Electronic Theses and Dissertations

Date of Award


Document Type


Degree Name

Ph.D. in Pharmaceutical Sciences


Biomolecular Sciences

First Advisor

Robert J. Doerksen

Second Advisor

Yixin Chen

Third Advisor

Abby L. Parrill

Relational Format



Glycogen synthase kinase-3 (GSK-3) is a multifunctional serine/threonine protein kinase which is engaged in a variety of signaling pathways, regulating a wide range of cellular processes. Due to its distinct regulation mechanism and unique substrate specificity in the molecular pathogenesis of human diseases, GSK-3 is one of the most attractive therapeutic targets for the unmet treatment of pathologies, including type-II diabetes, cancers, inflammation, and neurodegenerative disease. Recent advances in drug discovery targeting GSK-3 involved extensive computational modeling techniques. Both ligand/structure-based approaches have been well explored to design ATP-competitive inhibitors. Molecular modeling plus dynamics simulations can provide insight into the protein-substrate and protein-protein interactions at substrate binding pocket and C-lobe hydrophobic groove, which will benefit the discovery of non-ATP-competitive inhibitors. To identify structurally novel and diverse compounds that effectively inhibit GSK-3â, we performed virtual screening by implementing a mixed ligand/structure-based approach, which included pharmacophore modeling, diversity analysis, and ensemble docking. The sensitivities of different docking protocols to the induced-fit effects at the ATP-competitive binding pocket of GSK-3â have been explored. An enrichment study was employed to verify the robustness of ensemble docking compared to individual docking in terms of retrieving active compounds from a decoy dataset. A total of 24 structurally diverse compounds obtained from the virtual screening experiment underwent biological validation. The bioassay results shothat 15 out of the 24 hit compounds are indeed GSK-3â inhibitors, and among them, one compound exhibiting sub-micromolar inhibitory activity is a reasonable starting point for further optimization. To further identify structurally novel GSK-3â inhibitors, we performed virtual screening by implementing another mixed ligand-based/structure-based approach, which included quantitative structure-activity relationship (QSAR) analysis and docking prediction. To integrate and analyze complex data sets from multiple experimental sources, we drafted and validated hierarchical QSAR, which adopts a multi-level structure to take data heterogeneity into account. A collection of 728 GSK-3 inhibitors with diverse structural scaffolds were obtained from published papers of 7 research groups based on different experimental protocols. Support vector machines and random forests were implemented with wrapper-based feature selection algorithms in order to construct predictive learning models. The best models for each single group of compounds were then selected, based on both internal and external validation, and used to build the final hierarchical QSAR model. The predictive performance of the hierarchical QSAR model can be demonstrated by an overall R2 of 0.752 for the 141 compounds in the test set. The compounds obtained from the virtual screening experiment underwent biological validation. The bioassay results confirmed that 2 hit compounds are indeed GSK-3â inhibitors exhibiting sub-micromolar inhibitory activity, and therefore validated hierarchical QSAR as an effective approach to be used in virtual screening experiments. We have successfully implemented a variant of supervised learning algorithm, named multiple-instance learning, in order to predict bioactive conformers of a given molecule which are responsible for the observed biological activity. The implementation requires instance-based embedding, and joint feature selection and classification. The goal of the present project is to implement multiple-instance learning in drug activity prediction, and subsequently to identify the bioactive conformers for each molecule. The proposed approach was proven not to suffer from overfitting and to be highly competitive with classical predictive models, so it is very powerful for drug activity prediction. The approach was also validated as a useful method for pursuit of bioactive conformers.


Emphasis: Medicinal Chemistry



To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.