Sections
You are here: Home » Meet » OpenTox 2011 » Abstracts » Applications of QSAR in Drug Discovery

Applications of QSAR in Drug Discovery

Jonna Stalring, AstraZeneca, Sweden

Within pharmaceutical research it is often relevant to make the distinction between descriptive and predictive quantitative structure activity relationships (QSARs). Descriptive QSAR modeling is used extensively to understand structure activity relationships with respect to various endpoints within a chemical series and to guide structural changes driving the biological activity in a desired direction. Predictive QSAR modeling is used mainly for biological responses and physical properties relevant to all pharmaceutical projects such as modeling of Absorption, Distribution, Metabolism, Excretion and Toxicity (ADMET).

Economical necessities and the concern for laboratory animals constantly drive the pharmaceutical industry towards replacing in vivo studies with in vitro experiments and in silico methods. Hence, predictive QSAR modeling is becoming increasingly important within drug discovery, in particular for ADMET characterization. ADMET modeling is used throughout the pre-clinical discovery and development process, from hit prioritization to selection of compounds for in vivo testing. 

Developing a QSAR model, intended to be applicable to all pharmaceutically relevant parts of chemical space, is exceedingly challenging. In addition, many ADMET endpoints and in particular toxicological endpoints, are dependent upon a multitude of molecular mechanisms. Many of these mechanisms may remain unknown and even with a clear mechanistic understanding, the underlying physical processes are complex and structural knowledge often not available.

The chemical diversity amongst compounds to be predicted by a global ADMET model is vast, while the availability of relevant experimental data is usually relatively scarce. Though the amount of data within the public domain is constantly growing, the concordance between experiments is often insufficient to merge data of different origin. Development of new methods to successfully merge data from various sources is a necessary advancement within QSAR modeling to assure compliance with legislative demands for reduction of the use of laboratory animals while simultaneously adapting to the increased regulatory requirements on safety profiling of chemicals.

Assessment of the prediction reliability is another important tool to address the diversity of chemical space. The definition of an applicability domain for a model is generally recognized as a quality requirement by the OECD principals for validation of QSAR models. However, a scientifically justified metrics for the assessment of prediction reliability remains to be agreed upon by the QSAR community. A multitude of methods exists and they can be empirically assessed for a given QSAR data set by their correlation to the prediction error.

Global ADMET models are intended to be used for a long period of time in a vast range of pharmaceutical projects. During this time the structural characteristics of project compounds are generally migrating away from what was know while developing the ADMET model. The drift of chemical project space can be accounted for by automatically and frequently updating the QSAR models to incorporate into the training set all new compounds experimentally characterized with respect to the corresponding endpoint.

Document Actions