On Some Novel Similarity-Based Functions Used in the ML-Based q-RASAR Approach for Efficient Quantitative Predictions of Selected Toxicity End Points

Automated machine learning in nanotoxicity assessment: A comparative study of predictive model performance

Computational modeling has earned significant interest as an alternative to animal testing of toxicity assessment. However, the process of selecting an appropriate algorithm and fine-tuning hyperparameters for the developing of optimized models takes considerable time, expertise, and an intensive search. The recent emergence of automated machine learning (autoML) approaches, available as user-friendly platforms, has proven beneficial for individuals with limited knowledge in ML-based predictive model development. These autoML platforms automate crucial steps in model development, including data preprocessing, algorithm selection, and hyperparameter tuning. In this study, we used seven previously published and publicly available datasets for oxides and metals to develop nanotoxicity prediction models. AutoML platforms, namely Vertex AI, Azure, and Dataiku, were employed and performance measures such as accuracy, F1 score, precision, and recall for these autoML-based models were then compared with those of conventional ML-based models. The results demonstrated clearly that the autoML platforms produced more reliable nanotoxicity prediction models, outperforming those built with conventional ML algorithms. While none of the three autoML platforms significantly outperformed the others, distinctions exist among them in terms of the available options for choosing technical features throughout the model development steps. This allows users to select an autoML platform that aligns with their knowledge of predictive model development and its technical features. Additionally, prediction models constructed from datasets with better data quality displayed, enhanced performance than those built from datasets with lower data quality, indicating that future studies with high-quality datasets can further improve the performance of those autoML-based prediction models.

DeepRA: A novel deep learning-read-across framework and its application in non-sugar sweeteners mutagenicity prediction

Non-sugar sweeteners (NSSs) or artificial sweeteners have long been used as food chemicals since World War II. NSSs, however, also raise a concern about their mutagenicity. Evaluating the mutagenic ability of NSSs is crucial for food safety; this step is needed for every new chemical registration in the food and pharmaceutical industries. A computational assessment provides less time, money, and involved animals than the in vivo experiments; thus, this study developed a novel computational method from an ensemble convolutional deep neural network and read-across algorithms, called DeepRA, to classify the mutagenicity of chemicals. The mutagenicity data were obtained from the curated Ames test data set. The DeepRA model was developed using both molecular descriptors and molecular fingerprints. The obtained DeepRA model provides accurate and reliable mutagenicity classification through an independent test set. This model was then used to examine the NSSs-related chemicals, enabling the evaluation of mutagenicity from the NSSs-like substances. Finally, this model was publicly available at https://github.com/taraponglab/deepra for further use in chemical regulation and risk assessment.

Classification and regression machine learning models for predicting the combined toxicity and interactions of antibiotics and fungicides mixtures

Antibiotics and triazole fungicides coexist in varying concentrations in natural aquatic environments, resulting in complex mixtures. These mixtures can potentially affect aquatic ecosystems. Accurately distinguishing synergistic and antagonistic mixtures and predicting mixture toxicity are crucial for effective mixture risk assessment. We tested the toxicities of 75 binary mixtures of antibiotics and fungicides against Auxenochlorella pyrenoidosa. Both regression and classification models for these mixtures were developed using machine learning models: random forest (RF), k-nearest neighbors (KNN), and kernel k-nearest neighbors (KKNN). The KKNN model emerged as the best regression model with high values of determination coefficient (R2 = 0.977), explained variance in prediction leave-one-out (Q2LOO = 0.894), and explained variance in external prediction (Q2F1 = 0.929, Q2F2 = 0.929, and Q2F3 = 0.923). The RF model, the leading classifier, exhibited high accuracy (accuracy = 1 for the training set and 0.905 for the test set) in distinguishing the synergistic and antagonistic mixtures. These results provide crucial value for the risk assessment of mixtures.

