Prediction of dihydrofolate reductase inhibition and selectivity using computational neural networks and linear discriminant analysis

Analysis of Alzheimer's Disease Based on the Random Neural Network Cluster in fMRI

As Alzheimer’s disease (AD) is featured with degeneration and irreversibility, the diagnosis of AD at early stage is important. In recent years, some researchers have tried to apply neural network (NN) to classify AD patients from healthy controls (HC) based on functional MRI (fMRI) data. But most study focus on a single NN and the classification accuracy was not high. Therefore, this paper used the random neural network cluster which was composed of multiple NNs to improve classification performance. Sixty one subjects (25 AD and 36 HC) were acquired from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. This method not only could be used in the classification, but also could be used for feature selection. Firstly, we chose Elman NN from five types of NNs as the optimal base classifier of random neural network cluster based on the results of feature selection, and the accuracies of the random Elman neural network cluster could reach to 92.31% which was the highest and stable. Then we used the random Elman neural network cluster to select significant features and these features could be used to find out the abnormal regions. Finally, we found out 23 abnormal regions such as the precentral gyrus, the frontal gyrus and supplementary motor area. These results fully show that the random neural network cluster is worthwhile and meaningful for the diagnosis of AD.

Exploring the QSAR's predictive truthfulness of the novel N-tuple discrete derivative indices on benchmark datasets

Graph derivative indices (GDIs) have recently been defined over N-atoms (N = 2, 3 and 4) simultaneously, which are based on the concept of derivatives in discrete mathematics (finite difference), metaphorical to the derivative concept in classical mathematical analysis. These molecular descriptors (MDs) codify topo-chemical and topo-structural information based on the concept of the derivative of a molecular graph with respect to a given event (S) over duplex, triplex and quadruplex relations of atoms (vertices). These GDIs have been successfully applied in the description of physicochemical properties like reactivity, solubility and chemical shift, among others, and in several comparative quantitative structure activity/property relationship (QSAR/QSPR) studies. Although satisfactory results have been obtained in previous modelling studies with the aforementioned indices, it is necessary to develop new, more rigorous analysis to assess the true predictive performance of the novel structure codification. So, in the present paper, an assessment and statistical validation of the performance of these novel approaches in QSAR studies are executed, as well as a comparison with those of other QSAR procedures reported in the literature. To achieve the main aim of this research, QSARs were developed on eight chemical datasets widely used as benchmarks in the evaluation/validation of several QSAR methods and/or many different MDs (fundamentally 3D MDs). Three to seven variable QSAR models were built for each chemical dataset, according to the original dissection into training/test sets. The models were developed by using multiple linear regression (MLR) coupled with a genetic algorithm as the feature wrapper selection technique in the MobyDigs software. Each family of GDIs (for duplex, triplex and quadruplex) behaves similarly in all modelling, although there were some exceptions. However, when all families were used in combination, the results achieved were quantitatively higher than those reported by other authors in similar experiments. Comparisons with respect to external correlation coefficients (q²_ext) revealed that the models based on GDIs possess superior predictive ability in seven of the eight datasets analysed, outperforming methodologies based on similar or more complex techniques and confirming the good predictive power of the obtained models. For the q²_ext values, the non-parametric comparison revealed significantly different results to those reported so far, which demonstrated that the models based on DIVATI's indices presented the best global performance and yielded significantly better predictions than the 12 0-3D QSAR procedures used in the comparison. Therefore, GDIs are suitable for structure codification of the molecules and constitute a good alternative to build QSARs for the prediction of physicochemical, biological and environmental endpoints.

A chemoinformatics approach to the discovery of lead-like molecules from marine and microbial sources en route to antitumor and antibiotic drugs

The comprehensive information of small molecules and their biological activities in the PubChem database allows chemoinformatic researchers to access and make use of large-scale biological activity data to improve the precision of drug profiling. A Quantitative Structure-Activity Relationship approach, for classification, was used for the prediction of active/inactive compounds relatively to overall biological activity, antitumor and antibiotic activities using a data set of 1804 compounds from PubChem. Using the best classification models for antibiotic and antitumor activities a data set of marine and microbial natural products from the AntiMarin database were screened-57 and 16 new lead compounds for antibiotic and antitumor drug design were proposed, respectively. All compounds proposed by our approach are classified as non-antibiotic and non-antitumor compounds in the AntiMarin database. Recently several of the lead-like compounds proposed by us were reported as being active in the literature.

