of pesticides to honey bees via support vector machines with random walk graph kernels

Pesticides benefit agriculture by increasing crop yield, quality, and security. However, pesticides may inadvertently harm bees, which are valuable as pollinators. Thus, candidate pesticides in development pipelines must be assessed for toxicity to bees. Leveraging a dataset of 382 molecules with toxicity labels from honey bee exposure experiments, we train a support vector machine (SVM) to predict the toxicity of pesticides to honey bees. We compare two representations of the pesticide molecules: (i) a random walk feature vector listing counts of length- L walks on the molecular graph with each vertex- and edge-label sequence and (ii) the Molecular ACCess System (MACCS) structural key fingerprint (FP), a bit vector indicating the presence/absence of a list of pre-defined subgraph patterns in the molecular graph. We explicitly construct the MACCS FPs but rely on the fixed-length- L random walk graph kernel (RWGK) in place of the dot product for the random walk representation. The L-RWGK-SVM achieves an accuracy, precision, recall, and F1 score (mean over 2000 runs) of 0.81, 0.68, 0.71, and 0.69, respectively, on the test data set—with L = 4 being the mode optimal walk length. The MACCS-FP-SVM performs on par/marginally better than the L-RWGK-SVM, lends more interpretability, but varies more in performance. We interpret the MACCS-FP-SVM by illuminating which subgraph patterns in the molecules tend to strongly push them toward the toxic/non-toxic side of the separating hyperplane.

Predictive Quantitative Structure-Activity Relationship Modeling of the Antifungal and Antibiotic Properties of Triazolothiadiazine Compounds

Predictive models were developed using two-dimensional quantitative structure activity relationship (QSAR) methods coupled with B3LYP/6-311+G** density functional theory modeling that describe the antimicrobial properties of twenty-four triazolothiadiazine compounds against Aspergillus niger, Aspergillus flavus and Penicillium sp., as well as the bacteria Staphylococcus aureus, Bacillus subtilis, Escherichia coli, and Pseudomonas aeruginosa. B3LYP/6-311+G** density functional theory calculations indicated the triazolothiadiazine derivatives possess only modest variation between the frontier orbital properties. Genetic function approximation (GFA) analysis identified the topological and density functional theory derived descriptors for antimicrobial models using a population of 200 models with one to three descriptors that were crossed for 10,000 generations. Two or three descriptor models provided validated predictive models for antifungal and antibiotic properties with R² values between 0.725 and 0.768 and no outliers. The best models to describe antimicrobial activities include descriptors related to connectivity, electronegativity, polarizability, and van der Waals properties. The reported method provided robust two-dimensional QSAR models with topological and density functional theory descriptors that explain a variety of antifungal and antibiotic activities for structurally related heterocyclic compounds.

Chemoinformatics for the Safety of Energetic and Reactive Materials at Ineris

The characterization of physical hazards of substances is a key information to manage the risks associated to their use, storage and transport. With decades of work in this area, Ineris develops and implements cutting-edge experimental facilities allowing such characterizations at different scales and under various conditions to study all of the dreaded accident scenarios. This review presents the efforts engaged by Ineris more recently in the field of chemoinformatics to develop and use new predictive methods for the anticipation and management of industrials risks associated to energetic and reactive materials as a complement to experiments. An overview of the methods used for the development of Quantitative Structure-Property Relationships for physical hazards are presented and discussed regarding the specificities associated to this class of properties. A review of models developed at Ineris is also provided from the first tentative models on the explosivity of nitro compounds to the successful application to the flammability of organic mixtures. Then, a discussion is proposed on the use of QSPR models. Good practices for robust use for QSPR models are recalled with specific comments related to physical hazards, notably for regulatory purpose. Dissemination and training efforts engaged by Ineris are also presented. The potential offered by these predictive methods in terms of in silico design and for the development of new intrinsically safer technologies in safety-by-design strategies is finally discussed. At last, challenges and perspectives to extend the application of chemoinformatics in the field of safety and in particular for the physical hazards of energetic and reactive substances are proposed.

A Refractive Index Study of a Diverse Set of Polymeric Materials by QSPR with Quantum-Chemical and Additive Descriptors

Predicting the activities and properties of materials via in silico methods has been shown to be a cost- and time-effective way of aiding chemists in synthesizing materials with desired properties. Refractive index (n) is one of the most important defining characteristics of an optical material. Presented in this work is a quantitative structure–property relationship (QSPR) model that was developed to predict the refractive index for a diverse set of polymers. A number of models were created, where a four-variable model showed the best predictive performance with R² = 0.904 and Q²_LOO = 0.897. The robustness and predictability of the best model was validated using the leave-one-out technique, external set and y-scrambling methods. The predictive ability of the model was confirmed with the external set, showing the R²_ext = 0.880. For the refractive index, the ionization potential, polarizability, 2D and 3D geometrical descriptors were the most influential properties. The developed model was transparent and mechanistically explainable and can be used in the prediction of the refractive index for new and untested polymers.

