Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
A compact modeling approach fordeterministic biological systems
Luis M. Torres1 Annegret K. Wagler2
1Ecuadorian Research Center on Mathematical Modeling – ModeMatEscuela Politécnica Nacional, Quito
2LIMOS (UMR 6158 CNRS)University Blaise Pascal, Clermont-Ferrand
6th Int. Workshop on Biological Processes & Petri NetsBrussels, June 22nd
BioPPN 2015 1 / 51
Modeling phenomena in cellular networks
Adequate models for the structure and dynamics of biologicalsystems?
BioPPN 2015 2 / 51
Modeling phenomena in cellular networks
Adequate models for the structure and dynamics of biologicalsystems?
BioPPN 2015 2 / 51
Modeling phenomena in cellular networks
Systems biology aims at the integrated experimental andtheoretical analysis of cellular networks
I network inferencereconstructing interaction networks of biological entities
I network analysismining the information content of the network
I dynamic modelingconnecting interaction network and dynamic behaviour
BioPPN 2015 3 / 51
Application areas
I study biological processes “in silico”I response of cells/organs/organisms to environmental
changesI immune responses to virus infectionI effects of gene defectsI internal activities of a cell: proliferation, differentiation,
motility
I identify good conditions for growth
I design intervention strategies (pharmaceutical)
BioPPN 2015 4 / 51
Outline
IntroductionPetri NetsApplications and limitations
Modeling deterministic systemsThe transition conflict graphValid orientations
Model reconstruction from experimental dataPartial orientations and the MVTPThe Optimal Test Set Extension Problem
Summary
BioPPN 2015 5 / 51
Outline
IntroductionPetri NetsApplications and limitations
Modeling deterministic systemsThe transition conflict graphValid orientations
Model reconstruction from experimental dataPartial orientations and the MVTPThe Optimal Test Set Extension Problem
Summary
Introduction BioPPN 2015 6 / 51
Petri nets
The network is modeled as a weighted bipartite digraphG = (N ∪ T ,A,w) withI N = {1, . . . ,n} places (e.g. metabolites©)I T = {1, . . . , τ} transitions (e.g. reactions �)I arc weights stoichiometric coefficients
The network can be represented through its incidence matrixM ∈ Zn×τ :
0
0
−1
−1
0
2
−1
−2
1
0
0
0
0
0
−1
0
1
02
2
1
2
3
4
5
6
M =
Introduction BioPPN 2015 7 / 51
Petri nets: dynamics
I states of the system are modeled through assignments oftokens to places [“markings”]
I some places may have bounded capacities
I a transition is enabled at a certain state if firing it leads toanother valid state
I many transitions may be enabled at the same state
0
0
−1
−1
0
22
2
M =
−1
−2
1
0
0
0
0
0
−1
0
1
0
Introduction BioPPN 2015 8 / 51
Petri nets: dynamics
I states of the system are modeled through assignments oftokens to places [“markings”]
I some places may have bounded capacitiesI a transition is enabled at a certain state if firing it leads to
another valid state
I many transitions may be enabled at the same state
0
0
−1
−1
0
22
2
M =
−1
−2
1
0
0
0
0
0
−1
0
1
0
Introduction BioPPN 2015 8 / 51
Petri nets: dynamics
I states of the system are modeled through assignments oftokens to places [“markings”]
I some places may have bounded capacitiesI a transition is enabled at a certain state if firing it leads to
another valid stateI many transitions may be enabled at the same state
2
2
M =
−1
−2
1
0
0
0
0
0
−1
0
1
0
0
0
−1
−1
0
2
Introduction BioPPN 2015 8 / 51
The state digraph
I the potential space state is given by:
X := {x ∈ Zn : 0 ≤ xp ≤ up, ∀p ∈ NB}
I the state digraph G = (X ,A) contains arcs (x , y) ∈ A for ally obtained from x by firing one single transition
I dynamic processes are modeled as sequences oftransition fires starting from an initial state x0:
I movement of tokens in GI paths in reachability (marking) digraph G(x0)
Introduction BioPPN 2015 9 / 51
Some central problems
I Reachability:Can the system reach one of a set of target states startingfrom an initial state x0?
I Boundedness:Are there sequences of transition fires that lead tounlimited token accumulation at some place?
I Existence of deadlocks:Can the system reach a state at which no transitions areenabled?
