Spring Meeting 2003

Lilly Research Centre
Windlesham, Surrey


  • Ron Knegtel (Vertex, UK)
    A structure-based approach to the design of genuinely broad Caspase inhibitors [Abstract]
  • Celine Lebailly (U. Catholique de Louvain, Belgium)
    A fast exchange algorithm for designing focused library design in lead optimisation [Slides] [Abstract]
  • Mark Mackey (Cresset, UK)
    Structureless virtual screening for novel small molecule leads [Slides] [Abstract]
  • Dave Morley (RiboTargets, UK)
    rDock: A virtual screening platform for hit identification and lead optimisation [Slides] [Abstract]
  • Nick Stiefl (Universitaet Wuerzburg, Germany)
    MaP: A 3D-QSAR technique based on the distribution of molecular surface properties – Applications, validation, and parameter settings [Slides] [Abstract]
  • Francesca Toschi (University of Southampton, UK)
    The computational investigation of protein/ligand complexes: ligand binding induced-fit [Slides] [Abstract]
  • Trudie Wright (University of Sheffield, UK)
    Multiobjective Optimisation of Combinatorial Libraries [Slides]


Presentation: Ron Knegtel
A structure-based approach to the design of genuinely broad Caspase inhibitors

Vertex, UK

Programmed cell death or apoptosis, plays a key role in normal development and a wide range disease states such as stroke, myocardial infarction, sepsis and many others. The Caspase family of cysteine proteases initiates and executes apoptosis in mammalian cells when triggered by receptor mediated signals or external stress stimuli [1]. Due to their pivotal role in programmed cell death, Caspase inhibition represents an attractive target for pharmacological intervention in diseases where apoptosis is the cause of tissue damage. To date, some 14 distinct Caspases have been identified [2] and the crystal structures of Caspase-1, -3, -7 and -8 have been determined [3,4,5].

The fact that multiple, homologous proteases cooperate to induce apoptosis raises the question as to which Caspase should be inhibited for optimal therapeutic effect. One solution to this dilemma is to design broad-spectrum, irreversible Caspase inhibitors that target the entire Caspase family rather than any single member. In contrast to what is commonly assumed, peptidic Caspase inhibitors such as ZVAD-FMK show a strong degree of preference for certain Caspases. In addition, small, structurally simple irreversible inhibitors can also show show marked selectivity within the Caspase family. We have used insights gained from our crystallographic studies on different Caspases to determine the structural requirements for obtaining broad Caspase inhibition. This approach has led to the discovery of potent small molecule inhibitors with a broad Caspase inhibitory profile that can prevent cell death both in vitro and in vivo models of apoptosis.

[1] S.M. Schwartz, Circulation 97, 227-229 (1998).

[2] M. Van de Craen., G. Van Loo et al. Cell Death Differ. 5, 838-846 (1998).

[3] K.P. Wilson, J.A. Black et al., Nature 370, 270-275 (1994).

[4] Y.Wei., T. Fox et al., Chem. Biol. 7, 423-432 (2000).

[5] J.M. Golec, M.D. Mullican et al., Bioorg. Med. Chem. Lett. 7, 2181-2186 (1997).

Presentation: Celine Lebailly
A fast exchange algorithm for designing focused library design in lead optimisation

U. Catholique de Louvain, Belgium [Slides]

Combinatorial chemistry is widely used in drug discovery. Once a lead compound has been identified, a series of R-groups and reagents can be selected and combined to generate new potential drugs. The combinatorial nature of this problem leads to chemical libraries containing usually a very large number of virtual compounds, far too large to permit their chemical synthesis. Therefore, one often wants to select a subset of “good” reagents for each R-group of reagents and synthesise all their possible combinations. In this research, one encounters some difficulties. First, the selection of reagents has to be done such that the compounds of the resulting sub-library simultaneously optimise a series of chemical properties. For each compound, we use a desirability function, proposed by Derringer and Suich, to summarise those propertiesin one fitness value. Then a loss function is used as objective criteria to globally quantify the quality of a sub-library. Secondly , there are a huge number of possible sub-libraries and one has to explore the solutions space as fast as possible. In our algorithm, we begin with a random solution and iterate by applying exchanges, a simple method proposed by Federov and often used in the generation of D-optimal designs. Those exchanges are guided by a weighting of the reagents adapted recursively as the solutions space is explored. Our algorithm is applied on a real database and appears to converge rapidly. Other algorithms presented in the combinatorial chemistry literature (Piccolo algorithm of SmithKline Beecham Pharmaceuticals and Ultrafast algorithm of D. Agrafiotis) are also applied and results are compared.

Keywords:drug discovery, combinatorial chemistry, hit-to-lead, library design, reagent selection, multiobjective optimisation.


[1] D.K. Agrafiotis. Multiobjective optimization of combinatorial libraries. IBM J. Res. Develop.45(3/4):545-566

[2] D.K. Agrafiotis and V.S. Lobanov. Ultrafast algorithm for designing focused combinatorial arrays. J. Chem. Info. Comput. Sci., 40:1030-1038, 2000.

[3] G.C. Derringer and R. Suich. Simultaneous optimization of several response variables. Journal of Quality Technology, 12(4):214-219, 1980.

[4] V.V. Federov. Theory of optimal experiments. Academic Press, NY, 1972.

[5] J.T. Saunders W. Zheng, S.T. Hung and G.L. Seibel. Multiobjective optimization of combinatorial libraries. Pac Symp Biocomput. pages 588-599, 2000.