Prediction of Lower Flammability Limits for Binary Hydrocarbon Gases by Quantitative Structure-A Property Relationship Approach

The lower flammability limit (LFL) is one of the most important parameters for evaluating the fire and explosion hazards of flammable gases or vapors. This study proposed quantitative structure−property relationship (QSPR) models to predict the LFL of binary hydrocarbon gases from their molecular structures. Twelve different mixing rules were employed to derive mixture descriptors for describing the structures characteristics of a series of 181 binary hydrocarbon mixtures. Genetic algorithm (GA)-based multiple linear regression (MLR) was used to select the most statistically effective mixture descriptors on the LFL of binary hydrocarbon gases. A total of 12 multilinear models were obtained based on the different mathematical formulas. The best model, issued from the norm of the molar contribution formula, was achieved as a six-parameter model. The best model was then rigorously validated using multiple strategies and further extensively compared to the previously published model. The results demonstrated the robustness, validity, and satisfactory predictivity of the proposed model. The applicability domain (AD) of the model was defined as well. The proposed best model would be expected to present an alternative to predict the LFL values of existing or new binary hydrocarbon gases, and provide some guidance for prioritizing the design of safer blended gases with desired properties.

Large scale comparison of QSAR and conformal prediction methods and their applications in drug discovery

Structure–activity relationship modelling is frequently used in the early stage of drug discovery to assess the activity of a compound on one or several targets, and can also be used to assess the interaction of compounds with liability targets. QSAR models have been used for these and related applications over many years, with good success. Conformal prediction is a relatively new QSAR approach that provides information on the certainty of a prediction, and so helps in decision-making. However, it is not always clear how best to make use of this additional information. In this article, we describe a case study that directly compares conformal prediction with traditional QSAR methods for large-scale predictions of target-ligand binding. The ChEMBL database was used to extract a data set comprising data from 550 human protein targets with different bioactivity profiles. For each target, a QSAR model and a conformal predictor were trained and their results compared. The models were then evaluated on new data published since the original models were built to simulate a “real world” application. The comparative study highlights the similarities between the two techniques but also some differences that it is important to bear in mind when the methods are used in practical drug discovery applications.

Evaluating Molecular Properties Involved in Transport of Small Molecules in Stratum Corneum: A Quantitative Structure-Activity Relationship for Skin Permeability

The skin permeability (Kp) defines the rate of a chemical penetrating across the stratum corneum. This value is widely used to quantitatively describe the transport of molecules in the outermost layer of epidermal skin and indicate the significance of skin absorption. This study defined a Kp quantitative structure-activity relationship (QSAR) based on 106 chemical substances of Kp measured using human skin and interpreted the molecular interactions underlying transport behavior of small molecules in the stratum corneum. The Kp QSAR developed in this study identified four molecular descriptors that described the molecular cyclicity in the molecule reflecting local geometrical environments, topological distances between pairs of oxygen and chlorine atoms, lipophilicity, and similarity to antineoplastics in molecular properties. This Kp QSAR considered the octanol-water partition coefficient to be a direct influence on transdermal movement of molecules. Moreover, the Kp QSAR identified a sub-domain of molecular properties initially defined to describe the antineoplastic resemblance of a compound as a significant factor in affecting transdermal permeation of solutes. This finding suggests that the influence of molecular size on the chemical’s skin-permeating capability should be interpreted with other relevant physicochemical properties rather than being represented by molecular weight alone.

Solubility of organic compounds in octanol: Improved predictions based on the geometrical fragment approach

Two new models are introduced to predict the solubility of chemicals in octanol (S_oct), taking advantage of the extensive character of log(S_oct) through a decomposition of molecules into so-called geometrical fragments (GF). They are extensively validated and their compliance with regulatory requirements is demonstrated. The first model requires just a molecular formula as input. Despite an extreme simplicity, it performs as well as an advanced random forest model involving 86 descriptors, with a root mean square error (RMSE) of 0.64 log units for an external test set of 100 molecules. For the second one, which requires the melting point T_m as input, introducing GF descriptors reduces the RMSE from about 0.7 to <0.5 log units, a performance that could previously be obtained only through the use of Abraham descriptors. A script is provided for easy application of the models, taking into account the limits of their applicability domains.

