Facial reconstruction

Search LJMU Research Online

Browse Repository | Browse E-Theses

Increasing the Confidence of In Silico Modelling in Toxicology

Belfield, S (2024) Increasing the Confidence of In Silico Modelling in Toxicology. Doctoral thesis, Liverpool John Moores University.

2023SamuelJohnBelfieldPhD.pdf - Published Version
Available under License Creative Commons Attribution Non-commercial.

Download (5MB) | Preview


Consideration of all chemicals that we are exposed to on a daily basis is a daunting task, which has been traditionally assessed through animal testing procedures. However, the ethical and financial considerations associated with such testing has long been a topic of concern, with the desire to pursue alternative methods evident. Towards this, the vision of 21st century toxicology actively promoted the use of new approach methodologies (NAMs) that avoid the usage of animal testing, as well as fostering a more efficient means for toxicological assessment. Captured within these NAMs are in silico methods which include a range of in silico (or computational) approaches, one of the most popular being Quantitative Structure- Activity Relationships (QSARs). Although it is acknowledged that the majority of these in silico methods are by no means novel, it is the consideration of such within regulatory decisionmaking frameworks that is. Whilst these methods are being promoted for usage within regulatory settings, fundamental issues regarding assessment of confidence as well as knowledge sharing need to be addressed to further promote acceptance. Therefore, the aim of this thesis was to provide detailed analysis of methods for in silico model validation, and knowledge-sharing efforts that incorporate the state-of-the-art practices, which could potentially bolster their acceptance within regulatory settings. Recently developed uncertainty assessment criteria for the evaluation of QSARs were analysed with a particular focus on how they can be employed to demonstrate fitness-for-purpose. These uncertainty assessment criteria were subsequently developed further, with considerations of challenges in QSAR, such as mixture assessment and machine learning (ML) approaches. To facilitate this, a review was conducted of the key characteristics of QSAR methods applied to mixtures, using the knowledge gathered to identify areas for additional consideration within the criteria. ML approaches were studied, with six models developed to address ML-specific considerations within the criteria. The concept of model sharing has been promoted through the application of the FAIR (Findable, Accessible, Interoperable, Reusable) principles to in silico methods. Outcomes from each chapter and the overall thesis promote the advancement of regulatory acceptance of QSAR models and predictions, through development of improved reporting strategies and sharing methodologies. The thesis additionally benefits the field thorough considerations of the most challenging aspects of QSARs, and how these subfields, such as mixture assessment and ML approaches, can gain credibility.

Item Type: Thesis (Doctoral)
Uncontrolled Keywords: QSAR; Best practice; FAIR; Framework; In silico; Machine learning; Mixture toxicity; Modelling; New approach methodologies; Next generation risk assessment; Regulatory use; Review; Toxicity prediction; Toxicology; Uncertainty
Subjects: R Medicine > RS Pharmacy and materia medica
Divisions: Pharmacy & Biomolecular Sciences
SWORD Depositor: A Symplectic
Date Deposited: 26 Apr 2024 09:52
Last Modified: 26 Apr 2024 09:52
DOI or ID number: 10.24377/LJMU.t.00022954
Supervisors: Cronin, M, Madden, J and Enoch, S
URI: https://researchonline.ljmu.ac.uk/id/eprint/22954
View Item View Item