Presentation: Mark Mackey
Structureless virtual screening for novel small molecule leads

Cresset, UK [Slides]

A new method for the rapid identification of novel small molecule inhibitors of biological targets will be presented. The surface and shape properties of a molecule are described as ‘field points’. Electrostatic, steric and hydrophobic field points are used in a specific 3D arrangement that reflects the shape of the protein binding site. These descriptors can be encoded as a 1D vector, which when combined with a similarity metric allows molecules’ field properties to be rapidly compared. This technique should be applicable for any target for which a putative binding conformation can be determined: no protein structure is necessary.

In a recent example, virtual screening of 600,000 commercially available molecules against a GPCR target gave a hit list of 88 compounds which were purchased and tested. Of these, 27 (30%) were shown to have an activity better than 10uM. Most of the hits had no significant structural similarity to any known class of actives at the target. Virtual screening using fields represents a major advance in the rapid identification of new small molecule leads for medicinal chemistry programs.

Presentation: Dave Morley
rDock: A virtual screening platform for hit identification and lead optimisation

RiboTargets, UK [Slides]

Virtual screening (VS) provides a viable route to structure-based hit identification and lead optimisation. In VS large numbers of compounds (1M) are docked rapidly to the structure of the active site of a target protein . The goal is to find as many diverse hit compounds as possible to provide choice of series for lead optimisation and to provide additional information about molecular scaffolds that satisfy the chemical and spatial requirements of the binding site. We describe the major components of a virtual screening system with particular reference to our proprietary docking platform, rDock:

1. Management of a large database of available, lead-like compounds. At RiboTargets our 3.4M compound collection is reduced to a 1.2M compound docking library via substructure, ADMET, and lead-like filtering.

2. Preparation of the docking site. This critical step includes selection of appropriate side-chain conformations and protonation states around the active site of the protein, and mapping of the cavity to identify the extent of the docking calculation.

3. Docking and scoring. Efficient high-throughput docking protocols have been developed that combine GA, Monte Carlo and Simplex methods with empirical scoring functions.

4. Detailed analysis and post-docking filtering of the best scoring virtual hits. Compounds are selected for assay based on score components, docking mode, and diversity characteristics. We have developed a graphical filtering tool to assist in this task, that implements a set of filters for automated removal of likely false positives.

5. Acquisition of (typically about) 1000 compounds for assay. Because of the small numbers, we can use more sensitive and biologically relevant assays than would be possible in conventional HTS.

These methodological developments will be presented together with results demonstrating the effectiveness of the calculations in drug discovery.

Presentation: Nick Stiefl
MaP: A 3D-QSAR technique based on the distribution of molecular surface properties – Applications, validation, and parameter settings

Universitaet Wuerzburg, Germany [Slides]

An alignment-independent molecular descriptor called MaP (Mapping Property distributions onto the molecular surface) is presented. A three step procedure is used to calculate the MaP descriptor. First, an approximation to the molecular surface with equally distributed surface points is computed. Next, molecular properties are projected onto this surface. Finally, the distribution of surface properties is encoded into a translationally and rotationally invariant molecular descriptor which is based on distance dependent count statistics. The calculated descriptor is then correlated with biological data through chemometric regression techniques, in combination with a method of identifying variables that are highly relevant for the model and hence for its interpretation. These variables are back-projected into the original molecular space which allows a straightforward interpretation of the computed models.

A short introduction to the new descriptor and several applications are given. All models built with MaP are not only highly predictive but interpretation of the back-projected variables led to biologically and chemically relevant conclusions. Next, techniques for the validation of the variable selection procedure are outlined. Finally, an expansion of the MaP parameter set is presented. It can be shown that the default parameter set of MaP sufficiently describes relevant features of biologically active compounds. Moreover, MaP is robust with respect to parameter changes and generates consistent models for different parameter sets. However, additional parameters can improve interpretation of the resulting model

Presentation: Francesca Toschi
The computational investigation of protein/ligand complexes: ligand binding induced-fit

University of Southampton, UK [Slides]

Ligand binding may involve a wide range of structural changes in the receptor protein, from hinge movements of entire domains to small side-chain rearrangements in the binding pocket residues. While changes in the backbone of proteins are sometimes negligible, side-chain rearrangements are generally observed. Knowledge of the extent to which side-chain conformational changes occur is very important to improve drug design and docking prediction algorithms, particularly since most of these algorithms adopt a rigid receptor hypothesis. Different methods to characterize side-chain conformational changes occurring in protein-ligand complexes were employed. A dataset of PDB structures was chosen and apo-/holo- and holo-/holo- protein pairs’ torsion angles compared. Side-chain conformational changes were defined on the basis of both constant[1] and environment- and residue- dependent[2] thresholds. Also, recently published rotamer libraries[3] were employed and the probabilities of the different rotamers found compared. The results of the analysis provide insights into the intrinsic flexibilities of protein active sites, and the extent of side-chain conformational change on ligand binding. The general patterns and features are potentially useful for the prediction of conformational changes occurring in proteins upon ligand binding.

[1] R. Najmanovich, J. Kuttner, V. Sobolev, and M. Eldman, Prot. Struct. Funct. Genet., 39, 261, (2000).

[2] S. Zhao, D. S. Goodsell, and A. J. Olson, Prot. Struct. Funct. Genet., 43, 271, (2001).

[3] R. L. Dunbrack, and F. E. Cohen, Prot. Sci., 6, 1661, (1997).

Presentation: Trudie Wright
Multiobjective Optimisation of Combinatorial Libraries

University of Sheffield, UK [Slides]