Developing Techniques for The Detection and Prediction of Test Article Related Hepatic Findings in Mouse Livers

Abstract

Within the agrochemical development field newly designed agrochemicals must be tested to determine efficacy and safety including the effect on any organism that encounters the agrochemical through live study trials across multiple dose groups conducted upon rodents. An agrochemical is failed if it is determined to be harmful, for example, if it is proven to cause lesion development within a vital organ and this is traditionally done at the histopathological level. Currently detection and grading histopathological practices rely on the visual determination of both the lesion presence and the extent of the lesion. This is conducted by trained toxicological pathologists at very high detail, requiring visual interrogation of tissue down to the individual cells. The visual aspect of this process can be considered a limiting factor, restricting pathologists to only seeing the lesions when they appear. This can also lead to differences in opinion as to how to label said lesion’s severity. A lesion is different from normal tissue as it is the damaged tissue resultant of effects of the agrochemical on cells in different organs and will therefore differ at a chemical level. This difference in chemical make-up can be determined at a molecular level, providing the potential to both detect and predict lesion development in tissue prior to their appearance. Vibrational spectroscopy provides an avenue to interrogate tissues and determine harm potentials of agrochemicals at early time points of animal studies. It is theorised that current histopathological practices can be supplemented using the pairing of vibrational spectroscopy, specifically Fourier Transform Infrared Spectroscopy, and Machine Learning techniques to improve the decisions made within agrochemical development processes, alongside greatly expanding the amount of information available to the pathologist to make their decisions. This ultimately reduces the number of live animals required for live studies, reduces the time to failure for failing agrochemicals (because tissue damage is identified earlier), and brings about cost and efficiency benefits to the agrochemical industry and to society. In this study, the application of Machine Learning techniques on data obtained through Fourier-Transform Infrared Spectroscopy scans of mice liver tissue from samples exposed to an agrochemical for 14-days stage across multiple dose groups was explored. Machine learners are shown to be able to replicate the pathologist’s decisions regarding the diagnosis of lesion type, allowing for the determination of lesion characteristics at the chemical level.

Date of Award	1 Aug 2023
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Peter Gardner (Supervisor) & Jonathan Shapiro (Supervisor)