A new procedure for improving the predictiveness of CoMFA models and its application to a set of dihydrofolate reductase inhibitors

A new automated procedure to improve the predictive quality of CoMFA models for both training and test sets is described. A model of greater consistency is generated by performing small reorientations of the underlying molecules for which too low activities are calculated. In order to predict activities of test compounds, the most similar molecules in the previously optimized model are identified and used as a basis for the prediction. This method has been applied to two independent sets of dihydrofolate reductase inhibitors (80 compounds each, serving as training sets), resulting in a significant increase of the cross-validated r2 value. For both models, the predictive r2 value for a test set consisting of 70 compounds was improved substantially.     

Validation of EGSITE2, a mixed integer program for deducing objective site models for experimental binding data

EGSITE2 represents a substantial advance in a long series of methods for calculating receptor site models given only specific binding data. Compared to our most recently reported technique, EGSITE [Schnitker et al. J. Comput.-Aided Mol. Des. 1997, 11, 93-110] the user no longer has to simplify the structures of the molecules in the training set by clustering the atoms into a few superatoms. The only remaining source of subjectivity is the user's choice of compounds for the training set, which can be surprisingly few in number. Then EGSITE2 automatically produces typically several different models that explain the observed binding without outliers. The models are remarkably simple but have substantial predictive power for any sort of test compound, with an estimation of the uncertainty of the prediction. Validation of the method is reported for four standard test cases: triazines and pyrimidines binding to dihydrofolate reductase, steroids binding to corticosteroid-binding globulin and to testosterone-binding globulin, and peptides binding to angiotensin-converting enzyme.     

Static and fatigue failure properties of thoracic and lumbar vertebral bodies and their relation to regional density

This study investigated (1) whether a characterization of the macroscopic architecture within the vertebral centrum would improve predictions of vertebral strength, (2) if regions in the centrum where least bone loss with age occurs are more predictive of vertebral strength, and (3) whether different patterns of the macroscopic architecture are predictive of static as compared to fatigue strength. To characterize the vertebral macroscopic architecture, a regional bone mineral density (rBMD) technique was used that estimated the cancellous density distribution (in 18 specific regions of the vertebral centrum) for vertebrae T7-L4, from spines of 20 female cadavers. Static and fatigue failure properties of whole vertebrae were obtained, and predictive models of static and fatigue failure properties of whole vertebrae were examined. We found that (1) vertebral failure properties were better predicted by combinations of vertebral regional cancellous density (multiple linear regressions) rather than by any individual region of cancellous density alone (simple linear regressions); (2) models using regions of density that demonstrated minimum decline with age [from the data of Flynn and Cody (Calcif. Tissue Int. 53, S170-S175 (1993))] resulted in better correlations with ex vivo vertebral static failure properties than models using density regions that showed maximum decline with age, and (3) static and fatigue characteristics required different density regions to reach significance. (A comparison of models predictive of static and fatigue failure properties revealed that anterior density regions were most often included in predictive models of the static properties while posterior regions were more predictive of the fatigue properties).(ABSTRACT TRUNCATED AT 250 WORDS)     

Strategic item selection to reduce survey length: Reduction in validity?

The practical constraints of conducting rigorous research in applied settings often limit the ability to include well-validated, but lengthy, measures in survey designs. In the current study we tested the utility of using a strategic item selection process to alleviate this issue. We evaluated the measurement and structural model fit of items selected using a strategic item selection format and assessed the predictive validity of those abbreviated scales. We also compared fit and predictive validity to the full version item sets to assess the comparability of the abbreviated scales. Results indicated that measurement models and structural models for the a priori items produced similar fit when compared with full version items across both measures. When comparing the proportion of variance explained in the criterion, there were no differences between the strategically selected items and the full versions of the scales. We conclude that the a priori selection of items based on previous research may be a useful technique for developing short measures of psychological constructs that mitigate practical, analytical, and methodological issues associated with lengthy surveys. (PsycINFO Database Record (c) 2011 APA, all rights reserved)     

