Genetic Algorithm-multiple Linear Regression Research Articles

A Chemometric–Assisted Voltammetric Method for Simultaneous Determination of Acetaminophen and Diclofenac

AbstractDiclofenac (DCF) and Acetaminophen (ACT) are among the most important nonsteroidal anti–inflammatory drugs (NSAIDs) that sought after for their antipyretic, analgesic, and anti‐inflammatory properties. In addition, ACT and DCF show synergistic activity toward pain relief. Therefore, appropriate analytical strategies are required for the simultaneous quantification of these NSAIDs in prescription drugs. This study introduces a method for the direct and simultaneous determination of ACT and DCF using the voltammetric approach in conjunction with chemometric tools such as partial least squares (PLS) and genetic algorithm–multiple linear regressions (GA−MLR). The calibration set was established using differential pulse voltammetry for 25 distinct mixtures of ACT and DCF. The calibration curve exhibited linearity within the concentration ranges of 0.1–1000 μM for ACT and 0.1–100 and 100–1000 for DCF. The effectiveness of chemometric methods was evaluated by analyzing synthetic mixtures of ACT and DCF and applying these methods to real samples containing a commercial pharmaceutical preparation of the specified drugs. Both methods yielded favorable results.

Quantitative structure-electrochemistry relationship modeling of a series of anticancer agents using MLR and ANN approaches

Abstract QSPR is a powerful tool for elucidating the correlation between chemical structure and property for both natural and synthesized compounds. In the present work, the half-wave reduction potential for a set of aziridinylquinones (Anticancer Agents [AA]) is modelled using a quantitative structure-electrochemistry relationship (QSER) based on multilinear regression (MLR) and artificial neural network (ANN). Molecular descriptors introduced in this work were computed using the Dragon software (V5). Before the model’s generation, using the Kennard and Stone algorithm, the data set of 84 aziridinylquinones was divided into training and prediction sets consisting of 70 % and 30 % of data points. Quantitative Structure Electrochemistry Relationship (QSER) models were developed using the Genetic Algorithm Multiple Linear Regressions (GA-MLR) and an Artificial Neural Network (ANN). The coefficient of determination (R 2) and Root Mean Squared Error of prediction (RMSE) were mentioned to demonstrate the QSER model’s prediction abilities. Calculated R 2 and RMSEval values for the MLR model were 0.858 and 0.054, respectively. The R 2 and RMSEval values for the ANN training set were calculated to be 0.914 and 0.050, respectively. Findings show that GA is a powerful tool for selecting variables in QSER analysis. Comparing the two employed regression methods showed that ANN is superior to MLR in predictive ability.

QSAR studies of quinoline alkaloids camptothecin derivatives for prediction anticancer activity using linear and nonlinear methods

AbstractQuinoline alkaloid camptothecin (CPT) derivatives are compounds with a wide range of inhibitory activities and they are the basis of several groups of drugs. CPT is one of the prominent lead compounds in anticancer drug development. It inhibits DNA topoisomerase I (topo I) enzyme and has shown remarkable anticancer activity against lung, colon, rectum, ovarian, breast, and stomach cancers. Quantitative structure–activity relationship (QSAR) is basically a statistical approach correlating the response activity data with descriptors encoding chemical information. In the current research, a QSAR study was performed for modeling and predicting the anticancer activity (pIC50) of 76 CPT derivatives as an inhibitor of DNA topo I using the genetic algorithm multiple linear regression (GA‐MLR) and back propagation artificial neural network (BP‐ANN) methods. The results of this study indicate that the use of constitutional and geometrical descriptors provide good estimate for pIC50. The obtained results of statistical criteria, internal and external validation show the superiority of BP‐ANN model than GA‐MLR to predict the pIC50 values of the investigated compounds. QSAR model developed in this study can be used for design and development of novel potent CPT derivatives and for predicting their anticancer activity.

