You are on page 1of 4

Systems Biology Problem Set 11

Vincent Huang, December 6th, 2019

1 Luria-Delbruck Experiment
Part a: The probabilities given don’t take into account that the probability of each new cell being green is not
1 but instead 1 − a. Therefore the probabilities should instead be (1 − a)3 , a, a(1 − a)2 , a(1 − a)2 , a2 (1 − a). Thus
the mean is µ = 0 · (1 − a)3 + 2 · a + 1 · a(1 − a)2 + 1 · a(1 − a)2 + 2 · a2 (1 − a) = 2a(2 − a) . Now the variance is
σ 2 = E[X 2 ]−σ 2 = 0·(1−a)3 +4·a+1·a(1−a)2 +1·a(1−a)2 +4·a2 (1−a)−4a2 (2−a)2 = −4a4 + 14a3 − 16a2 + 6a .
−4a4 + 14a3 − 16a2 + 6a −2a3 + 7a2 − 8a + 3
Part b: The ratio of variance to mean is lima→0 = lima→0 =
2a(2 − a) 2−a
3
. If they only acquire the resistance in the final generation, the mean will be 2a and the variance is 2a(1 − a),
2
so the ratio in this case is 1 − a, which as a → 0 approaches 1 .
Part c:
Part 1: Since growth rate is one per unit of time, at time t there are 2t cells.

Part 2: At the start of the interval there are 2t cells, and mutation rate per cell is µ, so we expect 2t µ dt total
mutations.
Part 3: For each cell the expected number of mutations is a Poisson distribution of mean λ = µ dt, so
since the sum of Poissons is still a Poisson this means that the the total number of mutations should be a
Poisson distribution with mean λ = 2t µ dt . In particular this means the variance is also λ = 2t µ dt .

Part 4: T − t time has passed since then, so the mean has increased by a factor of 2T −t , resulting in an ending
mean of 2T µ dt . Similarly, the variance increases by 4T −t , so the resulting variance is 22T −t µ dt .
Part 5: The mean of the sum of two independent random variables is the sum of the means, so in the continuous
version this means the mean should the integral over time of the individual means, so we have that the mean
RT
should be 0 2T µ dt = T 2T µ . Similarly, variance is also additive over independent variables, so the variance
RT RT 0.5T − 1 2T − 1
is 0
22T −t µ dt = 22T µ 0
2−t dt = 22T µ · = 2T µ · . Thus the ratio of variance to mean is
ln 0.5 ln 2
2T − 1
.
T ln 2
Part 6: In the acquired mutation hypothesis each of the 2T individuals at the end acquires a mutation at rate µ
more or less independently, so the mean is 2T µ and the variance is 2T (µ−µ2 ) (we just add the mean and variances
2T (µ − µ2 )
per individual). Thus the ratio of variance to mean, as µ → 0, is limµ→0 = limµ→0 1 − µ = 1. Thus
2T µ
like in the simpler model, the ratio of variance to mean is much higher assuming the mutations are acquired
over time and selected for by the environment as opposed to assuming the mutations are acquired at the last
step in response to the environment.

2 Maximum of n IID Random Variables


Part a: If Sn ≤ z, this implies s1 , s2 , . . . , sn ≤ z , since Sn is the maximum of the si . Therefore we have
n
Y
from independence that P (Sn ≤ z) = P (si ≤ z) = P (s1 ≤ z)n . This tells us that the density of Sn is the
i=1
Z z
n n−1
derivative of P (s1 ≤ z) with respect to z, which is just np(z)P (s1 ≤ z) = np(z) p(s) ds .
−∞

1
Part b: The histogram and CDF for the normal distribution are shown below:

The histogram and CDF for the exponential distribution are shown below:

Both distributions end up having pretty similar shapes which peak relatively quickly but are have a long tail
to the right, making them right-skewed. This tells us that when clonal interference is important, there will
be many mutations which lead to moderate positive fitness growths and a few mutations which lead to large
changes in fitness.

3 Fixation Probability in the Moran Process


Consider the next time the number of mutants in the population changes. This happens either if a mutant
replaces a non-mutant or a non-mutant replaces a mutant. The probability of a mutant replacing a non-mutant
N −n rn N −n
is · since the individual being replaced has a chance of being a non-mutant and the
N N − n + nr N
rn
chance the new individual is a mutant is because the n mutants have an advantage of r. Similarly,
(N − n) + rn
n N −n
the probability of a non-mutant replacing a mutant is · .
N N − n + rn
This shows that the probability of a mutant replacing a non-mutant is r times the probability of a non-mutant
r 1
replacing a mutant, so we have a chance of going to n + 1 mutants and a chance of going to n − 1
r+1 r+1
Fn−1 rFn+1
mutants, hence Fn = + as desired. We know from extreme values that F0 = 0, FN = 1 .
r+1 r+1
Fn−1 rFn+1
Now note that the recurrence Fn = + has characteristic polynomial rλ2 − (r + 1)λ + 1 = 0, so
r+1 r+1
1 1
λ = 1, , hence it follows that Fn = α · 1n + β · n for some constants α, β. From F0 = 0, FN = 1 we find that
r r
rN rN
α=− N
,β = so we get
1−r 1 − rN

rN rN 1 − r−n
Fn = − + = .
1 − rN (1 − rN )rn 1 − r−N

2
4 Evolution in Finite Populations
Part a:
1 − 1r 1
1 − 1.001
Part 1: Since N = 100 and s = .001, we know xi = = ≈ .011 . Note that approximation
1 − r1N 1
1 − 1.001 100
1
with s doesn’t work in this regime since sN << 1, so 1 − N is not approximately 1.
r
1
1 − 1.001 20
Part 2: This time we have xi = 1 ≈ .208 .
1 − 1.001 100

Part b: Since population A has a much higher relative growth rate than populations B, C, we can treat rB , rC
1 − 21 1
as approximately 1 when computing the probability that individual A fixes. Then xA = 1 ≈ . Now
1 − 2100 2
the probability that B fixes is the probability
 that A1 doesn’t establish multiplied by the probability that A
1 − 1.01

1
establishes, which is therefore xB = 1 − × ≈ .0078 . The probability that C fixes is essentially
2 1 − 1.011100
xC = 0 since C is weaker than the population and has population size 1. To formally compute this, note the
probability that C fixes is the probability that A and B don’t establish multiplied by the probability that C
1
1 1 − .99
fixes, which becomes xC ≈ × 1 × ≈ .0029 .
2 1 − .991100
Part c:
Part 1:
1
1−
Part i: The equation should be approximately y ≈ 1000 1 + s , shown below. The 1000 is there to
1
1−
(1 + s)10
rescale the distribution to have integral 1.

1
1−
Part ii: The equation should be approximately y ≈ 20000 1 + s , shown below. The 20000 is there
1
1−
(1 + s)1000
to rescale the distribution to have integral 1. When s ≈ 0 this is approximately a line (in particular, y ≈
20(1 + 999s)).

3
Part 2:
1
1− 1+s
Part i: The equation should be approximately y ≈ .01e−.01s · 1 , which integrates to roughly 1 and is
1− (1+s)10
shown below.

1
1− 1+s
Part ii: The equation should be approximately y ≈ .01e−.01s · 1 , which integrates to roughly 1 and
1− (1+s)1000
is shown below.

You might also like