Thesis icon

Thesis

Simulation for steering evolution

Abstract:

Whilst conceptually simple, evolution is a powerful force that pushes biological objects toward superior variants. Or it would, if the world were more simple. In reality, there is no objective notion of “superior”: natural selection is immensely complex, representing an interplay between organisms, their environment, and the underlying genetic landscape, gently favouring organisms which can survive and reproduce in a particular environment, at a particular point in time. This complexity makes evolution difficult to study, but advances in both high-throughput experimental systems and deep learning have expanded our understanding of the space of genetic possibilities more than ever before. Given these insights, a natural question arises: can we not only understand evolution, but actively steer it?

This thesis is devoted to the question of how evolutionary systems can be manipulated to behave to our advantage. The investigation is carried out in silico, leveraging modern hardware and scientific computing tools to enable large-scale computational experiments. The work proceeds in two parts. The first focuses on accelerating evolution for the development of improved proteins; the second on constraining evolution, specifically by limiting the rate of viral antigenic escape through appropriate antibody choice.

In the first part, on directed evolution, we introduce the problem setting in Optimisation strategies for directed evolution without sequencing, demonstrating how alternative experimental strategies can yield superior outcomes in fewer generations. This is complemented by Sequence-free landscape inference for directed evolution, where we develop methods to infer properties of the underlying genetic landscape in order to guide strategic decision-making.

The second part consists of ADIOS: Antibody Development via Opponent Shaping, where ideas from multi-agent reinforcement learning are applied to the problem of viral evolution. In a mechanistic model of viral adaptation, we show that antibody choice can dramatically influence the rate of viral evolution, and that an antibody that performs well over long time horizons may differ substantially from one that performs best against the initial virus.

Together, these works show how evolution can be steered in the systems studied, providing a first step towards understanding how to control real-world evolutionary systems. They offer concrete computational methods and experimentally testable strategies, opening a path towards steering increasingly realistic models and, ultimately, real-world systems, including evolving viruses.

Actions

Access Document

Files:

Authors

More by this author
Institution:
University of Oxford
Division:
MPLS
Department:
Engineering Science
Oxford college:
Reuben College
Role:
Author

Contributors

Institution:
University of Oxford
Division:
MPLS
Department:
Engineering Science
Oxford college:
Reuben College
Role:
Supervisor
Institution:
University of Oxford
Division:
MPLS
Department:
Engineering Science
Oxford college:
Reuben College
Role:
Supervisor


Engineering and Physical Sciences Research Council More from this funder
Funding agency for:
Towers, SR
Grant:
EP/S024050/1
Programme:
EPSRC Centre for Doctoral Training in Autonomous Intelligent Machines and Systems (AIMS CDT)


DOI:
Type of award:
DPhil
Level of award:
Doctoral
Awarding institution:
University of Oxford

Terms of use


Views and Downloads

Views and downloads will return soon






If you are the owner of this record, you can report an update to it here: Report update to this record

TO TOP