A strategy for the incorporation of water molecules present in a ligand binding site into a three-dimensional quantitative structure--activity relationship analysis

Water present in a ligand binding site of a protein has been recognized to play a major role in ligand-protein interactions. To date, rational drug design techniques do not usually incorporate the effect of these water molecules into the design strategy. This work represents a new strategy for including water molecules into a three-dimensional quantitative structure-activity relationship analysis using a set of glucose analogue inhibitors of glycogen phosphorylase (GP). In this series, the structures of the ligand-enzyme complexes have been solved by X-ray crystallography, and the positions of the ligands and the water molecules at the ligand binding site are known. For the structure-activity analysis, some water molecules adjacent to the ligands were included into an assembly which encompasses both the inhibitor and the water involved in the ligand-enzyme interaction. The mobility of some water molecules at the ligand binding site of GP gives rise to differences in the ligand-water assembly which have been accounted for using a simulation study involving force-field energy calculations. The assembly of ligand plus water was used in a GRID/GOLPE analysis, and the models obtained compare favorably with equivalent models when water was excluded. Both models were analyzed in detail and compared with the crystallographic structures of the ligand-enzyme complexes in order to evaluate their ability to reproduce the experimental observations. The results demonstrate that incorporation of water molecules into the analysis improves the predictive ability of the models and makes them easier to interpret. The information obtained from interpretation of the models is in good agreement with the conclusions derived from the structural analysis of the complexes and offers valuable insights into new characteristics of the ligands which may be exploited for the design of more potent inhibitors.     

Model Development and Validation for Intelligent Data Collection for Lateral Spread Displacements

The geotechnical earthquake engineering community often adopts empirically derived models. Unfortunately, the community has not embraced the value of model validation, leaving practitioners with little information on the uncertainties present in a given model and the model’s predictive capability. In this study, we present a machine learning technique known as support vector regression (SVR) together with rigorous validation for modeling lateral spread displacements and outline how this information can be used for identifying gaps in the data set. We demonstrate the approach using the free face lateral displacement data. The results illustrate that the SVR has relatively better predictive capability than the commonly used empirical relationship derived using multilinear regression. Moreover, the analysis of the SVR model and its support vectors helps in identifying gaps in the data and defining the scope for future data collection.     

Place of in vitro models in preclinical evaluation of anticancer drugs

In vitro models have been intensively developed for several years for selecting new anticancer agents. The National Cancer Institute has even chosen as a primary screen of new molecules a panel of 60 human tumor cell lines. However, it may seem hazardous to rely too much on in vitro models for the discovery and selection of new anticancer drugs: (1 because no metabolism of the compounds occurs in cell culture; (2 because an in vitro cell line cannot be representative of an in situ tumor cell population; (3 because antiproliferative activity is only part of antitumor activity; (4 because the toxicologic properties of the molecules are not taken into account by in vitro systems; (5 because cell cultures do not allow any selectivity study between tumor cells and normal cells. With examples drawn from three different therapeutic classes, anthracyclines, taxoids and camptothecin derivatives, we show that in vitro tests are insufficiently predictive of antitumor potential. The excess of confidence allowed to these models may lead to premature decisions which are not after that justified by clinical trials.     

DIAMOD: display and modeling of DNA bending

MOTIVATION: DIAMOD (Displayandmodeling ofDNA) was created as a user-friendly software for exploring and better understanding DNA structural variations, particularly DNA bending. It was intended to be as open as possible so that any of the existing or future predictive models can be used with it. RESULTS: DIAMOD features graphic display and interactive manipulation of DNA molecules on the screen. Since it works with di-, tri- or tetranucleotide models supplied as external files of angular parameters, it was recently used to evaluate critically all available predictive models for DNA bending. The program has a unique option to insert bends at defined positions in DNA sequence independently of the currently used model, which enables the simulation of both intrinsic and protein-induced kinking. Finally, many output file formats facilitate the sharing of data with other programs and the creation of visually pleasing images. AVAILABILITY: The program is available on request to academic users free of charge. It will be distributed via the WWW (http://www-personal.umich.edu/ mensur/software.html). Users with no network access can get a copy directly from the author. CONTACT: mensur@umich.edu     

