Michael Fernández

Genetic algorithm optimization in drug design QSAR: Bayesian-regularized genetic neural networks (BRGNN) and genetic algorithm-optimized support vectors machines (GA-SVM)

Many articles in “in silico” drug design implemented genetic algorithm (GA) for feature selection, model optimization, conformational search, or docking studies. Some of these articles described GA applications to quantitative structure–activity relationships (QSAR) modeling in combination with regression and/or classification techniques. We reviewed the implementation of GA in drug design QSAR and specifically its performance in the optimization of robust mathematical models such as Bayesian-regularized artificial neural networks (BRANNs) and support vector machines (SVMs) on different drug design problems. Modeled data sets encompassed ADMET and solubility properties, cancer target inhibitors, acetylcholinesterase inhibitors, HIV-1 protease inhibitors, ion-channel and calcium entry blockers, and antiprotozoan compounds as well as protein classes, functional, and conformational stability data. The GA-optimized predictors were often more accurate and robust than previous published models on the same data sets and explained more than 65% of data variances in validation experiments. In addition, feature selection over large pools of molecular descriptors provided insights into the structural and atomic properties ruling ligand–target interactions.

Modeling of cyclin-dependent kinase inhibition by 1H-pyrazolo[3,4-d]pyrimidine derivatives using artificial neural network ensembles.

Artificial neural network ensembles were used for modeling the cyclin-dependent kinase inhibition of 1H-pyrazolo[3,4-d]pyrimidine derivatives. The structural characteristics of these inhibitors were encoded in relevant 3D-spatial descriptors extracted by genetic algorithm feature selection. Bayesian-regularized multilayer neural networks, trained by the back-propagation algorithm, were developed using these variables as inputs. The predictive power of the model was tested by leave-one-out cross validation. In addition, for a more rigorous measure of the predictive capacity, multiple validation sets were randomly generated as members of neural network ensembles, which makes doing averaged predictions feasible. In this way, the predictive power was analyzed accounting for the averaged test set R values and test set mean-square errors. Otherwise, Kohonen self-organizing maps were used as an additional tool for the same modeling. The location of the inhibitors in a map facilitates the analysis of the connection between compounds and serves as a useful tool for qualitative predictions.

Genome-wide enhancer prediction from epigenetic signatures using genetic algorithm-optimized support vector machines

The chemical modification of histones at specific DNA regulatory elements is linked to the activation, inactivation and poising of genes. A number of tools exist to predict enhancers from chromatin modification maps, but their practical application is limited because they either (i) consider a smaller number of marks than those necessary to define the various enhancer classes or (ii) work with an excessive number of marks, which is experimentally unviable. We have developed a method for chromatin state detection using support vector machines in combination with genetic algorithm optimization, called ChromaGenSVM. ChromaGenSVM selects optimum combinations of specific histone epigenetic marks to predict enhancers. In an independent test, ChromaGenSVM recovered 88% of the experimentally supported enhancers in the pilot ENCODE region of interferon gamma-treated HeLa cells. Furthermore, ChromaGenSVM successfully combined the profiles of only five distinct methylation and acetylation marks from ChIP-seq libraries done in human CD4+ T cells to predict ∼21 000 experimentally supported enhancers within 1.0 kb regions and with a precision of ∼90%, thereby improving previous predictions on the same dataset by 21%. The combined results indicate that ChromaGenSVM comfortably outperforms previously published methods and that enhancers are best predicted by specific combinations of histone methylation and acetylation marks.

Amino Acid Sequence Autocorrelation vectors and ensembles of Bayesian-Regularized Genetic Neural Networks for prediction of conformational stability of human lysozyme mutants.

Development of novel computational approaches for modeling protein properties from their primary structure is a main goal in applied proteomics. In this work, we reported the extension of the autocorrelation vector formalism to amino acid sequences for encoding protein structural information with modeling purposes. Amino Acid Sequence Autocorrelation (AASA) vectors were calculated by measuring the autocorrelations at sequence lags ranging from 1 to 15 on the protein primary structure of 48 amino acid/residue properties selected from the AAindex database. A total of 720 AASA descriptors were tested for building predictive models of the thermal unfolding Gibbs free energy change of human lysozyme mutants. In this sense, ensembles of Bayesian-Regularized Genetic Neural Networks (BRGNNs) were used for obtaining an optimum nonlinear model for the conformational stability. The ensemble predictor described about 88% and 68% variance of the data in training and test sets, respectively. Furthermore, the optimum AASA vector subset was shown not only to successfully model unfolding thermal stability but also to distribute wild-type and mutant lysozymes on a stability Self-organized Map (SOM) when used for unsupervised training of competitive neurons.

