David T. Stanton

EVALUATION AND USE OF BCUT DESCRIPTORS IN QSAR AND QSPR STUDIES

The recently developed BCUT metrics of Pearlman were evaluated to determine their utility as measures of molecular structure in quantitative structure−activity relationship (QSAR) and quantitative structure−property relationship (QSPR) studies. These metrics have been found to provide unique information regarding molecular structure and have been found to make significant contributions to resulting equations. The importance of these descriptors is illustrated in a study of inhibitors of dihydrofolate reductase and in a study of the normal boiling points of a diverse set of polar heterocyclic compounds.

The Effect of Room-Temperature Storage on the Stability of Compounds in DMSO

The stability of approximately 7200 compounds stored as 20-mM DMSO solutions under ambient conditions was monitored for 1 year. Compound integrity was measured by flow injection analysis using positive and negative electrospray ionization mass spectrometry. Each sample was assessed at the beginning of the study, after 12 months of storage, and at a randomized time point between the initial and final time points of the study. The relationship between length of storage and the probability of observing the compound was described by a repeated-measures logistic regression model. The probability of observing the compound was 92% after 3 months of storage at room temperature, 83% after 6 months, and 52% after 1 year in DMSO. An acceptable limit for compound loss and corresponding maximum storage time for samples in DMSO can be determined based on these results.(Journal of Biomolecular Screening 2003:205-209)

On the physical interpretation of QSAR models.

Multidimensional quantitative structure-activity models (QSAR) developed using molecular structure descriptors and regression analysis techniques have found wide utility and acceptance. However, it is often difficult to extract a physical interpretation of such models because of the types of descriptors involved and the multidimensional nature of the model. The work described here illustrates a method of model interpretation that employs partial least squares (PLS) analysis. Structure-activity relationship information is derived from the positions of specific sets of structures in the PLS score plots and the weights for each variable in the PLS components. Using these data, information regarding major structure-activity trends, trend exceptions, and unique or outlying observations is easily obtained. Examples of this methodology are illustrated using QSAR equations developed for the inhibition of quinolone-resistant bacterial DNA gyrase and human topoisomerase-II inhibition by a series of quinolone antibacterial agents.

Interpreting Computational Neural Network Quantitative Structure−Activity Relationship Models: A Detailed Interpretation of the Weights and Biases

In this work, we present a methodology to interpret the weights and biases of a computational neural network (CNN) quantitative structure-activity relationship model. The methodology allows one to understand how an input descriptor is correlated to the predicted output by the network. The method consists of two parts. First, the nonlinear transform for a given neuron is linearized. This allows us to determine how a given neuron affects the downstream output. Next, a ranking scheme for neurons in a layer is developed. This allows us to develop interpretations of a CNN model similar in manner to the partial least squares (PLS) interpretation method for linear models described by Stanton. The method is tested on three datasets covering both physical and biological properties. The results of this interpretation method correspond well to PLS interpretations for linear models using the same descriptors as the CNN models, and they are consistent with the generally accepted physical interpretations for these properties.

Understanding the underlying dimensions in perfumers’ odor perception space as a basis for developing meaningful odor maps

Various low-dimensional perceptual maps of fragrances have been proposed in the literature, as well as sensory maps for the odor descriptors most frequently applied in perfumery. To reach a consensus, however, seems difficult, if at all possible. In the present study, we applied principal components analysis to two databases. The first contains numeric odor profiles of 309 compounds based on 30 descriptors. The loading plot corresponding to the relevant components was strikingly similar to the odor effects diagram proposed by P. Jellinek (1951), primarily on the basis of his long experience as a perfumer. We obtained similar results in our analysis of the second database, which comprises 66 descriptors and contains the semantic descriptions of 119 perfume materials. On the basis of the results of both analyses, a commercial map of fragrances is discussed. Our findings suggest that it is possible to develop standard sensory maps of perfumery odor descriptors, if a consensus is first reached regarding which odorants best represent particular odor qualities.

Development of a quantitative structure--property relationship model for estimating normal boiling points of small multifunctional organic molecules.

Computer-assisted quantitative structure-property relationship techniques are applied in the development of a robust and accurate model of normal boiling points (boiling at 760 mmHg) for a very diverse set of 268 small organic molecules. Most of the molecules included in this study contain two or more functional groups. The final model yields a tight fit to the training set data (R2 = 0.963), with a fit error of 6.5%. More importantly, the model is also shown to perform well in external prediction. The mean prediction error for boiling points for a 78-member external test set was 12.3 degrees C, or 8.3%. A detailed analysis of the small number of compounds that were either outliers or not well predicted illustrates areas for potential improvement of the methodology used.

