Cheng, Weixiao
(2021)
Development of In Silico Tools to Predict the Behavior of Per- and Polyfluoroalkyl Substances (PFAS) in Biological Systems.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Per and polyfluoroalkyl substances (PFAS) are a group of chemicals that have been widely used in industrial and consumer products for decades. Recent estimates suggest there are over 4000 PFAS on the global market. However, many of these have very little information available about their potential hazards. Given the vast number of PFAS, a three-level hierarchical framework that includes permeability-limited physiologically based toxicokinetic (PBTK) model, molecular dynamics (MD) based workflow and machine learning (ML) based quantitative structure-activity relationships (QSAR) was proposed to inform the toxicokinetics, bioaccumulation and toxicity of PFAS. The PBTK model was developed to estimate the toxicokinetic and tissue distribution of perfluorooctanoic acid (PFOA) in male rats; the hierarchical Bayesian analysis was used to reduce the uncertainty of parameters and improve the robustness of the PBTK model. By comparing with different experimental studies, most of the predicted plasma toxicokinetic (e.g., half-life) and tissue distribution fell well within a factor of 2.0 of the measured data.
Moreover, a modeling workflow that combines molecular docking and MD simulation techniques was developed to estimate the binding affinity of PFAS for liver-type fatty acid binding protein (LFABP). The results suggest that EEA and ADONA are at least as strongly bound to rat LFABP as perfluoroheptanoic acid (PFHpA), and to human LFABP as PFOA; both F-53 and F-53B have similar or stronger binding affinities than perfluorooctane sulfonate (PFOS). In addition, human, rat, chicken, and rainbow trout had similar binding affinities to one another for each tested PFAS, whereas Japanese medaka and fathead minnow had significantly weaker LFABP binding affinity for some PFAS.
Finally, the ML-based QSAR model was developed to predict the bioactivity of around 4000 PFAS from the OECD report. Based on the collected PFAS dataset, a total of 5 different machine learning models were trained and validated that cover a variety of conventional models (i.e., logistic regression, random forest and multitask neural network) and advanced graph-based models (i.e., graph convolutional network and weave model). The model indicated that most of the biologically active PFAS have perfluoroalkyl chain lengths less than 12 and are categorized into fluorotelomer-related compounds and perfluoroalkyl acids.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
3 September 2021 |
| Date Type: |
Publication |
| Defense Date: |
29 June 2021 |
| Approval Date: |
3 September 2021 |
| Submission Date: |
2 July 2021 |
| Access Restriction: |
No restriction; Release the ETD for access worldwide immediately. |
| Number of Pages: |
164 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
Swanson School of Engineering > Civil and Environmental Engineering |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
PFAS, Toxicokinetic Model, Molecular Dynamics Simulation, Machine Learning, QSAR |
| Date Deposited: |
03 Sep 2021 15:46 |
| Last Modified: |
03 Sep 2021 15:46 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/41382 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}