QSPR analysis to predict some quantum chemical properties of 2‐phenylindol derivatives as anticancer drugs using molecular descriptor and genetic algorithm multiple linear regression

AbstractThermodynamic properties of molecules provide fundamental information for binding in fragment‐based drug discovery programs and for understanding complex interactions. Experimental measurements of some thermodynamic properties are not always feasible. Quantum mechanical methods and group contribution methods are the major toolboxes to obtain some theoretical quantities. In drug discovery and development, computational studies can reduce the costs and risks of bringing a new medicine to market. Computational methods have been widely used to optimize promising new ligands by estimating their binding affinity to proteins using the thermodynamic ligand binding parameters (Gibbs energy, enthalpy, entropy, and heat capacity). In this article, the relationship between molecular descriptors and quantum chemical properties, such as the Gibbs free energy (ΔG kJ mol−1) and enthalpy of formation (ΔHf kJ mol−1) of 2‐phenylindole derivatives as anticancer drugs is studied. These properties were calculated at the DFT‐B3LYP/6‐311G (d,p) level of quantum chemistry by Gaussian 09 software. Molecular descriptors were calculated by Dragon 5.4 software, and the stepwise multiple linear regression (MLR) and the Genetic algorithm (GA) techniques were used to select the best descriptors and build QSPR models. To evaluate the predictive ability of developed quantitative structure‐property relationship (QSPR) models, different internal and external validation methods were adopted. The predictive powers of the models were found to be satisfactory. The models revealed that 3D matrix‐based descriptors (H3D, SEig) are useful to predict the Gibbs free energy and enthalpy of formation of the investigated compounds respectively.

QSPR study to predict some of quantum chemical properties of anticancer imidazo[4,5‐b]pyridine derivatives using genetic algorithm multiple linear regression and molecular descriptors

AbstractPyridine and its derivatives have been applied clinically for the treatment of a wide range of diseases and in the synthesis of novel drugs. In the present work, imidazo[4,5‐b]pyridine derivatives as anticancer drugs were exhibited to select the important descriptor for quantum chemical properties. Two of the fundamental thermodynamic properties are heat capacity (Cv) and entropy (S), which are important in the field of chemical kinetics and are key in the understanding and design of chemical processes involving chemical reactions. A Quantitative Structure–Property Relationship (QSPR) study was used to predict the quantum chemical properties like Cv and S of 105 imidazole derivatives using molecular descriptors and the genetic algorithm–multiple linear regression (GA–MLR). The best QSPR models were selected using criteria coefficients such as R2, R2adj, RMSE and Fisher ratio. Different internal and external validation metrics were adopted to evaluate the stability, fit and predictive power of the QSPR models. The validation results and statistical analysis show that the models possess good prediction power and robustness, and the total size (TS) and Sanderson electronegativity(RDF060e) and total information content index(TIC1) of imidazo[4,5‐b]pyridine derivatives are increasingly related to the studied properties.

An alternative biomimetic tool – Dual hydrophilic/reversed-phase interaction mode

During the early stages of drug discovery, it becomes increasingly important to acquire a basic physico-chemical and pharmacokinetic profile of drug candidates using efficient and cost-effective laboratory and computational models. In this study, the possibility of applying the dual hydrophilic interaction liquid chromatography/reversed-phase mode (HILIC/RP) as a alternative biomimetic tool for the preliminary profiling of imidazoline and piperazine-related compounds was investigated. Their retention behaviour on the mixed-mode HILIC/RP column was described as a dependence of the logarithm of the retention factor (logk) in relation to the volume fraction of the water component in the mobile phase (φ). The retention was quantified by integration of HILIC (logkH), RP (logkR) and the total HILIC/RP (logkA) area under the logk = f(φ) curve. The biomimetic capabilities of HILIC/RP mode were observed through physico-chemical and pharmacokinetic interpretation of chromatographic behaviour. The physico-chemical characteristics of the HILIC/RP mode were defined based on linear solvation energy relationships, genetic algorithm-multiple linear regression and Boruta-XGBoost predictive modeling. The created models showed reliable predictive performances for defining the logkA, logkR, and logkH values. The mixed-mode logkA, logkR, and logkH parameters correlate well with physico-chemical (e.g., lipophilicity indices) and basic pharmacokinetic properties of the examined compounds. The models created have confirmed that the simultaneous contribution of both HILIC/RP mechanisms can be used to accurately evaluate the key properties of newly synthesized imidazoline and piperazine-related compounds while reducing costs and selecting suitable candidates in the early stages of drug development and discovery.

