Computing Interpretable Representations of Cell Morphodynamics
Author(s)
Cavanagh, Harry
Type
Thesis
Abstract
Shape changes (morphodynamics) are one of the principal ways cells interact with their environments and perform key intrinsic behaviours like division. These dynamics arise from a myriad of complex signalling pathways that often organise with emergent simplicity to carry out critical functions including predation, collaboration and migration. A powerful method for analysis can therefore be to quantify this emergent structure, bypassing the low-level complexity. Enormous image datasets are now available to mine. However, it can be difficult to uncover interpretable representations of the global organisation of these heterogeneous dynamic processes. Here, such representations were developed for interpreting morphodynamics in two key areas: mode of action (MoA) comparison for drug discovery (developed using the economically devastating Asian soybean rust crop pathogen) and 3D migration of immune system T cells through extracellular matrices (ECMs). For MoA comparison, population development over a 2D space of shapes (morphospace) was described using two models with condition-dependent parameters: a top-down model of diffusive development over Waddington-type landscapes, and a bottom-up model of tip growth. A variety of landscapes were discovered, describing phenotype transitions during growth, and possible perturbations in the tip growth machinery that cause this variation were identified. For interpreting T cell migration, a new 3D shape descriptor that incorporates key polarisation information was developed, revealing low-dimensionality of shape, and the distinct morphodynamics of run-and-stop modes that emerge at minute timescales were mapped. Periodically oscillating morphodynamics that include retrograde deformation flows were found to underlie active translocation (run mode). Overall, it was found that highly interpretable representations could be uncovered while still leveraging the enormous discovery power of deep learning algorithms. The results show that whole-cell morphodynamics can be a convenient and powerful place to search for structure, with potentially life-saving applications in medicine and biocide discovery as well as immunotherapeutics.
Version
Open Access
Date Issued
2022-01
Date Awarded
2022-04
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Endres, Robert
Sponsor
Biotechnology and Biological Sciences Research Council
Syngenta iCASE funding
Grant Number
BB/M011178/1
Publisher Department
Life Sciences
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)