QSAR studies for the computational prediction of HMG-CoA reductase inhibitors by genetic function approximation technique

Two-dimensional quantitative structure−activity relationship (2D-QSAR) models are useful in understanding how chemical structure is related to the biological activity of natural and synthetic chemicals. Also, they could be usefully employed for designing newer and better therapeutics. A 2D-QSAR study was performed for 52 compounds of a series of thiophenyl quinolines and α-asarone derivatives as potential hypocholesterolemic inhibitors using different types of physicochemical descriptors, which correlated significantly with the activity. Linear QSAR models were developed using multiple linear regression, where the genetic algorithm (genetic function approximation technique) was adopted for selecting the most appropriate descriptors. The results are discussed on the basis of regression data and the cross-validation technique. Model A is the best 2D-QSAR model describing the inhibition efficiency of HMG-CoA reductase with cross-validated squared correlation coefficient (Q 2 = 0.700) and the squared correlation coefficient (R 2 = 0.752), which is able to describe 70% of the variance in the experimental activity. The good agreement between the experimental and the predicted values of pIC50 (micromoles per litre) (R = 0.876) confirms the reliability and the predictability of the proposed model. The results obtained from the present QSAR study explained the importance of the electronic, structural, spatial, and electrotopological descriptors in enhancing the biological activity of the investigated inhibitors.

A comparison of different electrostatic potentials on prediction accuracy in CoMFA and CoMSIA studies

Computational chemistry is playing an increasingly important role in drug design and discovery, structural biology, and quantitative structure-activity relationship (QSAR) studies. For QSAR work, selecting an appropriate and accurate method to assign the electrostatic potentials of each atom in a molecule is a critical first step. So far several commonly used methods are available to assign charges. However, no systematic comparison of the effects of electrostatic potentials on QSAR quality has been made. In this study, twelve semi-empirical and empirical charge-assigning methods, AM1, AM1-BCC, CFF, Del-Re, Formal, Gasteiger, Gasteiger-Hückel, Hückel, MMFF, PRODRG, Pullman, and VC2003 charges, have been compared for their performances in CoMFA and CoMSIA modeling using several standard datasets. Some charge assignment models, such as Del-Re, PRODRG, and Pullman, are limited to specific atom and bond types, and, therefore, were excluded from this study. Among the remaining nine methods, the Gasteiger-Hückel charge, though commonly used, performed poorly in prediction accuracy. The AM1-BCC method was better than most charge-assigning methods based on prediction accuracy, though it was not successful in yielding overall higher cross-validation correlation coefficient (q(2)) values than others. The CFF charge model worked the best in prediction accuracy when q(2) was used as the evaluation criterion. The results presented should help the selection of electrostatic potential models in CoMFA and CoMSIA studies.

CoMFA analysis of tgDHFR and rlDHFR based on antifolates with 6-5 fused ring system using the all-orientation search (AOS) routine and a modified cross-validated r(2)-guided region selection (q(2)-GRS) routine and its initial application

We report the development of CoMFA analysis models that correlate the 3D chemical structures of 80 compounds with 6-5 fused ring system synthesized in our laboratory and their inhibitory potencies against tgDHFR and rlDHFR. In addition to conventional CoMFA analysis, we used two routines available in the literature aimed at the optimization of CoMFA: all-orientation search (AOS) and cross-validated r(2)-guided region selection (q(2)-GRS) to further optimize the models. During this process, we identified a problem associated with q(2)-GRS routine and modified using two strategies. Thus, for the inhibitory activity against each enzyme (tgDHFR and rlDHFR), five CoMFA models were developed using the conventional CoMFA, AOS optimized CoMFA, the original q(2)-GRS optimized CoMFA and the modified q(2)-GRS optimized CoMFA using the first and the second strategy. In this study, we demonstrate that the modified q(2)-GRS routines are superior to the original routine. On the basis of the steric contour maps of the models, we designed four new compounds in the 2,4-diamino-5-methyl-6-phenylsulfanyl-substituted pyrrolo[2,3-d]pyrimidine series. As predicted, the new compounds were potent and selective inhibitors of tgDHFR. One of them, 2,4-diamino-5-methyl-6-(2',6'-dimethylphenylthio)pyrrolo[2,3-d]pyrimidine, is the first 6-5 fused ring system compound with nanomolar tgDHFR inhibitory activity. The HCl salt of this compound was also prepared to increase solubility. Both forms of the drug were tested in vivo in a Toxoplasma gondii infection mouse model. The results indicate that both forms were active with the HCl salt significantly more potent than the free base.

