graph theory and epidemics - math circlemathcircle.berkeley.edu/.../2020/15-04_bmc-handout.pdf ·...
TRANSCRIPT
![Page 1: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/1.jpg)
Graph theory and epidemicsBerkeley Math Circle
15 April, 2020
![Page 2: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/2.jpg)
About me
Mark PenneyPerimeter Institute /UWaterlooVisitor at MSRI
1 30
![Page 3: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/3.jpg)
About me
From Newfoundland, CanadaIts nice enough
2 30
![Page 4: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/4.jpg)
About me
From Newfoundland, CanadaIts nice enough... Except in the winter
3 30
![Page 5: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/5.jpg)
Mathematics of infectious disease
I don’t study infectious diseasesI Topology (knots, surfaces,...) applied to theoretical physics(particle physics, quantum computing)
Wanted to understand how we model epidemicsI Personal reasons: Knowledge is powerI Maybe contribute to COVID-19 research
4 30
![Page 6: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/6.jpg)
COVID-19: Flattening the curve
5 30
![Page 7: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/7.jpg)
Fundamental number: Reproductive rate
R0 = average number of newinfectionsPhase transition at R0 = 1:I R0 < 1: disease dies o�I R0 > 1: epidemic outbreak
Determines the trajectory of outbreakFlattening the curve⇔ reducing R0
6 30
![Page 8: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/8.jpg)
How do we determine R0?
Depends on assumptions of modelSimplest model: R0 = µ · TI µ = average number of contactsI T = probability of spread
To reduce R0:I Reduce µ: social distancingI Reduce T: Handwashing, masks
What are the assumptions of this model?Main assumption: ‘Homogeneous mixing’. People interact witheach other at random.Does not take into account social structure!
7 30
![Page 9: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/9.jpg)
Contact networks
People are not like molecules of a gas!More realistic: Understand the network or graph describingcontacts
8 30
![Page 10: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/10.jpg)
Contact networks
Recall: GraphsA graph is a collection of vertices joined by edges
Coronaviruses spread by closeproximity interactions (CPIs)Build contact network for group ofpeople:I Vertices: People in groupI Edges: connect people having CPIs
9 30
![Page 11: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/11.jpg)
Examples of contact networks
What sort of social structure leads to the following contactnetworks?
5 people all sharing CPIsI Household, group of friends, small sports team
6 people having CPI with 1 otherI Doctor visiting patients, tutor
7 people in 2 clustersI The households of 2 friends
10 30
![Page 12: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/12.jpg)
Degrees of contact networks
Recall: DegreeThe degree of a vertex in the number of edges connected to it
All vertices have degree 4I Every family member is in contact with theremaining 4
Central vertex has degree 6, remainder degree 1I Doctor as 6 contacts, patients have 1
Within each household, each family member is incontact with the remainder. Friends have higherdegree
11 30
![Page 13: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/13.jpg)
Degree distribution
Takeaway messageSocial structures lead to varying degrees in contact networkHelpful to plot number of vertices having each degree
12 30
![Page 14: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/14.jpg)
Degree distribution
De�nitionThe degree distribution of a graph is
d 7→ pd =# vertices of degree d
# vertices
Probability that randomly chosen vertex has degree dObtained from previous graphs by dividing by # vertices
ExerciseCompute the average degree of our 3 examples. Can you �nd asimple, general formula?
13 30
![Page 15: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/15.jpg)
Average degree
Recall: Average of probability distributionGiven probability distribution pd, the average is:
µ = 〈d〉 = 0 · p0 + 1 · p1 + 2 · p2 + . . . =∑dd · pd
The average degrees are:‘Household’ network: µ = 4‘Doctor’ network: µ = 1 · 67 + 6 · 17 =
127 = 1.7...
‘Friends’ network: µ = 2 · 27 + 3 · 47 + 4 · 17 =207 = 2.9...
14 30
![Page 16: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/16.jpg)
Average degree formula
PropositionThe average degree of a graph is
µ =2 ·# edges# vertices
Proof
µ =
∑d d · # vertices of degree d
# vertices∑dd · # vertices of degree k =
∑vertices v
# edges meeting v
This sum counts each edge twice
15 30
![Page 17: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/17.jpg)
Regular graphs and homogeneous mixing
Regular graphA graph is called d-regular if each vertex has degree d.
Homogeneous mixingAssumption lead to R0 = µ · T. Assumes that contact network isd-regular with d ≈ µ
16 30
![Page 18: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/18.jpg)
Constructing regular graphs
Can you make a 3-regular graph with 5 vertices?7?No! Must have 3 = 2·# edges
# vertices
Must d ·# vertices is even!
QuestionCan you make a 4-regular graph with 6 vertices?
Arrange vertices in circle
17 30
![Page 19: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/19.jpg)
Constructing regular graphs
Can you make a 3-regular graph with 5 vertices?7?No! Must have 3 = 2·# edges
# vertices
Must d ·# vertices is even!
