Programme, abstracts and slides
The UKQSAR and ChemoInformatics Group Autumn 2011 Meeting, jointly hosted by Accelrys and Astex, was held on Thursday November 3rd at the Accelrys European Headquarters in Cambridge.
- 09:00 Registration and Coffee
- 09:30 Brad Sherborne
- 09:45 Jon Essex
- Fragment-based simulations – Slides
- Abstract not available
- 10:15 Richard Hall
- Improved docking accuracy using 3D restraints derived from X-ray crystallography – Slides
- This presentation describes a novel algorithm that can be used to improve ligand docking performance during a structure based optimization campaign. At Astex all optimization projects start with a fragment screen in which multiple fragments are co-complexed with a protein target. The solved crystallographic geometry of fragments enables the identification of the key binding regions of an active site. All available ligands are then used to automatically build pharmacophore models that reward docking into these key regions. As the project progresses, these models are continually updated with new structures, reflecting the direction of the medicinal chemistry effort. The technique has been applied retrospectively to a number of Astex projects and we shall present results that demonstrate a substantial improvement in cross docking accuracy.
- 10:45 Coffee break
- 11:15 Paul Ashford
- Visualisation of variable binding pockets on protein surfaces by probabilistic analysis of related structure sets – Slides
- Computational tools can predict surface pockets that are of suitable size and shape to accommodate complementary small-molecule drugs. However, proteins are dynamic entities, so prediction against single, static structures may miss features of pockets that vary over time. Sets of related structures (e.g., computationally generated conformers, solution NMR ensembles, multiple crystal structures, homologues or homology models) provide insight into the conformational variation of an unbound protein. Such variation may include transiently populated ligand-bound conformations.We have developed a computational method (Provar) that provides a consistent representation of predicted binding pockets across sets of related protein structures by outputting the probabilities that each atom or residue of the protein borders a predicted pocket. These outputs can be readily visualised, and examples will be presented as case studies employing either sets of homologues (kinases; IL-2), or sets of simulated conformations (alpha-1-antitrypsin, Bcl-2). Particular consideration will be given to persistent or variable pockets near regions of interest, such as known small-molecule binding sites.
- 11:45 Various
- Lightning Poster Talk
- 12:15 Lunch and Posters
- 13:45 Chris Luscombe
- Application of the QSAR Workbench to support Medicinal Chemistry – Slides
- Abstract not available
- 14:15 Dave Wood
- AutoQSAR and the Estimation of Prediction Confidence – Slides
- AutoQSAR is a system for automated QSAR that has been developed at AstraZeneca in collaboration with Accelrys. AutoQSAR improves prediction accuracy by automatically keeping the models up-to-date with new measured data as they become available, and by selecting the most predictive model for a given local domain from a model pool of both global and local models.An important, yet occasionally overlooked aspect of QSAR modeling is the provision of reliable confidence intervals with the predictions. We will outline an information theoretic framework for assessing the accuracy of prediction confidence intervals based on the Kullback-Leibler divergence, and we will present recent work on the development of an approach for the estimation of prediction confidence that is suitable for use in automated QSAR.
- 14:45 Coffee Break
- 15:15 Richard Lonsdale
- Insights into Reactivity of Cytochrome P450 Enzymes from QM/MM Methods: Applications to Drug Metabolism – Slides
- Cytochrome P450 enzymes play a central role in the metabolism of drugs, and contribute significantly to adverse drug reactions. Prediction of their metabolites is hence an obvious goal for computational drug discovery. Reliable predictions require understanding of the complex chemistry that takes place at the active site haem, as well as binding. Density functional theory (DFT) based quantum mechanics/molecular mechanics (QM/MM) calculations are ideally suited for detailed investigations of factors involved in reactivity, because they allow the modelling of reactions in proteins including the effects of the protein environment. An example is P450-catalysed aromatic hydroxylation, an important mechanism in drug metabolism, for which such calculations have been useful tools in understanding the mechanism of this process. They have allowed structure-function relationships to be developed, relating activation barriers to the ring substituents in model systems of substituted benzene molecules. It is now possible to model key steps in P450-mediated catalysis of drug metabolism, to characterise intermediates and transition states, as well as get accurate activation energy barriers and hence identify the most likely metabolites. This has been applied to several widely-used pharmaceutical compounds, including dextromethorphan, ibuprofen and diclofenac to analyse factors important in reactivity and selectivity, demonstrating subtle effects in reactivity.
- 15:45 Tony Long
- Mammalian Drug Metabolism In Silico: Strategies for Increasing Specificity of Prediction – Slides
- The use of the expert system for toxicology prediction in drug discovery is now mainstream and that of the expert system for metabolism prediction gaining rapid acceptance – this, despite the fact that it is extremely difficult (and, indeed, probably inappropriate) to validate an expert system in the same rigorously statistical way as for a quantitative method – factual and heuristic knowledge cannot be treated as such.
In this talk, a formal evaluation method for a knowledge-based xenobiotic metabolism prediction system will be described.
A combination of high sensitivity (defined here as the ratio of experimentally confirmed biotransformation predictions to false negative biotransformation predictions, a high ratio corresponding to a high sensitivity) and high specificity (defined here as the ratio of experimentally unconfirmed to confirmed biotransformation predictions, a high ratio corresponding to a low specificity) is a highly desirable feature of such a predictive system.
- 16:15 Nikolas Fechner
- Graph matching-motivated molecular similarity measures for scaffold hops in virtual screening – Slides not available
- Similarity-based virtual screening often fails to suggest new molecular scaffolds due to the topological bias imposed by the query compound. Various approaches have been applied to overcome this limitation, many of which are either motivated by molecular interaction fields such as CoMSIA  or FieldScreen , or by abstractions of the topology, such as feature trees .In this talk, it will be demonstrated that Graph Matching based similarity measures  can be successfully applied for virtual screening tasks and moreover can be extended to improve their scaffold hopping capability. A short introduction of the approach will be presented as well as several extensions of the original idea capable of taking conformational spaces [5,6] or interaction fields  into account. The differences of the variants as well as their advantages and drawbacks will be highlighted by virtual screening experiments using publicly available data sets.
- 16:45 Meeting ends