Mohammad Goodarzi

Feature Selection and Linear/Nonlinear Regression Methods for the Accurate Prediction of Glycogen Synthase Kinase-3β Inhibitory Activities

Few variables were selected from a pool of calculated Dragon descriptors through three different feature selection methods, namely genetic algorithm (GA), successive projections algorithm (SPA), and fuzzy rough set ant colony optimization (fuzzy rough set ACO). Each set of selected descriptors was regressed against the bioactivities of a series of glycogen synthase kinase-3beta (GSK-3beta) inhibitors, through linear and nonlinear regression methods, namely multiple linear regression (MLR), artificial neural network (ANN), and support vector machines (SVM). The fuzzy rough set ACO/SVM-based model gave the best estimation/prediction results, demonstrating the nonlinear nature of this analysis and suggesting fuzzy rough set ACO, first introduced in chemistry here, as an improved variable selection method in QSAR for the class of GSK-3beta inhibitors.

New Hybrid Genetic Based Support Vector Regression as QSAR Approach for Analyzing Flavonoids-GABA(A) Complexes

Several studies were conducted in past years which used the evolutionary process of Genetic Algorithms for optimizing the Support Vector Regression parameter values although, however, few of them were devoted to the simultaneously optimization of the type of kernel function involved in the established model. The present work introduces a new hybrid genetic-based Support Vector Regression approach, whose statistical quality and predictive capability is afterward analyzed and compared to other standard chemometric techniques, such as Partial Least Squares, Back-Propagation Artificial Neural Networks, and Support Vector Machines based on Cross-Validation. For this purpose, we employ a data set of experimentally determined binding affinity constants toward the benzodiazepine binding site of the GABA (A) receptor complex on 78 flavonoid ligands.

QSRR modeling for diverse drugs using different feature selection methods coupled with linear and nonlinear regressions

A Quantitative Structure-Retention Relationship (QSRR) is proposed to estimate the chromatographic retention of 83 diverse drugs on a Unisphere poly butadiene (PBD) column, using isocratic elutions at pH 11.7. Previous work has generated QSRR models for them using Classification And Regression Trees (CART). In this work, Ant Colony Optimization is used as a feature selection method to find the best molecular descriptors from a large pool. In addition, several other selection methods have been applied, such as Genetic Algorithms, Stepwise Regression and the Relief method, not only to evaluate Ant Colony Optimization as a feature selection method but also to investigate its ability to find the important descriptors in QSRR. Multiple Linear Regression (MLR) and Support Vector Machines (SVMs) were applied as linear and nonlinear regression methods, respectively, giving excellent correlation between the experimental, i.e. extrapolated to a mobile phase consisting of pure water, and predicted logarithms of the retention factors of the drugs (logk(w)). The overall best model was the SVM one built using descriptors selected by ACO.

Similarity analyses of chromatographic fingerprints as tools for identification and quality control of green tea.

Similarity assessment of complex chromatographic profiles of herbal medicinal products is important as a potential tool for their identification. Mathematical similarity parameters have the advantage to be more reliable than visual similarity evaluations of often subtle differences between the fingerprint profiles. In this paper, different similarity analysis (SA) parameters are applied on green-tea chromatographic fingerprint profiles in order to test their ability to identify (dis)similar tea samples. These parameters are either based on correlation or distance measurements. They are visualised in colour maps and evaluation plots. Correlation (r) and congruence (c) coefficients are shown to provide the same information about the similarity of samples. The standardised Euclidean distance (ds) reveals less information than the Euclidean distance (de), while Mahalanobis distances (dm) are unsuitable for the similarity assessment of chromatographic fingerprints. The adapted similarity score (ss*) combines the advantages of r (or c) and de. Similarity analysis based on correlation is useful if concentration differences between samples are not important, whereas SA based on distances also detects concentration differences well. The evaluation plots including statistical confidence limits for the plotted parameter are found suitable for the evaluation of new suspected samples during quality assurance. The ss* colour maps and evaluation plots are found to be the best tools (in comparison to the other studied parameters) for the distinction between deviating and genuine fingerprints. For all studied data sets it is confirmed that adequate data pre-treatment, such as aligning the chromatograms, prior to the similarity assessment, is essential. Furthermore, green-tea samples chromatographed on two dissimilar High-Performance Liquid Chromatography (HPLC) columns provided the same similarity assessment. Combining these complementary fingerprints did not improve the similarity analysis of the studied data set.

Near-infrared spectroscopy for in-line monitoring of protein unfolding and its interactions with lyoprotectants during freeze-drying.

This work presents near-infrared spectroscopy (NIRS) as an in-line process analyzer for monitoring protein unfolding and protein-lyoprotectant hydrogen bond interactions during freeze-drying. By implementing a noncontact NIR probe in the freeze-drying chamber, spectra of formulations containing a model protein immunoglobulin G (IgG) were collected each process minute. When sublimation was completed in the cake region illuminated by the NIR probe, the frequency of the amide A/II band (near 4850 cm(-1)) was monitored as a function of water elimination. These two features were well correlated during protein dehydration in the absence of protein unfolding (desired process course), whereas consistent deviations from this trend to higher amide A/II frequencies were shown to be related to protein unfolding. In formulations with increased sucrose concentrations, the markedly decreased amide A/II frequencies seen immediately after sublimation indicated an increased extent of hydrogen bond interaction between the proteins backbone and surrounding molecules. At the end of drying, there was evidence of nearly complete water substitution for formulations with 1%, 5%, and 10% sucrose. The presented approach shows promising perspectives for early fault detection of protein unfolding and for obtaining mechanistic process information on actions of lyoprotectants.