I Liveness:Is it (im)possible to reach a state where some transition ispermanently disabled?
Introduction BioPPN 2015 10 / 51
Outline
IntroductionPetri NetsApplications and limitations
Modeling deterministic systemsThe transition conflict graphValid orientations
Model reconstruction from experimental dataPartial orientations and the MVTPThe Optimal Test Set Extension Problem
Summary
Introduction BioPPN 2015 11 / 51
Application fields
Petri nets have been widely used to model several kinds ofdynamic systemsI asynchronous hardware circuits
[Yakovlev, Koelmans, Semenov & Kinniment 1996]I production and workflow systems
[Adam, Atluri & Huang 1998]I batch processes
[Gu & Bahri 2002]I distributed algorithms
[Reisig 1998]I biological networks
[Hardy & Robillard 2004; Chaouiya, Remy & Thieffry 2008]
Introduction BioPPN 2015 12 / 51
Petri Nets for cellular networks
Some (non-exhaustive) references:I Blätke MA, Heiner M, Marwan W (2015)BioModel Engineering with Petri
Nets. In: Algebraic and Discrete Mathematical Methods for ModernBiology. Elsevier
I Chen M, Hofestaedt R (2003)Quantitative Petri net model of generegulated metabolic networks in the cell. In Silico Biol 3:347-365
I Goss PJE, Peccoud J (1998)Quantitative modeling of stochasticsystems in molecular biology by using stochastic Petri nets. Proc NatlAcad Sci USA 95:6750-6755
I Hardy S, Robillard PN (2005)Phenomenological and molecular-levelPetri net modeling and simulation of long-term potentiation. BioSystems82:26-38
I Hardy S, Robillard PN (2004)Modeling and simulation of molecularbiology systems using Petri nets: modeling goals of variousapproaches. Journal of Bioinformatics and Computational Biology 2(4),619-637
Introduction BioPPN 2015 13 / 51
Petri Nets for cellular networks
Some (non-exhaustive) references:I Heiner M, Gilbert D, Donaldson R (2008a)Petri nets for systems and
synthetic biology. In: Bernardo M, Degano P, Zavattoro G (eds) Formalmethods for computational systems biology, LNCS 5016. Springer,Heidelberg, pp 215-264
I Hofestädt R (1994)A Petri net application of metabolic processes. SystAnal Model Simul 16:113-122
I Matsuno H, Tanaka Y, Aoshima H, Doi A, Matsui M, Miyano S(2003)Biopathways representation and simulation on hybrid functionalPetri net. In Silico Biol 3:389-404
I Pinney JW, Westhead RD, McConkey GA (2003)Petri Netrepresentations in systems biology. Biochem Soc Trans 31:1513-1515
Introduction BioPPN 2015 14 / 51
Limitations
I Petri Nets are well-suited for studying concurrent systemsI reachability graph G(x0) provides local point of view
I for biological systems a model aiming at a globalunderstanding of the (inferred) network is pursued
I G is not adequate for encoding dynamic behaviors:exponential on the size of the Petri Net
I some biological systems are deterministic:each state x has a unique successor x+, even if severaltransitions are enabled
Challenge:Find a compact adequate way to encode system dynamics!!
Introduction BioPPN 2015 15 / 51
Limitations
I Petri Nets are well-suited for studying concurrent systemsI reachability graph G(x0) provides local point of viewI for biological systems a model aiming at a global
understanding of the (inferred) network is pursuedI G is not adequate for encoding dynamic behaviors:
exponential on the size of the Petri NetI some biological systems are deterministic:
each state x has a unique successor x+, even if severaltransitions are enabled
Challenge:Find a compact adequate way to encode system dynamics!!
Introduction BioPPN 2015 15 / 51
Limitations
I Petri Nets are well-suited for studying concurrent systemsI reachability graph G(x0) provides local point of viewI for biological systems a model aiming at a global
understanding of the (inferred) network is pursuedI G is not adequate for encoding dynamic behaviors:
exponential on the size of the Petri NetI some biological systems are deterministic:
each state x has a unique successor x+, even if severaltransitions are enabled
Challenge:Find a compact adequate way to encode system dynamics!!
