ABSTRACT
It is logical to expand decision-analytic/mathematical modeling beyond individual patient or population decisions to encompass the whole drug discovery and development process. Successful biomedical drug discovery, to a great extent, requires us to answer what is the key biological target(s) that will change the disease outcome, and how much of the biological target(s) can my drug afford to perturb without causing intolerable harm to the patient population being treated? The focus has perhaps for too long been on the former, finding validated clinical targets, while the latter question relating to dose and toxicity has been avoided. At the molecular level, a coordinated system of proteins, including transporters, channels, receptors, and enzymes, act as gatekeepers to foreign molecules with toxicology implications. The understanding of small molecule-protein interactions would enable us to predict these interactions with targets of interest. Simultaneously, this would also improve our ability to predict the toxic consequences responsible for withdrawal of numerous
15.1 Introduction ..................................................................................................209 15.2 Learning from Computer-Aided Molecular Design ..................................... 211
15.2.1 Quantitative Structure-Activity Relationships (QSARs) .................. 211 15.2.2 Target-Based Methods ...................................................................... 214
15.3 Learning from Pharmacoeconomics: Computer-Aided Decision Making .......................................................................................................... 214 15.3.1 Types of Decision-Making Models .................................................. 215 15.3.2 Applying the Models Throughout Drug Discovery .......................... 216
15.3 Conclusion .................................................................................................. 221 Acknowledgments .................................................................................................. 221 References .............................................................................................................. 221
marketed drugs and late-stage clinical failures that are having a devastating effect on bringing safe and effective treatments to patients.1 Because of the complexities of different model systems, whether in vitro or in vivo, currently in use, better predictive approaches are needed overall. Accurate predictions for toxicity mechanisms are also complicated as the whole organism is involved, with possibly hundreds to thousands of endogenous (endobiotic) and foreign (xenobiotic, e.g., drug) molecules interacting in different cellular organelles of tissues. Species differences in protein expression and the small endobiotic or xenobiotic molecules that will bind to these (ligand specificity), should also be considered as complicating our understanding, as what happens in a mouse model may not relate to humans. It is also important to understand that both the parent molecule and the products of metabolic pathways may also be involved in favorable or unfavorable drug interactions, where they interfere with the metabolism of endogenous or other co-administered compounds. Such drug-drug interactions, or other adverse drug reactions, can have potentially fatal consequences for the patient or be very costly for health care providers.2-6
The above is just one small but critical part of the drug discovery and development process that generates a huge amount of data and has done so for decades. Information systems for data analysis and management within pharmaceutical companies have had to evolve to answer more complex questions as the data has flooded in, and these systems themselves have the potential to deliver increased value to the organization.7,8 We should be learning from our past experiences and using computational tools extensively to make decisions in this inherently highly dimensional space. The increasing generation of biological data using highly parallel automated screening and analytical systems (such as high throughput methods) in drug discovery complements the use of computational technologies and provides a foundation for model generation. Computational models are becoming more widely available based upon quantitative structure activity relationships (QSAR), or docking methods9 (see Section 15.2.2) with individual proteins known to be important therapeutic targets or that have some relationship to toxicity.
