Meetings

Autumn Meeting 2004

AstraZeneca R&D
Charnwood, UK

The programme contains a number of well known speakers covering topics ranging from classical QSAR analysis to cutting edge chemoinformatics.

Presentations

  • Charles Andrianjara (Pfizer, UK)
    Biological Profiling using Bioprint [Abstract]
  • John Dearden (John Moors University Liverpool, UK)
    QSAR of the hERG channel [Slides] [Abstract]
  • John Delaney (Syngenta, UK)
    Solubility estimation [Abstract]
  • Mark Earll (Umetrics, UK)
    Getting better Y’s for QSAR [Slides] [Abstract]
  • Dave Rogers (SciTegic, USA)
    PipelinePilot [Abstract]
  • Nick Tomkinson (AstraZeneca, UK)
    Practical applications of PipelinePilot [Slides] [Abstract]
  • David Whitley (Portsmouth University, UK)
    Molecular Surface Property Graphs [Slides]

Abstracts

Presentation: Charles Andrianjara
Biological Profiling using Bioprint

Pfizer, UK

In order to better characterize the relationship between the compound interaction with biological targets and its in vivo effects, it is essential to analyse the compound activity profile in biological space. The Bioprint technology provides a tool for exploring the compound biological space defined by fingerprints of activities against a broad range of in vitro pharmacological, pharmaceutical and ADME/Safety targets.

The compound biological fingerprint presents relevant descriptor for many drug discovery needs which include: prediction of in-vivo effects, insights into target relationship and selectivity issues and the risk of attrition.

Different examples on the impact of using biofingerprint as both compound and target descriptors in the drug design process will be presented.

Presentation: John Dearden
QSAR of the hERG channel

John Moors University Liverpool, UK [Slides]

An increasing number of non-antiarrhythmic drugs are associated with prolongation of the QT interval of the electrocardiogram, leading to significant adverse effects such as bradycardia, electrolyte imbalance and impaired hepatic and renal function. A number of such drugs (e.g. terfenadine, cisapride, astemizole) have been withdrawn from the market because of this.

The hERG (human ether-a-go-go-related gene) potassium channel is expressed in the human heart; it is a major contributor to cardiac repolarisation and contributes to the QT interval. Its inhibition (hERG K+ CI) generally leads to prolongation of the QT interval. Several 3-dimensional (3D) QSAR studies of hERG K+ CI have been published (Ekins et al 2002, Cavalli et al 2002, Pearlstein et al 2003). However, the results from 3D QSAR studies are often difficult of interpretation, so we have carried out a low (0-2D) QSAR analysis of a diverse data-set of 60 drugs and drug candidates using published hERG K+ CI IC50 values determined in mammalian cells.

We calculated a total of 200 descriptors using QsarIS (now MDL QSAR; www.mdli.com), TSAR (www.accelrys.com) and MOLPRO (www.ibmh.msk.su/qsar). The step-wise linear regression routine in MINITAB ver. 13.1 statistics software was used to select the descriptors that best modelled the IC50 values. The following QSAR was obtained:

log IC50 = 0.411 nO – 1.20 Edmax – 0.683 (3c-4pc) + 0.000148 Iz + 0.000635 TE + 2.12

n = 60 r2 = 0.842 Q2 = 0.797 s = 0.614 F = 57.5

where nO = number of oxygen atoms, Edmax = maximum hydrogen bond donor energy, (3c-4pc) = difference between 3rd order cluster and 4th order path-cluster molecular connectivity, Iz = principal moment of inertia along z axis, TE = total molecular energy, n = number of compounds in training set, r = correlation coefficient, Q = cross-validated correlation coefficient (leave-one-out procedure), s = standard error of the estimate, and F = Fisher statistic.

We also validated the QSAR by removing 20% of the compounds from the training set, regenerating the QSAR on the remaining 48 compounds, and then predicting the IC50 values of the 12 compounds that had been removed. This procedure was repeated five times, so that all compounds were removed in turn. The predicted and measured IC50 values correlated well (r2 = 0.806), indicating that the QSAR has good predictive ability.