Unified QSAR approach to antimicrobials. Part 3: first multi-tasking QSAR model for input-coded prediction, structural back-projection, and complex networks clustering of antiprotozoal compounds

Several pathogen parasite species show different susceptibilities to different antiparasite drugs. Unfortunately, almost all structure-based methods are one-task or one-target Quantitative Structure-Activity Relationships (ot-QSAR) that predict the biological activity of drugs against only one parasite species. Consequently, multi-tasking learning to predict drugs activity against different species by a single model (mt-QSAR) is vitally important. In the two previous works of the present series we reported two single mt-QSAR models in order to predict the antimicrobial activity against different fungal (Bioorg. Med. Chem.2006, 14, 5973-5980) or bacterial species (Bioorg. Med. Chem.2007, 15, 897-902). These mt-QSARs offer a good opportunity (unpractical with ot-QSAR) to construct drug-drug similarity Complex Networks and to map the contribution of sub-structures to function for multiple species. These possibilities were unattended in our previous works. In the present work, we continue this series toward other important direction of chemotherapy (antiparasite drugs) with the development of an mt-QSAR for more than 500 drugs tested in the literature against different parasites. The data were processed by Linear Discriminant Analysis (LDA) classifying drugs as active or non-active against the different tested parasite species. The model correctly classifies 212 out of 244 (87.0%) cases in training series and 207 out of 243 compounds (85.4%) in external validation series. In order to illustrate the performance of the QSAR for the selection of active drugs we carried out an additional virtual screening of antiparasite compounds not used in training or predicting series; the model recognized 97 out of 114 (85.1%) of them. We also give the procedures to construct back-projection maps and to calculate sub-structures contribution to the biological activity. Finally, we used the outputs of the QSAR to construct, by the first time, a multi-species Complex Networks of antiparasite drugs. The network predicted has 380 nodes (compounds), 634 edges (pairs of compounds with similar activity). This network allows us to cluster different compounds and identify on average three known compounds similar to a new query compound according to their profile of biological activity. This is the first attempt to calculate probabilities of antiparasitic action of drugs against different parasites.

QSAR modeling of matrix metalloproteinase inhibition by N-hydroxy-alpha-phenylsulfonylacetamide derivatives

The main molecular features which determine the selectivity of a set of 80 N-hydroxy-alpha-phenylsulfonylacetamide derivatives (HPSAs) in the inhibition of three matrix metalloproteinases (MMP-1, MMP-9, and MMP-13) have been identified by using linear and nonlinear predictive models. The molecular information has been encoded in 2D autocorrelation descriptors, obtained from different weighting schemes. The linear models were built by multiple linear regression (MLR) combined with genetic algorithm (GA), and a robust QSAR mapping paradigm. The Bayesian-regularized genetic neural network (BRGNN) was employed for nonlinear modeling. In such approaches each model could have its own set of input variables. All models were predictive according to internal and external validation experiments; but the best results correspond to nonlinear ones. The 2D autocorrelation space brings different descriptors for each MMP inhibition, and suggests the atomic properties relevant for the inhibitors to interact with each MMP active site. On the basis of the current results, the reported models have the potential to discover new potent and selective inhibitors and bring useful molecular information about the ligand specificity for MMP S(1)(') and S(2)(') subsites.

Systematic Statistical Comparison of Comparative Molecular Similarity Indices Analysis Molecular Fields for Computer-Aided Lead Optimization

Comparative molecular similarity indices analysis (CoMSIA) is a 3D quantitative structure-activity relationship technique used to determine structural and electronic features influencing biological activity. This proves particularly useful for facilitating lead optimization projects. This study aimed to compare CoMSIA models produced using different subsets of the CoMSIA molecular fields (steric, electrostatic, hydrophobic, hydrogen-bond donor, and hydrogen-bond acceptor) in a systematic and statistically valid manner. A total of 23 data sets sourced from the literature were used to compare molecular field contribution and model predictivity using leave-one-out cross-validated R2 values. Predictive ability varied in a highly statistically significant manner depending on the set of CoMSIA molecular fields used. In general, the greater the number of CoMSIA molecular fields included in the analysis, the better the model predictivity was. There is great redundancy in the information contained in the different CoMSIA molecular fields. When all five CoMSIA molecular fields are included, the hydrophobic and electrostatic fields had the largest and the steric field the smallest contribution. Data sets were clustered into four groups on the basis of the utility of molecular field sets to generate predictive models.