ARKA: a framework of dimensionality reduction for machine-learning classification modeling, risk assessment, and data gap-filling of sparse environmental toxicity data

Due to the lack of experimental toxicity data for environmental chemicals, there arises a need to fill data gaps by in silico approaches. One of the most commonly used in silico approaches for toxicity assessment of small datasets is the Quantitative Structure-Activity Relationship (QSAR), which generates predictive models for the efficient prediction of query compounds. However, the reliability of the predictions from QSARs derived from small datasets is often questionable from a statistical point of view. This is due to the presence of a larger number of descriptors as compared to the number of training compounds, which reduces the degree of freedom of the developed model. To reduce the overall prediction error for a particular QSAR model, we have proposed here the computation of the novel Arithmetic Residuals in K-groups Analysis (ARKA) descriptors. We have reduced the number of modeling descriptors in a supervised manner by partitioning them into K classes (K = 2 here) depending on the higher mean normalized values of the descriptors to a particular response class, thus preventing the loss of chemical information. A scatter plot of the data points using the values of two ARKA descriptors (ARKA_2 vs. ARKA_1) can potentially identify activity cliffs, less confident data points, and less modelable data points. We have used here five representative environmentally relevant endpoints (skin sensitization, earthworm toxicity, milk/plasma partitioning, algal toxicity, and rodent carcinogenicity of hazardous chemicals) with graded responses to which the ARKA framework was applied for classification modeling. On comparing the performance of the models generated using conventional QSAR descriptors and the ARKA descriptors, the prediction quality of the models derived from ARKA descriptors was found, based on multiple graded-data validation metrics-derived decision criteria, much better than the models derived from QSAR descriptors signifying the potential of ARKA descriptors in ecotoxicological classification modeling of small data sets. Additionally, this holds true for the Read-Across approach as well, since the Read-Across predictions using ARKA descriptors supersede the predictions generated from QSAR descriptors. For the ease of users, a Java-based expert system has been developed that computes the ARKA descriptors from the input of QSAR descriptors.

Chemometrics-driven prediction and prioritization of diverse pesticides on chickens for addressing hazardous effects on public health

The extensive use of various pesticides in the agriculture field badly affects both chickens and humans, primarily through residues in food products and environmental exposure. This study offers the first quantitative structure-toxicity relationship (QSTR) and quantitative read-across-structure toxicity relationship (q-RASTR) models encompassing the LOEL and NOEL endpoints for acute toxicity in chicken, a widely consumed protein. The study's significance lies in the direct link between chemical toxicity in chicken, human intake, and environmental damage. Both the QSTR and the similarity-based read-across algorithms are applied concurrently to improve the predictability of the models. The q-RASTR models were generated by combining read-across derived similarity and error-based parameters, alongside structural and physicochemical descriptors. Machine Learning approaches (SVM and RR) were also employed with the optimization of relevant hyperparameters based on the cross-validation approach, and the final test set prediction results were compared. The PLS-based q-RASTR models for NOEL and LOEL endpoints showed good statistical performance, as traced from the external validation metrics Q2F1: 0.762-0.844; Q2F2: 0.759-0.831 and MAEtest: 0.195-0.214. The developed models were further used to screen the Pesticide Properties DataBase (PPDB) for potential toxicants in chickens. Thus, established models can address eco-toxicological data gaps and development of novel and safe eco-friendly pesticides.

Quantitative read-across structure-activity relationship (q-RASAR): A novel approach to estimate the subchronic oral safety (NOAEL) of diverse organic chemicals in rats

We have developed a quantitative safety prediction model for subchronic repeated doses of diverse organic chemicals on rats using the novel quantitative read-across structure-activity relationship (q-RASAR) approach, which uses similarity-based descriptors for predictive model generation. The experimental -Log (NOAEL) values have been used here as a potential indicator of oral subchronic safety on rats as it determines the maximum dose level for which no observed adverse effects of chemicals are found. A total of 186 data points of diverse organic chemicals have been used for the model generation using structural and physicochemical (0D-2D) descriptors. The read-across-derived similarity, error, and concordance measures (RASAR descriptors) have been extracted from the preliminary 0D-2D descriptors. Then, the combined pool of RASAR and the identified 0D-2D descriptors of the training set were employed to develop the final models by using the partial least squares (PLS) algorithm. The developed PLS model was rigorously validated by various internal and external validation metrics as suggested by the Organization for Economic Co-operation and Development (OECD). The final q-RASAR model is proven to be statistically sound, robust and externally predictive (R2 = 0.85, Q2LOO = 0.82 and Q2F1 = 0.94), superseding the internal as well as external predictivity of the corresponding quantitative structure-activity relationship (QSAR) model as well as previously reported subchronic repeated dose toxicity model found in the literature. In a nutshell, the q-RASAR is an effective approach that has the potential to be used as a good alternative way to improve external predictivity, interpretability, and transferability for subchronic oral safety prediction as well as ecotoxicity risk identification.

