ABSTRACT
Computational methods are increasingly used to streamline and enhance the lead discovery and optimization process. However, accurate prediction of absorption, distribution, metabolism and excretion (ADME) and adverse drug reactions (ADR) is often difficult, due to the complexity of underlying physiological mechanisms. Modeling approaches have been hampered by the lack of large, robust and standardized training datasets. In an extensive effort to build such a dataset, the BioPrint database was constructed by systematic profiling of nearly all drugs available on the market, as well as numerous reference compounds. The database is composed of several large datasets: compound structures and molecular descriptors, in vitro ADME and pharmacology profiles, and complementary clinical data including therapeutic use information, pharmacokinetics profiles and ADR profiles. These data have allowed the development of computational tools designed to integrate a program of computational chemistry into library design and lead development. Models based on chemical structure are strengthened by in vitro results that can be used as additional compound descriptors to predict complex in vivo endpoints. The BioPrint pharmacoinformatics platform represents a systematic effort to accelerate the process of drug discovery, improve quantitative structure-activity relationships and develop in vitro/in vivo associations. In this review, we will discuss the importance of training set size and diversity in model development, the implementation of linear and neighborhood modeling approaches, and the use of in silico methods to predict potential clinical liabilities.
