complex networks and data mining on genetic databases

KnowledgeDiscovery inDatabases

throughComplex

Networks:application tophylodynamics

Luiz Max F.de Carvalho

ScientificComputingProgramme(PROCC),

FiocruzPan American

Center forFoot-and-

Mouth Disease(PAHO/WHO)

WaFiS 2012

KnowledgeDiscovery inDatabases(KDD)

ComplexNetworks

Example 1:Chitinpathwayphylogeny

Example 2:Foot-and-mouth diseasevirus in SouthAmerica

Who’s this guy?

Knowledge Discovery in Databases throughComplex Networks: application to

phylodynamics

Luiz Max F. de CarvalhoScientific Computing Programme (PROCC), FiocruzPan American Center for Foot-and-Mouth Disease

(PAHO/WHO)

WaFiS 2012

September 28, 2012


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Outline

1 Knowledge Discovery in Databases (KDD)

2 Complex Networks

3 Example 1: Chitin pathway phylogeny

4 Example 2: Foot-and-mouth disease virus in South America


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Knowledge Discovery in Databases (KDD)

Lots of data

human brain very limited processing capacity

Information → Knowledge

Increasing number of molecular data (sequences, 3Dstructures, antigenicity,. . . )

Is it possible to explore these databases to discover usefulstuff?


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Well. . . Let’s see


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



[We may use] Complex Networks

Graphs → G = (V ,E )


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Yeah, but how?

We can explore the ”dynamic signature” of these ComplexNetworks, i.e., study and compare their structural properties.Some useful formulas:

Clustering Coefficient < c >: 3×#triangles#triples

Degree distribution PK =∑∞

K ′=K pK ′

Diameter: max(d(i , j))


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Ok, Let’s work then

1 Grab n sequences;

2 Create an n × n matrix using some kind of (normalized)distance (say, S);

3 For each σ ∈ [0, 1] build M(σ) such that:

mij(σ) =

{1 if Sij > σ,

0 if Sij < σ.

In a sense, we are transforming a single network in a family ofnetworks.


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Analysis

We shall explore the relationships between these networks:First, define a higher-order neighborhood indicator function,such that you binarize the adjacency matrix with regard thepath length `, obtaining a matrix M =

∑D`=1 `M(`). Then

δ(α, β) =1

N2

N∑i=1

N∑j=1

(mij(α)

D(α)−

mij(β)

D(β)) (1)

Evaluating δ(σ, σ + ∆σ) can give some interesting insights.


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Example 1: Chitin pathway phylogeny

Proteins related to the chitin metabolic pathway from1605 complete genomes;

BLAST distances (which are asymmetric);

Search for phylogenetic relationships


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Example 1: Some results


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Example 1: Some more results


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Example 1: The expected Network(s)


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Example 2: Foot-and-mouth disease virus in SouthAmerica

S was built with phylogenetic (TN93) distances for NTand JTT distances for AA;

Try to make sense of a somewhat big data set (167 seqs);

Extract some nice patterns;


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Indexes × σ

(a) (b)


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



A nice network


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Some more developments


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Related Work

Identify transmission clusters (HIV, HCV) (Lewis et al,2008,Plos Medicine)

Explore scale-free behavior in phylodynamics (Shiino,2012, Frontiers in Microbiology)


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Future Directions

Explore the spatial aspect in the construction of SMaybe S = µ+ S(G )α

Power law analysis

Implement assortativity

Suggestions. . .


throughComplex




FiocruzPan American

Center forFoot-and-


WaFiS 2012


ComplexNetworks



Thank You!

complex networks and data mining on genetic databases

Education