QuestionCan you make a 4-regular graph with 6 vertices?
Arrange vertices in circleJoin each vertex with 2 nearest on each side
18 30
![Page 20: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/20.jpg)
Constructing regular graphs
Can you make a 3-regular graph with 5 vertices?7?No! Must have 3 = 2·# edges
# vertices
Must d ·# vertices is even!
QuestionCan you make a 4-regular graph with 6 vertices?
Arrange vertices in circleJoin each vertex with 2 nearest on each side
19 30
![Page 21: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/21.jpg)
Constructing regular graphs
QuestionCan you make a 5-regular graph with 8 vertices?
Arrange vertices in circleJoin each vertex with 2 nearest on each side
20 30
![Page 22: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/22.jpg)
Constructing regular graphs
QuestionCan you make a 5-regular graph with 8 vertices?
Arrange vertices in circleJoin each vertex with 2 nearest on each sideJoin opposite vertices
ObservationCan use these constructions to make d-regular graph with Vvertices, assuming d · V is even and V is large enough
21 30
![Page 23: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/23.jpg)
Realistic degree distributions
Simplest model assumes regular contact networkSaw that even simple contact networks are far from regular
QuestionWhat are the degree distributions of realistic contact networks?
Number of daily CPIs in UShigh schoolBlue = students, pink =teacher, green = sta�What other statistical ideasmight be useful here?
22 30
![Page 24: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/24.jpg)
Standard deviation
Standard deviation σ measure of ‘width’ of distribution
σ2 = (0−µ)2p0+(1−µ)2p1+(2−µ)2p2+ · · · =∑d(d−µ)2pd
σ2 = Average of (distance from µ)2
23 30
![Page 25: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/25.jpg)
Computing standard deviation
Formulaσ2 = (0− µ)2p0 + (1− µ)2p1 + (2− µ)2p2 + · · · =
∑d(d− µ)2pd
µ = 4I σ2 = (4− 4)2 55 = 0
µ = 127
I σ2 =(1− 12
7
)267 +
(6− 12
7
)217 =
15049 = 3.06...
µ = 207
I σ2 =(2− 20
7
)227 +
(3− 20
7
)247 +
(4− 20
7
)217 =
2049 = 0.408...
24 30
![Page 26: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/26.jpg)
Reproductive rate v2
Better model for R0Correction factor accounting for width
R0 = Tµ(1+ σ2
µ2
)
‘Household’ network: σ2
µ2 =042 = 0
I For regular graphs σ = 0 so simple = better model
‘Doctor’ network: σ2
µ2 =(15049
)2 ( 712)2
= 625196 = 3.19...
I Better model gives R0 4 times larger!
‘Friends’ network: σ2
µ2 =(2049
)2 ( 720)2
= 149 = 0.02...
I Better model is only slightly larger than simple model
25 30
![Page 27: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/27.jpg)
Reproductive rate v2
σ2
µ2 = 0.118Better model for R0 isalmost 12% higher
Takeaway messages
Since σ2
µ2 ≥ 0, simple R0 ≤ better R0Wider degree distribution⇒ faster spread of disease
26 30
![Page 28: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/28.jpg)
How popular are your friends?
How to make a random graph
1. Create vertices with chosendegrees
2. Connect the edges at random
TheoremOn average, a contact has µ
(1+ σ2
µ2
)contacts
⇒ since ≥ µ, your friends have more friends than you!
27 30
![Page 29: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/29.jpg)
Proof
Question:What is the probability that an edge meets a vertex of degree d?
Answer:
#edges meeting degree d2E =
d ·#vertices of degree d2E
=
(d
2E/V
)(#vertices of degree d
V
)=d · pdµ
The average degree of a vertex meeting a random edge is∑dd ·(d · pdµ
)=1µ
∑dd2 · pd
28 30
![Page 30: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/30.jpg)
Proof
Observe that,
σ2 =∑d(d− µ)2pd =
∑dd2 · pd − 2µ
∑dd · pd + µ2
∑dpd
=∑dd2 · pd − µ2
Therefore,
µ
(1+ σ2
µ2
)=µ2 + σ2
µ=1µ
∑dd2 · pd
29 30
![Page 31: Graph theory and epidemics - Math circlemathcircle.berkeley.edu/.../2020/15-04_BMC-handout.pdf · Graph theory and epidemics Berkeley Math Circle 15 April, 2020. About me Mark Penney](https://reader033.vdocument.in/reader033/viewer/2022053015/5f14143b1830ce42287623ba/html5/thumbnails/31.jpg)
Final words
Reproductive rate R0 determinesspread of diseaseGood model: R0 = T
[µ(1+ σ2
µ2
)]I Second factor is average degreeof a neighbor
To �atten the curve need to alsoreduce ‘width’ of contact networkI e.g., ‘super-spreaders’, largegatherings, schools
Thank you very much for your attention!
30 / 30