Autumn Meeting 2007

Stevenage, UK

Registration is now open for the autumn meeting of the UK-QSAR and Chemoinformatics group which is to be hosted by GlaxoSmithKline at their Stevenage Research Centre in Hertfordshire.

The meeting will cover a broad range of topics, but the two themes of ADMET and Automated/rapid (Q)SAR are very strong.

Giampa Bravi (GSK) and Jos Lommerse (Organon) have agreed to present the SAR tools they have been developing to allow their Medicinal Chemists rapid views/understanding of SAR data; Bernd Wendt (Tripos) will be discussing real-life experiences with the new Topomeric CoMFA, a quick and possibly painless 3D QSARmethod; Val Gillet (University of Sheffield) will be presenting on methods to design compound libraries with good developability properties; Stephen Curry (Imperial College) will be describing his use ofcrystallisation studies to explore the molecular detail of drug-HSA interactions; Prof. David Leahy (Newcastle University) and Olga Obrezanova (Biofocus) will each describe the work they have done to developautomated methods for QSAR, and Paul Gleeson (GSK) will conclude the meeting with a large analysis of developability data and his resulting “rules of thumb” to help balance Potency with ADMET.

For all general enquiries about the meeting please contact the organisers using


  • Francis Atkinson (Computational and Strcutural Sciences, GlaxoSmithKline, Harlow, UK)
    The SAR Toolkit: simple tools for the analysis of Structure-Activity Relationships [Slides] [Abstract]
  • Stephen Curry (Biophysics Section, Imperial College, UK)
    Crystallographic analysis of drug binding to human serum albumin [Slides] [Abstract]
  • Val Gillet (University of Sheffield)
    Designing Arrays Optimised on Multiple Properties [Slides]
  • Paul Gleeson (GSK, UK)
    Target Affinity, ADMET & PhysChem Properties: The need for a better balance [Slides] [Abstract]
  • David E Leahy (Molecular Informatics Group, Newcastle University, UK,
    Automating QSAR Modelling [Slides] [Abstract]
  • Jos Lommerse (Organon, The Netherlands)
    SeeSAR, an automated SAR analysis tool based on pairwise comparisons [Slides] [Abstract]
  • Olga Obrezanova (BioFocus DPI, Cambridge, UK)
    Gaussian Processes: A method for automatic QSAR modelling of ADME properties? [Slides] [Abstract]
  • Bernd Wendt (Tripos, UK)
    Novel procedures for 3D-QSAR analysis [Slides] [Abstract]


Presentation: Francis Atkinson , Gianpaolo Bravi, Colin Edge, Gavin Harper, Daniel Lowe, Nicola Richmond, Martin Saunders, Stefan Senger, Ian Wall, Shane Weaver
The SAR Toolkit: simple tools for the analysis of Structure-Activity Relationships

Computational and Strcutural Sciences, GlaxoSmithKline, Harlow, UK [Slides]

The ability to effectively manage the SAR generated in a Lead Optimisation program is of paramount importance to drug discovery. However, this is not a trivial task as the number of molecules typically produced in a modern Medicinal Chemistry program can grow very rapidly. In addition to the size of the problem, current Medicinal Chemistry approaches invariably exploit, simultaneously, several related, but different, chemical series and thus increase the complexity of the problem dramatically.

A number of tools have been developed recently at GSK to assist medicinal and computational chemists with data extraction, collation and analysis of structure activity relationships. These tools have all been embedded within Spotfire and form part of an SAR toolkit. The objective is to simulate, on a large scale, the way Medicinal Chemists build SAR associations when confronted with a small number of structures on a sheet of paper. The toolkit should assist when the expert eye can no longer cope with the numbers that are being handled, and when the extraction of basic SAR trends becomes a challenge. We believe that these tools are simple and intuitive, and should be used prior to any sophisticated analysis (e.g. QSAR).

Central to this SAR toolkit is an enhanced version of the commonly used R-group analysis. R-group analysis is a very efficient way of summarising information: breaking down the complexity of the problem and focusing on the key points of diversity. It is a core approach used by the Medicinal Chemist, but a notoriously difficult process to automate effectively. We have developed a versatile, highly interactive, decomposition algorithm using the Daylight Toolkit. Routines have been developed to allow for a swift comparison of the R-groups (even in the presence of multiple substitution sites), easy navigation, extraction and quantitative interpretation of the basic trends. Finally, the SAR toolkit can be used to propose novel combinations which are predicted achieve the activity/property objective.

The SAR toolkit has been used to assist many active projects and successful examples will be described during this presentation.