Discerning of isatin-based monoamine oxidase (MAO) inhibitors for neurodegenerative disorders by exploiting 2D, 3D-QSAR modelling and molecular dynamics simulation

Almost a billion people worldwide suffer from neurological disorders, which pose public health challenges. An important enzyme that is well-known for many neurodegenerative illnesses is monoamine oxidase (MAO). Although several promising drugs for the treatment of MAO inhibition have recently been examined, it is still necessary to identify the precise structural requirements for robust efficacy. Atom-based, field-based, and GA-MLR (genetic algorithm multiple linear regression) models were created for this investigation. All of the models have strong statistical (R 2 and Q 2) foundations because of both internal and external validation. Our dataset’s molecule has a higher docking score than safinamide, a well-known and co-crystallized MAO-B inhibitor, as we also noticed. Using the SwissSimilarity platform, we further inquired which of our docked molecules would be the best for screening. We chose ZINC000016952895 as the screen molecule with the best binding docking score (XP score = −13.3613). Finally, the 100 ns for the ZINC000016952895-MAO-B complex in our MD investigations is stable. For compounds that we hit, also anticipate ADME properties. Our research revealed that the successful compound ZINC000016952895 might pave the way for the future development of MAO inhibitors for the treatment of neurological disease. Communicated by Ramaswamy H. Sarma

In silico study to recognize novel angiotensin-converting-enzyme-I inhibitors by 2D-QSAR and constraint-based molecular simulations

Cardiovascular diseases (CVD) such as heart failure, stroke, and hypertension affect 64.3 million people worldwide and are responsible for 30% of all deaths. Primary inhibition of the angiotensin-converting enzyme (ACE) is significant in the management of CVD. In the present study, the genetic algorithm-multiple linear regressions (GA-MLR) method is used to generate highly predictive and statistically significant (R 2 = 0.70–0.75, Q 2 LOO=0.67–0.73, Q 2 LMO=0.66–0.72, CCCex=0.70–0.78) quantitative structure-activity relationships (QSAR) models conferring to OECD requirements using a dataset of 255 structurally diverse and experimentally validated ACE inhibitors. The models contain simply illustratable Padel, Estate, and PyDescriptors that correlate structural scaffold requisite for ACE inhibition. Also, constraint-based molecular docking reveals an interaction profile between ligands and enzymes which is then correlated with the essential structural features associated with the QSAR models. The QSAR-based virtual screening was utilized to find novel lead molecules from a designed database of 102 thiadiazole derivatives. The Applicability domain (AD), Molecular Docking, Molecular dynamics, and ADMET analysis suggest two compound D24 and D40 are inflexibly linked to the protein binding site and follows drug-likeness properties. Communicated by Ramaswamy H. Sarma

Quantum Chemical GA-MLR, Cluster Model, and Conceptual DFT Descriptors Studies on the Binding Interaction of Estrogen Receptor Alpha with Endocrine Disrupting Chemicals

In the present study, the predication of the binding affinity (log RBA) of estrogen receptor alpha with three categories of environmental endocrine disrupting chemicals (EDCs), namely, PCB, phenol, and DDT, is performed by the quantum chemical genetic algorithm multiple linear regression (GA-MLR) method. The result of the optimal model indicates that log RBA increases with increasing the electrophilicity and hydrophobicity of EDCs. However, by using the quantum chemical cluster model approach, the modeling results reveal that electrostatic interaction and hydrogen bonding play a significant role. The chemical reactivity descriptors calculated based on the conceptual density functional theory also indicate that the binding mechanism of charge-controlled interaction is superior to that of frontier-controlled interaction.

