Meetings

Autumn Meeting 2006

Novartis Institutes for BioMedical Research
Horsham, West Sussex

Presentations

  • Kristian Birchall (Chemoinformatics Research Group, Department of Information Studies, the University of Sheffield, Sheffield, S10 2TN, UK)
    Extracting SARs using a Multi-Objective SMARTS Evolutionary System [Slides] [Abstract]
  • David E Clark (Argenta Discovery Ltd., 8/9 Spire Green Centre, Flex Meadow, Harlow, Essex, CM19 5TR, UK)
    Discovery and optimisation of novel, nonsteroidal glucorticoid receptor modulators [Abstract]
  • John Delaney (Syngenta, UK)
    Modern agrochemical research – A missed opportunity for drug discovery? [Slides] [Abstract]
  • Mike Devereux (University of Manchester, UK)
    An ab initio fragment database for use predicting replacements in drug design [Slides] [Abstract]
  • Peter Ertl (Novartis Insitutes for BioMedical Research, Basel, Switzerland)
    Quest for the Rings – In Silico Exploration of theRing Universe to Identify Novel Bioactive Heteroaromatic Scaffolds [Slides] [Abstract]
  • Dave Evans (Lilly, UK)
    Commercial 3D QSAR Methods: Phase and Catalyst Compared [Slides] [Abstract]
  • David J Livingstone (ChemQuest, Sandown, UK)
    The use of F statistics in regression models from large pools of variables [Slides] [Abstract]
  • Michael Sternberg (Division of Molecular Biosciences, Imperial College London, London, SW7 2AZ)
    Logic-based drug discovery [Abstract]

Abstracts

Presentation: Kristian Birchall
Extracting SARs using a Multi-Objective SMARTS Evolutionary System

Chemoinformatics Research Group, Department of Information Studies, the University of Sheffield, Sheffield, S10 2TN, UK [Slides]

Reduced Graphs (RGs) summarise a chemical structure by grouping atoms into nodes based on properties likely to be important for bioactivity (H-bond donors, aromatic rings etc.). RG queries can be represented using SMARTS notation, where RG nodes replace the normal atoms. A Genetic Program (GP) has been used to derive SMARTS type RG queries for identifying SARs. The GP is provided with a list of RG node types and SMARTS syntactic constraints and is trained on a labelled mixture of active and inactive RGs for a specific activity class. From this it evolves queries that maximise the precision and recall of the actives. Their predictive power is then validated on datasets not used in deriving the queries. Furthermore, the queries inherent SAR information reveals the key features that identify a particular activity class, and which RG nodes are functionally similar in terms of their importance for bioactivity.

Presentation: David E Clark
Discovery and optimisation of novel, nonsteroidal glucorticoid receptor modulators

Argenta Discovery Ltd., 8/9 Spire Green Centre, Flex Meadow, Harlow, Essex, CM19 5TR, UK

Patients with some forms of psychiatric illness have been found to have increased levels of cortisol, a steroid that acts by binding to an intracellular glucocorticoid receptor (GR). Some sufferers from depression, particularly those with psychotic features, can be responsive to treatment with GR antagonists, such as mifepristone, which block the effect of cortisol. In the search for improved GR antagonists, we have used ligand-based virtual screening to discover a novel series of potent and selective GR antagonists in a very short time-frame. The most potent compound identified by the virtual screening possessed a Ki value of 16 nM. This presentation will describe the virtual screening approach undertaken and the subsequent optimization of the series.

Presentation: John Delaney
Modern agrochemical research – A missed opportunity for drug discovery?

Syngenta, UK [Slides]

The word “agrochemical” has often taken on a pejorative character in the public mind. Some of the negative tone may have coloured the perception of the industry by pharma, together with views on the chemical nature of agrochemicals that seem to be based on older pesticides dating back to the fifties and sixties. In this talk we try to address some of these concerns, draw out the similarities between agrochemical and pharmaceutical research and highlight opportunities for drug discovery offered by pesticide related compounds, particularly with regard to herbicides and compounds with “lead-like” physical properties.

