Spring Meeting 2004

Pfizer Global R&D
Sandwich, UK


  • Gavin Harper (GlaxoSmithKline, UK)
    Successes and pitfalls in the data mining of HTS data [Slides] [Abstract]
  • Chris Higgs (Argenta, UK)
    Hit-finding Against GPCR Targets: Discovery of Potent MCH-1 Receptor Antagonists by Virtual Screening [Slides] [Abstract]
  • Adam I. Ibrahim (University College London, UK)
    Prediction of Drug Tissue-Distribution in Human and Rats [Slides] [Abstract]
  • Mark Johnson (Pannanugget, USA)
    Specifying and Searching for Ring System and Functional Group Environments Using Concepts of Molecular Equivalence Analysis [Slides] [Abstract]
  • Daniel F. Ortwine (Pfizer Ann Arbor, USA)
    Modelling on the Web: Enabling Chemists to Pursue Drug Design Calculations [Slides] [Abstract]
  • Robert S. Pearlman (University of Texas and Optive Research, Austin TX, USA)
    Distinguishing between the concepts of chemical compound and chemical structure [Abstract]
  • Matt Wood (AstraZeneca Alderley Park, UK)
    Construction of QSAR models for passive permeability from a Parallel Artificial Membrane Permeability Assay (PAMPA) dataset [Slides] [Abstract]


Presentation: Gavin Harper
Successes and pitfalls in the data mining of HTS data

GlaxoSmithKline, UK [Slides]

High-throughput screening is the mainstay of lead discovery efforts in the pharma industry and the vast volumes of data generated provide a large resource for data mining. However, high-throughput screening is a highly complex process involving many stages and departments. In this presentation we discuss the whole high-throughput screening process, from determining which compounds should enter the screening collection, through compound supply and screening, to analysis at the chemist’s desktop. We examine issues at each stage and some of the solutions put in place at GSK to improve the quality of the high-throughput screening process and the quality of leads that result. Methodologies developed to help chemists visualise and interpret the results are discussed. Examples from GSK screening campaigns will be used to highlight the specific issues discussed.

Presentation: Chris Higgs
Hit-finding Against GPCR Targets: Discovery of Potent MCH-1 Receptor Antagonists by Virtual Screening

Argenta, UK [Slides]

Melanin-concentrating hormone (MCH) has been known to be an appetite-stimulating peptide for a number of years. However, it has only been since the discovery that MCH is the ligand for an orphan G-protein coupled receptor, designated MCH-1R, which has been shown to mediate the effects of MCH on appetite and body weight, that drug discovery programmes have been able to exploit this information to look for MCH-1R antagonists, in particular for the treatment of obesity. In this presentation, we report the discovery of multiple, drug-like and structurally distinct series of MCH-1R antagonists using multiple virtual screening techniques. The most potent of the initial hit compounds demonstrated an IC50 value of 55nM in the primary screen and exhibited antagonist properties in a functional assay measuring Ca2+ release. Follow-up similarity searches around this compound yielded additional hits, some with increased potency. A homology model of the MCH-1R was constructed and a binding mode proposed for a representative hit compound. The time from the commencement of the project to the start of hit-to-lead optimization was only 6 months, demonstrating that virtual screening is a rapid and cost-effective approach to hit-finding for GPCR targets.

Presentation: Adam I. Ibrahim
Prediction of Drug Tissue-Distribution in Human and Rats

University College London, UK [Slides]

The present project involves the collection of drug blood/ tissue distribution data obtained from a wide range of literature papers published in various journals. These published data are analysed carefully in order to obtain values of the distribution partition coefficients of drugs between blood/ plasma and various tissues. Once enough data for distribution to a given tissue have been collected, they are used to obtain predictive equations using the Abraham salvation equation. A necessary requirement for this is the determination of descriptors for the drugs in question. These predictive methods using the Abraham equation may than be used in the first stages of drug design. These predictions can be made from structure before compounds are synthesised. Compounds with poor tissue distribution, can then be identified. The Abraham salvation equation is stated below as:

SP = c + e.E + s.S + a.A + b.B + v.V

Here SP is a dependent variable of the system (for example log P, the blood to tissue partition coefficient for series of solutes), the coefficients of the equation are c, e, s, a, b, v and E, S, A, B and V are the solute descriptors (see table below). The equation constant is defined by c.

E: The excess molar refraction in units of (cm3 mol-1)/10, which reflects solute polarisabilty. S: The solute dipolar/ polarisabilty parameter. A: The solute hydrogen bond acidity summation parameter. B: The solute hydrogen bond basicity summation parameter. V: McGowan’s characteristic volume in units of (cm3 mol-1)/100 Represents the three-dimensional space occupied by the solute.

At the moment blood/ tissue data in human subjects have been obtained for a number of tissues.

Presentation: Mark Johnson
Specifying and Searching for Ring System and Functional Group Environments Using Concepts of Molecular Equivalence Analysis

Pannanugget, USA [Slides]

Nilakantan et al. (JCICS, 30(1990)65) presented a method for searching for ring systems using hash codes. This concept of recognizing disjoint components of molecular structures and assigning “names” to those components is basic to the idea of molecular equivalence analysis. A list of the ring-system names for a structure constitutes a “value” of a variable we term a molecular equivalence index. One searches for a ring-system by performing a substring search over the values of the index using the “name” of the desired ring system for the query substring. This concept is easily extended to other families of structural components such as cyclic systems, side chains and functional groups (See, for example, Bemis & Murcko, J. Med. Chem., 39(1996) 2887; Xu & Johnson, JCICS, 42(2002)912) . After presenting an overview of these extensions, we illustrate how local and global environmental information can be incorporated into the “names” assigned the components. We show how molecular equivalence searching complements substructure and similarity searching, and we illustrate the incorporation of local and global environmental information into ones ring system and functional group searches. We briefly demo a program (Meqi), easily incorporated into other software such as batch files and Pipeline Pilot, that we have developed for specifying an endless variety of molecular equivalence indices and use it to illustrate how one can quickly gain a useful overview to the most frequent or most infrequent ring system or functional group environments in a collection of structures.

Presentation: Daniel F. Ortwine
Modelling on the Web: Enabling Chemists to Pursue Drug Design Calculations

Pfizer Ann Arbor, USA [Slides]

As part of an initiative to place modeling tools usually available only on UNIX machines directly in the hands of medicinal chemists and biologists, we have developed a number of web-based tools for drug design. ChemSelect facilitates ‘smart’ multi-database searches for chemical intermediates. WebPK performs physicochemical and molecular property calculations useful for predictions of ADME profiles. WebDock runs protein/ligand docking experiments ‘behind the scene’ on servers, followed by visualization of results in 3D. WebAlign performs small molecule alignments to reference template structure(s) extracted from pharmacophore models, followed by 3D visualization in the active and excluded volumes from the model. WebQSAR predicts the potency of input structures using predefined HQSAR and CoMFA models. All applications are run from PC desktops using standard internet browsers, and all contain import/export and local save capabilities. The overall setup and examples of the use of each will be shown.

Presentation: Robert S. Pearlman
Distinguishing between the concepts of chemical compound and chemical structure

University of Texas and Optive Research, Austin TX, USA

We all know that chemical compounds can exist in various protonation states and in various tautomeric states. We all know that a compound’s environment (solvent, membrane, receptor, etc.) determines the protomeric state which the compound is most likely to adopt. Experimentally measured properties reflect Mother Nature’s choice of which protomeric state(s) – i.e., which structure(s) – predominate and determine the measured property value. Regrettably, we often tend to forget these facts when attempting to predict behaviors of compounds by performing calculations based on the structure we or some other human chose to associate with the compound.

We need to be able to enumerate and consider the various structures which a compound might exhibit in different Natural environments. We need to be able to associate measured data with compounds and we need to associate computed data with the particular structures used for the computations. This presentation will introduce algorithms and software tools for both bench chemists and cheminformaticians which address these needs and others.

Presentation: Matt Wood
Construction of QSAR models for passive permeability from a Parallel Artificial Membrane Permeability Assay (PAMPA) dataset

AstraZeneca Alderley Park, UK [Slides]

It is now well established in the Pharmaceutical industry that drug candidates possessing inadequate pharmacokinetic properties have a high risk of failure during the development process. The mechanisms of drug discovery have consequently been reorganised to facilitate early incorporation of property-based lead optimisation. One component of this process is a refined in-silico or in-vitro model designed to make an accurate prediction of intestinal absorption for orally delivered compounds. There are numerous examples of such models in the literature and this work focuses on our version of the in-vitro Parallel Artificial Membrane Permeability Assay (PAMPA).

The PAMPA allows the passive permeability of compounds to be measured through a membrane mimic. Its high level of versatility means that research into many assay derivatives and applications has been published. Our version has been extensively validated and optimised towards the correlation between apparent PAMPA permeability and the fraction of oral dose absorbed in humans (%Fa).

More than 1000 compounds from the AstraZeneca compound collection have been screened for passive permeability via the optimised PAMPA system. The resulting dataset has been used to construct QSAR models for the in-silico prediction of passive PAMPA permeability. In agreement with early QSAR work on permeability and pharmacological models, PAMPA permeability was observed to exhibit a bi-linear dependence upon measured logD7.4. As a result, the best models have been produced by means of the binary recursive partitioning method in CART. The RMSE values of a consensus CART model of three independent descriptor sets are 0.32 in training and 0.52 in test from a randomised split of the available data. However, a less accurate level of prediction for low permeability compounds highlights the importance of high property space coverage in the source dataset. This important disparity is due to a relative lack of training, and therefore compounds, in the low permeability region.