Chemoinformatics at IFP Energies Nouvelles: Applications in the Fields of Energy, Transport, and Environment

The objective of the present paper is to summarize chemoinformatics based research, and more precisely, the development of quantitative structure property relationships performed at IFP Energies nouvelles (IFPEN) during the last decade. A special focus is proposed on research activities performed in the "Thermodynamics and Molecular Simulation" department, i. e. the use of multiscale molecular simulation methods in responses to projects. Molecular simulation techniques can be envisaged to supplement dataset when experimental information lacks, thus the review includes a section dedicated to molecular simulation codes, development of intermolecular potentials, and some of their possible applications. Know-how and feedback from our experiences in terms of machine learning application for thermophysical property predictions are included in a section dealing with methodological aspects. The generic character of chemoinformatics is emphasized through applications in the fields of energy, transport, and environment, with illustrations for three IFPEN business units: "Transports", "Energy Resources", and "Processes". More precisely, the review focus on different challenges such as the prediction of properties for alternative fuels, the prediction of fuel compatibility with polymeric materials, the prediction of properties for surfactants usable in chemical enhanced oil recovery, and the prediction of guest-host interactions between gases and nanoporous materials in the frame of carbon dioxide capture or gas separation activities.

Development of Simple QSPR Models for the Prediction of the Heat of Decomposition of Organic Peroxides

Quantitative structure-property relationships represent alternative method to experiments to access the estimation of physico-chemical properties of chemicals for screening purpose at R&D level but also to gather missing data in regulatory context. In particular, such predictions were encouraged by the REACH regulation for the collection of data, provided that they are developed respecting the rigorous principles of validation proposed by OECD. In this context, a series of organic peroxides, unstable chemicals which can easily decompose and may lead to explosion, were investigated to develop simple QSPR models that can be used in a regulatory framework. Only constitutional and topological descriptors were employed to achieve QSPR models predicting the heat of decomposition, which could be used without any time consuming preliminary structure calculations at quantum chemical level. To validate the models, the original experimental dataset was divided into a training and a validation set according to two methods of partitioning, one based on the property value and the other based on the structure of the molecules by the mean of PCA. Four QSPR models were developed upon the type of descriptors and the methods of partitioning. The 2 models issuing from the PCA based method were highlighted as they presented good predictive power and they are easier to apply than our previous quantum chemical based model, since they do not need any preliminary calculations.

The Evolution of Process Safety: Current Status and Future Direction

The advent of the industrial revolution in the nineteenth century increased the volume and variety of manufactured goods and enriched the quality of life for society as a whole. However, industrialization was also accompanied by new manufacturing and complex processes that brought about the use of hazardous chemicals and difficult-to-control operating conditions. Moreover, human-process-equipment interaction plus on-the-job learning resulted in further undesirable outcomes and associated consequences. These problems gave rise to many catastrophic process safety incidents that resulted in thousands of fatalities and injuries, losses of property, and environmental damages. These events led eventually to the necessity for a gradual development of a new multidisciplinary field, referred to as process safety. From its inception in the early 1970s to the current state of the art, process safety has come to represent a wide array of issues, including safety culture, process safety management systems, process safety engineering, loss prevention, risk assessment, risk management, and inherently safer technology. Governments and academic/research organizations have kept pace with regulatory programs and research initiatives, respectively. Understanding how major incidents impact regulations and contribute to industrial and academic technology development provides a firm foundation to address new challenges, and to continue applying science and engineering to develop and implement programs to keep hazardous materials within containment. Here the most significant incidents in terms of their impact on regulations and the overall development of the field of process safety are described.

A quantitative approach to polar organic reactivity

A method is presented which allows one to predict toxic effects which are triggered by the formation of covalent bonds between electron-deficient (electrophilic) compounds and biological electron-rich (nucleophilic) targets, as proteins or nucleic acids. It is based on our comprehensive nucleophilicity and electrophilicity scales, which we constructed as an aid for the planning of organic syntheses. For the construction of these scales, rate constants for the reactions of benzhydrylium ions (aryl2CH(+)) and structurally related quinone methides with nucleophiles have been measured and correlated by the equation lg k(20 °C) = sN(E + N), which yields absolute rate constants k (L mol(-1) s(-1)) from one parameter for electrophiles (the electrophilicity E) and two for nucleophiles (the nucleophilicity parameter N and the susceptibility sN). A freely accessible database (http://www.cup.uni-muenchen.de/oc/mayr/DBintro.html) is described, which presently comprises data for 1000 nucleophiles and 260 electrophiles and provides links to the original literature reports. The kinetic scales are complemented by a thermodynamic counterpart, which enables one to calculate association constants K (L mol(-1)) of electrophiles with nucleophiles from the empirical Lewis acidity parameters LA and Lewis basicity parameters LB by the equation lg K (20°C) = LA + LB.