Novel phenolic inhibitors of the sarco/endoplasmic reticulum calcium ATPase: identification and characterization by quantitative structure–activity relationship modeling and virtual screening

Abstract Inhibitors of the sarco/endoplasmic reticulum calcium ATPase (SERCA) are valuable research tools and hold promise as a new generation of anti-prostate cancer agents. Based on previously determined potencies of phenolic SERCA inhibitors, we created quantitative structure–activity relationship (QSAR) models using three independent development strategies. The obtained QSAR models facilitated virtual screens of several commercial compound collections for novel inhibitors. Sixteen compounds were subsequently evaluated in SERCA activity inhibition assays and 11 showed detectable potencies in the micro- to millimolar range. The experimental results were then incorporated into a comprehensive master QSAR model, whose physical interpretation by partial least squares analysis revealed that properly positioned substituents at the central phenyl ring capable of forming hydrogen bonds and of undergoing hydrophobic interactions were prerequisites for effective SERCA inhibition. The established SAR was in good agreement with findings from previous structural studies, even though it was obtained independently using standard QSAR methodologies.

On the importance of topological descriptors in understanding structure–property relationships

It has been generally observed in our work that molecular descriptors derived from a molecular graph theory or topological representation of structure play an important and often key role in many QSAR and QSPR models we have developed. These descriptors do not only provide the means to generate a good fit to the observed data used to train the models, but they also provide information that is needed to generate a clear physical interpretation of the underlying structure–activity or property relationships. In addition, these descriptors provide a conformation-independent method of measuring the key features of molecular structure that affect the observed properties of the molecules. These characteristics are exemplified in a model developed to predict critical micelle concentration (CMC). A model is described that exhibits excellent predictive strength, is independent of conformation of the structures used, and that yields a great deal of detail regarding the underlying structure–property relationship driving the observed CMC.

Development and use of a high-throughput bacterial DNA gyrase assay to identify mammalian topoisomerase II inhibitors with whole-cell anticancer activity.

A high-throughput screen (HTS)was developed and used to identify inhibitors of bacterial DNA gyrase. Among the validated hits were 53 compounds that also inhibited mammalian topoisomerase II with IC50 values of <12.5 µg/mL for 51 of them. Using computational methods, these compounds were subjected to cluster analysis to categorize them according to their chemical and structural properties. Nine compounds from different clusters were tested for their whole-cell inhibitory activity against 3 cancer cell lines—NCI-H460 (lung), MCF7 (breast), and SF-268 (CNS)—at a concentration of 100 µM. Five compounds inhibited cell growth by >50% for all 3 cell lines tested. These compounds were tested further against a panel of 53 to 57 cell lines representing leukemia, melanoma, colon, CNS, ovarian, renal, prostate, breast, and non–small cell lung cancers. In this assay, PGE-7143417 was found to be the most potent compound, which inhibited the growth of all the cell lines by 50% at a concentration range of 0.31 to 2.58 µM, with an average of 1.21 µM. An additional 17 compounds were also tested separately against a panel of 10 cell lines representing melanoma, colon, lung, mammary, ovarian, prostate, and renal cancers. In this assay, 4 compounds—PGE-3782569, PGE-7411516, PGE-2908955, and PGE-3521917—were found to have activity with concentrations for 50% cell growth inhibition in the 0.59 to 3.33, 22.5 to 59.1, 7.1 to >100, and 24.7 to >100 µM range. (Journal of Biomolecular Screening 2003:157-163)

Profiling of drugs for membrane activity using liposomes as an in vitro model system.

ABSTRACT The increasing size of chemical libraries being analyzed by high-throughput screening results in a growing number of active compounds that need to be assessed before moving forward in the drug development process. As a consequence, more rapid and highly sensitive strategies are required to accelerate the process of drug discovery without increasing the cost. Due to the fact that significant numbers of compounds from combinatorial libraries are hydrophobic in nature, approaches are needed to evaluate the potential for these compounds to interfere with the functions of biological membranes. The liposome system was used to detect agents that act as follows: (i) ionophores able to induce specific ion permeability, e.g., valinomycin for K+ and protonophoric uncouplers for H+; (ii) ion antiporters which exchange H+ for other ions, e.g., nigericin; (iii) agents that form low specificity ion channels in the membrane, e.g., gramicidin; and (iv) detergents and other membrane-disrupting agents. We propose using this liposome assay during the drug development process to identify compounds that have membrane activity and, as a consequence, produce a biological effect by altering the physico-chemical properties of the cell membrane rather than interacting with a protein target. Screening of a representative set of biologically-active compounds (198) indicated that the majority of systemic antimicrobial drugs, but not topical drugs, lack membrane activity in this model system.