Introduction BioPPN 2015 15 / 51
Example
Light-induced sporulation of Physarum polycephalum
I entering the sporulation pathway is controlled byenvironmental factors like visible light
I a photoreversible photoreceptor occurs in twoconformational states PFR and PR
I PFR + far-red light FR converted into PR, causessporulation
I PR + red light R converted back to PFR, preventssporulation
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4
Introduction BioPPN 2015 16 / 51
Example
Light-induced sporulation of Physarum polycephalum
I entering the sporulation pathway is controlled byenvironmental factors like visible light
I a photoreversible photoreceptor occurs in twoconformational states PFR and PR
I PFR + far-red light FR converted into PR, causessporulation
I PR + red light R converted back to PFR, preventssporulation
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4
Introduction BioPPN 2015 16 / 51
Example
Light-induced sporulation of Physarum polycephalum
I entering the sporulation pathway is controlled byenvironmental factors like visible light
I a photoreversible photoreceptor occurs in twoconformational states PFR and PR
I PFR + far-red light FR converted into PR, causessporulation
I PR + red light R converted back to PFR, preventssporulation
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4
Introduction BioPPN 2015 16 / 51
Example
Light-induced sporulation of Physarum polycephalum
I entering the sporulation pathway is controlled byenvironmental factors like visible light
I a photoreversible photoreceptor occurs in twoconformational states PFR and PR
I PFR + far-red light FR converted into PR, causessporulation
I PR + red light R converted back to PFR, preventssporulation
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4
Introduction BioPPN 2015 16 / 51
Example
Light-induced sporulation of Physarum polycephalum
I entering the sporulation pathway is controlled byenvironmental factors like visible light
I a photoreversible photoreceptor occurs in twoconformational states PFR and PR
I PFR + far-red light FR converted into PR, causessporulation
I PR + red light R converted back to PFR, preventssporulation
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4
Introduction BioPPN 2015 16 / 51
Example
Light-induced sporulation of Physarum polycephalum
I entering the sporulation pathway is controlled byenvironmental factors like visible light
I a photoreversible photoreceptor occurs in twoconformational states PFR and PR
I PFR + far-red light FR converted into PR, causessporulation
I PR + red light R converted back to PFR, preventssporulation
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4
Introduction BioPPN 2015 16 / 51
Example
Potential system states X :
FR 0 0 0 0 0 0 0 0 1 1 1 1R 0 0 0 0 1 1 1 1 0 0 0 0PFR 1 0 1 0 1 0 1 0 1 0 1 0PR 0 1 0 1 0 1 0 1 0 1 0 1Spo 0 0 1 1 0 0 1 1 0 0 1 1
x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4
Introduction BioPPN 2015 17 / 51
Outline
IntroductionPetri NetsApplications and limitations
Modeling deterministic systemsThe transition conflict graphValid orientations
Model reconstruction from experimental dataPartial orientations and the MVTPThe Optimal Test Set Extension Problem
Summary
Modeling deterministic systems BioPPN 2015 18 / 51
Modeling determinism via transition priorities
Deterministic systems can be modeled via a successor oracle:
x ∈ X 7→ x+ (successor state)
We consider those systems where the oracle can beimplemented through a transition selection function:
X → Tx 7→ t∗(x) ∈ T (x),
where T (x) is the set of enabled transitions at state x ,so that...I firing t∗(x) puts the system in state x+
I t∗(x) is called the highest-priority transition at x
Modeling deterministic systems BioPPN 2015 19 / 51
Modeling determinism via transition priorities
Deterministic systems can be modeled via a successor oracle:
x ∈ X 7→ x+ (successor state)
We consider those systems where the oracle can beimplemented through a transition selection function:
X → Tx 7→ t∗(x) ∈ T (x),
Idea:Find a compact encoding for t∗.
Modeling deterministic systems BioPPN 2015 19 / 51
Modeling determinism via transition priorities
Can all deterministic systems be modelled in this way?
No!
1 2
3 4
t1 t2
t3
t4
x0
1 2
3 4
t1 t2
t3
t4
x1
1 2
3 4
t1 t2
t3
t4
x2
j branching state x j T (x j) x j+
0 (1, 1, 0, 0)T {t1, t2} (0, 0, 1, 1)T
1 (0, 1, 1, 0)T {t2, t3} (0, 1, 0, 1)T
2 (1, 0, 0, 1)T {t1, t4} (0, 1, 0, 0)T
3 (1, 1, 1, 0)T {t2, t3} (1, 1, 0, 1)T
Modeling deterministic systems BioPPN 2015 20 / 51
Modeling determinism via transition priorities
Can all deterministic systems be modelled in this way?No!
