The reduced graph descriptor in virtual screening and data-driven clustering of high-throughput screening data.

Virtual screening and high-throughput screening are two major components of lead discovery within the pharmaceutical industry. In this paper we describe improvements to previously published methods for similarity searching with reduced graphs, with a particular focus on ligand-based virtual screening, and describe a novel use of reduced graphs in the clustering of high-throughput screening data. Literature methods for reduced graph similarity searching encode the reduced graphs as binary fingerprints, which has a number of issues. In this paper we extend the definition of the reduced graph to include positively and negatively ionizable groups and introduce a new method for measuring the similarity of reduced graphs based on a weighted edit distance. Moving beyond simple similarity searching, we show how more flexible queries can be built using reduced graphs and describe a database system that allows iterative querying with multiple representations. Reduced graphs capture many important features of ligand-receptor interactions and, in conjunction with other whole molecule descriptors, provide an informative way to review HTS data. We describe a novel use of reduced graphs in this context, introducing a method we have termed data-driven clustering, that identifies clusters of molecules represented by a particular whole molecule descriptor and enriched in active compounds.

Three- and four-dimensional-quantitative structure activity relationship (3D/4D-QSAR) analyses of CYP2C9 inhibitors.

The interaction of competitive type inhibitors with the active site of cytochrome P450 (CYP) 2C9 has been predicted using three- and four-dimensional quantitative structure activity relationship (3D-/4D-QSAR) models constructed using previously unreported and literature-derived data. 3D-QSAR pharmacophore models of the common structural features of CYP2C9 inhibitors were built using the program Catalyst and compared with 3D- and 4D-QSAR partial least-squares models, which use molecular surface-weighted holistic invariant molecular descriptors of the size and shape of inhibitors. The Catalyst models generated from multiple conformers of competitive inhibitors of CYP2C9 activities contained at least one hydrophobic and two hydrogen bond acceptor/donor regions. Catalyst model 1 was constructed with Ki(apparent) values for inhibitors of tolbutamide and diclofenac 4'-hydroxylation (n = 9). Catalyst model 2 was generated from literature Ki(apparent) values for (S)-warfarin 7-hydroxylation (n = 29), and Catalyst model 3 from literature IC50 values for tolbutamide 4-hydroxylation (n = 13). These three models illustrated correlation values of observed and predicted inhibition for CYP2C9 of r = 0.91, 0.89, and 0.71, respectively. Catalyst pharmacophores generated with Ki(apparent) values were validated by predicting the Ki(apparent) value of a test set of CYP2C9 inhibitors also derived from the literature (n = 14). Twelve of fourteen of these Ki(apparent) values were predicted to be within 1 log residual of the observed value using Catalyst model 1, whereas Catalyst model 2 predicted 10 of 14 Ki(apparent) values. The corresponding partial least-squares molecular surface-weighted holistic invariant molecular 3D- and 4D-QSAR models for all CYP2C9 data sets yielded predictable models as assessed using cross-validation. These 3D- and 4D-QSAR models of CYP inhibition will aid in future prediction of drug-drug interactions.

Global 3D-QSAR methods: MS-WHIM and autocorrelation.

The recently proposed MS-WHIM indices, a set of theoretical descriptors containing information about size, shape and electrostatic distribution of a molecule, have been further investigated. The main objectives of this work were: (i) to confirm the descriptive power of MS-WHIM in modelling specific biological interactions, (ii) to analyse the dependence of MS-WHIM on the type of atomic charges used for computing electrostatic potential and (iii) to compare the performances of MS-WHIM with those provided by other global 3D molecular descriptors. The spatial autocorrelation of atomic and molecular surface properties were selected for comparison purposes. WHIM-based and autocorrelation-based vectors were calculated for two molecular sets from the literature, namely a series of 18 HIV-1 reverse transcriptase inhibitors and a set of 36 sulphonamide endothelin inhibitors. PLS was adopted to derive statistical predictive models that were validated by means of cross-validation. The reported results confirmed that MS-WHIM indices are able to provide meaningful statistical correlations with biological activity. MS-WHIM descriptors are sensitive to the type of partial atomic charges applied and improved models were obtained using more accurate charges. Moreover for both the datasets, MS-WHIM results, in terms of fitting and predictive power of PLS models, were superior to those from autocorrelation. Finally, the strengths/weaknesses of global 3D-QSAR descriptors over local CoMFA-like methods, as well as the main differences between WHIM-based and autocorrelation-based vectors, are discussed.

