Team:ETH Zurich/Model
MODEL
OVERVIEW
Due to the complexity of our system, an intuitive design would not be sufficient. Several components must be tuned specifically for our application, and the correct functioning of the candidate designs is not obvious. We developed a hybrid deterministic-stochastic model of our system covering the known mechanics of the components we use. The main goals include a model-based design of the system, tuning of the system components for our application and prediction of the system behavior.
The stochastic approach of our model allows us to exploit the intrinsic noise at our advantage: our model describes the
The model matches the modular sructure of the system, making easier to exchange components and test their behavior in the full system. Thanks to this feature we were able to quickly integrate and tune alternative parts like the lactate sensor and the CRISPR switch.
The parameters of the model are strictly based on literature or experimental characterization of the single components. The estiamtion of the parameters uses INSIGHT, a recent method based on Approximate Bayesian Computation (ABC) and flow cytomerty data.1
CONTRIBUTIONS
EARLY DESIGN INSIGHTS
Thanks to qualitative simulations of the system during the design phase, we managed to get early information about the behavior of the the system. In particular the model advised us on the design of the reporter system, the placement of the integrase gene and the choice of plasmid ORIs.
Moreover we were able to identify potential problems (like leakiness and sensitivity) and predict their influence on the final system. We designed and constructed the system under awareness of those critical points.
SYSTEM TUNING
Since our application targets specific time and concentration ranges we used the model to tune the components of the system. Using the sensor module was tuned for the relevant concentrations of quorum sensing molecule and the switch kinetics were adapted for the expected time scale of the measurements.
SYSTEM SIMULATION
After the characterization of the single components we were able to simulate the full system and validate its behavior. This allowed us to model the recorded signal as function of the measured fluorescence distribution, which is a critical operation during actual investigation. The stochastic nature of our model allows us to obtain additional information by taking advantage of the population effect.
MAIN RESULTS
EARLY DESIGN INSIGHTS
SYSTEM TUNING
Bottom-up tuning. 1) Reporters, 2) Switch, 3) Sensors
SYSTEM SIMULATION
MODEL STRUCTURE
For simplicity and modularity the model is divided in three parts. Those modules are designed to be easily interchangeable. Although here the model is explained in the context of IBD investigation, its structure is general enough to be employed in other associative learning applications.
Figure 1: Schematic view of the model structure.
Figure 1 shows a schematic view of the model structure. In order to maximize flexibility we choosed promoter activities as interfaces between modules. Since we tune the system in a bottom-up fashion we explicitely avoided backward dependencies.
SENSOR MODULE
The sensor module is composed by an AND gate for the association of the desired events and a regulated promoter for the induction of the output. It is the interface between circuit and external environment and models the activation of the promoters in response to the inputs.
This component is modeled deterministically as a set of ODEs using leaky Hill functions, since noise is negligible. This is intuitively supported by the high number of molecules involved and the frequency of the binding/unbinding events, which are much faster than gene expression. The hypothesis has been verified by comparing determinitic and stochastic models.
SWITCH MODULE
REPORTER MODULE
Thanks to the sponsors that supported our project:
