Spring Meeting 2008

Lilly UK
Windlesham, Surrey

Molecular recognition is a complex mix of interacting molecules and events. Chris Hunter (University of Sheffield) will present a semiquantitiative approach to understanding solvent effects on molecular interactions, while Fred Ludlow (University of Cambridge) will provide fundamental insights into the organisational principles of molecular networks using his modeling of dynamic combinatorial libraries. The dynamic nature of proteins and the ability to deduce better protein-ligand interaction models using inverse structure-based design will be discussed by Linda Hirons (Lilly). Describing purely small molecule interactions Jon Goodman (University of Cambridge) under the title “change we can believe in: what is your reaction?” will describe computational tools that give us more confidence in predicting and understanding chemical reactions while Hina Patel (University of Sheffield) will present a knowledge-based approach to de novo design using reaction vectors gleaned from reaction data. Leading the session on application in Pharma, Paul Leeson (AZ) will share his analysis of the influence of drug-like concepts on decision-making in medicinal chemistry. Peter Johnson (University of Leeds/Symbiosys) will describe LASSO – ligand activity by surface similarity order. The meeting will be wrapped up by Andrew Leach (AZ) who discusses his matched molecular pair analysis as a tool for lead optimisation.


  • Jon Goodman (University of Cambridge)
    Change we can believe in: What is your reaction? [Slides] [Abstract]
  • Linda Hirons (Lilly)
    Inverse structure based design [Slides] [Abstract]
  • Chris Hunter (University of Sheffield)
    Solvent effects on molecular interactions [Slides] [Abstract]
  • Peter Johnson (SimBioSys)
    LASSO – ligand activity by surface similarity order: a new tool for ligand based virtual screening applied to 18 million structures [Slides] [Abstract]
  • Andrew Leach (AstraZeneca, Alderley Park)
    The Effects of Structural Changes on Molecular Properties: Matched Molecular Pair Analysis as a Tool for Lead Optimisation [Slides] [Abstract]
  • Paul Leeson (AstraZeneca)
    The influence of drug like concepts on decision making in medicinal chemistry [Abstract]
  • Fred Ludlow (University of Cambridge)
    Dynamic combinatorial chemistry – models and applications [Abstract]
  • Hina Patel (University of Sheffield)
    Knowledge based de novo design using reaction vectors [Slides] [Abstract]


Presentation: Jon Goodman
Change we can believe in: What is your reaction?

University of Cambridge [Slides]

Organic chemistry has been studied using a reasonably consistent language for more than a century. Why is it still so hard? We have studied ways of automatically gathering and analysing chemical information, in order to discover how we can best make use of it, and how certain we can be about data from the literature. We have analysed detailed reaction mechanisms to explain reactivity and stereoselectivity, including 1,5 stereoinduction and counter-intuitive effects in polymerisation. We have discovered there are simple molecules that can never be synthesised. We are now trying to put all of these results together to produce computational tools that give us more confidence in predicting and understanding chemical changes.

Presentation: Linda Hirons
Inverse structure based design

Lilly [Slides]

Obtaining a predictive QSAR from structure-based design is often challenging and can be compounded by the dynamic nature of proteins (1). Different ligand series tend to stabilise different receptor conformations and strategies such as ‘induced fit’ or molecular simulation have been developed that target this problem. However, such techniques can ultimately be too time consuming. An Inverse Structure Based Design (iSBD) approach has been developed that uses SAR data to drive towards and identify the most appropriate conformation of a protein target for a particular ligand series. This novel Protein Prediction Pipeline has been built using the Konstanz Information Miner (2).

Docking programs are highly successful at generating the correct binding mode. However, when analysing protein-ligand interactions often little correlation exists between the docking score and activity. Correlations with binding affinity remain poor, even when the scores are calculated directly from the experimentally determined protein-ligand structures (3). A minute structural difference between molecules can be responsible for a dramatic shift in activity, which a single score cannot be expected to accurately encode and recognise. An alternative method of describing the ligand poses is clearly needed, in order to build predictive SARs. This work uses structural interaction fingerprints (4) to encode many of the necessary interactions present between the ligand and the protein.

Protein-ligand interaction tree models are built from the fingerprint-activity data and provide a useful visual aid that medicinal chemists can utilise. Even when only small conformational differences are present, the pipeline is able to identify the most appropriate conformation of a protein target to which a particular ligand series binds.


  1. Wildman and Kern. (2007) Nature 450, 966-972
  2. Konstanz Information Miner.
  3. Warren et al. (2006) J. Med. Chem., 49, 5912-5931
  4. Deng et al. (2004) J. Med. Chem., 47, 337-344


Presentation: Chris Hunter
Solvent effects on molecular interactions

University of Sheffield [Slides]

Molecular recognition events in solution are affected by many different factors that have hampered the development of an understanding of intermolecular interactions at a quantitative level. Our tendency is to partition these effects into discrete phenomenological fields that are classified, named, and divorced: aromatic interactions, cation-p interactions, CH_O hydrogen bonds, short strong hydrogen bonds, and hydrophobic interactions to name a few. To progress in the field, we need to develop an integrated quantitative appreciation of the relative magnitudes of all of the different effects that might influence the molecular recognition behavior of a given system. In an effort to navigate undergraduates through the vast and sometimes contradictory literature on the subject, I have developed an approach that treats theoretical ideas and experimental observations about intermolecular interactions in the gas phase, the solid state, and solution from a single simplistic viewpoint. The key features are outlined, and although many of the ideas will be familiar, the aim is to provide a semiquantitative thermodynamic ranking of these effects in solution at room temperature.