Amino acid sequence autocorrelation vectors and bayesian‐regularized genetic neural networks for modeling protein conformational stability: Gene V protein mutants

Development of novel computational approaches for modeling protein properties from their primary structure is the main goal in applied proteomics. In this work, we reported the extension of the autocorrelation vector formalism to amino acid sequences for encoding protein structural information with modeling purposes. Amino acid sequence autocorrelation (AASA) vectors were calculated by measuring the autocorrelations at sequence lags ranging from 1 to 15 on the protein primary structure of 48 amino acid/residue properties selected from the AAindex data base. A total of 720 AASA descriptors were tested for building predictive models of the change of thermal unfolding Gibbs free energy change (ΔΔG) of gene V protein upon mutation. In this sense, ensembles of Bayesian‐regularized genetic neural networks (BRGNNs) were used for obtaining an optimum nonlinear model for the conformational stability. The ensemble predictor described about 88% and 66% variance of the data in training and test sets respectively. Furthermore, the optimum AASA vector subset not only helped to successfully model unfolding stability but also well distributed wild‐type and gene V protein mutants on a stability self‐organized map (SOM), when used for unsupervised training of competitive neurons. Proteins 2007.

Thermal stabilization of trypsin by enzymic modification with β-cyclodextrin derivatives

Streptoverticillum sp. transglutaminase was used as catalyst for the attachment of several β‐cyclodextrin derivatives to the glutamine residues in bovine pancreatic trypsin. The modifying agents used were mono‐6‐ethylenediamino‐6‐deoxy‐β‐cyclodextrin, mono‐6‐propylenediamino‐6‐deoxy‐β‐cyclodextrin, mono‐6‐butylenediamino‐6‐deoxy‐β‐cyclodextrin and mono‐6‐hexylenediamino‐6‐deoxy‐β‐cyclodextrin. The transformed trypsin preparations contained about 3 mol of oligosaccharides/mol of protein. The specific esterolytic activity of trypsin was increased by about 4–21% after conjugation. The Km values for cyclodextrin–trypsin complexes represented about 58–87% of that corresponding to the native enzyme. The optimum temperature for esterolytic activity of trypsin was increased by about 5–10 °C after enzymic modification with the cyclodextrin derivatives. The thermostability was increased by 16 °C for the modified trypsin. Thermal inactivation at different temperatures ranging from 45 to 60 °C was markedly increased for the oligosaccharide–trypsin complexes. This modification also protected the enzyme against autolysis at alkaline pH.

Artificial Neural Networks from MATLAB® in Medicinal Chemistry. Bayesian-Regularized Genetic Neural Networks (BRGNN): Application to the Prediction of the Antagonistic Activity Against Human Platelet Thrombin Receptor (PAR-1)

Artificial neural networks (ANNs) have been widely used for medicinal chemistry modeling. In the last two decades, too many reports used MATLAB environment as an adequate platform for programming ANNs. Some of these reports comprise a variety of applications intended to quantitatively or qualitatively describe structure-activity relationships. A powerful tool is obtained when there are combined Bayesian-regularized neural networks (BRANNs) and genetic algorithm (GA): Bayesian-regularized genetic neural networks (BRGNNs). BRGNNs can model complicated relationships between explanatory variables and dependent variables. Thus, this methodology is regarded as useful tool for QSAR analysis. In order to demonstrate the use of BRGNNs, we developed a reliable method for predicting the antagonistic activity of 5-amino-3-arylisoxazole derivatives against Human Platelet Thrombin Receptor (PAR-1), using classical 3D-QSAR methodologies: Comparative Molecular Field Analysis (CoMFA) and Comparative Molecular Similarity Indices Analysis (CoMSIA). In addition, 3D vectors generated from the molecular structures were correlated with antagonistic activities by multivariate linear regression (MLR) and Bayesian-regularized neural networks (BRGNNs). All models were trained with 34 compounds, after which they were evaluated for predictive ability with additional 6 compounds. CoMFA and CoMSIA were unable to describe this structure-activity relationship, while BRGNN methodology brings the best results according to validation statistics.

