Quasi-SMILES for predicting toxicity of Nano-mixtures to Daphnia Magna

A comprehensive machine learning-based models for predicting mixture toxicity of azole fungicides toward algae (Auxenochlorella pyrenoidosa)

Quantitative structure-activity relationships (QSARs) have been used to predict mixture toxicity. However, current research faces gaps in achieving accurate predictions of the mixture toxicity of azole fungicides. To address this gap, the application of machine learning (ML) algorithms has emerged as an effective strategy. In this study, we applied 12 algorithms, namely, k-nearest neighbor (KNN), kernel k-nearest neighbors (KKNN), support vector machine (SVM), random forest (RF), stochastic gradient boosting (GBM), cubist, bagged multivariate adaptive regression splines (Bagged MARS), eXtreme gradient boosting (XGBoost), boosted generalized linear model (GLMBoost), boosted generalized additive model (GAMBoost), bayesian regularized neural networks (BRNN), and recursive partitioning and regression trees (CART) to build ML models for 225 mixture toxicity of azole fungicides towards Auxenochlorella pyrenoidosa. A total of 36 single ML models and 12 consensus models were developed. The results indicated that models employing concentration addition (CA), independent action (IA), and molecular descriptors (MD) as variables demonstrated superior predictive abilities. The consensus model combining SVM and RF algorithms (labeled as CM0) demonstrated the highest level of accuracy in fitting the data, with a coefficient of determination of 0.980. Additionally, it showed strong predictive abilities when tested with external data, achieving an external R2 value of 0.945 and a Concordance Correlation Coefficient of 0.967. This study provides a positive contribution to the ecological risk assessment of a mixture of azole fungicides.

In silico aquatic toxicity prediction of chemicals toward Daphnia magna and fathead minnow using Monte Carlo approaches

The fast-increasing use of chemicals led to large numbers of chemical compounds entering the aquatic environment, raising concerns about their potential effects on ecosystems. Therefore, assessment of the ecotoxicological features of organic compounds on aquatic organisms is very important. Daphnia magna and Fathead minnow are two aquatic species that are commonly tested as standard test organisms for aquatic risk assessment and are typically chosen as the biological model for the ecotoxicology investigations of chemical pollutants. Herein, global quantitative structure-toxicity relationship (QSTR) models have been developed to predict the toxicity (pEC(LC)50) of a large dataset comprising 2106 chemicals toward Daphnia magna and Fathead minnow. The optimal descriptor of correlation weights (DCWs) is calculated using the notation of simplified molecular input line entry system (SMILES) and is used to construct QSTR models. Three target functions, TF1, TF2, and TF3 are utilized to generate 12 QSTR models from four splits, and their statistical characteristics are also compared. The designed QSTR models are validated using both internal and external validation criteria and are found to be reliable, robust, and excellently predictive. Among the models, those generated using the TF3 demonstrate the best statistical quality with R2 values ranging from 0.9467 to 0.9607, Q2 values ranging from 0.9462 to 0.9603 and RMSE values ranging from 0.3764 to 0.4413 for the validation set. The applicability domain and the mechanistic interpretations of generated models were also discussed.

Models for the No-Observed-Effect Concentration (NOEC) and Maximal Half-Effective Concentration (EC50)

Typical in silico models for ecotoxicology focus on a few endpoints, but there is a need to increase the diversity of these models. This study proposes models using the NOEC for the harlequin fly (Chironomus riparius) and EC50 for swollen duckweed (Lemna gibba) for the first time. The data were derived from the EFSA OpenFoodTox database. The models were based on the correlation weights of molecular features used to calculate the 2D descriptor in CORAL software. The Monte Carlo method was used to calculate the correlation weights of the algorithms. The determination coefficients of the best models for the external validation set were 0.74 (NOAEC) and 0.85 (EC50).

The system of self-consistent models based on quasi-SMILES as a tool to predict the potential of nano-inhibitors of human lung carcinoma cell line A549 for different experimental conditions

The different features of the impact of nanoparticles on cells, such as the structure of the core, presence/absence of doping, quality of surface, diameter, and dose, were used to define quasi-SMILES, a line of symbols encoded the above physicochemical features of the impact of nanoparticles. The correlation weight for each code in the quasi-SMILES has been calculated by the Monte Carlo method. The descriptor, which is the sum of the correlation weights, is the basis for a one-variable model of the biological activity of nano-inhibitors of human lung carcinoma cell line A549. The system of models obtained by the above scheme was checked on the self-consistence, i.e., reproducing the statistical quality of these models observed for different distributions of available nanomaterials into the training and validation sets. The computational experiments confirm the excellent potential of the approach as a tool to predict the impact of nanomaterials under different experimental conditions. In conclusion, our model is a self-consistent model system that provides a user to assess the reliability of the statistical quality of the used approach.

In Silico Simulation of Impacts of Metal Nano-Oxides on Cell Viability in THP-1 Cells Based on the Correlation Weights of the Fragments of Molecular Structures and Codes of Experimental Conditions Represented by Means of Quasi-SMILES

A simulation of the effect of metal nano-oxides at various concentrations (25, 50, 100, and 200 milligrams per millilitre) on cell viability in THP-1 cells (%) based on data on the molecular structure of the oxide and its concentration is proposed. We used a simplified molecular input-line entry system (SMILES) to represent the molecular structure. So-called quasi-SMILES extends usual SMILES with special codes for experimental conditions (concentration). The approach based on building up models using quasi-SMILES is self-consistent, i.e., the predictive potential of the model group obtained by random splits into training and validation sets is stable. The Monte Carlo method was used as a basis for building up the above groups of models. The CORAL software was applied to building the Monte Carlo calculations. The average determination coefficient for the five different validation sets was R² = 0.806 ± 0.061.