Component Characterization in a Growth-Dependent Physiological Context: Optimal Experimental Design

Abstract

Synthetic biology design challenges have driven the use of mathematical models to characterize genetic components and to explore complex design spaces. Traditional approaches to characterization have largely ignored the effect of strain and growth conditions on the dynamics of synthetic genetic circuits, and have thus confounded intrinsic features of the circuit components with cell-level context effects. We present a model that distinguishes an activated gene’s intrinsic kinetics from its physiological context. We then demonstrate an optimal experimental design approach to identify dynamic induction experiments for efficient estimation of the component’s intrinsic parameters. Maximally informative experiments are chosen by formulating the design as an optimal control problem; direct multiple-shooting is used to identify the optimum. Our numerical results suggest that the intrinsic parameters of a genetic component can be more accurately estimated using optimal experimental designs, and that the choice of growth rates, sampling schedule, and input profile each play an important role. The proposed approach to coupled component–host modelling can support gene circuit design across a range of physiological conditions.

Deep Reinforcement Learning for Optimal Experimental Design in Biology

“…For linear models with Gaussian measurement error, this goal is equivalent to minimising the volume of the confidence ellipsoid of the resulting parameter estimates [24]. This approach has been demonstrated to be useful even for non-linear systems [7, 8, 9, 10]. Fig 1A shows the expected outcome from a hypothetical OED application.…”

“…We then introduced algorithm refinements that focussed on eliminating the dependence on the prior knowledge of parameter values. The dependence on the true parameter values is a limitation of other OED works [9, 10, 17, 18, 19], which require ad hoc verification [9] or other workarounds [17, 18, 19]. To decouple the RL controller from the true parameters, we used a recurrent neural network to enable it to make experimental decisions based on a full experimental history of past measurements and experimental inputs.…”

“…Bayesian approaches allow global optimisation and the incorporation of prior knowledge, but can be computationally expensive. Here we build on previous work that has demonstrated the applicability of Fisher information based methods to non-linear biological systems [9, 10], including online OED of fed batch bioreactors [18, 19]. We have shown the capability to encode prior parameter knowledge and therefore optimise globally with respect to the Fisher information by training an RL controller over samples from a parameter distribution.…”

“…Applications of OED to nonlinear biological systems is challenging; techniques for efficient OED in this context have been developed via Bayesian methods [4, 5, 6] and methods based on optimisation of the Fisher information [7, 8, 9, 10]. Here, we use reinforcement learning (RL), a branch of machine learning, to develop a novel Fisher information-based OED method for parametrisation of non-linear models of biological systems.…”

Computational Methods Enabling Next-Generation Bioprocesses

“…Bandiera and colleagues [1] show how iterative optimization of chemical stimuli leads to faster and more accurate model calibration of a building block of synthetic biology: an inducible promoter. Braniff and colleagues [2] instead use Optimal Experimental Design to extract the dependence of model parameters on the physiology of the host organism.…”