1 2
3 4
t1 t2
t3
t4
x0
1 2
3 4
t1 t2
t3
t4
x1
1 2
3 4
t1 t2
t3
t4
x2
j branching state x j T (x j) x j+
0 (1, 1, 0, 0)T {t1, t2} (0, 0, 1, 1)T
1 (0, 1, 1, 0)T {t2, t3} (0, 1, 0, 1)T
2 (1, 0, 0, 1)T {t1, t4} (0, 1, 0, 0)T
3 (1, 1, 1, 0)T {t2, t3} (1, 1, 0, 1)T
Modeling deterministic systems BioPPN 2015 20 / 51
The transition conflict graph
I two transitions are in conflict if they are both enabled atsome state
I they are in dynamic conflict if firing one disables the other
Lemma (TW 2011)Transitions t , t ′ ∈ T are in conflict if and only if
wpt ≤ up − wt ′p ∀p ∈ P−(t) ∩ P+(t ′) ∩ B,andwpt ′ ≤ up − wtp ∀p ∈ P−(t ′) ∩ P+(t) ∩ B.
Modeling deterministic systems BioPPN 2015 21 / 51
The transition conflict graph
Transition conflicts can be represented by a graph K := (T ,E)
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
I for every x ∈ STS, T (x) induces a clique in K
I not all cliques correspond to states
Modeling deterministic systems BioPPN 2015 22 / 51
The transition conflict graph
Transition conflicts can be represented by a graph K := (T ,E)
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
I for every x ∈ STS, T (x) induces a clique in K
I not all cliques correspond to states
Modeling deterministic systems BioPPN 2015 22 / 51
The transition conflict graph
Transition conflicts can be represented by a graph K := (T ,E)
?FR
?
PFR
?PR
?R
?
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
I for every x ∈ STS, T (x) induces a clique in K
I not all cliques correspond to states
Modeling deterministic systems BioPPN 2015 22 / 51
The transition conflict graph
The structure of K is very difficult to characterize from a graphtheoretical point of view!
Theorem (TW 2011)Let H = (V ,E) be an (arbitrary) undirected graph and Q thefamily containing all cliques in H. Then there exists a networkG = (N ∪ T ,A,w) that satisfiesI H = KI Q := {T (x) : x ∈ X}
Modeling deterministic systems BioPPN 2015 23 / 51
Outline
IntroductionPetri NetsApplications and limitations
Modeling deterministic systemsThe transition conflict graphValid orientations
Model reconstruction from experimental dataPartial orientations and the MVTPThe Optimal Test Set Extension Problem
Summary
Modeling deterministic systems BioPPN 2015 24 / 51
Valid orientations
IdeaEncode transition priorities by orienting the edges of K
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Definition (Valid orientation)An orientation D of the edges of K is valid ifI ∀x ∈ X , T (x) induces a clique with a unique sink
Modeling deterministic systems BioPPN 2015 25 / 51
Valid orientations
IdeaEncode transition priorities by orienting the edges of K
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Definition (Valid orientation)An orientation D of the edges of K is valid ifI ∀x ∈ X , T (x) induces a clique with a unique sink
Modeling deterministic systems BioPPN 2015 25 / 51
Valid orientations
Theorem (TW 2009)There is a valid orientation of K iff for any pair of statesx , x ′ ∈ X with t∗(x) ∈ T (x) ∩ T (x ′),I either t∗(x ′) = t∗(x) or t∗(x ′) ∈ T (x ′) \ T (x)
I in this case D encodes the dynamics of the system.I the successor oracle can be implemented as
I select t∗(x) as unique sink in clique in K induced by x
I obtain x+ by swtiching t∗(x)
Modeling deterministic systems BioPPN 2015 26 / 51
Successor oracle
Input: (G,u,D), x ∈ XOutput: x+
3: Construct the set T (x) of enabled transitionsif T (x) = ∅ then
return x6: end if
{Compute out-degree of transitions in the subgraph of Dinduced by T (x)}for t ∈ T (x) do
9: δ−(t) := |{tt ′ ∈ A : t ′ ∈ T (x)}|end for{Return successor state}
12: t∗ ← t ∈ T (x) with δ−(t) = 0return x + M·t∗
Modeling deterministic systems BioPPN 2015 27 / 51
Predecesor oracle
Input: (G,u,D), x ∈ XOutput: pred(x)
3: {Construct set of transition candidates}
cand(x)←{t ∈ T : xp + wpt ≤ up, ∀p ∈ P−(t) ∩ B;
xp − wtp ≥ 0,∀p ∈ P+(t)}
{Call successor-oracle to construct sets of predecessors}6: pred(x)← ∅
for t ∈ cand(x) doy ← x −M·t
9: if t∗(y) = t thenpred(x)← pred(x) ∪ {y}
end if12: end for
{Return set of possible predecessors}return pred(x)
Modeling deterministic systems BioPPN 2015 28 / 51
Example
Light-induced sporulation of Physarum polycephalum
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Modeling deterministic systems BioPPN 2015 29 / 51
Example
Light-induced sporulation of Physarum polycephalum
