In this section

Research

Biological systems - including the simplest cells - exhibit a broad range of functions to thrive in their environment. Research in the Imperial College Centre for Synthetic Biology is focused on the possibility of engineering the underlying biochemical processes to solve many of the challenges facing society, from healthcare to sustainable energy. In particular, we model, analyse, design and build biological and biochemical systems in living cells and/or in cell extracts, both exploring and enhancing the engineering potential of biology.

As part of our research we develop novel methods to accelerate the celebrated Design-Build-Test-Learn synthetic biology cycle. As such research in the Centre for Synthetic Biology highly multi- and interdisciplinary covering computational modelling and machine learning approaches; automated platform development and genetic circuit engineering ; multi-cellular and multi-organismal interactions, including gene drive and genome engineering; metabolic engineering; in vitro/cell-free synthetic biology; engineered phages and directed evolution; and biomimetics, biomaterials and biological engineering.

Publications

Citation

BibTex format

@article{Poole:2025,
author = {Poole, W and Ouldridge, T and Gopalkrishnan, M},
journal = {Journal of the Royal Society Interface},
title = {Autonomous learning of generative models with chemical reaction network ensembles},
year = {2025}
}

Download

RIS format (EndNote, RefMan)

TY  - JOUR
AB  - Can a micron sized sack of interacting molecules autonomously learn an internalmodel of a complex and fluctuating environment? We draw insights from controltheory, machine learning theory, chemical reaction network theory, and statisticalphysics to develop a general architecture whereby a broad class of chemical systemscan autonomously learn complex distributions. Our construction takes the form ofa chemical implementation of machine learning’s optimization workhorse: gradientdescent on the relative entropy cost function which we demonstrate can be viewedas a form of integral feedback control. We show how this method can be applied tooptimize any detailed balanced chemical reaction network and that the constructionis capable of using hidden units to learn complex distributions.
AU  - Poole,W
AU  - Ouldridge,T
AU  - Gopalkrishnan,M
PY  - 2025///
SN  - 1742-5662
TI  - Autonomous learning of generative models with chemical reaction network ensembles
T2  - Journal of the Royal Society Interface
ER  -

Download