Statistical evaluation of the role of Helicobacter pylori in stress gastritis: applications of splines and bootstrapping to the logistic model

Stress gastritis is a serious problem in the intensive care unit population. The recent discovery of the causal nature of Helicobacter pylori (H. pylori) in the development of gastric ulcers led us to examine its relationship with stress gastritis. We investigated this relationship in 874 veterans admitted to intensive care units who were tested for the presence of H. pylori and followed for 6 weeks for the development of stress gastritis. We fit spline models to assess functional relationships and used the logistic model to determine the association between H. pylori and stress gastritis. The predictive ability of the model was assessed with receiver operating characteristic (ROC) curve analysis and validated with the bootstrapping technique. Increased anti-H. pylori immunoglobulin A concentrations were found to be an important predictor of stress gastritis independent of other known risk factors.     

On the Rotational Analysis of the Ground Vibrational State of CH3D Molecule

Three-dimensional quantitative structure-activity relationship (3D-QSAR) models have been developed using comparative molecular field analysis (CoMFA) on a large data set (118 compounds) of diverse cyclic urea derivatives as protease inhibitors against the human immunodeficiency virus type 1 (HIV-1). X-ray crystal structures of HIV-1 protease bound with this class of inhibitors were used to derive the most probable bioactive conformations of the inhibitors. The enzyme active site was used as a constraint to limit the number of possible conformations that are sterically accessible. The test sets have been created keeping in mind structural diversity as well as the uniform simple statistical criteria (mean, standard deviation, high and low values) of the protease inhibitory activities of the molecules compared to the training sets. Multiple predictive models have been developed with the training sets (93 compounds in each set) and validated with the corresponding test sets (25 compounds in each set). All the models yielded high predictive correlation coefficients (q2 from 0.699 to 0.727), substantially high fitted correlation coefficients (r2 from 0.965 to 0.973), and reasonably low standard errors of estimates (S from 0. 239 to 0.265). The steric and electrostatic effects have approximately equal contributions, 45% and 55% (approximately), respectively, toward explaining protease inhibitory activities. This analysis yielded models with significant information on steric and electrostatic interactions clearly discerned by the respective coefficient contour plots when overlapped on the X-ray structure of the HIV-1 protease. The HINT CoMFA study revealed significant contribution of hydrophobicity toward protease inhibitory activity. The 3D visualization technique utilizing these contour plots as well as the receptor site geometry may significantly improve our understanding of the inhibitor-protease (HIV-1) interactions and help in designing compounds with improved activity.     

Can we learn to distinguish between "drug-like" and "nondrug-like" molecules?

We have used a Bayesian neural network to distinguish between drugs and nondrugs. For this purpose, the CMC acts as a surrogate for drug-like molecules while the ACD is a surrogate for nondrug-like molecules. This task is performed by using two different set of 1D and 2D parameters. The 1D parameters contain information about the entire molecule like the molecular weight and the the 2D parameters contain information about specific functional groups within the molecule. Our best results predict correctly on over 90% of the compounds in the CMC while classifying about 10% of the molecules in the ACD as drug-like. Excellent generalization ability is shown by the models in that roughly 80% of the molecules in the MDDR are classified as drug-like. We propose to use the models to design combinatorial libraries. In a computer experiment on generating a drug-like library of size 100 from a set of 10 000 molecules we obtain at least a 3 or 4 order of magnitude improvement over random methods. The neighborhoods defined by our models are not similar to the ones generated by standard Tanimoto similarity calculations. Therefore, new and different information is being generated by our models, and so it can supplement standard diversity approaches to library design.     