  1. Spotfire DecisionSite, Spotfire, 212 Elm Street, Somerville, MA 02144
  2. Daylight Toolkit, DaylightChemical Information Systems Inc., 120 Vantis – Suite 550 – Aliso Viejo, CA 92656


Presentation: Stephen Curry
Crystallographic analysis of drug binding to human serum albumin

Biophysics Section, Imperial College, UK [Slides]

HSA is an abundant multi-domain plasma protein normally involved in the transport of lipophilic anions such as fatty acids, heme and bilirubin. The protein can also bind a wide variety of drug compounds and can have a significant impact on pharmacokinetics and pharmacodynamics. The effect of the protein on drug distribution is particularly severe for those drugs that bind with high affinity.

To date most efforts to identify drug properties that confer high affinity albumin binding have relied solely on binding measurements, the interpretation of which is non-trivial given the complex architecture of the protein which contains multiple binding sites for endogenous and exogenous ligands. To address this problem, we have solved the crystal structures of a wide range of HSA-ligand complexes using natural ligands such as fatty acids and a wide variety of drugs. This work has revealed the nature of the protein-ligand interactions at several binding sites in unprecedented detail. It gives us a new molecular understanding of the drug binding capabilities of the protein, the interactions between drugs and natural ligands and suggests new ways of probing drug binding.

Presentation: Val Gillet
Designing Arrays Optimised on Multiple Properties

University of Sheffield [Slides]


Presentation: Paul Gleeson
Target Affinity, ADMET & PhysChem Properties: The need for a better balance

GSK, UK [Slides]

Analyses have been performed on (a) a significant number of GSK ADMET measurements to investigate the underlying properties fundamental to all ADMET liabilities and (b) the relationship between in-vitro potency and dose for oral drugs. From these analyses the following issues are discussed:

  1. The key physico-chemical factors influencing all ADMET liabilities
  2. Do we require compounds with in vitro nanomolar activities for in vivo efficacy and what are the implications for drug discovery programmes?


Presentation: David E Leahy
Automating QSAR Modelling

Molecular Informatics Group, Newcastle University, UK, [Slides]

The Discovery Bus (, a multi-agent software system designed for automating aspects of Molecular Design, particularly expert decision making, is described. It extends approaches aimed at automating the processing of drug discovery information but where control remains with the human expert, to automating the “tacit knowledge” of the expert and best practice, which we model as a workflow, and experience, which we model as alternative, competing processing nodes in the workflow. An example application of this architecture to automating QSAR best practice will be described with examples of specific models as well as performance metrics for large numbers of QSAR datasets. Recent extensions of the approach to multi-objective, reverse QSAR will also be covered.

Presentation: Jos Lommerse , Markus Wagener, Siem Heisterkamp
SeeSAR, an automated SAR analysis tool based on pairwise comparisons

Organon, The Netherlands [Slides]

The results obtained in the course of a lead optimization project are frequently summarized in a so-called SAR table, a listing of all the different chemical modifications applied to the lead structure and the corresponding changes in biological activities. The large amount of information in SAR tables is often summarized by presenting a list of substituents at a give site ranked by their usefulness or “performance”. One way to obtain such a rank-order is based on the coefficients of the classical Free-Wilson analysis. In order to account for non-linear and dependent behavior in the structure-activity relationship we are currently investigating alternative approaches that are based on pairs-wise analyses of substituents.

Presentation: Olga Obrezanova , Joelle Gola, Matthew Segal
Gaussian Processes: A method for automatic QSAR modelling of ADME properties?

BioFocus DPI, Cambridge, UK [Slides]

In the presentation we will discuss the application of the Gaussian Processes (GPs) method to prediction of ADME properties. GPs are often applied in machine learning and geostatistics but have not yet been widely used in QSAR studies. The method is suitable for modelling nonlinear relationship, does not require subjective a priori determination of parameters such as variable importance or network architectures. GPs have proved to compare and exceed artificial neural networks in performance and, as GPs are based on a Bayesian approach, they have an intrinsic ability to evaluate the confidence in each prediction. These features make Gaussian Processes a potentially powerful modelling tool for QSAR and ADME problems.

Here we will describe the concepts of the GPs method for regression problems and illustrate application of the method to model several ADME properties. Because Gaussian Processes do not need any subjective input from the user, this technique is perfect for automatic model generation. We will show results of the method being used as part of an automatic process and make a comparative analysis of Gaussian processes with other modelling techniques.

Presentation: Bernd Wendt
Novel procedures for 3D-QSAR analysis

Tripos, UK [Slides]

The presentation will propose a similarity distance matrix display and a series trajectory analysis for 3D-QSAR modelling. Both procedures have been developed on the basis of topomer technologies. The new procedures have been applied to 16 published datasets and revealed critical information for the assembly of structures into meaningful QSAR datasets. The results will be discussed and further applications outlined.