Identification of a potent and selective σ1 receptor agonist potentiating NGF-induced neurite outgrowth in PC12 cells

Herein we report the synthesis, drug-likeness evaluation, and in vitro studies of new sigma (σ) ligands based on arylalkenylaminic scaffold. For the most active olefin the corresponding arylalkylamine was studied. Novel arylalkenylamines generally possess high σ(1) receptor affinity (K(i) values <25 nM) and good σ(1)/σ(2) selectivity (K(i)σ(2) >100). Particularly, the piperidine derivative (E)-17 and its arylalkylamine analog (R,S)-33 were observed to be excellent σ(1) receptor ligands (K(i)=0.70 and 0.86 nM, respectively) and to display significantly high selectivity over σ(2), μ-, and κ-opioid receptors and phencyclidine (PCP) binding site of the N-methyl-d-aspartate (NMDA) receptors. Moreover in PC12 cells (R,S)-33 promoted the nerve growth factor (NGF)-induced neurite outgrowth and elongation. Co-administration of the selective σ(1) receptor antagonist BD-1063 totally counteracted this effect, confirming that σ(1) receptors are involved in the (R,S)-33 modulation of the NGF effect in PC12 cells and suggesting a σ(1) agonist profile. As a part of our work, a threedimensional σ(1) pharmacophore model was also developed employing GALAHAD methodology. Only active compounds were used for deriving this model. The model included two hydrophobes and a positive nitrogen as relevant features and it was able to discriminate between molecules with and without affinity toward σ(1) receptor subtype.

Structural requirements of pyrido[2,3-d]pyrimidin-7-one as CDK4/D inhibitors : 2D autocorrelation, CoMFA and CoMSIA analyses

2D autocorrelation, comparative molecular field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA) were undertaken for a series of pyrido[2,3-d]pyrimidin-7-ones to correlate cyclin-dependent kinase (CDK) cyclin D/CDK4 inhibition with 2D and 3D structural properties of 60 known compounds. QSAR models with considerable internal as well as external predictive ability were obtained. The relevant 2D autocorrelation descriptors for modeling CDK4/D inhibitory activity were selected by linear and nonlinear genetic algorithms (GAs) using multiple linear regression (MLR) and Bayesian-regularized genetic neural network (BRGNN) approaches, respectively. Both models showed good predictive statistics; but BRGNN model enables better external predictions. A weight-based input ranking scheme and Kohonen self-organized maps (SOMs) were carried out to interpret the final net weights. The 2D autocorrelation space brings different descriptors for CDK4/D inhibition, and suggests the atomic properties relevant for the inhibitors to interact with CDK4/D active site. CoMFA and CoMSIA analyses were developed with a focus on interpretative ability using coefficient contour maps. CoMSIA produced significantly better results. The results indicate a strong correlation between the inhibitory activity of the modeled compounds and the electrostatic and hydrophobic fields around them.

Prediction of dinucleotide-specific RNA-binding sites in proteins

BackgroundRegulation of gene expression, protein synthesis, replication and assembly of many viruses involve RNA–protein interactions. Although some successful computational tools have been reported to recognize RNA binding sites in proteins, the problem of specificity remains poorly investigated. After the nucleotide base composition, the dinucleotide is the smallest unit of RNA sequence information and many RNA-binding proteins simply bind to regions enriched in one dinucleotide. Interaction preferences of protein subsequences and dinucleotides can be inferred from protein-RNA complex structures, enabling a training-based prediction approach.ResultsWe analyzed basic statistics of amino acid-dinucleotide contacts in protein-RNA complexes and found their pairing preferences could be identified. Using a standard approach to represent protein subsequences by their evolutionary profile, we trained neural networks to predict multiclass target vectors corresponding to 16 possible contacting dinucleotide subsequences. In the cross-validation experiments, the accuracies of the optimum network, measured as areas under the curve (AUC) of the receiver operating characteristic (ROC) graphs, were in the range of 65-80%.ConclusionsDinucleotide-specific contact predictions have also been extended to the prediction of interacting protein and RNA fragment pairs, which shows the applicability of this method to predict targets of RNA-binding proteins. A web server predicting the 16-dimensional contact probability matrix directly from a user-defined protein sequence was implemented and made available at: http://tardis.nibio.go.jp/netasa/srcpred.