Presentation: Peter Johnson , Darryl Reid, Aniko Simon, Zsolt Zsoldos, Darryl Reid, Antony Williams
LASSO – ligand activity by surface similarity order: a new tool for ligand based virtual screening applied to 18 million structures

SimBioSys [Slides]

LASSO is a similarity searching tool that uses descriptors to find molecules with diverse chemical scaffolds but similar surface properties. Based on the idea that ligands must have surface properties compatible with the target site in order to bind, LASSO uses a descriptor using 23 different surface point types, ranging from hydrogen bond donors/acceptor, to hydrophobic sites, to pi-stacking interactions. This descriptor contains no 2D skeletal information about the molecule, no 3D geometrical relationships (shape) and very little volume information. However it is this “fuzziness” that allows LASSO to perform well as a scaffold hopping tool. LASSO will retrieve molecules, from a screened database, that have similar surface properties to those of a query set of known actives, regardless of the underlying scaffolds.

ChemSpider is a free access online database of over 18 million molecules provided as the basis of a structure centric community for chemists. LASSO has been applied to provide the virtual screening results for 40 target receptor families taken from the Database of Useful Decoys (DUD) and chosen to cover a wide range of receptor classes due to their interest in drug discovery. Similarity screening was performed on the full ChemSpider database across all targets and the similarity scores for each structure/target pair has been made available via the ChemSpider website providing a similarity score (based on surface properties) relative to actives of each of the 40 target receptors. In addition to allowing instant ranking results for your particular target of interest (retrieving molecules that are likely to be active for your receptor) this matrix of screening results can be used to find molecules that have high predicted affinity for your target but low predicted affinity for all other targets. Performing such searches promises to improve selectivity and can be a guide to reducing toxicity concerns. This presentation will provide an overview of the technology underlying both LASSO and ChemSpider and discuss the value of this offering to scientists in identifying potential leads via this virtual screening approach. We will discuss validation of the approach and early results from providing such information to the public.

Presentation: Andrew Leach
The Effects of Structural Changes on Molecular Properties: Matched Molecular Pair Analysis as a Tool for Lead Optimisation

AstraZeneca, Alderley Park [Slides]

Predicting many of the properties of a molecule that are of interest during lead optimisation is extremely challenging. Furthermore, it is often of limited value to a medicinal chemistry programme where the interest is directed towards knowing the likely effect of making changes to a lead compound’s structure. Matched molecular pairs analysis of databases of property measurements yields the most general idea of what effect a given structural change has made to that property. This historical track record provides an estimate of what effect that same structural change is likely to have when applied to a lead compound. Just as critically, the spread in the range of changes observed previously sets a range for the prediction. The outliers in the historical distribution can also be instructive as they frequently indicate ways in which the general trend can be overcome. The approach has been applied to a number of in-house and public domain databases and some of the more surprising observations will be presented.

Presentation: Paul Leeson
The influence of drug like concepts on decision making in medicinal chemistry


Despite the wide acceptance of drug-like principles such as the ‘rule of five’, this analysis of molecules currently being synthesized in leading pharmaceutical companies reveals that their physical properties differ significantly from those of recently discovered oral drugs. The marked increase in lipophilicity in particular could increase the likelihood of attrition in drug development. The application of guidelines linked to the concept of drug-likeness, such as the ‘rule of five’, has gained wide acceptance as an approach to reduce attrition in drug discovery and development. However, despite this acceptance, analysis of recent trends reveals that the physical properties of molecules that are currently being synthesized in leading drug discovery companies differ significantly from those of recently discovered oral drugs and compounds in clinical development. The consequences of the marked increase in lipophilicity – the most important drug-like physical property – include a greater likelihood of lack of selectivity and attrition in drug development. Tackling the threat of compound-related toxicological attrition needs to move to the mainstream of medicinal chemistry decision-making.

Presentation: Fred Ludlow
Dynamic combinatorial chemistry – models and applications

University of Cambridge

Dynamic combinatorial chemistry is a relatively new technique which has been applied to, amongst other things, the discovery of synthetic receptors for small molecules and of ligands for biomolecules. A dynamic combinatorial library of interconverting candidate molecules is prepared under thermodynamic control – if a target molecule is added to this library, those candidates which bind to it will be stabilised, and the equilibrium will shift to generate more of them. By analysing the concentrations of library members before and after addition of a target, the good binders can be identified.

We have applied this technique to the development of a nanomolar affinity receptor for spermine which is capable of decomplexing spermine from it’s biological host, DNA. We have also demonstrated that, by building models of these equilibrium systems and fitting them to experimental data, the affinities of the library members for a target can be determined directly from analysis of the whole library without the need for isolation of individual library members.

Presentation: Hina Patel
Knowledge based de novo design using reaction vectors

University of Sheffield [Slides]

A number of de novo design tools have been described with the aim of generating novel molecules for drug design, however, they are limited in their ability to propose molecules which are synthetically feasible. Here we describe how a novel method that utilises reaction vectors from databases of known reactions can generate structures of interest in the pipelining environment KNIME (1).

The reaction vector can capture the changes that take place at the reaction centre, without the need for complex reaction mapping procedures (2). By first describing the individual components of a reaction using descriptors such as atom pairs, the overall reaction vector is generated using:

Reaction Vector = [Sum of product vectors] – [Sum of reactant vectors]

We are able to show how reaction vectors can be used in both simple transformations involving, for example, a simple functional group substitution, to more complex multi-component reactions of the form (R1 + R2 >> P1 + P2), to generate novel molecules for synthesis. We demonstrate the application of the method to the design of known drugs from simple starting materials and a ‘cleaned’ reaction dataset, via mixing and matching of reaction transforms and reactants.

We also describe the how the method can be developed into an automated multi-objective application for de novo design.


  1. Konstanz Information Miner.
  2. Broughton, H. B. et al. Methods for Classifying and Searching Chemical Reactions. United States Patent Application 367550, 25 Sept, 2000.