A geometry-based suite of molecular docking processes

We have developed a geometry-based suite of processes for molecular docking. The suite consists of a molecular surface representation, a docking algorithm, and a surface inter-penetration and contact filter. The surface representation is composed of a sparse set of critical points (with their associated normals) positioned at the face centers of the molecular surface, providing a concise yet representative set. The docking algorithm is based on the Geometric Hashing technique, which indexes the critical points with their normals in a transformation invariant fashion preserving the multi-element geometric constraints. The inter-penetration and surface contact filter features a three-layer scoring system, through which docked models with high contact area and low clashes are funneled. This suite of processes enables a pipelined operation of molecular docking with high efficacy. Accurate and fast docking has been achieved with a rich collection of complexes and unbound molecules, including protein-protein and protein-small molecule associations. An energy evaluation routine assesses the intermolecular interactions of the funneled models obtained from the docking of the bound molecules by pairwise van der Waals and Coulombic potentials. Applications of this routine demonstrate the goodness of the high scoring, geometrically docked conformations of the bound crystal complexes.     

Personality, chronic medical morbidity, and health-related quality of life among older persons

We report the design of optimal linker geometries for the synthesis of stapled DNA-minor-groove-binding molecules. Netropsin, distamycin, and lexitropsins bind side-by-side to mixed-sequence DNA and offer an opportunity for the design of sequence-reading molecules. Stapled molecules, with two molecules covalently linked side-by-side, provide entropic gains and restrain the position of one molecule relative to its neighbor. Using a free-atom simulated annealing technique combined with a discrete mutable atom definition, optimal lengths and atomic composition for covalent linkages are determined, and a novel hydrogen bond 'zipper' is proposed to phase two molecules accurately side-by-side.     

On the usefulness of epitaxy experiments in evaluating interface models

New results on low energy interfaces for the system gold on sodium chloride are presented together with a short review of the various versions of the sphere-plate technique which are used to study interface systems. The observed low energy orientation relationships as well as those recently found for the systems silver on sodium chloride and silver on silicon are compared with preferred orientations observed in epitaxy experiments. It is concluded that results for epitaxy are extremely useful when evaluating interface models for their predictive power.     

Predictive model selection for repeated measures random effects models using Bayes factors

The random effects model fit to repeated measures data is an extremely common model and data structure in current biostatistical practice. Modern data analysis often involves the selection of models within broad classes of prespecified models, but for models beyond the generalized linear model, few model-selection tools have been actively studied. In a Bayesian analysis, Bayes factors are the natural tool to use to explore these classes of models. In this paper, we develop a predictive approach for specifying the priors of a repeated measures random effects model with emphasis on selecting the fixed effects. The advantage of the predictive approach is that a single predictive specification is used to specify priors for all models considered. The methodology is applied to a pediatric pain data analysis.     

Predicting diabetes. Moving beyond impaired glucose tolerance

We developed predictive models for type II diabetes using stepwise multiple logistic regression analyses of a cohort of 844 Mexican Americans and 641 non-Hispanic whites who were nondiabetic at baseline and who were then followed for 8 yr. Models were developed for the overall population and separately for each sex and ethnic group. For optimal models, the multiple logistic regression program selected potential risk factors from a panel of 5 categorical and 14 continuous demographic, anthropometric, metabolic, and hemodynamic variables. For reduced models, the list of candidate variables was restricted to those commonly used in ordinary clinical practice, i.e., skinfolds, and serum insulin and postural glucose load variables were excluded. For all models, the stepwise process selected a mixture of anthropometric, glucose, lipid, and hemodynamic variables. The top 15% of the risk continuum for each model was defined as high risk to compare the performance of the models with the performance of impaired glucose tolerance (15% prevalence) as a predictor of diabetes. The relative risk of being high risk ranged from 12.16 to 35.29, whereas the relative risk of having impaired glucose tolerance ranged from 7.11 to 10.0. The sensitivity of the multiple logistic regression models ranged from 67.7 to 83.3% compared with 56.5 to 62.1% for impaired glucose tolerance. The results indicate that multivariate predictive models perform at least as well, if not better than impaired glucose tolerance in predicting type II diabetes but need not require an oral glucose load. Moreover, the models highlight the complex metabolic and hemodynamic syndrome that precedes diabetes.     