Towards safer pesticide management: A quantitative structure-activity relationship based hazard prediction model

Pesticides are recognized as common environmental contaminants. The potential pesticide hazard to non-target organisms, including various mammal species, is a global concern. The global problem requires a comprehensive risk assessment. To assess the toxic effects of pesticides at the early stage, a toxicological risk analysis is conducted to determine pesticide hazard levels. World Health Organization (WHO) has established five pesticide hazard classes based on lethal dose (LD50) values to perform these assessments. In this paper, we have developed one-vs-all quantitative structure-activity relationship (OvA-QSAR) models using five machine-learning techniques with the selected optimum molecular descriptors. Descriptor selection was conducted based on correlation to evaluate the relevance and significance of individual features in our dataset. Our OvA-QSAR model was built using a dataset obtained from the WHO, covering a wide range of chemical pesticides. These models can predict the hazard category for a pesticide within the five available categories. Notably, our experiments demonstrate the outstanding performance and robustness of the Random Forest (RF) model in addressing the challenge of multi-class classification with the selected descriptors.

Protein-protein interaction network analysis for the identification of novel multi-target inhibitors and target miRNAs against Alzheimer's disease

This study presents a strategy for extracting significant gene complexes and then provides prospective therapeutics for AD. In this research, a total of 7905 reports published from 1981 to 2022 were retrieved. Following a review of all those articles, only the genetic association studies on AD were considered. Finally, there is a list of 453 Alzheimer-related genes in our dataset for network analysis. To this end, an experimentally derived protein-protein interaction (PPI) network from the String database was utilized to extract four meaningful gene complexes functionally interconnected using Cytoscape v3.9.1 software. The acquired gene complexes were subjected to an enrichment analysis using the ClueGO v2.5.9 tool to emphasize the most significant biological processes and pathways. Afterward, extracted gene complexes were used to extract the drugs related to AD from DGI v3.0 database and introduce some new drugs which may be helpful for this disease. Finally, a comprehensive network that included every gene connected to each gene complex group as well as the drug targets for each gene has been shown. Moreover, molecular docking studies have been performed with the selected compounds to identify the interaction pattern with the respective targets. Finally, we proposed a list of 62 compounds as multi-targeted directed drug-like compounds with a degree value between 2 and 5 and 30 compounds as target-specific drug-like compounds, which have not been proclaimed as AD-related drugs in prior scientific and medical investigations. Then, new drugs were suggested that can be experimentally examined for future work. In addition to this, four bipartite networks representing each group's genes and target miRNAs were established to introduce target miRNAs by using the miRWalk v3 server.

In silico soil degradation and ecotoxicity analysis of veterinary pharmaceuticals on terrestrial species: first report

With the aim of persistence property analysis and ecotoxicological impact of veterinary pharmaceuticals on different terrestrial species, different classes of veterinary pharmaceuticals (n = 37) with soil degradation property (DT50) were gathered and subjected to QSAR and q-RASAR model development. The models were developed from 2D descriptors under organization for economic cooperation and development guidelines with the application of multiple linear regressions along with genetic algorithm. All developed QSAR and q-RASAR were statistically significant (Internal = R2adj: 0.721-0.861, Q2LOO: 0.609-0.757, and external = Q2Fn = 0.597-0.933, MAEext = 0.174-0.260). Further, the leverage approach of applicability domain assured the model's reliability. The veterinary pharmaceuticals with no experimental values were classified based on their persistence level. Further, the terrestrial toxicity analysis of persistent veterinary pharmaceuticals was done using toxicity prediction by computer assisted technology and in-house built quantitative structure toxicity relationship models to prioritize the toxic and persistent veterinary pharmaceuticals. This study will be helpful in estimation of persistence and toxicity of existing and upcoming veterinary pharmaceuticals.

