Paul M. Selzer

Library Design for Fragment Based Screening

According to Hanns model of molecular complexity an increased probability of detection binding to a target protein can be expected when small, low complex molecular fragments are screened with high sensitivity instead of full-sized ligands with lower sensitivity. Analysis of the HTS summary data of Novartis and comparison with NMR screening results obtained on generic fragment libraries indicate this expectation to be true with hitrates of 0.001% - 0.151% observed in the identification of ligands with an IC(50) threshold in the micromolar range in an HTS setup and hitrates above or equal to 3% observed in NMR screening of fragments with an affinity threshold in the millimolar range. It is however necessary to keep in mind that the sets of target studied were not identical for both method and the experience in NMR screening is too limited for a final conclusion. The term hitrate as used here reflects only the success rate in the observation of ligand binding event. It must not be confused with the overall success rate of fragment and high throughput screening in the lead finding process, which can be entirely different, since the steps required to follow-up a ligand binding event to a lead are different for both methods. A survey of fragment-based lead discovery case studies given in the literature shows that in approximately half of the cases the initial hit fragment was discovered by screening a generic library, whereas in the other cases some knowledge about an initial ligands or the protein binding site has been used, whereas systematic virtual screening of fragment databases has been only rarely reported. As comparatively high hitrates were obtained, further consideration to optimize the generic fragment screening library were directed to the chemical tractability of the fragment. As several functional groups preferred by chemists for modification and linking of the fragments are also preferentially involved in interactions between the fragments and the target protein, a set of screening fragments was derived from chemical building blocks by masking its linker group by a chemical transformation which can be later on used in the chemical follow-up of the fragment hit. For example primary amines can be masked as acetamides. If the screening fragment is active the related building block can then be used for synthesis of a follow-up library.

Relationships between Molecular Complexity, Biological Activity, and Structural Diversity

Following the theoretical model by Hann et al. moderately complex structures are preferable lead compounds since they lead to specific binding events involving the complete ligand molecule. To make this concept usable in practice for library design, we studied several complexity measures on the biological activity of ligand molecules. We applied the historical IC50/EC50 summary data of 160 assays run at Novartis covering a diverse range of targets, among them kinases, proteases, GPCRs, and protein-protein interactions, and compared this to the background of inactive compounds which have been screened for 2 years but have never shown any activity in any primary screen. As complexity measures we used the number of structural features present in various molecular fingerprints and descriptors. We found generally that with increasing activity of the ligands, their average complexity also increased, and we could therefore establish a minimum number of structural features in each descriptor needed for biological activity. Especially well suited in this context were the Similog keys and circular substructure fingerprints. These are those descriptors, which also perform especially well in the identification of bioactive compounds by similarity search, suggesting that structural features encoded in these descriptors have a high relevance for bioactivity. Since the number of features correlates with the number of atoms present in the molecule, also the number of atoms serves as a reasonable complexity measure and larger molecules have, in general, higher activities. Due to the relationship between feature counts and densities on one hand and biological activity on the other, the size bias present in almost all similarity coefficients becomes especially important. Diversity selections using these coefficients can influence the overall complexity of the resulting set of molecules, which has an impact on the biological activity that they exhibit. Using sphere-exclusion based diversity selection methods, such as OptiSim together with the Tanimoto dissimilarity, the average feature count distribution of the resulting selections is shifted toward lower complexity than that of the original set, particularly when applying tight diversity constraints. This size bias reduces the fraction of molecules in the subsets having the complexity required for a high, submicromolar activity. None of the diversity selection methods studied, namely OptiSim, divisive K-means clustering, and self-organizing maps, yielded subsets covering the activity space of the IC50 summary data set better than subsets selected randomly.

Estimation of pKa for druglike compounds using semiempirical and information-based descriptors.

A pragmatic approach has been developed for the estimation of aqueous ionization constants (pKa) for druglike compounds. The method involves an algorithm that assigns ionization constants in a stepwise manner to the acidic and basic groups present in a compound. Predictions are made for each ionizable group using models derived from semiempirical quantum chemical properties and information-based descriptors. Semiempirical properties include the partial charge and electrophilic superdelocalizabilty of the atom(s) undergoing protonation or deprotonation. Importantly, the latter property has been extended to allow predictions to be made for multiprotic compounds, overcoming limitations of a previous approach described by Tehan et al. The information-based descriptions include molecular-tree structured fingerprints, based on the methodology outlined by Xing et al., with the addition of 2D substructure flags indicating the presence of other important structural features. These two classes of descriptor were found to complement one another particularly well, resulting in predictive models for a range of functional groups (including alcohols, amidines, amines, anilines, carboxylic acids, guanidines, imidazoles, imines, phenols, pyridines, and pyrimidines). A combined RMSE of 0.48 and 0.81 was obtained for the training set and an external test set compounds, respectively. The predictive models were based on compounds selected from the commercially available BioLoom database. The resultant speed and accuracy of the approach has also enabled the development of Web application on the Novartis intranet for pKa prediction.