Comparative QSAR modeling of 2‐phenylindol derivatives for predicting the anticancer activity using genetic algorithm multiple linear regression and back‐propagation‐artificial neural network techniques

AbstractQuantitative structure–activity relationship (QSAR) studies on a series of 2‐phenylindole derivatives as anticancer drugs were performed to choice the important descriptor, which is responsible for their anticancer activity (expressed as pIC50). The geometry optimizations were performed on the structures using Gaussian software with the density functional B3LYP and 6‐311G(d,p) basis sets. Dragon software was used to calculate molecular descriptors, and the genetic algorithm (GA) procedure and backward regression were used to proper selection of the most relevant descriptors. The backward multiple linear regression (BW‐MLR) and backpropagation‐artificial artificial neural network (BP‐ANN) were carried out to design QSAR models. The squared correlation coefficient (R2) and the root mean squared error (RMSE) values of the GA‐MLR model were calculated to be 0.2843 and 0.7001, respectively. The BP‐ANN model was the most powerful, with the square of predictive correlation coefficient R2pred, root mean square error (RMSE), and absolute average deviation (AAD) which was equal to 0.9416, 0.0238, and 0.0099, respectively.

Identification of hydantoin based Decaprenylphosphoryl-β-d-Ribose Oxidase (DprE1) inhibitors as antimycobacterial agents using computational tools

Tuberculosis (TB) is one of the emerging infectious diseases in the world. DprE1 (Decaprenylphosphoryl-β-d-ribose 2′-epimerase), an enzyme accountable for mycobacterial cell wall synthesis was the first drug gable target based on discoveries of inhibitors via HTS (high throughput screening). Since then, many literature reports have been published so far enlightening varieties of chemical scaffolds acting as inhibitors of DprE1. Herein, in our present study, we have developed statistically robust GA-MLR (genetic algorithm multiple linear regression), atom-based as well as field based-3D-QSAR models. Both atom-based as well as field based-3D-QSAR models (internally as well as externally validated) were obtained with robust Training set, R2 > 0.69 and Test set, Q2 > 0.50. We have also developed top ranked 5 point hypothesis AAAHR_1 among 14 CPHs (common pharmacophore hypotheses). We found that our dataset molecule had more docking score (XP mode = − 9.068 kcal/mol) than the standards isoniazid and ethambutol; when docked into binding pockets of enzyme 4P8C with Glide module. We further queried our best docked dataset molecule 151 for ligand based virtual screening using “SwissSimilarity” platform. Among 9 identified hits, we found ZINC12196803 had best binding energies and docking score (docking score = − 9.437 kcal/mol, MMGBSA dgBind = − 70.508 kcal/mol). Finally, our molecular dynamics studies for 1.2–100 ns depicts that these complexes are stable. We have also carried out in-silico ADMET predictions, Cardiac toxicity, ‘SwissTargetPredictions’ and Molecular Mechanics/Generalized Born Surface Area (MM/GBSA) binding energy calculations for further explorations of dataset as well as hit molecules. Our current studies showed that the hit molecule ZINC12196803 may enlighten the path for future developments of DprE1 inhibitors.

QSPR models to predict quantum chemical properties of imidazole derivatives using genetic algorithm–multiple linear regression and back‐propagation–artificial neural network