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Modeling deterministic systems BioPPN 2015 29 / 51
Example
Light-induced sporulation of Physarum polycephalum
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Modeling deterministic systems BioPPN 2015 29 / 51
Example
Light-induced sporulation of Physarum polycephalum
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Modeling deterministic systems BioPPN 2015 29 / 51
Example
Light-induced sporulation of Physarum polycephalum
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Modeling deterministic systems BioPPN 2015 29 / 51
Strongly valid orientations
I in general, valid orientations are difficult to characterize
I an orientation is strongly valid, if it is valid for all Petri Netssharing the same conflict graph
I strongly valid orientations have a unique sink for eachclique
Theorem (TW 2009)I D is strongly valid if it does not contain a directed cycle of
length 3I in particular, acyclic orientations are strongly valid
Modeling deterministic systems BioPPN 2015 30 / 51
Strongly valid orientations
I in general, valid orientations are difficult to characterize
I an orientation is strongly valid, if it is valid for all Petri Netssharing the same conflict graph
I strongly valid orientations have a unique sink for eachclique
Theorem (TW 2009)I D is strongly valid if it does not contain a directed cycle of
length 3I in particular, acyclic orientations are strongly valid
Modeling deterministic systems BioPPN 2015 30 / 51
Outline
IntroductionPetri NetsApplications and limitations
Modeling deterministic systemsThe transition conflict graphValid orientations
Model reconstruction from experimental dataPartial orientations and the MVTPThe Optimal Test Set Extension Problem
Summary
Model reconstruction from experimental data BioPPN 2015 31 / 51
The Minimum Valid Test Set Problem
ProblemFor a deterministic system whose dynamic behavior isprescribed by an unknown valid orientation D,I determine D by observing some pairs (x , x+)
I use as few observations as possible
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Model reconstruction from experimental data BioPPN 2015 32 / 51
The Minimum Valid Test Set Problem
ProblemFor a deterministic system whose dynamic behavior isprescribed by an unknown valid orientation D,I determine D by observing some pairs (x , x+)
I use as few observations as possible
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Model reconstruction from experimental data BioPPN 2015 32 / 51
The Minimum Valid Test Set Problem
ProblemFor a deterministic system whose dynamic behavior isprescribed by an unknown valid orientation D,I determine D by observing some pairs (x , x+)
I use as few observations as possible
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Model reconstruction from experimental data BioPPN 2015 32 / 51
The Minimum Valid Test Set Problem
ProblemFor a deterministic system whose dynamic behavior isprescribed by an unknown valid orientation D,I determine D by observing some pairs (x , x+)
I use as few observations as possible
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Model reconstruction from experimental data BioPPN 2015 32 / 51
The Minimum Valid Test Set Problem
ProblemFor a deterministic system whose dynamic behavior isprescribed by an unknown valid orientation D,I determine D by observing some pairs (x , x+)
I use as few observations as possible
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Model reconstruction from experimental data BioPPN 2015 32 / 51
The Minimum Valid Test Set Problem
ProblemFor a deterministic system whose dynamic behavior isprescribed by an unknown valid orientation D,I determine D by observing some pairs (x , x+)
I use as few observations as possible
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Model reconstruction from experimental data BioPPN 2015 32 / 51
The Minimum Valid Test Set Problem
ProblemFor a deterministic system whose dynamic behavior isprescribed by an unknown valid orientation D,I determine D by observing some pairs (x , x+)
I use as few observations as possible
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4 t1
t2
t3
t5
t4
Model reconstruction from experimental data BioPPN 2015 32 / 51
The Minimum Valid Test Set Problem
DefinitionA set X ′ ⊂ X is a valid test set if knowledge about{
(x , x+) : x ∈ X ′}
is sufficient for determining D.
