Bioinformatics: Assignment 3

Michael Sonneyboy Gboneh ([email protected])

March 14, 2016

Question 1
Given the following reaction system
1k
A −→X
2 k
2X + Y −→3X
3 k
B + X −→Y +D
4k
X −→E

Differential equation for the mass action kinetics

A (-) is when a species is used up while a (+) is used when a species is gained.The differential equation
for the mass action is

d[X]
=k1 [A] + k2 [X]2 [Y ] − k3 [B][X] − k4 [X]
dt

d[Y ]
= − k2 [X]2 [Y ] + k3 [B][X]
dt

1
Question 2
Explicit definition of the propensity function.

Given a reaction
k
s1 x1 + s2 x2 + · · · + sn xn −
→Y

The explicit definition of the propensity function is written as

n  
Y xi
aj =cj
i=1
si

where
n
k Y
cj = sj !,
(nA V )n−1 j=1

aj is the propensity function

cj is the rate constant
xi is the reactant for a particular reaction
si is the coefficient of reactant xi
nA is the Avogadro‘s number
and V is the volume of the cell with the species
From out reactions we get
A.
1 k
A −→ X

k1
C1 =
(nA V )1−1
=k1

n  
Y A
=A
i=1
1

a1 = k1 A Propensity function

B.
2 k
2X + Y −→3X

2
 k2
C2 = 2!1!
(nA V )3−1

2k2
=
(nA V )2

n   
Y X Y X(X − 1)Y
=
i=1
2 1 2

2k2 X(X − 1)Y

a2 =
(nA V )2 2

k2 (X)(X − 1)Y
= Propensity function
(nA V )2

C.
3 k
B + X −→Y +D

k3
C3 = 1!1!
(nA V )2−1

k3
=
nA V

n   
Y B X
= BX
i=1
1 1

k3 BX
a3 = Propensity function
nA V

4.
4k
X −→E

k4
C4 =
(nA V )1−1
=k4

3
 n  
Y X
=X
i=1
1

a4 = k4 X Propensity function

4
Question 3
Probability that no reaction fires over time interval [s,t]

This probability is given by

Pr(no reaction fires in [s,t]) = e−a0 (x(s))(t−s)

where
The sum of all propensities in state x(s) is given by a0 (x(s)).
The probability that a reaction will fire on the time interval (t, t + dτ ) is given byP (τ, j).
The probability that no reaction occurs on the interval (s, t)is given by P0 (τ )
The probability that the reaction will occur on the interval (t, t + dτ ) is given by aj .

P (τ, j)dτ = P0 (τ )aj dτ

The probability that no reaction fires on the interval (t, t + dτ ) is

M
X
P0 (τ + dτ ) =P0 (τ )(1 − aj dτ )
k=1
=P0 (τ )(1 − a0 dτ )

This gives

P0 (τ + dτ ) − P0 (τ ) = − P0 (τ )a0
P0 (τ + dτ ) − P0 (τ )
= − P0 (τ )a0
dτ

Taking the limit as dτ −→ 0 :

dP0 (τ ) = − P0 (τ )a0 dτ
dP0 (τ )
= − a0 dτ
P0 (τ )

Integrating with respect to dτ

lnP0 (τ ) =a0 τ
P0 (τ ) =ea0 τ , where τ =t−s

5
Question 4
Proportional relationship for the likelihood of a path in the complete data case:
( M ) ( Z T )
Y
L(c, x) ∝ crj i exp −ci gi (x(t))dt
i=1 0

Solution
From slide 14 of week three and day three the lecture notes,we have the following:
  ( Z )
Y r  T
L(x) = apj (x(tj−1 )) exp − a0 (x(t))dt (1)
 
j=1 0

From this, we can write

  ( Z )
Yr  T
L(c; x) = apj (x(tj−1 ), cj ) exp − a0 (x(t), c)dt (2)

j=1
 0

But we know that

M
X
a0 (x, c) = ai (x, ci )
i=1

Then   ( Z )
Yr  M
T X
L(c; x) = apj (x(tj−1 ), cj ) exp − ai (x, ci )dt (3)

j=1
 0 i=1

Let ai (x, ci ) = ci gi (x), i = 1, ..., M Substituting this into equation (3) and simplifying, we have that
  ( )
Yr  M Z T
X
L(c; x) = cpj gpj (x(tj−1 ), cj ) exp − ci gi (x, ci )dt (3)

j=1
 0 i=1
( M
) ( M Z
)
Y X T
∝ crj i exp − ci gi (x(t))dt
i=1 i=1 0

M
( )
Y Z T
= crj i exp −ci gi (x(t))dt simplifying further we generate the below poduct
i=1 0

M
Y
= Li (ci , x) QED
i=1

Where c = (c1 , c2 , ..., cM ) is the vector of stochastic rate constant and x a full sample path of the model.
The component likelihoods are defined by
( )
Z T
Li (ci , x) = crj i exp −ci gi (x(t))dt i = 1, 2, ..., M
0

6
Question 5
The maximum likelihood estimator for each rate constant.
ri
cbi = Pr
g
j=0 i (x(tj ))(tj+1 − tj )

for i = 1,· · · ,M
M is the number of reactions in the model.
From question 4,
RT
Li (ci , x) = (ci )ri eci 0
gi (x(t))dt

We take log both sides

Z T
logLi (ci , x) =ri log(ci ) − ci gi (x(t))dt
0

Differentiating with respect to ci , we obtain

T
L0i (ci , x) ri
Z
= − gi (x(t))dt
Li (ci , x) ci 0
 Z T 
0 ri
Li (ci , x) =Li (ci , x) − gi (x(t))dt
ci 0

Inorder to maximised the likelihood, we equate L0i (ci , x) to 0

Li (ci , x) =0 or
Z T
ri
− gi (x(t))dt =0
ci 0
Z T
ri
= gi (x(t))dt
ci 0
ri
ci = R T
0
gi (x(t))dt

gi (x(t)) is a piecewise constant, therefore

ri
cbi = Pr
j=0 gi (x(tj ))(tj+1 − tj )