Problem Set 5 Systems Biology (7.32/7.81J/8.

591J)

Problem Set 5

1 Waiting times for chemical reactions (8 points)

In the previous assignment, we saw that for a chemical reaction occurring at rate r, the distribution
of waiting times τ between reaction events is given by:

p(τ )dτ = re−rτ dτ

a. By integrating over time, verify that the distribution is normalized

b. Here we will see a method for obtaining a random variable from a given distribution, starting
from a uniformly distributed random variable:

Suppose u is a random variable with some distribution p(u), and θ(u) is a function of u.
∗
Then the probability that θ lies between θ(u) and θ(u + du) is the same as the probability
∗
that u lies between u and u + du. This gives us a recipe for calculating the distribution p(θ):

du
p(θ)dθ = p(u)du =⇒ p(θ) = p(u)| |
dθ
Assuming u is uniformly distributed between 0 and 1, show that the number

1 1
θ(u) = ln( )
r u
is distributed precisely as required for the waiting times above.

2 Stochastic simulation of an auto-regulatory system (22 points)

In this problem you will learn how to implement a stochastic simulation. You will also use the
simulation to unveil the role of negative auto-regulation in reducing fuctuations.

a. Consider a simple diferential equation that describes the rate of change of some molecule x:
dx
= β − αx
dt

\hat is the solution of the diferential equation? Let's assume β = 6, α = 3 and the initial
condition x(t = 0) = x0 .

b. Unfortunately, the solution of the diferential equation is deterministic and fails when fuc-
tuations become important. Specifcally, when the system has reached the steady state, the
deterministic solution says that x will no longer change. But if x were to describe a real
mRNA, then x should fuctuate about its mean value, since translation and degradation are
stochastic processes. Consistency with reality requires a stochastic description that respects
both the probabilistic nature of events and the discreteness of molecule numbers. From now

Problem Set 5 Systems Biology (7.32/7.81J/8.591J)

on, x will represent the molecule numbers instead of molecule concentration, and the reaction
rates (β and αx) will be regarded as probabilities per unit time of an reaction taking place.
β will be the probability per unit time of the creation of a molecule x, and αx will be the
probability per unit time of the degradation of a molecule x. So let's start by returning to
the actual underlying reactions:

β
[DN A] → [DN A] + X
α
X → ∅

1. Consider a system that has x0 = 3 at time t = 0. \hat is the probability that the
system will transition to the state x = 2 during the next transition? (i.e., what is the
probability that degradation occurs before transcription?)

2. \hich aspect of the stochastic process does the deterministic diferential equation de-
scribe successfully?

c. Let's consider how to perform stochastic simulations of chemical reactions. A rather inef-
cient (and potentially inaccurate) method to perform the stochastic simulation is by deter-
mining what process takes place during infnitesimal time intervals dt. Say the system starts
at the state x0 = 3. The simulation examines what happens during the frst "infnitesimal"
time interval dt = 10−10 .The probability of degradation is pdeg = αx × 10
−10
= 9 × 10−10
−10
while the probability of transcription is ptrans = β × 10 = 6 × 10−10 . A random number
m is drawn from a uniform distribution between 0 and 1. \hat happens next depends on
the value of m as follows:

(1) If m < pdeg → one molecule of x is degraded, reaction rates are updated and the simula-
tion proceeds to the next time step.
(2) If pdeg < m < pdeg + ptrans → one molecule of x is transcribed, reaction rates are updated
and the simulation proceeds to the next time step.
(3) If pdeg +ptrans < m → nothing happens and the simulation proceeds to the next time step.

A better way to approach the simulation is using Gillespie's "frst-reaction" method. This
method is much more efcient and is analytically exact. The main idea is to determine which
of the two competing processes (degradation or transcription) takes place frst and at what
time it occurs. Gillespie shows that you can do it in the following manner: for each process (i)
generate a waiting time (τi ) from the probability distribution (Pi (τi )): Pi (τi ) = ki exp[−ki τi ].
The process that occurs frst is the one with the shorter waiting time. The time at which it
occurs is just the waiting time.

Sketch an algorithm for simulating stochastic reactions using Gillespie's method. In this case
we are dealing with only one species, the output of a simulation run should be the number
of molecule at diferent time points.

d. COMPUTATION Now consider the following two systems:

System A: a system in which protein X is expressed constitutively. In a deterministic setting,

the concentration of X would evolve according to

dx
= A − γx.
dt

Problem Set 5 Systems Biology (7.32/7.81J/8.591J)

System B: a system in which protein X represses its own synthesis. In a deterministic

setting, it would evolve according to

dx K
=B − γx
dt K +x

Problem 2d Image and problem courtesy of Alexander

van Oudenaarden. Used with permission.

1. Compute the fxed points of the deterministic systems and their stability.

2. In the following, we will interpret the reaction rates as probabilities per unit time of an
reaction taking place, as we discussed in b). Based on the algorithm that you sketch
in c), implement a discrete stochastic simulation for System A, for both A = 10 and
A = 100. \ithout loss of generality, we set γ = 1. \e can also measure volume in
units of cell size, thus concentrations are equivalent to molecule numbers.

i. Plot a time series of your simulation results. Compare it to the numerical solution
of the corresponding diferential equation.

ii. Run the system for a long enough time so that it has reached a steady state. Plot a
histogram of x. Verify that the variance and mean of the steady state distribution
are consistent with Poisson statistics.

3. For System B, set K = 100 and choose B such that (x) = 100 is a steady state.
Implement a discrete stochastic simulation to demonstrate that the variance of the
auto-regulatory system is lower than that of the constitutive system given the same
mean.

3 Deterministic and stochastic descriptions of a system with

nonlinear degradation (10 points)
Let's consider a fxed volume system in which a protein X, which is constitutively expressed at
a rate k, acts as its own degradation enzyme, i.e. once there are enough proteins around, the
following irreversible reaction X +X →X can take place with a rate γ.
a. If there are n proteins, in how many diferent ways can one enzymatic degradation occur?

b. \rite the master equation for the process.

d�n) 2 2
c. \rite an equation for in terms of the mean (n) and the variance V = (n ) − (n) . Note
dt
that in this case the variance will not go away as it did in the cases shown in class where
reaction terms were linear.

d. \rite the diferential equation for the corresponding deterministic system.

Problem Set 5 Systems Biology (7.32/7.81J/8.591J)

e. Assuming that the variance of the steady state distribution is known, what conditions on the
variance will ensure that the steady state mean is similar to the prediction of the deterministic
system?

4 Growth of a population (10 points)

a. Show that for any birth-death Markov process

d (n2 )
= 2 (n(fn − gn )) + (fn + gn ) ,
dt

where fn and gn are birth and death probabilities.

In the questions below we will consider a population of cells, which is characterized by

cell density n.

b. Assume fn = αn and gn = βn. \rite down the equations governing the time evolution of
the mean (n) and the variance V of the population size n.

c. For the description of the evolution of population size, is it necessary to know both α and β
or knowing α−β is enough? Explain.

MIT OpenCourseWare
http://ocw.mit.edu

8.591J / 7.81J / 7.32 Systems Biology

Fall 2014

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.