Linear and nonlinear QSAR study of N-hydroxy-2-[(phenylsulfonyl)amino]acetamide derivatives as matrix metalloproteinase inhibitors

The inhibitory activity (IC50) toward matrix metalloproteinases (MMP-1, MMP-2, MMP-3, MMP-9, and MMP-13) of N-hydroxy-2-[(phenylsulfonyl)amino]acetamide derivatives (HPSAAs) has been successfully modeled using 2D autocorrelation descriptors. The relevant molecular descriptors were selected by linear and nonlinear genetic algorithm (GA) feature selection using multiple linear regression (MLR) and Bayesian-regularized neural network (BRANN) approaches, respectively. The quality of the models was evaluated by means of cross-validation experiments and the best results correspond to nonlinear ones (Q2>0.7 for all models). Despite the high correlation between the studied compound IC50 values, the 2D autocorrelation space brings different descriptors for each MMP inhibition. On the basis of these results, these models contain useful molecular information about the ligand specificity for MMP S'1, S1, and S'2 pockets.

Generation of QSAR sets with a self-organizing map

A Kohonen self-organizing map (SOM) is used to classify a data set consisting of dihydrofolate reductase inhibitors with the help of an external set of Dragon descriptors. The resultant classification is used to generate training, cross-validation (CV) and prediction sets for QSAR modeling using the ADAPT methodology. The results are compared to those of QSAR models generated using sets created by activity binning and a sphere exclusion method. The results indicate that the SOM is able to generate QSAR sets that are representative of the composition of the overall data set in terms of similarity. The resulting QSAR models are half the size of those published and have comparable RMS errors. Furthermore, the RMS errors of the QSAR sets are consistent, indicating good predictive capabilities as well as generalizability.

CoMFA and CoMSIA Analyses of Pneumocystis carinii Dihydrofolate Reductase, Toxoplasma gondii Dihydrofolate Reductase, and Rat Liver Dihydrofolate Reductase

In a continuing effort to develop potent and selective dihydrofolate reductase (DHFR) inhibitors against opportunistic pathogens, we developed three-dimensional quantitative structure-activity relationship (3D QSAR) models for the inhibitory activity against Pneumocystis carinii (pc) DHFR, Toxoplasma gondii (tg) DHFR, and rat liver DHFR, using a data set of 179 structurally diverse compounds. To ensure a balanced distribution of more potent and less potent drugs in the training set, three different 90-compound training sets taken from the main data set were used, one for each enzyme, while the remaining 89 compounds in the main data set in each case were used as the test set. Three methods, namely, conventional CoMFA, all orientation search (AOS) CoMFA, and CoMSIA were applied to the training sets. While the AOS CoMFA models gave the best internal predictions (cross-validated r(2) values from the training sets), which are satisfactory, CoMSIA models gave the best external predictions (predictive r(2) values from the test sets). Both AOS CoMFA and CoMSIA analyses were used to construct stdev*coefficient contour maps which can be used to design new compounds in an interactive fashion.

A comparison of methods for modeling quantitative structure-activity relationships

A large number of methods are available for modeling quantitative structure-activity relationships (QSAR). We examine the predictive accuracy of several methods applied to data sets of inhibitors for angiotensin converting enzyme, acetylcholinesterase, benzodiazepine receptor, cyclooxygenase-2, dihydrofolate reductase, glycogen phosphorylase b, thermolysin, and thrombin. Descriptors calculated with CoMFA, CoMSIA, EVA, HQSAR, and traditional 2D and 2.5D descriptors were used for developing models with partial least squares (PLS). In addition, the genetic function approximation algorithm, genetic PLS, and back-propagation neural networks were used for deriving models from 2.5D descriptors (i.e., 2D descriptors and 3D descriptors calculated from CORINA structures and Gasteiger-Marsili charges). Predictive accuracy was assessed using designed test sets. It was found that HQSAR generally performs as well as CoMFA and CoMSIA; other descriptor sets performed less well. When 2.5D descriptors were used, only neural network ensembles were found to be similarly or more predictive than PLS models. In addition, we show that many cross-validation procedures yield similar estimates of the interpolative accuracy of methods. However, the lack of correspondence between cross-validated and test set predictive accuracy for four sets underscores the benefit of using designed test sets.