A deep-learning approach for identifying prospective chemical hazards

With the aim of helping to set safe exposure limits for the general population, various techniques have been implemented to conduct risk assessments for chemicals and other environmental stressors; however, none of these tools facilitate the identification of completely new chemicals that are likely hazardous and elicit an adverse biological effect. Here, we detail a novel in silico, deep-learning framework that is designed to systematically generate structures for new chemical compounds that are predicted to be chemical hazards. To assess the utility of the framework, we applied the tool to four endpoints related to environmental toxicants and their impacts on human and animal health: (i) toxicity to honeybees, (ii) immunotoxicity, (iii) endocrine disruption via ER-α antagonism, and (iv) mutagenicity. In addition, we characterized the predicted potency of these compounds and examined their structural relationship to existing chemicals of concern. As part of the array of emerging new approach methodologies (NAMs), we anticipate that such a framework will be a significant asset to risk assessors and other environmental scientists when planning and forecasting. Though not in the scope of the present study, we expect that the methodology detailed here could also be useful in the de novo design of more environmentally-friendly industrial chemicals.

Definition of the Neurotoxicity-Associated Metabolic Signature Triggered by Berberine and Other Respiratory Chain Inhibitors

To characterize the hits from a phenotypic neurotoxicity screen, we obtained transcriptomics data for valinomycin, diethylstilbestrol, colchicine, rotenone, 1-methyl-4-phenylpyridinium (MPP), carbaryl and berberine (Ber). For all compounds, the concentration triggering neurite degeneration correlated with the onset of gene expression changes. The mechanistically diverse toxicants caused similar patterns of gene regulation: the responses were dominated by cell de-differentiation and a triggering of canonical stress response pathways driven by ATF4 and NRF2. To obtain more detailed and specific information on the modes-of-action, the effects on energy metabolism (respiration and glycolysis) were measured. Ber, rotenone and MPP inhibited the mitochondrial respiratory chain and they shared complex I as the target. This group of toxicants was further evaluated by metabolomics under experimental conditions that did not deplete ATP. Ber (204 changed metabolites) showed similar effects as MPP and rotenone. The overall metabolic situation was characterized by oxidative stress, an over-abundance of NADH (>1000% increase) and a re-routing of metabolism in order to dispose of the nitrogen resulting from increased amino acid turnover. This unique overall pattern led to the accumulation of metabolites known as biomarkers of neurodegeneration (saccharopine, aminoadipate and branched-chain ketoacids). These findings suggest that neurotoxicity of mitochondrial inhibitors may result from an ensemble of metabolic changes rather than from a simple ATP depletion. The combi-omics approach used here provided richer and more specific MoA data than the more common transcriptomics analysis alone. As Ber, a human drug and food supplement, mimicked closely the mode-of-action of known neurotoxicants, its potential hazard requires further investigation.

Ensemble Quantitative Read-Across Structure-Activity Relationship Algorithm for Predicting Skin Cytotoxicity

Read-across (RA) and quantitative structure-activity relationship (QSAR) are two alternative methods commonly used to fill data gaps in chemical registrations. These approaches use physicochemical properties or molecular fingerprints of source substances to predict the properties of unknown substances that have similar chemical structures or physicochemical properties. Research on RA and QSAR is essential to minimize the time, money, and animal testing needed to determine biological properties that are not currently known. This study developed a stacked ensemble quantitative read-across structure-activity relationship algorithm (enQRASAR) for predicting skin irritation toxicity based on negative log cell viability inhibition concentration at 50% (pIC50) against skin keratinocytes as the end point. The goodness-of-fit and predictability of this algorithm were validated using leave-one-out cross-validation and external test data sets. The results obtained were statistically reliable in terms of goodness-of-fit, robustness, and predictability metrics. Additionally, the developed model demonstrated a low prediction error when predicting FDA-approved drugs. These results confirm that the enQRASAR algorithm can be used to predict skin cytotoxicity of chemicals. Therefore, this model was publicly available to further facilitate toxicity predictions of unknown compounds in chemical registrations.