Biodiversity of small molecules – a new perspective in screening set selection

How is the diversity of a compound set defined and how is the most appropriate compound subset identified for assay when screening the entire HTS deck is not an option? A common approach has so far been to cover as much of the chemical space as possible by screening a chemically diverse set of compounds. We show that, rather than chemical diversity, the biologic diversity of a compound library is an essential requirement for hit identification. We describe a simple and efficient approach for the design of a HTS library based on compound-target diversity. Biodiverse compound subsets outperform chemically diverse libraries regarding hit rate and the total number of unique chemical scaffolds present among hits. Specifically, by screening ~19% of a HTS collection, we expect to discover ~50-80% of all desired bioactive compounds.

Web-based cheminformatics tools deployed via corporate Intranets

Abstract Web-based molecular processing tools installed on corporate Intranets bring easy to use cheminformatics and molecular modeling capabilities directly to the desks of synthetic chemists, giving them comfortable access to data and their visualization and analysis, considerably improving efficiency of the drug design and development process. User-friendly tools that use a standard Web browser as an interface allow users access to a broad range of expert molecular processing tools and techniques, without the need for extensive expertise in their use.

Clustering and rule-based classifications of chemical structures evaluated in the biological activity space.

Classification methods for data sets of molecules according to their chemical structure were evaluated for their biological relevance, including rule-based, scaffold-oriented classification methods and clustering based on molecular descriptors. Three data sets resulting from uniformly determined in vitro biological profiling experiments were classified according to their chemical structures, and the results were compared in a Pareto analysis with the number of classes and their average spread in the profile space as two concurrent objectives which were to be minimized. It has been found that no classification method is overall superior to all other studied methods, but there is a general trend that rule-based, scaffold-oriented methods are the better choice if classes with homogeneous biological activity are required, but a large number of clusters can be tolerated. On the other hand, clustering based on chemical fingerprints is superior if fewer and larger classes are required, and some loss of homogeneity in biological activity can be accepted.

Pyruvate Fermentation by Oenococcus oeni and Leuconostoc mesenteroides and Role of Pyruvate Dehydrogenase in Anaerobic Fermentation

ABSTRACT The heterofermentative lactic acid bacteria Oenococcus oeni and Leuconostoc mesenteroides are able to grow by fermentation of pyruvate as the carbon source (2 pyruvate → 1 lactate + 1 acetate + 1 CO2). The growth yields amount to 4.0 and 5.3 g (dry weight)/mol of pyruvate, respectively, suggesting formation of 0.5 mol ATP/mol pyruvate. Pyruvate is oxidatively decarboxylated by pyruvate dehydrogenase to acetyl coenzyme A, which is then converted to acetate, yielding 1 mol of ATP. For NADH reoxidation, one further pyruvate molecule is reduced to lactate. The enzymes of the pathway were present after growth on pyruvate, and genome analysis showed the presence of the corresponding structural genes. The bacteria contain, in addition, pyruvate oxidase activity which is induced under microoxic conditions. Other homo- or heterofermentative lactic acid bacteria showed only low pyruvate fermentation activity.

Applications of Self-Organizing Neural Networks in Virtual Screening and Diversity Selection†

Artificial neural networks provide a powerful technique for the analysis and modeling of nonlinear relationships between molecular structures and pharmacological activity. Many network types, including Kohonen and counterpropagation, also provide an intuitive method for the visual assessment of correspondence between the input and output data. This work shows how a combination of neural networks and radial distribution function molecular descriptors can be applied in various areas of industrial pharmaceutical research. These applications include the prediction of biological activity, the selection of screening candidates (cherry picking), and the extraction of representative subsets from large compound collections such as combinatorial libraries. The methods described have also been implemented as an easy-to-use Web tool, allowing chemists to perform interactive neural network experiments on the Novartis intranet.

Web-based cheminformatics and molecular property prediction tools supporting drug design and development at Novartis

Web-based tools offer many advantages for processing chemical information, most notably ease of use and high interactivity. Therefore more and more pharmaceutical companies are using web technology to deliver sophisticated molecular processing tools directly to the desks of their chemists, to assist them in the process of designing and developing new drugs. In this paper, the web-based cheminformatics system developed at Novartis and currently used by more than thousand users is described. The system allows various molecular modeling and molecular processing tasks, including the calculation of molecular and substituent properties, property-based virtual screening, visualization of molecules, bioisosteric design, diversity analysis, and support of combinatorial chemistry. The methodology to calculate various molecular properties relevant to drug design is described, including the prediction of intestinal absorption, blood–brain barrier penetration, efflux, and water solubility. Information about the web technology used is also provided.

Comparison of Multivariate Data Analysis Strategies for High-Content Screening

High-content screening (HCS) is increasingly used in biomedical research generating multivariate, single-cell data sets. Before scoring a treatment, the complex data sets are processed (e.g., normalized, reduced to a lower dimensionality) to help extract valuable information. However, there has been no published comparison of the performance of these methods. This study comparatively evaluates unbiased approaches to reduce dimensionality as well as to summarize cell populations. To evaluate these different data-processing strategies, the prediction accuracies and the Z′ factors of control compounds of a HCS cell cycle data set were monitored. As expected, dimension reduction led to a lower degree of discrimination between control samples. A high degree of classification accuracy was achieved when the cell population was summarized on well level using percentile values. As a conclusion, the generic data analysis pipeline described here enables a systematic review of alternative strategies to analyze multiparametric results from biological systems.