QSPR estimation models of normal boiling point and relative liquid density of pure hydrocarbons using MLR and MLP-ANN methods

Combined interaction of fungicides binary mixtures: experimental study and machine learning-driven QSAR modeling

Fungicide mixtures are an effective strategy in delaying the development of fungicide resistance. In this research, a fixed ratio ray design method was used to generate fifty binary mixtures of five fungicides with diverse modes of action. The interaction of these mixtures was then analyzed using CA and IA models. QSAR modeling was conducted to assess their fungicidal activity through multiple linear regression (MLR), support vector machine (SVM), and artificial neural network (ANN). Most mixtures exhibited additive interaction, with the CA model proving more accurate than the IA model in predicting fungicidal activity. The MLR model showed a good linear correlation between selected theoretical descriptors by the genetic algorithm and fungicidal activity. However, both ML-based models demonstrated better predictive performance than the MLR model. The ANN model showed slightly better predictability than the SVM model, with R2 and R2cv at 0.91 and 0.81, respectively. For external validation, the R2test value was 0.845. In contrast, the SVM model had values of 0.91, 0.78, and 0.77 for the same metrics. In conclusion, the proposed ML-based model can be a valuable tool for developing potent fungicidal mixtures to delay fungicidal resistance emergence.

Improving ADMET Prediction Accuracy for Candidate Drugs: Factors to Consider in QSPR Modeling Approaches

Quantitative Structure-Property Relationship (QSPR) employs mathematical and statistical methods to reveal quantitative correlations between the pharmacokinetics of compounds and their molecular structures, as well as their physical and chemical properties. QSPR models have been widely applied in the prediction of drug absorption, distribution, metabolism, excretion, and toxicity (ADMET). However, the accuracy of QSPR models for predicting drug ADMET properties still needs improvement. Therefore, this paper comprehensively reviews the tools employed in various stages of QSPR predictions for drug ADMET. It summarizes commonly used approaches to building QSPR models, systematically analyzing the advantages and limitations of each modeling method to ensure their judicious application. We provide an overview of recent advancements in the application of QSPR models for predicting drug ADMET properties. Furthermore, this review explores the inherent challenges in QSPR modeling while also proposing a range of considerations aimed at enhancing model prediction accuracy. The objective is to enhance the predictive capabilities of QSPR models in the field of drug development and provide valuable reference and guidance for researchers in this domain.

Predicting rejection of emerging contaminants through RO membrane filtration based on ANN-QSAR modeling approach: trends in molecular descriptors and structures towards rejections

In this work, a quantitative structure-activity relationship (QSAR) study was performed on a set of emerging contaminants (ECs) to predict their rejections by reverse osmosis membrane (RO). A wide range of molecular descriptors was calculated by Dragon software for 72 ECs. The QSAR data was analyzed by an artificial neural network method (ANN), in which four out of 3000 theoretical molecular descriptors were chosen and their significance was computed based on the Garson method. The significance trends of descriptors were as follows in descending order: ESpm14u > R2e > SIC1 > EEig03d. The selected descriptors were ranked based on their importance and then an explorative study was conducted on the QSAR data to show the trends in molecular descriptors and structures toward the rejections values of ECs. The MLR algorithm was used to make a linear model and the results were compared with those of the nonlinear ANN algorithm. The comparison results revealed it is necessary to apply the ANN model to this data with non-linear properties. For the whole dataset, the correlation coefficient (R2) and residual mean squared error (RMSE) of the ANN and MLR methods were 0.9528, 6.4224; and 0.8753, 11.3400, respectively. The comparison results showed the superiority of ANN modeling in the analysis of ECs' QSAR data.

Development of QSPR-ANN models for the estimation of critical properties of pure hydrocarbons

