Predictably Profitable Paths in Metabolic Networks

Ehsan Ullah, Prof. Soha HassounDepartment of Computer Science

Mark Walker, Prof. Kyongbum LeeDepartment of Chemical and Biological Engineering

Tufts University

Engineered Pathway Interventions

(Atsumi et al., 2008) (Trinh et al., 2006) (Steen et al., 2010)

Embedding

new pathways

Removing

pathways

Improving

existing

pathways

2

Enumeration ◦Elementary Flux Mode

(Schuster et al., 2000) Graph traversal

◦ Dominant-Edge Pathway Algorithm(Ullah et al., 2009)

◦ Favorite Path Algorithm*

Pathway Analysis

s

b

R1

c

R2

e

R4

t

R6

R5

d

R3

Dominant-Edge 1st

3rd

2nd

4th

3

*Unpublished

Flux variations arise from different conditions

Given a metabolic network graph G = (V,E), source vertex s and destination vertex t and a flux range associated with each edge, find the predictably profitable path in the graph

Problem: Pathway Analysis in Presence of Flux Variations

4

R5

(4)

d

R3

(4)

R5

(4)

d

R3

(4)

A network in which any path from s to t can carry at minimum vp amount of fluxGp = G(V,E)

such that we ≥ vp

vp is obtained from the best flux-limiting step

Profitable Network

s

b

R1

(10)

c

R2

(6)

e

R4

(6)

t

R6

(10)

5

R5

[3 11]

d

R3

[7 12]

R5

[3 11]

d

R3

[7 12]

A path in the network having reactions with smallest variations in flux

Predictable Path

s

b

R1

[10 15]

c

R2

[8 14]

e

R4

[6 10]

t

R6

[9 18]

6

1. Identification of profitable networka) Assign the lower limit of each flux range as edge

weightb) Find flux limiting step using favorite path algorithmc) Prune all edges having weight less than the flux

liming step found in (b)

2. Identification of predictable path in profitable network

a) Assign the flux ranges as edge weightb) Use favorite path algorithm to find predictably

profitable path

Approach to Find Predictably Profitable Path

7

Escherichia coli◦ 62 Reactions◦ 51 Compounds

Liver Cell◦ 121 Reactions◦ 126 Compounds

Test Cases

8

Escherichia coli

9

Production of ethanol from glucose in anaerobic state

Flux data generated from Carlson, R., Scrienc, F. 2004

10

glucose

ethanol

Escherichia coli

PEP

Pyruvate

Flux-limiting step

11

Flux Limiting

Step

glucose

ethanol

Escherichia coli

PEP

Pyruvate

Flux-limiting step Profitable network

12

Profitable

Network

glucose

ethanol

Escherichia coli

PEP

Pyruvate

Flux-limiting step Profitable network Predictably profitable

path Glycolysis is more

predictable than PPP Matches maximal

production path identified by (Trinh et al., 2006)

13

Glycolysis

glucose

ethanol

Escherichia coli

PEP

Pyruvate

Production of glutathione from glucose Flux data taken from HepG2 cultures* Two observed states

◦ Drug free state◦ Drug fed state (0.1mM of Troglitazone)

Liver Cell

*Unpublished results

14

Liver Cell

15

glucose

glu

cys

ala

gly

gluglutathione

akg

akg

lys

Liver Cell

Drug free state

16

glucose

glu

cys

ala

gly

gluglutathione

akg

akg

lys

Liver Cell

Drug free state◦ PPP, Alanine

biosynthesis, Lysine degradation

17

glucose

glu

cys

ala

gly

gluglutathione

akg

akg

lys

Liver Cell

Drug fed state

18

glucose

glu

cys

ala

gly

gluglutathione

akg

akg

lys

Liver Cell

Drug fed state◦ PPP, Cystine

biosynthesis

19

glucose

glu

cys

ala

gly

gluglutathione

akg

akg

lys

Efficient way of identifying target pathways for analyzing and engineering metabolic networks

Capable of handling variations in flux data Polynomial runtime

Conclusions

20