| Recent advances in High Throughput Screening (HTS) led to the rapid growth of chemical libraries of small molecules, which calls for improved computational tools and predictive models for Virtual Screening (VS). Thus this dissertation focuses on both the development and application of predictive Quantitative Structure-Activity Relationship (QSAR) models and aims to discover novel therapeutic agents for certain diseases.;First, this dissertation adopts the combinatorial QSAR framework created by our lab, including the first application of the Distance Weighted Discrimination (DWD) method that resulted in a set of robust QSAR models for the 5-HT 7 receptor. VS using these models, followed by the experimental test of identified compounds, led to the finding of five known drugs as potent 5-HT7 binders. Eventually, droperidol (Ki = 3.5 nM) and perospirone (Ki = 8.6 nM) proved to be strong 5-HT7 antagonists. Second, we intended to enhance VS hit rate. To that end, we developed a cost/benefit ratio as an evaluation performance metric for QSAR models. This metric was applied in the Decision Tree machine learning method in two ways: (1) as a benchmarking criterion to compare the prediction performances of different classifiers and (2) as a target function to build QSAR classification trees. This metric may be more suitable for imbalanced HTS data that include few active but many inactive compounds.;Finally, a novel QSAR strategy was developed in response to the polygenic nature of most psychotic disorders, related mainly to G-Protein-Coupled Receptors (GPCRs), one class of molecular targets of greatest interest to the pharmaceutical industry. We curated binding data for thousands of GPCR ligands, and developed predictive QSAR models to assess the GPCR binding profiles of untested compounds that could be used to identify potential drug candidates. This comprehensive study yielded a compendium of validated QSAR predictors (the GPCR QSARome), providing effective in silico tools to search for novel antipsychotic drugs.;The advances in results and procedures achieved in these studies will be integrated into the current computational strategies for rational drug design and discovery boosted by our lab, so that predictive QSAR modeling will become a reliable support tool for drug discovery programs. |
