E-Thesis 636 views 163 downloads
Investigating the use of Deep Learning Algorithms to Automatically Score Micronuclei in Human Cell Lines / RACHEL BARNES
Swansea University Author: RACHEL BARNES
Abstract
The in vitro micronucleus (MN) assay is a globally used test to quantify DNA damage induced by test chemicals from various industries such as pharmaceuticals, cosmetics and agriculture. Currently, manual scoring is used which is extremely time-consuming and scorer subjective so causes a significant...
| Published: |
Swansea, Wales, UK
2023
|
|---|---|
| Institution: | Swansea University |
| Degree level: | Master of Research |
| Degree name: | MSc by Research |
| Supervisor: | Johnson, George E. and Rees, Paul. |
| URI: | https://cronfa.swan.ac.uk/Record/cronfa63717 |
| Abstract: |
The in vitro micronucleus (MN) assay is a globally used test to quantify DNA damage induced by test chemicals from various industries such as pharmaceuticals, cosmetics and agriculture. Currently, manual scoring is used which is extremely time-consuming and scorer subjective so causes a significant bottleneck in the use of the MN assay. This project shows that imaging flow cytometry coupled with deep learning neural networks can be reliably and accurately used with inter-laboratory function, to automatically score micronucleus events in chemically exposed human B lymphoblastoid cells called TK6 cells. Images were taken from both the cytokinesis-block micronucleus (CBMN) assay and the mononucleate MN assay at Newcastle University. Six different chemicals were tested in this study which are known genotoxic agents and known non-genotoxic agents: aroclor, carbendazim, methyl methanosulphate (MMS), vinblastine, benzo(a)pyrene, D-mannitol. These images were then inputted into a “Deep Flow” neural network, coded in the MATLAB platform which was previously trained on human-scored images assembled from the CBMN assay conducted by Cardiff and Cambridge universities, using MMS and carbendazim treated TK6 cells. Using image data from multiple laboratories in this study provides evidence that the neural network can be used to score unseen data from any laboratory. The neural network correctly scores micronucleus events for both the CBMN and mononucleate MN assays at a percentage confidence of 70% and above. Dose response data for each chemical is parallel to ECVAM guidelines. The aneugen, carbendazim, was shown by the deep learning algorithm to increase the mean dose response by 3.4-fold which shows that as the dose of carbendazim increases, the abundance of micronuclei increases. Further optimisation of the ground truth will prevent underscoring of micronuclei in binucleated cells. It can be concluded that with further optimisation and development of the neural network, this automated platform offers a great potential for the use of the in vitro MN assay to be widened. This method has a higher throughput and has the capability to test greater numbers of compounds and chemicals, therefore, this method will be able to keep up with the increasing demand for genotoxicity testing in industrial and pharmaceutical settings. |
|---|---|
| Keywords: |
Genetic Toxicology, Artificial Intelligence, Machine Learning |
| College: |
Faculty of Medicine, Health and Life Sciences |