Three-Dimensional Quantitative Structure−Activity Relationship Analysis of a Set ofPlasmodium falciparumDihydrofolate Reductase Inhibitors Using a Pharmacophore Generation Approach

A 3D pharmacophore model able to quantitatively predict inhibition constants was derived for a series of inhibitors of Plasmodium falciparum dihydrofolate reductase (PfDHFR), a validated target for antimalarial therapy. The data set included 52 inhibitors, with 23 of these comprising the training set and 29 an external test set. The activity range, expressed as Ki, of the training set molecules was from 0.3 to 11 300 nM. The 3D pharmacophore, generated with the HypoGen module of Catalyst 4.7, consisted of two hydrogen bond donors, one positive ionizable feature, one hydrophobic aliphatic feature, and one hydrophobic aromatic feature and provided a 3D-QSAR model with a correlation coefficient of 0.954. Importantly, the type and spatial location of the chemical features encoded in the pharmacophore were in full agreement with the key binding interactions of PfDHFR inhibitors as previously established by molecular modeling and crystallography of enzyme-inhibitor complexes. The model was validated using several techniques, namely, Fisher's randomization test using CatScramble, leave-one-out test to ensure that the QSAR model is not strictly dependent on one particular compound of the training set, and activity prediction in an external test set of compounds. In addition, the pharmacophore was able to correctly classify as active and inactive the dihydrofolate reductase and aldose reductase inhibitors extracted from the MDDR database, respectively. This test was performed in order to challenge the predictive ability of the pharmacophore with two classes of inhibitors that target very different binding sites. Molecular diversity of the data sets was finally estimated by means of the Tanimoto approach. The results obtained provide confidence for the utility of the pharmacophore in the virtual screening of libraries and databases of compounds to discover novel PfDHFR inhibitors.

QSAR modeling based on the bias/variance compromise: a harmonious

Modeling quantitative structure-activity relationships (QSAR) is considered with an emphasis on prediction. An abundance of methods are available to develop such models. Using a harmonious approach that balances the bias and variance of predictions, the best calibration models are identified relative to the bias and variance criteria used. Criteria utilized to determine the adequacy of models are the root mean square error of calibration (RMSEC) and validation (RMSEV), respective R2 values, and the norm of the regression vector. QSAR data from the literature are used to demonstrate concepts. For these data sets and criteria used, it is suggested that models obtained by ridge regression (RR) are more harmonious and parsimonious than models obtained by partial least squares (PLS) and principal component regression (PCR) when the data is mean-centered. The most harmonious RR models have the best bias/variance tradeoff, reflected by the smallest RMSEC, RMSEV, and regression vector norms and the largest calibration and validation R2 values. The most parsimonious RR models have the smallest effective rank.

Development of QSAR Models To Predict and Interpret the Biological Activity of Artemisinin Analogues

This work presents the development of Quantitative Structure-Activity Relationship (QSAR) models to predict the biological activity of 179 artemisinin analogues. The structures of the molecules are represented by chemical descriptors that encode topological, geometric, and electronic structure features. Both linear (multiple linear regression) and nonlinear (computational neural network) models are developed to link the structures to their reported biological activity. The best linear model was subjected to a PLS analysis to provide model interpretability. While the best linear model does not perform as well as the nonlinear model in terms of predictive ability, the application of PLS analysis allows for a sound physical interpretation of the structure-activity trend captured by the model. On the other hand, the best nonlinear model is superior in terms of pure predictive ability, having a training error of 0.47 log RA units (R2 = 0.96) and a prediction error of 0.76 log RA units (R2 = 0.88).

Classification of Diverse Organic Compounds That Induce Chromosomal Aberrations in Chinese Hamster Cells

A data set of 297 diverse organic compounds that cause varying degrees of chromosomal aberrations in Chinese hamster lung cells is examined. Responses of an assay are categorized as clastogenic (>10% aberrant cells) and nonclastogenic (<5% aberrant cells). Each of the compounds is represented by calculated structural descriptors that encode topological, geometric, electronic, and polar surface features. A genetic algorithm (GA) employing a k-nearest neighbor (kNN) fitness evaluator is used to iteratively search a reduced descriptor space to find small, information-rich subsets of descriptors that maximize the classification rates for clastogenic and nonclastogenic responses. To further improve modeling, a similarity measure using atom-pair descriptors is employed to create more homogeneous data subsets. Three different data sets are examined. Results for a set of 297 compounds using the GA-kNN method were 86.5% and 80.0% correct classification in the training set and prediction set, respectively. Results for a subset of 279 compounds in model 2 are 85.7% and 85.7% for the training and prediction sets, respectively. Results for a subset of 182 compounds in model 3 are 91.5% and 94.4% for the training and prediction sets, respectively. Creating smaller, more topologically similar data sets result in improved classification rates.