AbstractImidazole derivatives are the foundation of different types of drugs with a wide range of biological activities. In this study, the genetic algorithm–multiple linear regression (GA–MLR), and backpropagation–artificial neural network (BP–ANN) were applied to design QSPR models to predict the quantum chemical properties like the entropy (S) and enthalpy of formation (∆Hf) of imidazole derivatives. In order to draw molecular structure of 84 derivative compounds Gauss View 05 program was used. These structures were optimized at DFT‐B3LYP/6‐311G* level with Gaussian09W. The Dragon software was used to calculate a set of different molecular descriptors, and the genetic algorithm procedure and backward stepwise regression were applied for the selection of descriptors. The resulting quantitative GA–MLR model of ∆Hf, showed that there is good linear correlation between the selected descriptors and ∆Hf of compounds. Also the results show that the BP–ANN model appeared to be superior to GA–MLR model for prediction of entropy. Different internal and external validation metrics were adopted to verify the predictive performance of QSPR models. The predictive powers of the models were found to be acceptable. Thus, these QSPR models may be useful for designing new series of imidazole derivatives and prediction of their properties.

QSER modeling of half-wave oxidation potential of indolizines by theoretical descriptors

Indolizine derivatives hold essential biological functions and have been researched for hypoglycemic, antibacterial, anti-inflammatory, analgesic, and anti-tumor actions. Indolizine scaffold has intrigued conjecture and continuous attention and has become an effective parent system for generating powerful novel medication candidates. This research focused on applying the quantitative structure-electrochemistry relationship (QSER) approach to the half-wave potential (E1/2) for Indolizine derivatives using theoretical molecular descriptors. After calculating the descriptors and splitting the data into both sets, training and prediction. The QSER model was constructed using the Genetic Algorithm/Multiple Linear Regression (GA/MLR) technique, which was used to choose the optimal descriptors for the model. A four-parameter model has been established. Many assessment procedures, including cross-validation, external validation, and Y-scrambling testing, were used to assess the model's performance. Furthermore, the applicability domain (AD) was investigated using the Williams and Insubria graphs to assess the correctness of the established model's predictions. The constructed model exhibits great goodness-of-fit to experimental data, as well as high stability (R²=0.893, Q²LOO= 0.851, Q²LMO=0.843 RMSEtr= 0.052, s= 0.056). Prediction results show a good agreement with the experimental data of E1/2 (R²ext= 0.912, Q²F1= 0.883, Q²F2= 0.883, Q²F3= 0.919, CCCext= 0.942, RMSEext=0.045).

Monte Carlo Method and GA-MLR-Based QSAR Modeling of NS5A Inhibitors against the Hepatitis C Virus.

Hepatitis C virus (HCV) is a serious disease that threatens human health. Despite consistent efforts to inhibit the virus, it has infected more than 58 million people, with 300,000 deaths per year. The HCV nonstructural protein NS5A plays a critical role in the viral life cycle, as it is a major contributor to the viral replication and assembly processes. Therefore, its importance is evident in all currently approved HCV combination treatments. The present study identifies new potential compounds for possible medical use against HCV using the quantitative structure–activity relationship (QSAR). In this context, a set of 36 NS5A inhibitors was used to build QSAR models using genetic algorithm multiple linear regression (GA-MLR) and Monte Carlo optimization and were implemented in the software CORAL. The Monte Carlo method was used to build QSAR models using SMILES-based optimal descriptors. Four splits were performed and 24 QSAR models were developed and verified through internal and external validation. The model created for split 3 produced a higher value of the determination coefficients using the validation set (R2 = 0.991 and Q2 = 0.943). In addition, this model provides interesting information about the structural features responsible for the increase and decrease of inhibitory activity, which were used to develop eight novel NS5A inhibitors. The constructed GA-MLR model with satisfactory statistical parameters (R2 = 0.915 and Q2 = 0.941) confirmed the predicted inhibitory activity for these compounds. The Absorption, Distribution, Metabolism, Elimination, and Toxicity (ADMET) predictions showed that the newly designed compounds were nontoxic and exhibited acceptable pharmacological properties. These results could accelerate the process of discovering new drugs against HCV.

