El Oakley, Omar; (2023) Investigating LC3 and influenza M2 trafficking via robust representation learning on high throughput microscopy datasets. Masters thesis (M.Phil), UCL (University College London).

Preview

Text El Oakley_10184447_thesis.pdf Download (6MB) | Preview

Abstract

Quantitative descriptors of bioimage features have been used to efficiently discriminate between different phenotypic populations within microscopy image datasets. However, trying to discriminate between the processes of canonical and non-canonical autophagy is difficult as they share a key patterning determinant - relocalisation of LC3 - and because both responses are subject to significant heterogeneity. Here, we apply both classical feature extraction and unsupervised representation learning to isolate and the describe the differences between canonical and non-canonical autophagy. In doing so, this thesis aims to demonstrate how using quantitative bioimage analysis in this context can increase the robustness of observations derived from hypothesis-driven high throughput imaging experiments. A texture descriptor approach is contrasted with two deep representation learning methods, the InfoGAN and BYOL, where it is seen that classical approaches are more amenable to statistical testing and allow for stronger inferences about the underlying biology. These classical approaches are then used to identify imaging phenotypes unique to different stimuli of non-canonical autophagy. From a methodological perspective, this thesis underlines the continued utility of classically-derived image features for scientific discovery.

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item