The current work aimed to predict three critical properties: critical temperature (Tc), critical volume (Vc), and critical pressure (Pc) of pure hydrocarbons. A multi-layer perceptron artificial neural network (MLP-ANN) has been adopted as a nonlinear modeling technique and computational approach based on a few relevant molecular descriptors. A set of diverse data points was used to build three QSPR-ANN models, including 223 points for Tc, Vc, and 221 for Pc. The entire database was randomly split into two subsets: 80% for the training set and 20% for the testing set. A large number of 1666 molecular descriptors were calculated and then reduced by a statistical methodology based on several phases to retain them into a reasonable number of relevant descriptors, wherein about 99% of initial descriptors were excluded. Thus, the Quasi-Newton backpropagation (BFGS) algorithm was applied to train the ANN structure. The results of three QSPR-ANN models showed good precision, confirmed by the high values of determination coefficient (R²) ranging from 0.9990 to 0.9945, and the low values of calculated errors, such as the Mean Absolute Percentage Error (MAPE) that ranged from 2.2497 to 0.7424% for the best three models of Tc, Vc, and Pc. The weight sensitivity analysis method was applied to know the contribution of each input descriptor individually or by class on each appropriate QSPR-ANN model. Moreover, the applicability domain (AD) method was also used with a strict limit of standardized residual values (d_i = ±2). However, the results were promising, with nearly 88% of the data points validated within the AD range. Finally, the results of the proposed QSPR-ANN models were compared with other well-known QSPR or ANN models for each property. Consequently, our three models provided satisfactory results, outperforming most of the models mentioned in this comparison. This computational approach can be applied in petroleum engineering and other related fields to accurately determine the critical properties of pure hydrocarbons: Tc, Vc, and Pc.

Intelligent soft computational models integrated for the prediction of potentially toxic elements and groundwater quality indicators: a case study

Reports have shown that potentially toxic elements (PTEs) in air, water, and soil systems expose humans to carcinogenic and non-carcinogenic health risks. In southeastern Nigeria, works that have used data-driven algorithms in predicting PTEs in groundwater are scarce. In addition, only a few works have simulated water quality indices using machine learning modelling methods in the region. Therefore, in this study, physicochemical analyses were carried out on groundwater samples in southeastern Nigeria. The laboratory results were used to compute two water quality indices: pollution index of groundwater (PIG) and the water pollution index (WPI), to ascertain groundwater quality. In addition, the physicochemical parameters served as input variables for multiple linear regression (MLR) and artificial neural network (ANN) modelling and prediction of Cr, Fe, Ni, NO3−, Pb, Zn, WPI, and PIG. The results of WPI and PIG computation showed that about 30–35% of the groundwater samples were unsuitable for human consumption, whereas 65–70% of the samples were deemed suitable. The insights from the PIG and WPI model also revealed that lead (Pb) was the most influential PTE that degraded the quality of groundwater resources in the research area. The findings of the MLR and ANN models indicated strong positive prediction accuracies (R 2 = 0.856–1.000) with low modeling errors. The predictive MLR and ANN models of the PIG and WPI generally outperformed those of the PTEs. The models produced in this study predicted the PTEs better compared to previous studies. Thus, this work provides insights into effective water sustainability, management, and protection.

Versatile in silico modelling of microplastics adsorption capacity in aqueous environment based on molecular descriptor and machine learning

To comprehensively evaluate the hazards of microplastics and their coexisting organic pollutants, the sorption capacity of microplastics is a major issue that is quantified through the microplastic-aqueous sorption coefficient (K_d). Almost all quantitative structure-property relationship (QSPR) models that describe K_d apply only to narrow, relatively homogeneous groups of reactants. Herein, non-hybrid QSPR-based models were developed to predict PE-water (K_PE-w), PE-seawater (K_PE-sw), PVC-water (K_PVC-w) and PP-seawater (K_PP-sw) sorption coefficients at different temperatures, with eight machine learning algorithms. Moreover, novel hybrid intelligent models for predicting K_d more accurately were innovatively developed by applying GA, PSO and AdaBoost algorithms to optimize MLP and ELM models. The results indicated that all three optimization algorithms could improve the robustness and predictability of the standalone MLP and ELM models. In all models trained with K_PE-w, K_PE-sw, K_PVC-w and K_PP-sw data sets, GBDT-1 and XGBoost-1 models, MLP-GA-2 and MLP-PSO-2 models, MLR-3 and MLR-4 models performed better in terms of goodness of fit (R_adj²: 0.907-0.999), robustness (Q_BOOT²: 0.900-0.937) and predictability (R_ext²: 0.889-0.970), respectively. Analyzing the descriptors revealed that temperature, lipophilicity, ionization potential and molecular size were correlated closely with the adsorption capacity of microplastics to organic pollutants. The proposed QSPR models may assist in initial environmental exposure assessments without imposing heavy costs in the early experimental phase.