[1] Cavalli, A. et al (2002) J. Med. Chem. 45: 3844-3853

[2] Ekins, S. et al (2002) J. Pharmacol. Exp. Ther. 301: 427-434

[3] Pearlstein, R.A. et al (2003) Bioorg. Med. Chem. Lett. 13: 1829-1835

Presentation: John Delaney
Solubility estimation

Syngenta, UK

This talk will show how chemometric techniques may be used to improve the quality of biological results used in QSAR studies and how different types of Y data may be used in multivariate models.

Following a brief description of Principal Components Analysis (PCA) its use in the visualisation of systematic variation in biological results will be demonstrated.

Statistical Design of Experiments is a method of maximising the information from a minimal set of well-constructed experiments. Its application in the improvement of the signal to noise ratio of a reporter gene assay will be shown.

Partial Least Squares Regression (PLS) is a versatile regression technique that can handle many Y variables at a time and so is useful in simultaneous optimisation of biological responses. Its role in Statistical Molecular Design (SMD) will be demonstrated using examples from the design of thrombin inhibitors and detergents with low aquatic toxicity.

Finally where complex Y data is encountered the use of Hierarchical PCA and PLS will be illustrated with cases involving very many Y responses and complex 3D time-resolved Y matrices.

Presentation: Mark Earll
Getting better Y’s for QSAR

Umetrics, UK [Slides]

This talk will show how chemometric techniques may be used to improve the quality of biological results used in QSAR studies and how different types of Y data may be used in multivariate models.

Following a brief description of Principal Components Analysis (PCA) its use in the visualisation of systematic variation in biological results will be demonstrated.

Statistical Design of Experiments is a method of maximising the information from a minimal set of well-constructed experiments. Its application in the improvement of the signal to noise ratio of a reporter gene assay will be shown.

Partial Least Squares Regression (PLS) is a versatile regression technique that can handle many Y variables at a time and so is useful in simultaneous optimisation of biological responses. Its role in Statistical Molecular Design (SMD) will be demonstrated using examples from the design of thrombin inhibitors and detergents with low aquatic toxicity.

Finally where complex Y data is encountered the use of Hierarchical PCA and PLS will be illustrated with cases involving very many Y responses and complex 3D time-resolved Y matrices.

Presentation: Dave Rogers
PipelinePilot

SciTegic, USA

Structure-activity relationship (SAR) studies are focused on finding the relationships between a set of measured or calculated variables and a property of interest, often a biological activity. As the quality of quantitative SAR (QSAR) methods improved, the models developed from their application were increasing suitable to use in the discovery of compounds that exhibited improved behavior. However, many of these compounds failed to be useful for reasons outside of the scope of the original model: for example, they were indeed highly-active, but were also too insoluble; or, they were not selective, and were also active in another activity class, leading to unwanted side-effects. For example, 5-hydroxytryptophan (5HT) drugs have a variety of effects, and a given compound may interact with multiple sites. Better understanding of these effects would assist in the construction of appropriate libraries for such complex optimizations.

This talk will discuss how information from multiple models (that is, models built for different activity endpoints or properties) can be combined to help guide the process of designing compounds that may satisfy multiple requirements. The process is suitable for “high noise” data such as that derived from combinatorial chemistry and high-throughput screening studies. The design elements extracted from the study will be structurally-based and easily visualized, allowing a chemist to understand critical “leverage points” in a chemical library.

Presentation: Nick Tomkinson
Practical applications of PipelinePilot

AstraZeneca, UK [Slides]

Pipeline pilot is a powerful and flexible environment which allows the programmer/informatician to sit down with the chemist and easily plan out an automation task. But as a new environment with many options and possibilities, it is daunting for the chemist to approach on their own. A task based approach operating within an environment already familiar to the chemist was required. Common tasks can then be made available within spotfire or excel for example as applications supporting soap communication with the server. New functionality can then be readily added with the full input of the customer. Examples will be given of applications delivered within spotfire and excel.

Presentation: David Whitley
Molecular Surface Property Graphs

Portsmouth University, UK [Slides]

HASH(0x36088c0)