National Technical University of Athens (NTUA)
Development of mechanistic models and machine learning approaches for supporting risk assessment of human exposure to chemicals and materials
Abstract
dc:descriptionThe ever-increasing variety of products and services available to humans is accompanied by a corresponding rise in the number of chemicals and materials released into the environment. Through various transport and biotransformation processes, these substances and their derivatives eventually interact with human biological systems. Depending on their toxicological mechanisms and the extent of accumulation, they may exert significant effects on human health. Risk assessment aims to evaluate the potential health risks associated with chemical exposure under defined scenarios. Traditionally, this process has relied on animal studies to establish safe exposure levels for humans. However, the vast number of new compounds entering the market each year makes comprehensive animal testing infeasible. In addition, growing ethical, scientific, and practical concerns regarding the use of animals in hazard and risk assessment have emphasized the urgent need to develop alternative approaches. Consequently, in silico methodologies have become essential tools for addressing the existing toxicity data gap and supporting next generation risk assessment. In response to the growing need for reliable and ethical alternatives to animal testing, this PhD thesis aims to develop in silico methodologies and modelling frameworks that support the risk assessment of chemicals and materials. These models are intended not only to predict the biological activity of new and untested structures but also to offer deeper mechanistic insights into the processes governing biokinetics and toxicity, with a strong focus on the development, calibration and evaluation of physiologically based kinetic (PBK) models, i.e., physiology-informed models that simulate the movement of substances through the body to estimate internal exposures over time. The research focuses on two major groups of compounds: nanomaterials (NMs) and per- and polyfluoroalkyl substances (PFAS). For both groups, conventional experimental testing is slow and resource-intensive relative to the vast array of structures with diverse physicochemical and biological characteristics. Therefore, in silico tools can play a crucial role in accelerating hazard identification, guiding experimental prioritisation, and bridging knowledge gaps in toxicity and exposure assessment. A fundamental component of mechanistic modelling, ranging from simple compartmental models to PBK models, is parameter estimation. A common challenge in model calibration is overparameterisation, particularly when limited in vivo data are available for estimating a large number of parameters. To address this issue, the first part of this thesis introduces two novel computational methodologies to enhance parameter estimation. The first presents a genetic algorithm-based parameter grouping approach that reduces model dimensionality and improves parameter identifiability during fitting. Its applicability is demonstrated through two case studies investigating the biodistribution of titanium dioxide (TiO2) nanoparticles and perfluorooctanoic acid (PFOA) in rats. The second methodology employs physics-informed neural networks (PINNs) to infer interindividual variability in model parameters using aggregated concentration data. This framework enables the translation of population-level mean and variability information into parameter distributions, which can subsequently be incorporated into probabilistic risk assessment workflows. In the second part of this work, three modelling applications that address key challenges in NM research and risk assessment are presented. First, a simple compartmental model is developed to integrate information from multiple sources and identify kinetic bottlenecks in nanoparticle-mediated drug delivery to target tissues. Second, a PBK model for inhalation exposure in rats is constructed and extrapolated to humans using allometric scaling. The PBK model is coupled with an exposure module predicting the spatiotemporal concentration of TiO2 particles in occupational environments, allowing the translation of source-level exposure into a lung-deposited dose. Finally, a toxicodynamic component is introduced through a dose–response QSAR model predicting cytotoxicity across a diverse set of NMs. This study incorporates in vitro dosimetry modelling to identify the most relevant dose metrics driving NM toxicity. The third part of this thesis focuses on developing models to predict and elucidate the biokinetics and toxicity of PFAS. The first application involves the development of a multi-route, permeability-limited PBK model to explain biodistribution patterns of PFOA in male and female rats. Constructed using a middle-out approach, the model is extensively validated against independent biodistribution datasets. Sensitivity analysis is employed to identify the most influential parameters governing PFOA biodistribution. In the second application, a QSAR model is developed to predict PPAR-α activation across PFAS congeners. A dose–response in vitro dataset is analysed using benchmark dose (BMD) modelling to derive a biologically relevant dose metric. A linear model combined with genetic algorithm–based feature selection identified key physicochemical descriptors driving receptor activation. The model reveals structural features associated with higher potency and also enables the screening of an extended PFAS library, identifying compounds within the applicability domain that may act as strong PPAR-α inducers.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tsiros, Periklis
- Τσίρος, Περικλής
Subjects
dc:subject × 16- Νανοϋλικά
- Κινητική μοντελοποίηση
- Μαθηματική μοντελοποίηση
- Ανάλυση ρίσκου
- Χημική έκθεση
- Περ- και πολυφθοροαλκυλικές ουσίες (PFAS)
- Physiologically based kinetic models
- Quantitative structure activity relationships
- Benchmark dose (BMD)
- Nanomaterials
- Per- and polyfluoroalkyl substances (PFAS)
- Επιστήμη Χημικού Μηχανικού
- Επιστήμες Μηχανικού και Τεχνολογία
- Χημική μηχανική
- Chemical Engineering
- Engineering and Technology
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/60841