PLUMS: a program for the rapid optimization of focused libraries.

PLUMS is a new method to perform rational monomer selection for combinatorial chemistry libraries. The algorithm has been developed to optimize focused libraries with specific two-dimensional and/or three-dimensional properties. A preliminary step is the identification of those molecules in the initial virtual library which satisfy the imposed property constraints; we define these molecules as the virtual hits. From the virtual hits, PLUMS generates a starting library, which is the true combinatorial library that includes all the virtual hits. Monomers are then removed in an iterative fashion, thus reducing the size of the library. At each iteration, the worst monomer is removed. Each sublibrary is selected using a global scoring function, which balances effectiveness and efficiency. The iterative process continues until one is left with a library that consists entirely of virtual hits. The optimal library, which is the best compromise between effectiveness and efficiency, can then be selected according to the score. During the iterative process, equivalent solutions may well occur and are taken into account by the algorithm, according to a user-defined parameter. The number of monomers for each substitution site and the size of the library are parameters that can be either optimized or used to constrain the selection. The results obtained on two test libraries are presented. PLUMS was compared with genetic algorithms (GA) and monomer frequency analysis (MFA), which are widely used for monomer selection. For the two test libraries, PLUMS and GA gave equivalent results. MFA is the fastest method, but it can give misleading solutions. Possible advantages and disadvantages of the different methods are discussed.

Three-dimensional-quantitative structure activity relationship analysis of cytochrome P-450 3A4 substrates.

To gain a better understanding of the active site of cytochrome P-450 (CYP) 3A4, a three-dimensional-quantitative structure activity relationship model was constructed using the structures and K(m (apparent)) values of 38 substrates of human liver microsomal CYP3A4. This pharmacophore was built using the program Catalyst and consisted of four features: two hydrogen bond acceptors, one hydrogen bond donor, and one hydrophobic region. The pharmacophore demonstrated a fit value (r) of observed and expected K(m(apparent)) value of 0.67. The validity of the CYP3A4 substrate model was tested by twice permuting (randomizing) the activity values and substrate structures. The results of this validation procedure indicated that the original model was a significant representation of the features required of CYP3A4 substrates. The second validation method used the Catalyst model to predict the K(m(apparent)) values of a test set of structurally diverse substrates for CYP3A4 not included in the 38 molecules used to build the model. Two fitting algorithms included in this software were examined: fast fit and best fit. The fast fitting method resulted in predictions for all 12 substrates that were within 1 log unit for the residual [i.e., the difference between predicted and observed K(m(apparent))]. In contrast, the best fit algorithm poorly predicted the K(m (apparent)) values (i.e., residual >1 log unit) of 4 of 12 substrates. These poor fits with the best fit function suggest that the fast fit method within Catalyst is more representative of the observed K(m(apparent)) values for CYP3A4 substrates and enables good in silico prediction of this activity. A Catalyst common features pharmacophore was also constructed from three molecules known to activate their own metabolism included in the 38 molecules of the initial CYP3A4 model. This demonstrated that activators of CYP3A4 possess multiple hydrophobic regions that might correspond with a region in the active site away from the metabolic site.

Three- and four-dimensional quantitative structure activity relationship analyses of cytochrome P-450 3A4 inhibitors.

The program Catalyst was used to build three-dimensional quantitative structure activity relationship (3D-QSAR) pharmacophore models of the structural features common to competitive-type inhibitors of cytochrome P-450 (CYP) 3A4. These were compared with 3D- and four-dimensional (4D)-QSAR partial least-squares (PLS) models built using molecular surface-weighted holistic invariant molecular (MS-WHIM) descriptors for size and shape of the inhibitor. The Catalyst pharmacophore model generated from multiple conformers of competitive inhibitors of CYP3A4-mediated midazolam 1'-hydroxylation (n = 14) yielded a high correlation of observed and predicted Ki values of r = 0.91. Similarly, PLS MS-WHIM was used to produce 3D- and 4D-QSARs for this data set and produced models that were statistically predictable after cross-validation. Two additional Catalyst pharmacophores were constructed from literature Ki values (n = 32) derived from the inhibition of CYP3A-mediated cyclosporin A metabolism and IC50 data (n = 22) from the inhibition of CYP3A4-mediated quinine 3-hydroxylation. These Catalyst pharmacophores illustrated correlations of observed and predicted inhibition for CYP3A4 of r = 0.77 and 0.92, respectively. The corresponding 4D-QSARs generated by PLS MS-WHIM for these data sets were of comparable quality as judged by cross-validation. Both Ki pharmacophores generated with Catalyst were also validated by predicting the Ki(apparent) values of a test set of eight CYP3A4 inhibitors not included in either model. In seven of eight cases, the residuals of the predicted Ki(apparent) values were within 1 log unit of the observed values. The 3D- and 4D-QSAR models produced in this study suggest the utility of future in silico prediction of CYP3A4-mediated drug-drug interactions.

