from reaction networks to information flow
TRANSCRIPT
From Reaction Networks to Information Flow
Using Modular Response Analysis to Track Information inSignalling Networks
Pascal Schulthess
Charite – Universitatsmedizin Berlin
October 15, 2010
Reaction networks → information flow?
MRA▶▶▶
Oda and Kitano (2006) Weinberg (2006)
2 / 1
Reaction networks → information flow?
B
A Ap
Bp
MRA▶▶▶ Ap
Bp
3 / 1
The dynamic behaviour of a biochemical reaction system isdetermined by
d
dt
c(t) =Nv(c(t))with
c ∈ Rm concentration vector
N ∈ Rm×n stoichiometric matrix
v ∈ Rn reaction rate vector.
4 / 1
Example – System
ESE
S
P
1
3
2
d
dt
c(t) =Nv(c(t))
d
dt
���������
S(t)E(t)
ES(t)P (t)
���������=���������
−1 0 1−1 1 01 −1 00 1 −1
���������
�������k+1ES(t) − k−1E(t)S(t)k+2E(t)P (t) − k−2ES(t)k+3S(t) − k−3P (t)
�������
5 / 1
Example – System
ESE
S
P
1
3
2
d
dt
c(t) =Nv(c(t))
d
dt
���������
S(t)E(t)
ES(t)P (t)
���������=���������
−1 0 1−1 1 01 −1 00 1 −1
���������
�������k+1ES(t) − k−1E(t)S(t)k+2E(t)P (t) − k−2ES(t)k+3S(t) − k−3P (t)
�������with
m0 = 26 / 1
Conservation Analysis
If conservation relations exist,one can separate N by
N = � NR
N0� ESE
S
P
(1) (2)
with
NR ∈ Rm0×n linearly independent metabolites
N0 ∈ R(m−m0)×n linearly dependent metabolites
andm0 = rank(N).
7 / 1
Conservation Analysis
From the separation it followsthat
N = � I⇤0�NR = ⇤NR
ESE
S
P
(1) (2)
from which the conservation relations can be determined as
� = � −⇤0 I �with
⇤ ∈ Rm×m0 link matrix
� ∈ R(m−m0)×m0 conservation matrix.
8 / 1
Conservation Analysis
In practice, one has to solve NT�T = 0 which is gettingnumerically expensive with increasing system size.Thus,
PNTQ = LU
where U is partitioned such that
U = � I M0 0
�from which ⇤, NR and � follow to
⇤ = � IMT � , NR = ⇤+N and � = � −MT I � ,
respectively.
9 / 1
Modular Response Analysis
The unscaled elasticity coe�cient matrix
✏ = �v(c(t))�c(t)
�����������cgives the sensitivities of all local reaction ratesto perturbations in all species concentrations.The dependencies among species are describedby the (reduced) Jacobian matrix which isdefined as
J = N✏
JR = NR✏⇤.
B
Ap
Bp
✏
vB > 0
✏
vAp> 0
✏
vBp< 0
10 / 1
Modular Response Analysis
After rescaling ✏� ✏ such that
✏ = ✏ c(t)v(c(t))
�����������c with 0 < ✏ < 1the local and global response matrices follow to
r =NR✏⇤
andR = r−1,
respectively.
11 / 1
Modular Response Analysis – Example
r =�������−1 2 00 −1 −31 0 −1
�������B
A
C
12 / 1
Modular Response Analysis – Example
r =�������−1 2 00 −1 −31 0 −1
�������B
A
C
13 / 1
Modular Response Analysis – Example
r =�������−1 2 00 −1 −31 0 −1
�������B
A
C
14 / 1
Modular Response Analysis – Example
r =�������−1 2 00 −1 −31 0 −1
�������B
A
C
15 / 1
Modular Response Analysis – Example
r =�������−1 2 00 −1 −31 0 −1
�������B
A
C
16 / 1
Modular Response Analysis
r =�������−1 2 00 −1 −31 0 −1
�������B
A
C
The influence between species in r and R can be categorised by(k, l)-th entry < 0 inactivation of k by l(k, l)-th entry > 0 activation of k by l(k, l)-th entry = 0 no (direct) influence between k by l
withk, l ∈ {1,2, ...,m0}
17 / 1
Wnt model
Wnt
Dshi
Dsha
GSK3
Axin
APC
APC•Axin
APC•Axin•GSK3 APCp•Axin
p•GSK3 ß-catenin•APC
p•Axin
p•GSK3
ß-cateninp•APC
p•Axin
p•GSK3
ß-cateninp
ß-catenin•APC
TCF
ß-catenin ß-catenin•TCF
DestructionCore Cycle
