Inferring bifurcations between phenotypes

Abstract

The gene-expression history of a cell together with its environment determine the its phenotype. The phenotype is an inherently qualitative state, deduced by relative biochemical measurements, increasingly collected with high-throughput single cell methods. In synthetic biology phenotypes are designed, often with qualitative constraints in mind that enforce a desired behaviour with respect to environmental conditions. The design of phenotypes can be cast as a problem in inverse bifurcation analysis when cells are modelled using differential equations. Recently, an emerging discipline called scientific machine learning (SciML) has been pushing for differentiability within simulations and structure-informed models. Bifurcation analysis, however, is yet to benefit from differentiability. In this thesis, bifurcation theory is leveraged to address questions in pattern formation with synthetic E. Coli. The inter-disciplinary collaboration gave rise a novel approach for estimating the parameters of differential equations. The emergent picture suggests that qualitative observations should be learned with bifurcations prior to any attempt at learning quantitative details that may be subject to inter-experiment variability. Concurrently, a focus on flow cytometry, one of the most abundant methodologies in which gating strategies are used to annotate cell phenotypes, gave rise to FlowAtlas.jl: a tool for navigating high-dimensional giga-scale cytometry in settings where manual annotation by domain experts becomes unfeasible. The ideas in this thesis are valuable to those wanting to build interactive and differentiable design–learn workflows for biomedical research.

Date of Award	1 Sept 2022
Original language	English
Awarding Institution	King's College London
Supervisor	Attila Csikasz-Nagy (Supervisor)