Significance of Theoretical Decomposition Enthalpies for Predicting Thermal Hazards

Much effort is currently put into the development of models for predicting decomposition enthalpies measured using differential scanning calorimetry (DSC). As an alternative to the purely empirical schemes reported so far, this work relies on theoretical values obtained on the basis of simple assumptions. For nitroaromatic compounds (NACs) studied in sealed sample cells, our approach proves clearly superior to previous ones. In contrast, it correlates poorly with data measured in pin-hole sample cells. Progress might be obtained through a combination of the present approach with the usual Quantitative Structure-Property Relationships (QSPR) methodologies. This work emphasizes the significance of the theoretical decomposition enthalpy as a fundamental descriptor for the prediction of DSC values. In fact, the theoretical value provides a valuable criterion to characterize thermal hazards, as a complement to experimental decomposition temperatures.

Prediction of the thermal decomposition of organic peroxides by validated QSPR models

Organic peroxides are unstable chemicals which can easily decompose and may lead to explosion. Such a process can be characterized by physico-chemical parameters such as heat and temperature of decomposition, whose determination is crucial to manage related hazards. These thermal stability properties are also required within many regulatory frameworks related to chemicals in order to assess their hazardous properties. In this work, new quantitative structure-property relationships (QSPR) models were developed to predict accurately the thermal stability of organic peroxides from their molecular structure respecting the OECD guidelines for regulatory acceptability of QSPRs. Based on the acquisition of 38 reference experimental data using DSC (differential scanning calorimetry) apparatus in homogenous experimental conditions, multi-linear models were derived for the prediction of the decomposition heat and the onset temperature using different types of molecular descriptors. Models were tested by internal and external validation tests and their applicability domains were defined and analyzed. Being rigorously validated, they presented the best performances in terms of fitting, robustness and predictive power and the descriptors used in these models were linked to the peroxide bond whose breaking represents the main decomposition mechanism of organic peroxides.

Predicting Flash Point of Organosilicon Compounds Using Quantitative Structure Activity Relationship Approach

The flash point (FP) of a compound is the primary property used in the assessment of fire hazards for flammable liquids and is amongst the crucial information that people handling flammable liquids must possess as far as industrial safety is concerned. In this work, the FPs of 236 organosilicon compounds were collected and used to construct a quantitative structure activity relationship (QSAR) model for predicting their FPs. The CODESSA PRO software was adopted to calculate the required molecular descriptors, and 350 molecular descriptors were developed for each compound. A modified stepwise regression algorithm was applied to choose descriptors that were highly correlated with the FP of organosilicon compounds. The proposed model was a linear regression model consisting of six descriptors. This 6-descriptor model gave anR2value of 0.9174,QLOO2value of 0.9106, andQ2value of 0.8989. The average fitting error and the average predictive error were found to be of 10.34 K and 11.22 K, respectively, and the average fitting error in percentage and the average predictive error in percentage were found to be of 3.30 and 3.60%, respectively. Compared with the known reproducibility of FP measurement using standard test method, these predicted results were of a satisfactory precision.

A multi-stakeholder perspective on the use of alternative test strategies for nanomaterial safety assessment

There has been a conceptual shift in toxicological studies from describing what happens to explaining how the adverse outcome occurs, thereby enabling a deeper and improved understanding of how biomolecular and mechanistic profiling can inform hazard identification and improve risk assessment. Compared to traditional toxicology methods, which have a heavy reliance on animals, new approaches to generate toxicological data are becoming available for the safety assessment of chemicals, including high-throughput and high-content screening (HTS, HCS). With the emergence of nanotechnology, the exponential increase in the total number of engineered nanomaterials (ENMs) in research, development, and commercialization requires a robust scientific approach to screen ENM safety in humans and the environment rapidly and efficiently. Spurred by the developments in chemical testing, a promising new toxicological paradigm for ENMs is to use alternative test strategies (ATS), which reduce reliance on animal testing through the use of in vitro and in silico methods such as HTS, HCS, and computational modeling. Furthermore, this allows for the comparative analysis of large numbers of ENMs simultaneously and for hazard assessment at various stages of the product development process and overall life cycle. Using carbon nanotubes as a case study, a workshop bringing together national and international leaders from government, industry, and academia was convened at the University of California, Los Angeles, to discuss the utility of ATS for decision-making analyses of ENMs. After lively discussions, a short list of generally shared viewpoints on this topic was generated, including a general view that ATS approaches for ENMs can significantly benefit chemical safety analysis.