Use of biomimetic chromatography and in vitro assay to develop predictive GA-MLR model for use in drug-property prediction among anti-depressant drug candidates

In recent years mental disorders, such as depression, have become an alarmingly frequent problem in modern society. Therefore, a need for novel anti-depressant agents is ever-growing. With low drug acceptance rate, industry is struggling to introduce new anti-depressant agents, and every single tool helping to change this state of things is of immense value. In this study, a set of potential novel anti-depressant derivatives with confirmed pharmacological activity was subjected to a number of assays. A group of four drug properties, assessed in early stages of drug development were chosen for those assays. Lipophilicity, membrane permeability and drug-protein binding were estimated using biomimetic chromatography methods, while metabolic stability was estimated using in-vitro enzymatic assay followed by Liquid chromatography coupled with mass spectrometry (LC-MS) analysis. Afterwards, 3D models of studied compounds were geometrically optimized using Density Functional Theory (DFT) molecular mechanics protocol in order to resemble real-state energies and structure of those derivatives. After calculating molecular descriptors from optimized 3D models, a matrix combining data from assays and molecular descriptors was created. Using Genetic Algorithm Multiple Linear Regression (GA-MLR) on this matrix allowed to create four mathematical models, one for each of the studied properties, which accurately predicts properties crucial in early drug-development stages. Such models can severely help in reducing costs of drug-development, by allowing exclusion of extremely unfavorable derivatives – even before the stage of synthesis.

Improved structural annotation of triterpene metabolites of traditional Chinese medicine in vivo based on quantitative structure-retention relationships combined with characteristic ions: Alismatis Rhizoma as an example

As a fast, sensitive and selective method, liquid chromatography-tandem high-resolution mass spectrometry (LC-HRMS) has been used for studying the in vivo metabolism of traditional Chinese medicine (TCM). However, the rapid discovery and characterization of metabolites, especially isomers, remain challenging due to their complexity and low concentration in vivo. This study proposed a strategy to improve the structural annotation of prototypes and metabolites through characteristic ions and a quantitative structure-retention relationship (QSRR) model, and Alismatis Rhizoma (AR) triterpenes were used as an example. This strategy consists of four steps. First, based on an in-house database reported previously, prototypes and metabolites in biosamples were preliminarily identified. Second, the candidate structures of prototype compounds and metabolites were determined by characteristic ions, databases or potential metabolic pathways. Then, a QSRR model was established to predict the retention times of the proposed structure. Finally, the structures of unknown prototypes and metabolites were determined by matching experimental retention times with the predicted values. The QSRR model built by the genetic algorithm-multiple linear regression (GA-MLR) has excellent regression correlation (R2 = 0.9966). Based on this strategy, a total of 118 compounds were identified, including 47 prototypes and 71 metabolites, among which 61 unknown compounds were reasonably characterized. The typical compound identified by this strategy was successfully validated using a triterpene standard. This strategy can improve the annotation confidence of in vivo metabolites of TCM and facilitate further pharmacological research.

Comparison of quantum mechanics protocols during the evaluation of quantitative structure-retention relationships supported by genetic-algorithm multiple linear regression

While developing Quantitative Structure-Retention Relationships (QSRR) models it is crucial to construct a data matrix, consisting of both experimentally designated values (the ones that are being modelled) and molecular descriptors, representing physicochemical properties in form of numbers. In order to designate those descriptors, studied compounds need to be modelled and optimized using molecular modelling methods. From a variety of possible choices, semi-empirical AM1 (being an often discredited method due to the fact that it is quick and provides various approximations) and DFT (being a complex, quantum mechanics-based method, which is more time-consuming yet still bases on approximations) are one of the most popular. In this study one took a set of carefully chosen compounds used to model retention on C18 column, and modelled their molecules using both of those techniques, which were afterwards subjected to Genetic-Algorithm Multiple Linear Regression (GA-MLR) in order to derive and compare achieved models and their parameters. Study found that for the tested chromatographic system (C18 column and a set of model fifteen compounds) there is no advantage in using DFT over AM1, despite common modelling principles. What is more important, the developed models are accurate and use very simple descriptors, which can be easily calculated without any need for complex 3D modelling of structures. The derived models enabled also to assess the physicochemical properties of the tested column finding slight similarities and dissimilarities between the applied models.