Three-dimensional quantitative structure-activity relationship analyses using comparative molecular field analysis and comparative molecular similarity indices analysis to elucidate selectivity differences of inhibitors binding to trypsin, thrombin, and factor Xa

Three-dimensional quantitative structure-activity relationship (3D QSAR) methods were applied using a training set of 72 inhibitors of the benzamidine type with respect to their binding affinities (Ki values) toward thrombin, trypsin, and factor Xa to yield statistically reliable models of good predictive power. Two methods were compared: the widely used comparative molecular field analysis (CoMFA) and the recently reported CoMSIA approach (comparative molecular similarity indices analysis). CoMSIA produced significantly better results for all correlations. Furthermore, in contrast to CoMFA, CoMSIA is not sensitive to changes in orientation of the superimposed molecules in the lattice. The correlation results obtained by CoMSIA were graphically interpreted in terms of field contribution maps allowing physicochemical properties relevant for binding to be easily mapped back onto molecular structures. The advantage of this feature is demonstrated using the maps to design new molecules. Finally, the CoMSIA method was applied to elucidate structural features among ligands which are responsible for affinity differences toward thrombin and trypsin. These selectivity-determining features were interpreted graphically in terms of spatial regions responsible for affinity discrimination. Such indicators are highly informative for the lead optimization process with respect to selectivity enhancement.     

Psychosocial factors and high-risk sexual behavior: race differences among urban adolescents

Adolescence is a period of sexual experimentation. We examined psychosocial predictors of high-risk sexual behavior and condom use. The sample included 824 ninth-graders, most of whom are African American. We conducted separate analyses for whites and African Americans. Predictors included alcohol and substance use, delinquency, prosocial behaviors, and family and peer influences. We found that problem behaviors predicted high-risk sexual behavior, but the effects were stronger for white youth. We also found that friends' behaviors were more predictive than family influences, except for family conflict. In general, the models explained more variance for white youths than for African-American youths. The results suggest that problem behavior theory and social interactions theory may be most relevant for white youth and that other models may be necessary to explain high-risk sexual behavior among African-American youths.     

Structural equation modeling in practice: A review and recommended two-step approach.

In this article, we provide guidance for substantive researchers on the use of structural equation modeling in practice for theory testing and development. We present a comprehensive, two-step modeling approach that employs a series of nested models and sequential chi-square difference tests. We discuss the comparative advantages of this approach over a one-step approach. Considerations in specification, assessment of fit, and respecification of measurement models using confirmatory factor analysis are reviewed. As background to the two-step approach, the distinction between exploratory and confirmatory analysis, the distinction between complementary approaches for theory testing versus predictive application, and some developments in estimation methods also are discussed. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Patenting computer-designed peptides

The problem of designing new peptides that possess specific properties, such as bactericidal activity, is of wide interest. Recently, attention has focused on the use of Computer-Aided Molecular Design techniques in parallel with more traditional 'synthesise and test' methods. These techniques may typically use Genetic Algorithms to optimise molecules based on Neural Network models that predict activity. In this paper we describe a successful application of this Molecular Design methodology that has resulted in novel bactericidal peptides of real value. A key issue for commercial utilisation of such results is the ability to protect the intellectual property rights associated with the discovery of new molecules. Typically peptide patents use structural templates of amino acid hydrophobicity-hydrophilicity that define highly regular peptide patent spaces. In an extension of established patenting practice we describe a patent application that uses a Neural Net predictive model to define the regions of peptide space that we claim within the patent. This formalism makes no a priori assumptions about the regularity of the patent space. A preliminary comparative investigation of the shape and size of this and other bactericidal peptide patent spaces is conducted.