Modeling and Analysis of Biological Networks

Final Project

Anh Le
anhle@mail.usf.edu

This document explains what I did for my final project. I took a
genetic circuit from Synthetic Biology from an article publised in
2013 and called Analysis and Design of a Genetic Circuit for
Dynamic Metabolic Engineering by Nikolaos Anesiadis, Hideki
Kobayashi, William R. Cluett, and Radhakrishnan Mahadevan. I
also used the iBioSims software to do the SBML model of the
circuit as well as other types of simulations.
Based on the Figure 1,2 I created 3 different SBML models
of the circuit. The first one is the SBML model of the whole
genetic circuit itself (Figure 2), the second one is the SBML
model of the genetic controller (Figure 3) and the last one is the
SBML model of the whole censor (Figure 4).

I. INTRODUCTION
Nowadays, synthetic biology has achieved significant
improvement and scientists have come to the point that we can
do bio simulation at the genetic and micro level using tools
such as iBioSims software. The genetic circuit I am studying
consists of 2 parts (the sensors and the genetic controller
plasmid). It is an integrated in designation of silico and it is
used to control gene expression dynamically. The theory is
based on a sensor unit that can sense the density and thus
toggle the genetic switch. The production of Escherichia coli
mutants benefit the most from this switch. In theory, this on-off
switch can lead to a production increase of nearly 30%
compared to the regular mutant.
Keywords: toggle switch, metabolic biology, mutants
My objectives of this project are to do different types of
simulations with the above SBML models using iBioSims. The
types of simulations that I am going to do are ordinary
differential equation (ODE) simulation and Stochastic Analysis
(SSA) simulation.

II. SIMULATIONS USING IBIOSIM

A. iBioSim
iBioSim is a software to create model, study and analyze
different types of genetic circuits. While its first priority is
genetic circuit, iBiosim can certainly work in the fields of
biochemical engineering, chemistry and other kinds of
network as well. With the help of iBioSims, we are able to
export and import Systems Biology Markup Language
(SBML) models. Furthermore, Synthetic Biology Open
Language (SBOL), a synthetic biology standard language is
also supported in iBioSim, 1.
B. SBML models
This is the image taken directly from the research paper
Analysis and Design of a Genetic Circuit for Dynamic
Metabolic Engineering.
Figure 2: SBML model of the genetic circuit.
Figure 3: SBML model of the controller
Figure 4: SBML model of the censor
C. ODE simulations
Figure 5: ODE simulation of the whole genetic circuit.
The Simulator I used for the ODE simulation is the Runge-
Kutta-Fehlberg (4,5) or rkf45. The analysis type was ODE and
there is no abstraction for this simulation. Initial time was 0,
time limit was set at 1000 with the print interval of 50. Random
seed was 314159 and the number of run is 1. The total time
taken for this simulation was about 2.6 seconds. The reason I
chose 1000-time limit and 50 print intervals is because at first I
put the default values from iBioSims, the TSD graph was very
small and very unclear. From the TSD graph above, we are
able to see that the amount of LacI and AHL both increased by
a large amount while luxI increased by a small amount. This
behavior is totally expected and it fits the characteristics of the
plasmids,3. CI repressed the productions of lacI.
 increased to a much higher extent compared to AHL. This
means that both SSA and ODE simulations of the circuits
show the same result but the number of runs might cause the
differences in the amount of AHL increased.

Figure 6: ODE simulation of the censor.

Similarly, I also used rkf 45 for the ODE simulation of the

censor. However, this time the time limit was only 100 with
print interval of 1. These numbers show the best and clearest
TSD graph. LuXR increased as well as LuxI. LacI increased as Figure 9: SSA simulation of the censor.
well but to a much smaller extent. This ODE simulation also
fits the characteristics of the serine producing E. coli mutant, 2. For the censor SSA simulation, the same parameters
were used. From the TSD graph above, we can see that the
LuxR and LuxI increased the most. This fits perfectly with the
ODE simulations of the censor. This again shows the perfect
characteristics of the substances.

Figure 7: ODE simulation of the controller.

The ODE simulation of the controller produces a rather

interesting TSD graph. The only thing increased was LacI and
it increased significantly. I used the same parameter with the
censor (100-time limit and 1.0 print interval). This behavior
does not suit the characteristics of the substance,3. This
suggests that we cannot separate the controller and the sensor
and expecting them to perform normally as if they are in the
same model.
D. SSA simulations
Figure 9: SSA simulation of the controller.
Again, the TSD graph is similar to the ODE simulations TSD
graph. The parameters used are the same as the 2 SSA
simulations above. It also proves that the controller cannot be
separated from the sensor and produce the expected behaviors.
E. LearnView Data
For LearnView Data, I use the 50th run of all SSA
simulations to plot the graph. The reason is because from the
above graphs, we can see that ODE and SSA simulations are
somewhat showing similar result so I chose the medium run to
plot the LearnView Data

Figure 8: SSA simulation of the circuit.

For SSA simulation of the whole circuit, the analysis

type was Monte Carlo, there is no abstraction. The
simulators/analyzers were Gillespie, SSA- Direct method.
Initial time was 0.0, time limit 100, print interval 50.0 and
runs 100. I tested using different parameters and I think that
the number above gives me the best graph. Similarly, to ODE Figure 10: Learn View Data of the whole circuit.
simulation, both AHL and LacI increased. However, LacI
 parameters. The above simulations results were the best
amongst the simulations.
However, the ODE simulations for the censor could be
done better as the behaviors of the substances might change
later. The reason that I still pick it as I have compared the 2 and
the above one show the initial stages of the substances clearer.

B. Seperating the censor and it controller

This might not be ideal. We can see from above
illustrations that only the behaviors of the censor are similar to
Figure 11: Learn View Data of the censor. the whole circuit and not the controller. To show the ideal
behaviors the 2 parts should be in the same genetic circuit.

REFERENCES

[1] http://www.async.ece.utah.edu/ibiosim
[2] Nikolaos Anesiadis, Hideki Kobayashi, William R. Cluett, and
Radhakrishnan Mahadevan*
Department of Chemical Engineering and Applied Chemistry,
University of Toronto, Canada, M5S 3E5
Figure 12: Learn View Data of the controller. Japan Agency for Marine Earth Science and Technology, Japan
Institute of Biomaterials and Biomedical Engineering, University of
The Learn View Data of the ODE simulations are Toronto, Canada, M5S 3G9
like that of SSA simulations so they are not attached here. ACS Synth. Biol., 2013, 2 (8), pp 442452
[3] Yang, L., Cluett, W. R., and Mahadevan, R. (2011) EMILiO: a
fast algorithm for genome-scale strain design. Metab. Eng. 13, 272
III. ISSUES ENCOUNTERED AND ASSUMPTIONS MADE TO DEAL 281.
WITH THEM

A. Parameters of the simulations

The first difficulty was to find the best parameters that
show the best simulations results. One way to deal with this
problem is do the simulations with several different