Cheminformatics Approaches to Structure Based Virtual Screening: Methodology Development and ApplicationsPublic Deposited
Add to collection
You do not have access to any existing collections. You may create a new collection.
Downloadable ContentDownload PDF
MLAHsieh, Jui Hua. Cheminformatics Approaches to Structure Based Virtual Screening: Methodology Development and Applications. 2011. https://doi.org/10.17615/6p90-6521
APAHsieh, J. (2011). Cheminformatics Approaches to Structure Based Virtual Screening: Methodology Development and Applications. https://doi.org/10.17615/6p90-6521
ChicagoHsieh, Jui Hua. 2011. Cheminformatics Approaches to Structure Based Virtual Screening: Methodology Development and Applications. https://doi.org/10.17615/6p90-6521
- Last Modified
- March 22, 2019
- Affiliation: Eshelman School of Pharmacy, Division of Chemical Biology and Medicinal Chemistry
- Structure-based virtual screening (VS) using 3D structures of protein targets has become a popular in silico drug discovery approach. The success of VS relies on the quality of underlying scoring functions. Despite of the success of structure-based VS in several reported cases, target-dependent VS performance and poor binding affinity predictions are well-known drawbacks in structure-based scoring functions. The goal of my dissertation is to use cheminformatics approaches to address above problems of the existing structure-based scoring methods. In Aim 1, cheminformatics practices are applied to those problems which conventional structure-based scoring functions find difficult (anti-bacterial leads efflux study) or fail to address (AmpC β-lactamase study). Predictive binary classification QSAR models can be constructed to classify complex efflux properties (low vs. high) and to differentiate AmpC β-lactamase binders from binding decoys (i.e., the false positives generated by scoring functions). The above models are applied to virtual screening and many computational hits are experimentally confirmed. In Aim 2, novel statistical binding and pose scoring functions (or pose filter in Aim 3) are developed, to accurately predict protein-ligand binding affinity and to discriminate native-like poses of ligands from pose decoys respectively. In my approach, the proteinligand interface is represented at the atomic level resolution and transformed via a special computational geometry approach called Delaunay tessellation to a collection of atom quadruplet motifs. And individual atom members of the motifs are characterized by conceptual Density Functional Theory (DFT)-based atomic properties. The binding scoring function shows acceptable prediction accuracy towards Community Structure-Activity Resources (CSAR) data sets with diverse protein families. In Aim 3, a two-step scoring protocol for target-specific virtual screening is developed and validated using the challenging Directory of Useful Decoys (DUD) data sets. In the first step our target-specific pose (-scoring) filter developed in Aim 2 is used to filter out/penalize putative pose decoys for every compound. Then in the second step the remaining putative native-like poses are scored with MedusaScore, which is a conventional force-field-based scoring function. This novel screening protocol can consistently improve MedusaScore VS performance, suggesting it possible applications to practical pharmaceutically relevant targets.
- Date of publication
- May 2011
- Resource type
- Rights statement
- In Copyright
- ... in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the School of Pharmacy (Division of Medicinal Chemistry and Natural Products)
- Tropsha, Alexander
This work has no parents.