matickulous_modelling

Kinetic models provide an exciting opportunity for strain design, however, this tool is currently under utilized. We hope to turn these models into an essential piece of the Design-Build-Test-Learn cycle.

Introduction

• Bayesian Optimisation using Gaussian Processes --> In this instance it will be surrogate modelling.
• Construction of Kernel (defines the prior over the system) first order approximations can be incorporated.
• Feasibility of kinetic models in DBTL cycle.

Figures

• Error of validation datasets as model size increases
  • Central Carbon Metabolism in E. Coli/Yeast with increasing scope
  • Should give an indication of scaling of Gaussian Processes
• Comparison of learning rates with different optimization techniques
  • Optimizing enzymes directly in Kinetic Model
  • Training phase using random enzyme concentrations
  • Training phase using loss function centered on uncertainty

Results

• Tuning of Loss Function
  • Design of Experiments
  • Strain Design Application