Meetings

Autumn Meeting 2003

Cyprotex Discovery Ltd
Adlington Hall, Macclesfield, Cheshire

Presentations

  • Kamaldeep Chohan (AstraZeneca Charnwood, UK)
    Development of Local or Global Models for the Prediction of DMPK and other Properties [Abstract]
  • Mark Cronin (Liverpool John Moores University, UK)
    Information on: The 11th International Workshop on QSAR in the Human Health and Environmental Sciences
  • Martyn Ford (Portsmouth University, UK)
    Hits or Misses? – Selecting Compounds for Gene Family Targeted High Throughput Screening [Slides]
  • Brian Hudson (Portsmouth University, UK)
    Application of the TREPAN Rule Extraction Method to Chemoinformatic Data [Slides] [Abstract]
  • Pranas Japertas (PharmAlgorithms)
    Automated Approaches to QSAR Model Building [Slides] [Abstract]
  • Alexey Lagunin (Institute for Biomedical Chemistry, Moscow, Russia)
    Recent Developments in the PASS approach [Slides]
  • David Leahy (Cyprotex Discovery Ltd, UK)
    Opening Remarks
  • Ian Nabney (Aston University, UK)
    Data Visualisation and Chemometrics [Slides] [Abstract]
  • Russell Viner (Syngenta, UK)
    A Fragment-Based Design Approach to ACP-Enoyl Reductase Inhibitors [Abstract]

Abstracts

Presentation: Kamaldeep Chohan
Development of Local or Global Models for the Prediction of DMPK and other Properties

AstraZeneca Charnwood, UK

Whether to build a project specific (local) or project independent (global) QSAR model depends on the property in question. Nevertheless, both have their strengths and weaknesses. For instance, global models are useful for DMPK properties because these tend to be project independent. Discussion will be based on local versus global modelling, along with general modelling examples drawn from DMPK and other areas that are important in the drug discovery process.

Presentation: Mark Cronin
Information on: The 11th International Workshop on QSAR in the Human Health and Environmental Sciences

Liverpool John Moores University, UK

HASH(0x3601c00)

Presentation: Martyn Ford
Hits or Misses? – Selecting Compounds for Gene Family Targeted High Throughput Screening

Portsmouth University, UK [Slides]

HASH(0x35ffca4)

Presentation: Brian Hudson
Application of the TREPAN Rule Extraction Method to Chemoinformatic Data

Portsmouth University, UK [Slides]

TREPAN is an algorithm for the extraction of comprehensible rules from trained neural networks. The method has been applied successfully to biological sequence (bioinformatics) problems. It is currently being extended to handle chemoinformatics (QSAR) datasets. The method has been shown to have advantages over traditional symbolic rule induction methods such as C5. Results obtained for bioinformatics and chemoinformatics problems using the TREPAN algorithm will be presented.

Presentation: Pranas Japertas
Automated Approaches to QSAR Model Building

PharmAlgorithms [Slides]

A QSAR model is only meaningful if the underlying descriptors can be related to the “causal” chemical factors that determine the investigated properties or activities. For example, intrinsic solubility of crystalline compounds depends on the structural planarity and multipoint interactions that favour tight crystal packing. In logP we must consider internal interactions that make for example partitioning of salicylic acid different from other (non-ortho) oxy-benzoic acids. Most of these “causal factors” are substructure-specific, i.e. they cannot be accurately described by generic properties such as lipophilicity or polar surface area. This forces us to consider large functional groups and non-polar skeletons as individual descriptors. If larger structural constructs are considered, then clearer relationships with the causal factors can be established, and better accuracy can be achieved. On the other hand, this leads to a less general model and reduced predictive power when applied to structurally dissimilar molecules. An automated approach to building fragmental QSAR models (e.g for logP prediction) has been devised that will automatically build several models using chains and scaffolds of different sizes. These are then combined into an algorithm that estimates the reliability of predictive calculations based on the similarity of compounds to the training set. It is essential to achieve a balance between the accuracy and generality of predictions and the best solution is to build several models with descriptors of different size and structural specificity. If a new compound is similar to the training set, the most accurate model based on the largest structural skeletons can be used while if the new compound is not similar to the training set, the least accurate (but most general) model should be used. The procedure has been tested on a variety of experimental data, including log P, aqueous solubility, and Pgp transport. To exemplify the approach the development of an algorithm to predict solubility in DMSO will be described.

Presentation: Alexey Lagunin
Recent Developments in the PASS approach

Institute for Biomedical Chemistry, Moscow, Russia [Slides]

HASH(0x35fd64c)

Presentation: David Leahy
Opening Remarks

Cyprotex Discovery Ltd, UK

HASH(0x35fd580)

Presentation: Ian Nabney
Data Visualisation and Chemometrics

Aston University, UK [Slides]

Techniques for visualising high-dimensional data have moved on a long way from simple graphical plots and Principal Component Analysis. Non-linear and hierarchical visualisation techniques allow non-statistical users to ‘drill down’ into their data and understand it better. This talk will describe and demonstrate these novel algorithms and discuss the role that they can play in chemometrics. Examples will be drawn from QSAR, HTS and gene expression microarray experiments.

Presentation: Russell Viner
A Fragment-Based Design Approach to ACP-Enoyl Reductase Inhibitors

Syngenta, UK

ACP-Enoyl Reductase (ACP-ER) is an enzyme involved in fatty acid biosynthesis. It is a potential target for both anti-bacterial compounds and also herbicides. This talk describes the design of a novel class of ACP-ER inhibitors using a fragment-based approach. A range of low molecular weight compounds (typically <150 Da) were selected from our compound collection that contained structural features that could potentially form favourable interactions at the active site. NMR experiments were then used to identify those compounds that did indeed bind at the active site. For those that bound, soaking experiments were conducted in order to determine the precise binding mode crystallographically. These binding ‘fragments’ were then used as starting points in the design of larger molecules, which displayed inhibitory activity on a functional enzyme assay.