Predicting age at onset of type 1 diabetes in children using regression, artificial neural network and Random Forest: A case study in Saudi Arabia

The rising incidence of type 1 diabetes (T1D) among children is an increasing concern globally. A reliable estimate of the age at onset of T1D in children would facilitate intervention plans for medical practitioners to reduce the problems with delayed diagnosis of T1D. This paper has utilised Multiple Linear Regression (MLR), Artificial Neural Network (ANN) and Random Forest (RF) to model and predict the age at onset of T1D in children in Saudi Arabia (S.A.) which is ranked as the 7th for the highest number of T1D and 5th in the world for the incidence rate of T1D. De-identified data between (2010-2020) from three cities in S.A. were used to model and predict the age at onset of T1D. The best subset model selection criteria, coefficient of determination, and diagnostic tests were deployed to select the most significant variables. The efficacy of models for predicting the age at onset was assessed using multi-prediction accuracy measures. The average age at onset of T1D is 6.2 years and the most common age group for onset is (5-9) years. Most of the children in the sample (68%) are from urban areas of S.A., 75% were delivered after a full term pregnancy length and 31% were delivered through a cesarean section. The models of best fit were the MLR and RF models with R2 = (0.85 and 0.95), the root mean square error = (0.25 and 0.15) and mean absolute error = (0.19 and 0.11) respectively for logarithm of age at onset. This study for the first time has utilised MLR, ANN and RF models to predict the age at onset of T1D in children in S.A. These models can effectively aid health care providers to monitor and create intervention strategies to reduce the impact of T1D in children in S.A.

Prediction modeling of potentially toxic elements' hydrogeopollution using an integrated Q-mode HCs and ANNs machine learning approach in SE Nigeria

Machine learning techniques have proven to be very useful in environmental and engineering assessments, including water quality studies. This is because they have flexible linear and nonlinear forecasting functions that can efficiently and reliably estimate measurable and continuous variables. The aim of this paper was to classify the water quality and also predict potentially toxic anions (PTAs; e.g., Cl, SO₄, HCO₃, and NO₃) and potentially toxic heavy metals (PTHMs; e.g., Fe, Zn, Ni, Cr, and Pb) in water resources in Ojoto and its surroundings, Nigeria. Q-mode hierarchical clusters (HCs) and artificial neural networks (ANNs) were integrated to achieve the research objectives. Prior to the HCs and ANNs modeling, correlation-, unrotated principal component-, and varimax-rotated factor analyses were performed to flag the level of associations between the input water quality variables. With respect to pH, two water quality cluster groups were identified. However, three and four cluster groups were identified based on the PTAs and PTHMs concentrations, respectively. Four ANN models (two for each group) were used for predicting the PTAs and PTHMs in the waters resources. Using coefficient of determination (R²) and AUC (area under curve) values and direct comparison of parity plots, the performance and accuracy of the ANN models were substantiated. Overall, the results obtained reveal that the proposed ANN models suitably predicted the concentrations of the PTAs and PTHMs. Thus, this paper provides useful information for better monitoring, management, and protection of the water resources. However, more modeling studies are encouraged to validate and/or improve the findings of the current work.

Exploring QSPR models for predicting PUF-air partition coefficients of organic compounds with linear and nonlinear approaches

Partition coefficients are important parameters for measuring the concentration of chemicals by passive sampling devices. Considering the wide application of the polyurethane foam (PUF) in passive air sampling, an attempt for developing several quantitative structure-property relationship (QSPR) models was made in this work, to predict PUF-air partition coefficients (K_PUF-air) using linear (multiple linear regression, MLR) and non-linear (artificial neural network, ANN and support vector machine, SVM) methods by machine learning. All of the developed models were performed on a dataset of 170 compounds comprising 9 distinct classes. A series of statistical parameters and validation results showed that models had good prediction ability, robustness and goodness-of-fit. Furthermore, the underlying mechanisms of molecular descriptors emphasized that ionization potential, molecular bond, hydrophilicity, size of molecule and valence electron number had dominating influence on the adsorption process of chemicals. Overall, the obtained models were all established on the extensive applicability domains, and thus can be used as effective tools to predict the K_PUF-air of new organic compounds or those have not been synthesized yet which, in turn, could help researchers better understand the mechanistic basis of adsorption behavior of PUF.

