QSTR Modeling to Find Relevant DFT Descriptors Related to the Toxicity of Carbamates

Hydrodesulfurization of Dibenzothiophene: A Machine Learning Approach

The hydrodesulfurization (HDS) process is widely used in the industry to eliminate sulfur compounds from fuels. However, removing dibenzothiophene (DBT) and its derivatives is a challenge. Here, the key aspects that affect the efficiency of catalysts in the HDS of DBT were investigated using machine learning (ML) algorithms. The conversion of DBT and selectivity was estimated by applying Lasso, Ridge, and Random Forest regression techniques. For the estimation of conversion of DBT, Random Forest and Lasso offer adequate predictions. At the same time, regularized regressions have similar outcomes, which are suitable for selectivity estimations. According to the regression coefficient, the structural parameters are essential predictors for selectivity, highlighting the pore size, and slab length. These properties can connect with aspects like the availability of active sites. The insights gained through ML techniques about the HDS catalysts agree with the interpretations of previous experimental reports.

Several models combined with ultrasound techniques to predict breast muscle weight in broilers

The weight of breast muscle (WBM) is a highly monitored indicator in broiler breeding that can be obtained after slaughtering. Currently, due to the lack of accurate in vivo phenotypes for both genomic and phenotypic selection, genetic gains in WBM fall short of initial expectations. In this study, 1,006 market-age (42 d) broilers from 3 generations over 2 yr were randomly selected, and the breast width (BW), fossil bone length (FBL), breast muscle thickness (BMT), and live weight (LW) were measured exactly in vivo. Eight models, including multiple linear regression (MLR), ridge regression (RR), least absolute shrinkage and selection operator (LASSO), and elastic net (EN), were fitted to explore the best regression relationships between breast muscle weight and these indicators. Support vector machine (SVM) methods with both linear kernels and radial kernels were used to fit the models, while 2 decision tree-based machine learning algorithms, random forest (RF), and extreme gradient boosting (XGBoost), were used to establish the prediction model. The predictive effects of different combinations of independent variables were compared, leading to the conclusion that the EN model achieves the best predictive power when all 4 live features are used as inputs and is slightly better than the other models (R2 = 0.7696). This method could be applied in practical production and breeding work, leading to substantial cost savings and enhancements in the breeding process.

Conceptual DFT, machine learning and molecular docking as tools for predicting LD50 toxicity of organothiophosphates

CONTEXT

Several descriptors from conceptual density functional theory (cDFT) and the quantum theory of atoms in molecules (QTAIM) were utilized in Random Forest (RF), LASSO, Ridge, Elastic Net (EN), and Support Vector Machines (SVM) methods to predict the toxicity (LD₅₀) of sixty-two organothiophosphate compounds. The A-RF-G1 and A-RF-G2 models were obtained using the RF method, yielding statistically significant parameters with good performance, as indicated by R² values for the training set (R²_Train) and R² values for the test set (R²_Test), around 0.90.

METHODS

The molecular structure of all organothiophosphates was optimized via the range-separated hybrid functional ωB97XD with the 6-311 +  + G** basis set. Seven hundred and eighty-seven descriptors have been processed using a variety of machine learning algorithms: RF LASSO, Ridge, EN and SVM to generate a predictive model. The properties were obtained with Multiwfn, AIMALL and VMD programs. Docking simulations were performed by using AutoDock 4.2 and LigPlot + programs. All the calculations in this work are carried out in Gaussian 16 program package.

Quantitative Measurements of Pharmacological and Toxicological Activity of Molecules

Toxicity and pharmacological activity scales of molecules, in particular toxicants, xenobiotics, drugs, nutraceuticals, etc., are described by multiples indicators, and the most popular is the median lethal dose (LD50). At the molecular level, reversible inhibition or binding constants provide unique information on the potential activity of molecules. The important problem concerning the meaningfulness of IC50 for irreversible ligands/inhibitors is emphasized. Definitions and principles for determination of these quantitative parameters are briefly introduced in this article. Special attention is devoted to the relationships between these indicators. Finally, different approaches making it possible to link pharmacological and toxicological properties of molecules in terms of molecular interactions (or chemical reactions) with their biological targets are briefly examined. Experimental trends for future high-throughput screening of active molecules are pointed out.