Three and four dimensional-quantitative structure activity relationship (3D/4D-QSAR) analyses of CYP2D6 inhibitors.

Three- and four-dimensional quantitative structure activity relationship (3D/4D-QSAR) pharmacophore models of competitive inhibitors of CYP2D6 were constructed using data from our laboratory or the literature. The 3D-QSAR pharmacophore models of the common structural features of CYP2D6 inhibitors were built using the program Catalyst (Molecular Simulations, San Diego, CA, USA). These 3D-QSAR models were compared with 3D and 4D-QSAR partial least squares (PLS) models which were constructed using molecular surface-weighted holistic invariant molecular (MS-WHIM) descriptors of size and shape of inhibitors. The first Catalyst model was generated from multiple conformers of competitive inhibitors (n = 20) of CYP2D6 mediated bufurolol 1'-hydroxylation. This model demonstrated a correlation of observed and predicted Ki (apparent) values of r = 0.75. A second Catalyst model was constructed from literature derived Ki (apparent) values (n = 31) for the inhibition of CYP2D6. This model provided a correlation of observed and predicted inhibition for CYP2D6 of r = 0.91. Both Catalyst Ki pharmacophores were then validated by predicting the Ki (apparent) of a test set of known CYP2D6 inhibitors (n = 15). Ten out of 15 of these Ki (apparent) values were predicted to be within one log residual of the observed value using our CYP2D6 inhibitor model, while the literature model predicted nine out of 15 values. Similarly, 3D- and 4D-QSARs derived from PLS MS-WHIM for our dataset yielded predictable models as assessed using cross-validation. The corresponding cross-validated PLS MS-WHIM model for the literature dataset yielded a comparable 3D-QSAR and improved 4D-QSAR value. Such computational models will aid in future prediction of drug-drug interactions.

Three-dimensional quantitative structure activity relationship analyses of substrates for CYP2B6.

To begin to build an understanding of the interactions of CYP2B6 with substrates, two different 3-dimensional quantitative structure activity relationship (3D-QSAR) models were constructed using 16 substrates of B-lymphoblastoid expressed CYP2B6. A pharmacophore model was built using the program Catalyst, which was compared with a partial least-squares (PLS) model using molecular surface-weighted holistic invariant molecular (MS-WHIM) descriptors. The Catalyst model yielded a 3-dimensional model of the common structural features of CYP2B6 substrates, whereas PLS MS-WHIM generated a model based on statistical analyses of molecular descriptors for size and shape of the substrate. The pharmacophore model obtained with Catalyst consisted of three hydrophobes and one hydrogen bond acceptor region. The cross-validated PLS MS-WHIM model gave a good q2 value of 0.607. Size, positive electrostatic potential, hydrogen bonding acceptor capacity, and hydrophobicity were found to be the most relevant descriptors for the model. These models were then used to predict the Km (apparent) values of a test set of structurally diverse substrates for CYP2B6 not included in the model building, specifically lidocaine, amitriptyline, bupropion, arteether, and verapamil. Overall, both 3D-QSAR methods yielded satisfactory Km (apparent) value predictions for the majority of the molecules in the test set. However, PLS MS-WHIM was unable to reliably predict the Km (apparent) value for verapamil, whereas Catalyst did not predict the Km (apparent) value for lidocaine. In both of these cases the residual of the Km (apparent) value was greater than one log unit. The strengths and limitations of both of these 3D-QSAR approaches are discussed.

MS-WHIM, new 3D theoretical descriptors derived from molecular surface properties: a comparative 3D QSAR study in a series of steroids.