TranscriptionalActivation
Wntsignaling
Non-Axin dependentproteolysis
Axin dependentproteolysis
Axin turnover
18 / 1
Elasticity sampling
1. calculate ✏
2. store signs of ✏ = �v(c(t))�c(t) c(t)
v(c(t))3. sample � ✏ � randomly between 0 and 1
4. restore signs of ✏
5. calculate r for each sample
6. observe sign changes between samples
7. plot
19 / 1
Local interactions of the reduced model
Dshi
GSK3
Axin
APC TCF
APCp•Axin
p•GSK3 ß-catenin•APC
p•Axin
p•GSK3
ß-catenin•APC ß-catenin
ß-cateninp•APC
p•Axin
p•GSK3
ß-cateninp
20 / 1
Local interactions of the reduced model
Dshi GSK3
Axin
APC TCF
APCp•Axinp•GSK3 ß-catenin•APCp•Axinp•GSK3
ß-catenin•APC ß-catenin
⇒ Input inhibits output
21 / 1
Local interactions of the reduced model
Dshi GSK3
Axin
APC TCF
APCp•Axinp•GSK3 ß-catenin•APCp•Axinp•GSK3
ß-catenin•APC ß-catenin
⇒ Input activates output
22 / 1
Local interactions of the reduced model
Dshi GSK3
Axin
APC TCF
APCp•Axinp•GSK3 ß-catenin•APCp•Axinp•GSK3
ß-catenin•APC ß-catenin
⇒ Input activates output
23 / 1
Local interactions of the reduced model
Dshi GSK3
Axin
APC TCF
APCp•Axinp•GSK3 ß-catenin•APCp•Axinp•GSK3
ß-catenin•APC ß-catenin
⇒ Input inhibits output
24 / 1
Local interactions of the reduced model
Dshi GSK3
Axin
APC TCF
APCp•Axinp•GSK3 ß-catenin•APCp•Axinp•GSK3
ß-catenin•APC ß-catenin
⇒ Input inhibits output
25 / 1
Local interactions of the reduced model
Dshi GSK3
Axin
APC TCF
APCp•Axinp•GSK3 ß-catenin•APCp•Axinp•GSK3
ß-catenin•APC ß-catenin
⇒ Input activates output
26 / 1
Local interactions of the reduced model
Dshi
GSK3
Axin
APC TCF
APCp•Axin
p•GSK3 ß-catenin•APC
p•Axin
p•GSK3
ß-catenin•APC ß-catenin
E↵ect of Dshi on TCF varies with the use of local interactions.⇒ global interactions
R(Dshi,TCF) ≤ 0⇒ Wnt stimulus has no negative e↵ect on transcription
27 / 1
Elasticity sampling applied to 4 models
ESE
S
P
1
3
2
Figure: Example: 4 species,3 reactions
Wnt
Dshi
Dsha
GSK3
Axin
APC
APC•Axin
APC•Axin•GSK3 APCp•Axin
p•GSK3 ß-catenin•APC
p•Axin
p•GSK3
ß-cateninp•APC
p•Axin
p•GSK3
ß-cateninp
ß-catenin•APC
TCF
ß-catenin ß-catenin•TCF
Figure: Wnt: 15 species,17 reactions
x1
x2
x3
x4
x5
x6
x7
x8
x9
x10
x11
x12
x13
x14
x15
x16
x17
x18
x19
x20
x21
x22
x23
x24
x25
x26
x27
x28
x29
x30
x31
x32
x33
x34
x35
x36
x37
x38
x39
x40
x41
x42
x43
x44
x45
x46
x47
x48
x49
x50
x51
x52
x53
x54
x55
x56
x57
x58
x59
x60
x61
x62
x63
x64
x65
x66
x67
x68
x69
x70
x71
x72
x73
x74x75
x76
x77
x78
x79
x80
x81
x82
x83
x84
x85
x86
x87
x88
x89
x90
x91
x92
x93
x94
x99
x100x103
Figure: Schoberl: 97 species,148 reactions
Figure: Reactome: 6232 species,3652 reactions
28 / 1
Sign distribution after 1000 samples
0.1
1ExampleWntMapkReactome
always zero always positive always negative positive or negative0
0.0018
Fra
ction
s
29 / 1
Summary
Conservation and modular response analysis
� e↵ective algorithm based on sparse matrix operations
� transform reaction networks to interaction networks
Sampling experiments
� large definiteness (99%) of interactions
Even without knowing the reaction network in detail, itsinformation flow can be calculated.
30 / 1
Acknowledgements
Group of Systems Biology of Molecular Networks
� Nils Bluthgen
� Raphaela Fritsche
� Pawel Durek
� Anja Sieber
� Nadine Schmidt
� Franziska Witzel
� Bertram Klinger
� Jorn Schmiedel
Funding
31 / 1