Presentation: Mike Devereux
An ab initio fragment database for use predicting replacements in drug design

University of Manchester, UK [Slides]

In this presentation we will introduce the Quantum Isostere Database (QID), a system for prediction of optimal bioisosteric fragment replacements during lead optimization in Drug Discovery. The tool uses ab initio fragment data stored in an Oracle database and is accessed by the chemist through a web interface. The method takes advantage of the charge density, partitioned within the framework of Bader’s theory of “Atoms in Molecules”, as a novel means of obtaining physical and chemical information for each fragment. Several hundred chemical fragments have been generated from the world drug index (WDI) and a wide variety of conformers and ab initio properties produced for each.

We will outline the properties included and demonstrate the validity of our approach in terms of fragment transferability and results obtained. We will then discuss the value added by ab initio descriptors over more traditional empirical data.

Presentation: Peter Ertl
Quest for the Rings – In Silico Exploration of theRing Universe to Identify Novel Bioactive Heteroaromatic Scaffolds

Novartis Insitutes for BioMedical Research, Basel, Switzerland [Slides]

Bioactive molecules only contain a relatively limited number of unique ring types. In order to identify those ring properties and structural characteristics which are necessary for biological activity, a large virtual library of 580165 heteroaromatic scaffolds was created and characterized by calculated properties, including structural features, bioavailability descriptors and quantum-chemical parameters. A self-organizing neural network was used to cluster these scaffolds and to identify properties which best characterize bioactive ring systems. The analysis shows that bioactivity is very sparsely distributed within scaffold property and structural space forming only several relatively small, well defined “bioactivity islands”. The bioactivity is related to ring properties in a complex nonlinear way, with the most important properties responsible for separating active and inactive areas being the size of the scaffolds, their heteroatom composition and electronic stability. The analysis also provided a “ring bioactivity score” which may be used to rank molecules containing rings according to their bioactivity potential. Various possible applications of a large database of rings with calculated properties and bioactivity scores in the drug design and discovery process are discussed, including virtual screening, support for the design of combinatorial libraries, bioisosteric design and scaffold hopping.

Presentation: Dave Evans
Commercial 3D QSAR Methods: Phase and Catalyst Compared

Lilly, UK [Slides]

Rationalising the activity of a series of compounds in terms of their three-dimensional alignment remains an intuitively attractive approach, even in the absence of information about the receptor structure. Both Catalyst (Accelrys) and Phase (Schrodinger) produce alignments of active molecules and use these to construct QSAR models which can be used to predict the activity of further compounds. We present a method for automated preparation of appropriate training sets from large collections of activity data, measure the predictive power of the Phase and Catalyst programs on a series of data sets with a range of varying parameters, and discuss strategies for improving their performance.

Presentation: David J Livingstone
The use of F statistics in regression models from large pools of variables

ChemQuest, Sandown, UK [Slides]

Variable selection methods are routinely applied in regression modelling to identify a small number of descriptors which ‘best’ explain the variation in the response variable. Most statistical packages that perform regression have some form of stepping algorithm that can be used in this identification process. Unfortunately, when a subset of p variables measured on a sample of n objects are selected from a set of k (>p) to maximise the squared sample multiple regression coefficient, the significance of the resulting regression is upwardly biased. The extent of this bias is investigated by using Monte Carlo simulation and is presented as an inflation factor which when multiplied by the usual tabulated F-ratio gives an estimate of the true 5% critical value. The results show that selection bias can be very high even for moderate size data sets. Selecting 3 variables from 50 generated at random with 20 observations will almost certainly provide a significant result if the usual tabulated F values are used. An interpolation formula is provided for the calculation of the inflation factor for different combinations of (n,p,k).

Presentation: Michael Sternberg
Logic-based drug discovery

Division of Molecular Biosciences, Imperial College London, London, SW7 2AZ

The application of Quantitative Structure Activity Relationship (QSAR) is a central tool in drug discovery and development due to its role in identifying key structural features for activity or toxicity. A variety of methods have been developed and each has its merits and limitations. Over the last few years we have been using a form of logic-based machine learning known as Inductive Logic Programming (ILP) to derive QSARs. A major benefit of this method is that it readily provides intelligible rules to explain the activity of the molecules. However, being logic-based, it suffers from the drawback that it is primarily a qualitative rather than a quantitative approach.

Recently we have enhanced the ILP methodology by the addition of combining it with a quantitative support vector analysis in a methodology known as SVILP (Support Vector Inductive Logic Programming). This talk will present the successful application of SVILP to toxicity modeling and to modeling the activity of ligands bound to protein receptors of known structure. Preliminary results on the application of SVILP to screening will also be reported.