Read-across-based intelligent learning: development of a global q-RASAR model for the efficient quantitative predictions of skin sensitization potential of diverse organic chemicals

Environmental chemicals and contaminants cause a wide array of harmful implications to terrestrial and aquatic life which ranges from skin sensitization to acute oral toxicity. The current study aims to assess the quantitative skin sensitization potential of a large set of industrial and environmental chemicals acting through different mechanisms using the novel quantitative Read-Across Structure-Activity Relationship (q-RASAR) approach. Based on the identified important set of structural and physicochemical features, Read-Across-based hyperparameters were optimized using the training set compounds followed by the calculation of similarity and error-based RASAR descriptors. Data fusion, further feature selection, and removal of prediction confidence outliers were performed to generate a partial least squares (PLS) q-RASAR model, followed by the application of various Machine Learning (ML) tools to check the quality of predictions. The PLS model was found to be the best among different models. A simple user-friendly Java-based software tool was developed based on the PLS model, which efficiently predicts the toxicity value(s) of query compound(s) along with their status of Applicability Domain (AD) in terms of leverage values. This model has been developed using structurally diverse compounds and is expected to predict efficiently and quantitatively the skin sensitization potential of environmental chemicals to estimate their occupational and health hazards.

Machine learning - based q-RASAR modeling to predict acute contact toxicity of binary organic pesticide mixtures in honey bees

We have reported here a quantitative read-across structure-activity relationship (q-RASAR) model for the prediction of binary mixture toxicity (acute contact toxicity) in honey bees. Both the quantitative structure-activity relationship (QSAR) and the similarity-based read-across algorithms are used simultaneously for enhancing the predictability of the model. Several similarity and error-based parameters, obtained from the read-across prediction tool, have been put together with the structural and physicochemical descriptors to develop the final q-RASAR model. The calculated statistical and validation metrics indicate the goodness-of-fit, robustness, and good predictability of the partial least squares (PLS) regression model. Machine learning algorithms like ridge regression, linear support vector machine (SVM), and non-linear SVM have been used to further enhance the predictability of the q-RASAR model. The prediction quality of the q-RASAR models outperforms the previously reported quasi-SMILEs-based QSAR model in terms of external correlation coefficient (Q2F1 SVM q-RASAR: 0.935 vs. Q2VLD QSAR: 0.89). In this research, the toxicity values of several new untested binary mixtures have been predicted with the new models, and the reliability of the PLS predictions has been validated by the prediction reliability indicator tool. The q-RASAR approach can be used as reliable, complementary, and integrative to the conventional experimental approaches of pesticide mixture risk assessment.

Prediction-Inspired Intelligent Training for the Development of Classification Read-across Structure-Activity Relationship (c-RASAR) Models for Organic Skin Sensitizers: Assessment of Classification Error Rate from Novel Similarity Coefficients

The advancements in the field of cheminformatics have led to a reduction in animal testing to estimate the activity, property, and toxicity of query chemicals. Read-across structure-activity relationship (RASAR) is an emerging concept that utilizes various similarity functions derived from chemical information to develop highly predictive models. Unlike quantitative structure-activity relationship (QSAR) models, RASAR descriptors of a query compound are computed from its close congeners instead of the compound itself, thus targeting predictions in the model training phase. The objective of the present study is not to propose new QSAR models for skin sensitization but to demonstrate the enhancement in the quality of predictions of the skin-sensitizing potential of organic compounds by developing classification-based RASAR (c-RASAR) models. A diverse, previously curated data set was collected from the literature for which 2D descriptors were computed. The extracted essential features were then used to develop a classification-based linear discriminant analysis (LDA) QSAR model. Furthermore, from the read-across-based predictions, RASAR descriptors were calculated using the basic settings of the hyperparameters for the Laplacian Kernel-based optimum similarity measure. After feature selection, an LDA c-RASAR model was developed, which superseded the prediction quality of the LDA-QSAR model. Various other combinations of RASAR descriptors were also taken to develop additional c-RASAR models, all showing better prediction quality than the LDA QSAR model while using a lower number of descriptors. Various other machine learning c-RASAR models were also developed for comparison purposes. In this work, we have proposed and analyzed three new similarity metrics: gm_class, sm1, and sm2. The first one is an indicator variable used to generate a simple univariate c-RASAR model with good prediction ability, while the remaining two are similarity indices used to analyze possible activity cliffs in the training and test sets and are believed to play an important role in the modelability analysis of data sets.

