Bioinfo 7 Alignment - University of Illinois Urbana-Champaign
Random Graphs - Carnegie Mellon School of Computer...
Transcript of Random Graphs - Carnegie Mellon School of Computer...
![Page 1: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/1.jpg)
Random Graphs858L
![Page 2: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/2.jpg)
Descriptive Studies of Biological Graphs
• What to networks that arise in nature “look like”? What are their characteristics?
• Related to random graphs because you need to know what features can arise just by chance.
• Need to describe their properties so that:
- we can postulate how the graphs evolved.
- understand trends in how they are put together, and perhaps how they perform their functions.
- to create null models to distinguish between functional, conserved, important features, and those that are merely there by chance.
![Page 3: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/3.jpg)
Random ModelsErdos-Renyi (1960):
Create n verticesBetween every pair of vertices {u,v}, add an edge with probability p.
ShowGraphArray[Partition[Table[RandomGraph[15, p], {p, 0.1, 0.9, 0.1}], 3]]
Expected degree is (n-1)p
![Page 4: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/4.jpg)
Preferential Attachment
At time step t:• Node ut is added.• Edge {v, ut} added independently with probability:
kv�w kw
Proposed in the context of the WWW: popular pages become more popular.
Expect a scale-free degree distribution.
(Albert et al, Nature, 2000)
![Page 5: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/5.jpg)
Degree Distribution
ER has an “average” node.
SF networks don’t have a strong typical node
They have a longer tail: more nodes of higher degree.
(Jeong et al., Nature, 2000)
![Page 6: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/6.jpg)
Attack vs. Failure: Random Graph
10,000 nodes; 20,000 edges; average degree 4
(Albert et al, Nature, 2000)
Dia
met
er
![Page 7: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/7.jpg)
Attack vs. Failure on Real Graphs:
(Albert et al, Nature, 2000)
Dia
met
er
![Page 8: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/8.jpg)
The large-scale organization of metabolic networks
Jeong et al., Nature, 2000.
![Page 9: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/9.jpg)
Metabolic Network
• Nodes are substrates (small molecules, e.g.)
• Reactions are shown as black boxes. Reactions have input substrates and output substrates.
• Can think of reactions as “directed hyperedges”.
• Reactions labeled with enzyme that catalyzes the reaction.
• Actually, the analysis won’t really use the structure of the graph.
![Page 10: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/10.jpg)
Metabolic Data Sets
43 organisms
Reactions from literature
Which organisms have particular reactions predicted from the genes they have (WIT database)
Expect that metabolic networks are not at all like really “random” networks.
![Page 11: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/11.jpg)
Degree Distribution
P(k) is the fraction of nodes with in- or out- degree k.
(a) Archae(b) E.coli(c) Worm (c. elegans)(d) Average
Clearly more similar to scale-free distribution & not like an ER random graph.
P(k) ∼ k-γ, with γ = 2.2
True for all 43 organisms(Jeong et al., Nature, 2000)
![Page 12: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/12.jpg)
Network Diameter
Histogram of path lengths between substrates for E. coli
Average path length (diameter) for each of the
43 organisms.
• Surprising that diameter is the same no matter how many substrates.• More connections in complex organisms• Perhaps advantageous to maintain a small diameter so that changes in one substrate can propagate quickly.
(Jeong et al., Nature, 2000)
![Page 13: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/13.jpg)
Indeed, higher average degrees for organisms with more substrates:
(Jeong et al., Nature, 2000)
![Page 14: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/14.jpg)
Response to Mutations
But they are very robust to random knock-outs
Simulation on E coli.
Random knockouts simulate random mutations.
Except the simulation knocks out substrates, and typically it’s enzymes (edges) that are mutated.
Scale-free networks are easy to destroy: knock out the hubs (high-degree nodes)
(Jeong et al., Nature, 2000)
![Page 15: Random Graphs - Carnegie Mellon School of Computer Scienceckingsf/bioinfo-lectures/attack.pdfDescriptive Studies of Biological Graphs • What to networks that arise in nature “look](https://reader034.fdocuments.us/reader034/viewer/2022042218/5ec39e49957691153f31fdc0/html5/thumbnails/15.jpg)
Are The Same Substrates Always Hubs?
Average ranking (based on in- or out-degree)
Stan
dard
dev
iatio
n of
rank
ing
Variance increases as ranking increases: means that hubs are usually hubs, and non-hubs sometimes are more or less hubby.
Ties?
(Jeong et al., Nature, 2000)