The recently proposed WHIM (Weighted Holistic Invariant Molecular) approach [Todeschini, R., Lasagni, M. and Marengo, E., J. Chemometrics, 8 (1994) 263] has been applied to molecular surfaces to derive new 3D theoretical descriptors, called MS-WHIM. To test their reliability, a 3D QSAR study has been performed on a series of steroids, comparing the MS-WHIM description to both the original WHIM indices and CoMFA fields. The analysis of the statistical models obtained shows that MS-WHIM descriptors provide meaningful quantitative structure-activity correlations. Thus, the results obtained agree well with those achieved using CoMFA fields. The concise number of indices, the ease of their calculation and their invariance to the coordinate system make MS-WHIM an attractive tool for 3D QSAR studies.

A 3D QSAR CoMFA study of non-peptide angiotensin II receptor antagonists.

A series of non-peptide angiotensin II receptor antagonists was investigated with the aim of developing a 3D QSAR model using comparative molecular field analysis descriptors and approaches. The main goals of the study were dictated by an interest in methodologies and an understanding of the binding requirements to the AT1 receptor. Consistency with the previously derived activity models was always checked to contemporarily test the validity of the various hypotheses. The specific conformations chosen for the study, the procedures invoked to superimpose all structures, the conditions employed to generate steric and electrostatic field values and the various PCA/PLS runs are discussed in detail. The effect of experimental design techniques to select objects (molecules) and variables (descriptors) with respect to the predictive power of the QSAR models derived was especially analysed.

Substrate recognition by ribosome-inactivating protein studied by molecular modeling and molecular electrostatic potentials.

A computer model of dianthin 30, a type 1 ribosome-inactivating protein (RIP), is constructed by homology modeling using two known X-ray structures; a type 1 RIP, pokeweed antiviral protein (PAP), and chain A of a type 2 RIP, ricin. The 3D structure is refined by molecular dynamics and its binding site compared with those of PAP and ricin using molecular electrostatic potential mapping. The differences in the maps obtained clearly show how, despite the similarity of the topology of the binding site, differences in electrostatic potential can account for the experimentally observed differences in substrate recognition and binding. This demonstrates the potential of these techniques for guiding further experimental analyses.

Nonpeptide angiotensin II receptor antagonists. Synthesis, in vitro activity, and molecular modeling studies of N-[(heterobiaryl)methyl] imidazoles.

With the aim of explaining the influence of the structural changes on the biphenylic moiety on the activity, a series of N-[(heterobiaryl)methyl]imidazoles (I), constructed on the model of DuPont compounds by replacing either the central or terminal phenyl ring with a heteroaromatic one, such as furan, thiophene, thiazole, and pyridine, was synthesized. Compared to the reference DuPont compound (EXP-7711), all the heterobiaryl derivatives showed a reduced potency both in receptor binding (rat adrenal capsular membranes) and in the functional assay (angiotensin II-induced contraction of rabbit aorta strips). The lower activity was justified by the extensive molecular modeling studies, which took into consideration the conformational and electrostatic features of several heterobiaryl derivatives. On the basis of the results obtained, it was hypothesized that the central aromatic ring of the biarylic portion works as a spacer, orienting in the right way the terminal phenyl ring, whose electronic distribution is, instead, crucial to its fitting well with a lipophilic pocket at the receptor site.

A 3D QSAR approach to the search for geometrical similarity in a series of nonpeptide angiotensin II receptor antagonists.

A 3D QSAR methodology based on the combined use of conformational analysis and chemometrics was applied to perform a comparative analysis of the 3D conformational features of 13 nonpeptide angiotensin II receptor antagonists showing different levels of binding affinity. Conformational analysis by using a molecular mechanics MM2 method was carried out for each of these structures to obtain conformational minima. These minima were described by ten interatomic distances which define the relative spatial disposition of five significant atoms belonging to relevant functional groups present in all the 13 molecules. The structure-activity relationship between the interatomic distances and the biological activity was then assessed by using chemometric methods (cluster analysis, principal component analysis, classification methods). With our indirect approach based on the search for geometrical similarity it was possible, even though structural information on the receptor active site was lacking, to identify the 3D geometrical requirements for the binding affinity of nonpeptide angiotensin II receptor inhibitors.