Prioritizing Small Molecules for Drug Discovery or Chemical Safety Assessments using Ligand- and Structure-based Cheminformatics Approaches

Recent growth in the experimental data describing the effects of chemicals at the molecular, cellular, and organism level has triggered the development of novel computational approaches for the prediction of a chemical's effect on an organism. The studies described in this dissertation research predict chemical activity at three levels of biological complexity: binding of drugs to a single protein target, selective binding to a family of protein targets, and systemic toxicity. Optimizing cheminformatics methods that examine diverse sources of experimental data can lead to novel insight into the therapeutic use and toxicity of chemicals.;In the first study, a combinatorial Quantitative Structure-Activity Relationship (QSAR) modeling workflow was successfully applied to the discovery of novel bioactive compound against one specific protein target: histone deacetylase inhibitors (HDACIs). Four candidate molecules were selected from the virtual screening hits to be tested experimentally, and three of them were confirmed active against HDAC.;Next, a receptor-based protocol was established and applied to discover targetselective ligands within a family of proteins. This protocol extended the concept of protein/ligand interaction-guided pose selection by employing a binary classifier to discriminate poses of interest from a calibration set. The resulting virtual screening tools were applied for enriching beta2-adrenergic receptor (beta2AR) ligands that are selective against other subtypes in the betaAR family (i.e. beta1AR and beta3AR). Moreover, some computational 3D protein structures used in this study have exhibited comparative or even better performance in virtual screening than X-ray crystal structures of beta2AR, and therefore computational tools that use these computational structures could complement tools utilizing experimental structures.;Finally, a two-step hierarchical QSAR modeling approach was developed to estimate in vivo toxicity effects of small molecules. Besides the chemical structural descriptors, the developed models utilized additional biological information from in vitro bioassays. The derived models were more accurate than traditional QSAR models utilizing chemical descriptors only. Moreover, retrospective analysis of the developed models helped to identify the most informative bioassays, suggesting potential applicability of this methodology in guiding future toxicity experiments.;These studies contribute to the development of computational strategies for comprehensive analysis of small molecules' biological properties, and have the potential to be integrated into existing methods for modern rational drug design and discovery.