In silico identification of novel heterocyclic compounds combats Alzheimer's disease through inhibition of butyrylcholinesterase enzymatic activity

Increasing evidence indicates that heterocyclic molecules possess properties against butyrylcholinesterase (BChE) enzymatic activity, which is a potential therapeutic target for Alzheimer's disease (AD). Thus, this study aimed to further evaluate the relationship between heterocyclic molecules and their biological activities. A dataset of 38 selective and potent heterocyclic compounds (-log[the half‑maximal inhibitory concentration (pIC50)]) values ranging from 8.02 to 10.05) was applied to construct a quantitative structure-activity relationship (QSAR) study, including Bayesian model average (BMA), artificial neural network (ANN), multiple nonlinear regression (MNLR), and multiple linear regression (MLR) models. Four models met statistical acceptance in internal and external validation. The ANN model was superior to other models in predicting the pIC50 of the outcome. The descriptors put into the models were found to be comparable with the target-ligand complex X-ray structures, making these models interpretable. Three selected molecules possess drug-like properties (pIC50 values ranged from 9.19 to 9.54). The docking score between candidates and the BChE receptor (RCSB ID 6EYF) ranged from -8.4 to -9.0 kcal/mol. Remarkably, the pharmacokinetics, biological activities, molecular dynamics, and physicochemical properties of compound 18 (C20H22N4O, pIC50 value = 9.33, oxadiazole derivative group) support its protective effects on AD treatment due to its non-toxic nature, non-carcinogen, cholinergic nature, capability to penetrate the blood-brain barrier, and high gastrointestinal absorption.Communicated by Ramaswamy H. Sarma.

Hi-MGT: A hybrid molecule graph transformer for toxicity identification

Conventional toxicity testing methods that rely on animal experimentation are resource-intensive, time-consuming, and ethically controversial. Therefore, the development of alternative non-animal testing approaches is crucial. This study proposes a novel hybrid graph transformer architecture, termed Hi-MGT, for the toxicity identification. An innovative aggregation strategy, referred to as GNN-GT combination, enables Hi-MGT to simultaneously and comprehensively aggregate local and global structural information of molecules, thus elucidating more informative toxicity information hidden in molecule graphs. The results show that the state-of-the-art model outperforms current baseline CML and DL models on a diverse range of toxicity endpoints and is even comparable to large-scale pretrained GNNs with geometry enhancement. Additionally, the impact of hyperparameters on model performance is investigated, and a systematic ablation study is conducted to demonstrate the effectiveness of the GNN-GT combination. Moreover, this study provides valuable insights into the learning process on molecules and proposes a novel similarity-based method for toxic site detection, which could potentially facilitate toxicity identification and analysis. Overall, the Hi-MGT model represents a significant advancement in the development of alternative non-animal testing approaches for toxicity identification, with promising implications for enhancing human safety in the use of chemical compounds.

Machine learning-based approach for efficient prediction of toxicity of chemical gases using feature selection

Toxic gases can be fatal as they damage many living tissues, especially the nervous and respiratory systems. They can cause permanent damage for many years by harming environmental tissue and living organisms. They can also cause mass deaths when used as chemical weapons. These chemical agents consist of organophosphates, namely ester, amide, or thiol derivatives of phosphorus, phosphonic or phosphinic acids, or can be synthesized independently. In this study, machine learning models were used to predict the toxicity of chemical gases. Toxic and non-toxic gases, consisting of 144 gases, were identified according to the United States Environmental Protection Agency, Occupational Safety and Health Administration, and the Centers for Disease Control and Prevention. Six machine-learning models were used to predict the toxicity of these chemical gases. The performance of the models was verified through internal and external validation. The results showed that the model's internal validation accuracy was 86.96% with the Relief-J48 algorithm. The accuracy value of the model was 89.65% with the Bayes Net algorithm for external validation. Our results reveal that identifying the toxicity of existing and potential chemicals is essential for the early detection of these chemicals in nature.