Quantitative structure-toxicity relationship modeling of organic compounds and nanoparticles Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 22, 2019
Creator
  • Pu, Dongqiuye
    • Affiliation: Eshelman School of Pharmacy, Division of Pharmacoengineering and Molecular Pharmaceutics
Abstract
  • Safety issues are considered the single largest reason for today’s drug development failures. It is both costly and time-consuming for toxicological evaluation of materials. This dissertation focuses on computational modeling of specific toxicityrelated endpoints against chemical compounds and nanoparticles. We concentrate on the application of cheminformatic and QSAR approaches in predicting the toxicity profile for small molecules as well as nanoparticles. Extensive efforts have been made in terms of data collection, data curation, QSAR modeling and virtual screening of external libraries for biologically benign molecules or nanoparticles. Firstly, QSAR analysis has been applied to a group of organic molecules to predict their skin sensitization toxicities. Combinatorial QSAR analysis was utilized to boost the final model performance. 5-fold external cross-validation and y-randomization processes were also applied to validate the robustness of the models. The final models achieved prediction accuracy as high as 83% (for both kNN and RF models) after the implementation of applicability domain. Secondly, we illustrated successful application of QSAR in modeling nanoparticles with two case studies. In both cases, the object datasets consist of nanoparticles with same core structure yet different surface molecular modifiers. In the first study, computational models were developed for cellular uptake property of a series of nanoparticles possessing same core structure (cross-linked iron oxide) with different surface functional groups. Regression models were successfully developed with R02 as high as 0.77 with kNN method after the implementation of applicability domain. Descriptor analysis suggests that the hydrophobicity of the surface molecule may have significant impact on the cellular uptake of iron oxides by pancreatic cancer cells. The second study takes this concept a step further. Besides building statistically significant computational models for predicting the protein binding and acute toxicity properties of a series of carbon nanotubes, an external chemical library consisting of 240,000 molecules were virtually screened in seeking for biologically benign nanoparticles. Moreover, the virtual hit list resulting from the virtual screening exercise was shared with our collaborators for experimental testing. The final results confirm the high prediction accuracy (80% for acute toxicity and 85% for carbonic anhydrase binding endpoint) of the established models. This is also the first-ever study in the area of nanotoxicity to successfully utilizing computational models for prioritizing nanoparticles for experimental testing.
Date of publication
DOI
Resource type
Rights statement
  • In Copyright
Note
  • ... in partial fulfillment of the requirements for the degree of Master of Science in the Division of Molecular Pharmaceutics at Eshelman School of Pharmacy.
Advisor
  • Tropsha, Alexander
Language
Parents:

This work has no parents.

Items