Dr. Dixon has overseen Phase development since the program's inception, and is heavily involved in its ongoing enhancement. In this article, Steve discusses the design philosophy and capabilities of Phase.
Ligand-based drug design formally includes any approach that attempts to explain biological activity in terms of ligand structure alone, but it is most closely identified with pharmacophore model development and 3D QSAR. Software for ligand-based drug design has been available for decades and has enjoyed mixed success. In part this is because certain models that emerge are not sufficiently predictive, but compounding these problems are the inherent difficulties in adapting the software to constantly evolving drug discovery workflows.
We have been mindful of these challenges while designing Phase,
and have developed a suite of ligand-based tools that combine superior
science with flexibility and ease of use. We also recognize the
knowledge and expertise of the modeler as key ingredients to the
success of any ligand-based software package. Phase therefore
emphasizes the user as an integral part of the modeling process,
allowing researchers to tune database screens and model development
using an intuitive, stepwise interface.
Phase’s wizard-like interface for pharmacophore model development guides researchers through the process of identifying common pharmacophores among a group of actives, scoring those 3D configurations of features according to various geometric and heuristic criteria, and evaluating a set of ranked pharmacophore hypotheses. Pharmacophore scoring can rely on information from just an active set of compounds, or it may incorporate data from inactive compounds as well.
A unique consequence of the way Phase perceives pharmacophores is its inherent ability to identify distinct binding modes among a set of actives. While other software typically constructs a union-of-features pharmacophore model that attempts to cover all actives, using partial matching if necessary, Phase proposes a set of common-feature pharmacophores, each of which fully maps to some subset of the actives. By applying hierarchical clustering techniques to uncover patterns of association between subsets of ligands and hypotheses, Phase is able to identify groups of ligands that are most likely to be associated with distinct binding modes.
Once a pharmacophore hypothesis has been created, it can be augmented with a set of excluded volume spheres to map out regions of space not occupied by active ligands. Excluded volumes can be positioned manually or using a variety of automated techniques. The automated methods can consider the space occupied by active and inactive ligands, or the space occupied by the receptor to which a reference ligand is bound. Incorporating receptor information, when available, bridges the gap between ligand-based and structure-based disciplines and leverages the power of both.
When experimental activities are known, a 3D QSAR model may be created for each pharmacophore hypothesis. A QSAR model can consider the entire ligand structure, or just the pharmacophoric features that can be mapped to the hypothesis. The former is appropriate for congeneric series with limited flexibility, whereas the latter is recommended for datasets with significant structural diversity and/or flexibility. Phase QSAR incorporates automatic validation of each model against an external test set. This allows researchers to easily identify the most statistically robust models by directly comparing training and test set statistics.
Ultimately, a pharmacophore model can be used to search a 3D database to identify additional molecules that satisfy the hypothesis. If the model satisfactorily embodies characteristics that are critical for ligand binding, the database search should return active compounds. Phase provides workflows to create and search 3D databases, as well as workflows to search multi-structure files. Conformations can be created once and stored, or generated on-the-fly while searching. In the upcoming 2007 release of Phase (Version 2.5), a hybrid of these two approaches will be supported, allowing on-the-fly refinement of matches from pre-generated conformations. Many new options for matching and filtering will be featured, including OR and NOT logic; and overall searching speed against pre-generated conformations will be approximately five times faster. A new checkpoint mechanism will also be available, enabling users to monitor partial search results and restart any job that fails to complete due to hardware-related reasons.
Since its initial release, Phase development has been driven by customer feedback and requests, evolving and improving to meet the changing needs of computational scientists throughout the world. Readers interested in learning more about Phase will find a comprehensive description in our paper in the Journal of Computer-Aided Molecular Design.