COVID-19: In silico identification of potent α-ketoamide inhibitors targeting the main protease of the SARS-CoV-2.

The COVID-19 has been creating a global crisis, causing countless deaths and unbearable panic. Despite the progress made in the development of the vaccine, there is an urge need for the discovery of antivirals that may better work at different stages of SARS-CoV-2 reproduction. The main protease (Mpro) of the SARS-CoV-2 is a crucial therapeutic target due to its critical function in virus replication. The α-ketoamide derivatives represent an important class of inhibitors against the Mpro of the SARS-CoV. While there is 99% sequence similarity between SARS-CoV and SARS-CoV-2 main proteases, anti-SARS-CoV compounds may have a huge demonstration's prospect of their effectiveness against the SARS-CoV-2. In this study, we applied various computational approaches to investigate the inhibition potency of novel designed α-ketoamide-based compounds. In this regard, a set of 21 α-ketoamides was employed to construct a QSAR model, using the genetic algorithm-multiple linear regression (GA-MLR), as well as a pharmacophore fit model. Based on the GA-MLR model, 713 new designed molecules were reduced to 150 promising hits, which were later subject to the established pharmacophore fit model. Among the 150 compounds, the best selected compounds (3 hits) with greater pharmacophore fit score were further studied via molecular docking, molecular dynamic simulations along with the Absorption, distribution, metabolism, excretion, and toxicity (ADMET) analysis. Our approach revealed that the three hit compounds could serve as potential inhibitors against the SARS-CoV-2 Mpro target.

Application of Genetic Algorithm-Multiple Linear Regression and Artificial Neural Network Determinations for Prediction of Kovats Retention Index

This study aims to compare linear and nonlinear prediction methods in predicting the Kovats retention index of 126 compounds extracted from the Lippia origanoides plant. The retention index of each compound has been predicted based on its molecular descriptors. There have been 189 molecular descriptors for each compound in this study, and the best descriptors have been selected using the Genetic Algorithm (GA). It has succeeded in selecting the best five descriptors used to build the Multiple Linear Regression (MLR) and the Artificial Neural Network (ANN) model. MLR has obtained R2 of 0.959, 0.946, 0.955, and a Root Mean Square Error (RMSE) of 48.00, 50.84, 47.19 on training, validation, and testing, respectively. Meanwhile, ANN has obtained R2 of 0.963, 0.947, 0.962, and an RMSE of 45.45, 50.59, 43.20, respectively. Compared to MLR, there has been an increase in the ANN model's performance, with an increase in R2 of 0.004, 0.001, and 0.007 and a decrease in RMSE of 2.55, 0.25, and 3.99. Based on the prediction results obtained, it is known that in this case, the ANN method can provide better predictive results than MLR.

Quantitative structure-critical micelle concentration modeling of anionic gemini surfactants, comparison of MLR, PLS, WNN, and ANFIS models with eigenvalue and correlation ranking methods

Unique properties of gemini surfactants have attracted the attention of the cosmetics, paints, textile, and detergents industries. Therefore, design of new gemini surfactants is considered important. In this work, the critical micelle concentration of 85 anionic gemini surfactants was predicted by structural descriptors for the first time. Genetic algorithm-multiple linear regression (GA-MLR), genetic algorithm-partial least squares (GA-PLS), principal component-adaptive neuro-fuzzy inference systems (PC-ANFIS), and principal component-wavelet neural network (PC-WNN) were used as linear and nonlinear models. Between linear models, the PLS method had better root mean square errors (RMSE) and mean relative error (MRE) values than MLR model. The eigenvalue and correlation ranking methods were used to select the best PCs in nonlinear models such as PC-ANFIS and PC-WNN. The correlation ranking selection method has better results in both the ANFIS and WNN models. WNN has better results with respect to other models. The RMSE of the training, test, and validation sets for the best PC-WNN model with eight inputs were 0.250, 0.359, and 0.310, respectively.