In-silico driven design and development of spirobenzimidazo-quinazolines as potential DNA gyrase inhibitors

DNA gyrase and Topoisomerase IV are promising antibacterial drug targets as they regulate bacterial DNA replication and topology. In a quest for novel DNA topoisomerase inhibitors, a multidisciplinary approach was adopted that involves computational prediction of binding sites and molecular modelling followed by green synthesis and biological evaluation of antibacterial activity of spirobenzimidazo quinazolines derivatives. Using basic quantum chemistry principles, we evaluated spirobenzimidazo quinazolines derivatives with their pharmacokinetic profiles. Based on the results of the aforesaid in-silico studies, we synthesized a series of titled compounds using green synthetic methodology that were validated as potential antimicrobial agents. Quantum chemoinformatics based predicted activity for the synthesized compounds 9b, 9c, and 9j was concomitant with biological evaluation of broadspectrum antibacterial activity. Biological evaluation revealed that inhibition of biofilm formation was due to their potential antibacterial activity. We believe that the novel spirobenzimidazo quinazolines have the potential to be alternatives to aminocoumarins and classical quinazolines upon detailed target specific biological studies.

Application of artificial neural networks to predict the heavy metal contamination in the Bartin River

In this study, copper (Cu), iron (Fe), zinc (Zn), manganese (Mn), nickel (Ni), and lead (Pb) analyses were performed, and the results were modelled by artificial neural networks (ANN) and adaptive neuro-fuzzy inference system (ANFIS). Samples were taken from 3 stations selected on the Bartin River for 1 year between December 2012 and December 2013. Radial basis neural network (RBANN), multilayer perceptron (MLP) neural networks models, and adaptive neuro-fuzzy inference system (ANFIS) were applied to the data in order to predict the heavy metal concentrations. As a result of the study, the RMSE and MAE values of all the heavy metal models were found to have very low error values during the test phase, and it was found that the models created using MLP had R² values higher than 0.77 during the test phase; the test phase R² values of the models using RBN method were found to be ranging between 0.773 and 0.989, and the test phase R² value of the ANFIS model was higher than 0.80. If sorted from the best model to the worst by taking the MAE and RMSE values into consideration based on the test evaluation results, according to the heavy metal types, where all of the MLP, RBN, and ANFIS models were generally approximate to each other, RBN was successful for Cu, Zn, and Mn, while MLP model was successful for Ni and ANFIS model for Fe and Pb. According to the results, it can be inferred that the heavy metal contents can be estimated approximately with artificial intelligence models and relatively easy-to-measure parameters; it will be possible to detect heavy metals which are harmful to the viability of the rivers, both quickly and economically.

Accurate elemental boiling points from first principles

The normal boiling point (NBP) is a fundamental property of liquids and marks the intersection of the Gibbs energies of the liquid and the gas-phase at ambient pressure. This work provides the first comprehensive demonstration of the calculation of boiling points of atomic liquids through first-principles molecular-dynamics simulations. To this end, thermodynamic integration (TDI) and perturbation theory (TPT) are combined with a density-functional theory (DFT) Hamiltonian, which provides absolute Gibbs energies, internal energies, and entropies of atomic liquids with an accuracy of a few meV/atom. Linear extrapolation to the intersection with the Gibbs energy of a non-interacting gas-phase eventually pins-down the NBPs. While these direct results can already be quite accurate, they are susceptible to a systematic over or underbinding of the employed density functional. It is shown how this dependency can be strongly reduced and the robustness of the method increased through a simple linear correction termed λ-scaling. Eventually, by carefully tuning of the technical parameters of the approach, the walltime per element is reduced from weeks to about a day (10-20k core-hours), enabling extensive testing for B, Al, Na, K, Ca, Sr, Ba, Mn, Cu, Xe, and Hg. This comprehensive benchmark demonstrates the excellent performance and robustness of the approach with a mean absolute deviation (MAD) of less than 2% from experimental NBPs and very similar accuracy for liquid entropies (MAD 2.3 J (mol K)-1, 2% relative). In some cases, the uncertainties in the predictions are several times smaller than the variation between literature values, allowing us to clear out long-standing ambiguities in the NBPs of B and Ba.

