Aristotle University Of Thessaloniki (AUTH)
Ανάπτυξη και εφαρμογή ποσοτικών σχέσεων δομής - ενεργότητας χημικών ουσιών για εκτίμηση της έκθεσης του ανθρώπινου οργανισμού σε περιβαλλοντικούς ρύπους
Abstract
dc:descriptionThe main objective of this thesis was to develop unified QSAR models, which could be used to predict significant physicochemical and biochemical properties of industrial chemical compounds. These properties include the tissue/blood partition coefficient for adipose, kidney, liver, brain, heart, lung and muscle, the elimination half-life and the kinetic properties of Michaelis-Menten equation.The first step of QSAR modeling was data pre-proccesing to bring them into the appropriate format for statistical analysis. The initial data consisted of the experimental values of responses and two sets of molecular descriptors for industrial chemical compounds, the LFER and PaDEL set. The semi-constant and intercorrelated descriptors were excluded from the analysis. Principal Component Analysis (PCA) was then used for the distribution of chemical compounds and the determination of the most influential descriptors for QSAR analysis.Moreover, the datasets were divided into the training, validation and test set and were analyzed using two statistical methods, Genetic Algorithm based Multiple Linear Regression (GA-MLR) and Artificial Neural Networks (ANN). Levenberg-Marquardt (LM), Scaled Conjugate Gradient (SCG) and Quasi-Newton (BFG) were implemented as training algorithms regarding ANN technique. The developed models were evaluated according to the coefficient of determination and Mean Squared Error (MSE). The Applicability Domain (AD) of the models was determined using geometric, range-based, distance-based and Probability Density Function (PDF)-based methods.The results showed that LM-based ANN outperformed GA-MLR and SCG/BFG-based ANN methods. The analysis concluded to one PaDEL and one LFER model for each of the responses of interest. Specifically, the estimated values of coefficients of determination for both training, validation and test set for the developed best models exceeded the desirable criteria for QSAR acceptance, which are 0.6, 0.5 and 0.6, respectively. In addition, the calculated MSE values for all sets were significantly low, verifying that the models could be both reliable and predictive. In the majority of the predicted responses, the models that were developed using PaDEL descriptors offered a larger chemical space than the one obtained from the models using LFER descriptors.The proposed models were examined for their fitting capacity, validity and applicability. They were stable, reliable and capable to predict physicochemical and biochemical parameters of “data poor” chemical compounds that fall within the applicability domain. In this way, the broader use of Physiologically Based Biokinetic (PBBK) models in chemical risk assessment could be reinforced, especially when considering the current shift of toxicological research to non-animal testing as aptly shown by the large international efforts on high throughput screening. Finally, the Prevention through Design (PtD) concept for environmental chemicals is supported, by allowing the successful prediction of toxicokinetic behaviour based on molecular parameters.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2017
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Papadaki, Krystalia
- Παπαδάκη, Κρυσταλία
Subjects
dc:subject × 20- Ποσοτικές σχέσεις δομής-ενεργότητας
- βιομηχανικά χημικά
- Περιγραφικές μεταβλητές
- Μοντελοποίηση
- Τεχνητά νευρωνικά δίκτυα
- εκτίμηση έκθεσης
- Quantitative Structure-Activity Relationships
- environmental chemical compounds
- Descriptors
- Modeling
- Artificial neural networks
- Exposure assessment
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Χημικού Μηχανικού
- Natural Sciences
- Computer and Information Sciences
- Engineering and Technology
- Chemical Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/42403