Predicting Boiling Points of Aliphatic Alcohols through Multivariate Image Analysis Applied to Quantitative Structure-Property Relationships

The boiling points of a set of 58 aliphatic alcohols have been modeled through an image-based approach, in which descriptors are pixels (binaries) of 2D chemical structures. While some simple descriptors, such as molecular weight, do not account for some structural influences (e.g., in chain and position isomerism) on the studied property, the MIA-QSPR (multivariate image analysis applied to quantitative structure-property relationship) method, coupled to multilinear partial least-squares regression, correlated the chemical structures with the corresponding boiling points satisfactorily well.

Multimode Methods Applied on MIA Descriptors in QSAR

Since the introduction of physicochemical descriptors to derive useful QSAR (quantitative structure-activity relationship) models, some regression methods have been applied to linearly correlate dependent (bioactivities) and independent variables. Multiple linear regression (MLR) has been widely used when the number of samples (rows) exceed the amount of descriptors (columns), whilst partial least squares (PLS) is the most commonly applied regression method in 3D QSAR (e.g. CoMFA and related methods), where a large number of descriptors are generated. The recently implemented MIA-QSAR (Multivariate Image Analysis applied to QSAR) method is a especial (not only) case in which the descriptors (pixels) for each active compound result in a three-way array after grouping samples to give a data set. Such array may be properly treated by using N-way methods, such as multilinear PLS (N-PLS) and parallel factor analysis (PARAFAC). However, these methods have not been appropriately explored in QSAR studies, despite their supposed advantages over well established methods. Thus, this review formally details the MIA-QSAR approach prior to presenting two promising multimode methods to be applied on MIA descriptors, namely N-PLS and PARAFAC. Also, the suitability of such methods is discussed in terms of application to a case study (a series of anti-HIV compounds) and comparison to traditional (bilinear) PLS and docking studies.

Classification models for neocryptolepine derivatives as inhibitors of the β-haematin formation

This paper describes the construction of a QSAR model to relate the structures of various derivatives of neocryptolepine to their anti-malarial activities. QSAR classification models were build using Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), Classification and Regression Trees (CART), Partial Least Squares-Discriminant Analysis (PLS-DA), Orthogonal Projections to Latent Structures-Discriminant Analysis (OPLS-DA), and Support Vector Machines for Classification (SVM-C), using four sets of molecular descriptors as explanatory variables. Prior to classification, the molecules were divided into a training and a test set using the duplex algorithm. The different classification models were compared regarding their predictive ability, simplicity, and interpretability. Both binary and multi-class classification models were constructed. For classification into three classes, CART and One-Against-One (OAO)-SVM-C were found to be the best predictive methods, while for classification into two classes, LDA, QDA and CART were.

Linear and nonlinear quantitative structure-activity relationship modeling of the HIV-1 reverse transcriptase inhibiting activities of thiocarbamates.

For a series of thiocarbamates, non-nucleoside HIV-1 reverse transcriptase inhibitors, few descriptors have been selected from a large pool of theoretical molecular descriptors by means of the ant colony optimization (ACO) feature selection method. The selected descriptors were correlated with the bioactivities of the molecules using the well known multiple linear regression (MLR) and partial least squares (PLS) regression techniques, and, to account for nonlinearity, also PLS coupled to radial basis function (RBF) on the one hand and radial basis function neural network (RBFNN) on the other. In this case study, the RBF/PLS results were better than those from the other modeling techniques applied. The prediction ability of the ACO/RBF/PLS-based quantitative structure-activity relationship (QSAR) model was found to be significantly superior to comparative molecular field analysis (CoMFA) and comparative molecular similarity index analysis (CoMSIA) models previously established for this series of compounds. It was also demonstrated that RBF as a nonlinear approach is useful in deriving simple and predictive QSAR models, without the need to recourse to expeditious 3D methodologies.

Potentially antioxidant compounds indicated from Mallotus and Phyllanthus species fingerprints

The genera of Mallotus and Phyllanthus contain several species that are commonly used as traditional medicines in oriental countries. Some species show interesting pharmaceutical activities, such as an antioxidant activity. To produce clinically useful medicines or food supplements (nutraceuticals) from these herbs, the species should be identified and a thorough quality control should be implemented. Nowadays, the integration of chromatographic and chemometric approaches allows a high-throughput identification and activity prediction of medicinal plants. In this study, Principal Component Analysis (PCA) and Hierarchical Cluster Analysis (HCA) were applied and compared to distinguish Mallotus and Phyllanthus species. Moreover, peaks from their chromatographic fingerprints, which were responsible for their antioxidant activity were assigned. For the latter purpose, the relevant information was extracted from the chromatographic fingerprints using linear multivariate calibration techniques, i.e., Partial Least Squares (PLS) and Orthogonal Projections to Latent Structures (O-PLS). Results reveal that exploratory analysis using PCA shows somewhat diverging clustering tendencies between Mallotus and Phyllanthus samples than HCA. However, both approaches mainly confirm each other. Concerning the multivariate calibration techniques, both PLS and O-PLS models demonstrate good predictive abilities. By comparing the regression coefficients of the models with the chromatographic fingerprints, the peaks that are potentially responsible for the antioxidant activity of the extracts could be confirmed.