Development and validation of QSPR models for corrosion inhibition of carbon steel by some pyridazine derivatives in acidic medium

Statistical modeling of the corrosion inhibition process by twenty-one pyridazine derivatives for mild steel in acidic medium was investigated by the quantitative structure property relationship (QSPR) approach. This modeling was established by the correlation between the corrosion inhibition efficiency (IE %) and a number of the electronic and structural properties of these inhibitors such as: the E HOMO (highest occupied molecular orbital energy), the E LUMO (lowest unoccupied molecular orbital energy), the energy gap (E L-H ), the dipole moment (μ), the hardness (η), the softness (σ), the absolute electronegativity (χ), the ionization potential (IP), the electron affinity (EA), the fraction of electrons transferred (ΔN), the electrophilicity index ω the molecular volume (V m ), the logarithm of the partition coefficient (Log P), and the molecular mass (M), in addition to the inhibitor concentration (C i ). The structure electronic properties was calculated by the use of the density functional theory method (DFT), at B3LYP/6-31G (d, p) level of theory and the analysis of dimensionality and redundancy as well as the test of collinearity between descriptors are carried out using principal component analysis (PCA). Whereas, the correlation between EI % and molecular structure is performed through the development of tree mathematical models, based-QSPR approaches: the partial least squares regression (PLS), the principal component regression (PCR) and the artificial neural networks (ANN). Indeed, the statistical quantitative results revealed that PCR and ANN were the most relevant and predictive models in comparison with the PLS model. This pertinence was demonstrated by using leave one-out cross-validation as an efficient method for testing the internal stability and predictive capability of said models with a high cross-validated determination coefficient R 2 cv = 0.92 and predicted determination coefficient R 2 pred = 0.92 and R 2 pred = 0.90 for PCR and ANN respectively; in addition to an extrapolation test set as an external validation with a significant external coefficient of determination: R 2 test = 0.94 and R 2 test = 0.92, for the two correspondingly models.

Versatile modelling of polyoxymethylene-water partition coefficients for hydrophobic organic contaminants using linear and nonlinear approaches

Environmental fate or transport of hydrophobic organic contaminants (HOCs) depends on the partitioning properties of compounds within various environmental phases. Due to the wide application of polyoxymethylene (POM) in the passive sampling technique, several in silico models were developed to predict POM-water partition coefficients (K_POM-w) in accordance with the guidelines of the Organization for Economic Cooperation and Development (OECD). It is an attempt to combine conventional linear method (multiple linear regression, MLR) and popular nonlinear algorithm (artificial neural network, ANN) for estimating partition coefficients of HOCs. All models were performed on a dataset of 210 chemicals from 13 different classes. The polyparameter linear free energy relationship (pp-LFER) model included 5 molecular descriptors, namely, E, S, A, B and V, and predicted log K_POM-w with R²_adj of 0.825. The values of statistical parameters including R²_adj, Q²_ext, RMSE_tra and RMSE_ext for quantitative structure-property relationship (QSPR)-MLR and QSPR-ANN models with four descriptors (ALOGP, MeanDD, E1m and Mor24s) were: (0.928, 0.877, 0.498 and 0.649) and (0.943, 0.905, 0.443 and 0.571), with high similarity for both models, which confirmed the robustness, significance, and remarkable prediction accuracy of the QSPR models. Moreover, the mechanism interpretation revealed that the molecular volume and hydrophobicity had a major impact on distribution procedure of HOCs. The models developed herein, with the broad applicability domain (AD), provide suitable tools to fill the experimental data gap for untested chemicals and help researchers better understand the mechanistic basis of adsorption behavior of POM.

Predicting Methane Concentration in Longwall Regions Using Artificial Neural Networks

Methane, which is released during mining exploitation, represents a serious threat to this process. This is because the gas may ignite or cause an explosion. Both of these phenomena are extremely dangerous. High levels of methane concentration in mine headings disrupt mining operations and cause the risk of fire or explosion. Therefore, it is necessary to monitor and predict its concentration in the areas of ongoing mining exploitation. The paper presents the results of tests performed to improve work safety. The article presents the methodology of using artificial neural networks for predicting methane concentration values in one mining area. The objective of the paper is to develop an effective method for forecasting methane concentration in the mining industry. The application of neural networks for this purpose represents one of the first attempts in this respect. The method developed makes use of direct methane concentration values measured by a system of sensors located in the exploitation area. The forecasting model was built on the basis of a Multilayer Perceptron (MLP) network. The corresponding calculations were performed using a three-layered network with non-linear activation functions. The results obtained in the form of methane concentration prediction demonstrated minor errors in relation to the recorded values of this concentration. This offers an opportunity for a broader application of intelligent systems for effective prediction of mining hazards.