Minimum Valid Test Set Problem (MVTP)Determine a valid test set of minimum cardinality.
Model reconstruction from experimental data BioPPN 2015 33 / 51
The Minimum Valid Test Set Problem
DefinitionA set X ′ ⊂ X is a valid test set if knowledge about{
(x , x+) : x ∈ X ′}
is sufficient for determining D.
Minimum Valid Test Set Problem (MVTP)Determine a valid test set of minimum cardinality.
Model reconstruction from experimental data BioPPN 2015 33 / 51
Partial orientations
I a partial orientation D′ := (T ,A′,E′) is a mixed graphobtained by fixing directions of some edges of K
I D′ is extendible if it is possible to choose directions for theedges in E′ to obtain a valid orientation
t1
t2
t3
t4
non-extendible
t1
t2
t3
t4
extendible
t1
t2
t3
t4
sufficient
Model reconstruction from experimental data BioPPN 2015 34 / 51
Inferable, dominated, and essential arcs
I an edge tt ′ ∈ E′ is inferable as (t , t ′) if the digraph(T ,A′ ∪ {(t ′, t)},E′ \ {tt ′}) is not extendible
I an edge tt ′ ∈ E′ is dominated if
∀x ∈ X with t , t ′ ∈ T (x), t∗(x) 6∈ {t , t ′}
I D′ is sufficient if all edges in E′ are either inferable ordominated
I an arc (t , t ′) of D is essential ifI tt ′ is not dominatedI the orientation D2 := (T ,A \ {(t , t ′)} ∪ {(t ′, t)}) is valid
Model reconstruction from experimental data BioPPN 2015 35 / 51
Characterizing optimal solutions for MVTP
Theorem (TW 2009)Let A∗ be the set of all essential arcs. Then,I the partial orientation D∗ := (T ,A∗,E∗) is sufficientI for any x ∈ X , knowledge about (x , x+) allows orientation
of at most one essential arcI X ′ ⊂ X is an optimal solution for MVTP iff for every x ∈ X ′
{(t , t∗(x)) : t ∈ T (x), t 6= t∗(x)} ∩ A∗ 6= ∅
We consider in the following the special case when D isacyclic...
Model reconstruction from experimental data BioPPN 2015 36 / 51
Characterizing optimal solutions for MVTP
Theorem (TW 2009)Let A∗ be the set of all essential arcs. Then,I the partial orientation D∗ := (T ,A∗,E∗) is sufficientI for any x ∈ X , knowledge about (x , x+) allows orientation
of at most one essential arcI X ′ ⊂ X is an optimal solution for MVTP iff for every x ∈ X ′
{(t , t∗(x)) : t ∈ T (x), t 6= t∗(x)} ∩ A∗ 6= ∅
We consider in the following the special case when D isacyclic...
Model reconstruction from experimental data BioPPN 2015 36 / 51
Acyclic graphs
Lemma (TW 2015)If D is acyclic, then every clique Q has a unique sink, and adirected hamiltonian path PQ through its nodes.
Lemma (TW 2015)If D is acyclic, the essential arcs of a clique are exactly the arcscorresponding to PQ.
Model reconstruction from experimental data BioPPN 2015 37 / 51
Algorithm
Input: (G, u, t∗) {deterministic system with oracle for computing t∗}Output: D {valid orientation}
3: initialize Q as the set of all inclusion-wise maximal cliques in Kwhile Q 6= ∅ do
retrieve a clique Q from Q6: while |Q| > 1 and ∃x ∈ X with T (x) = Q do
call the oracle and determine t∗ := t∗(x)orient the arcs {(t , t∗) : t ∈ T (x), t 6= t∗}
9: deduce orientations for inferable edgesremove t∗ from Q
end while12: if Q contains more than one node then
compute Q′ := {T (x) ⊂ Q : x ∈ X}remove from Q′ cliques that are not inclusion-wise maximal
15: Q := Q∪Q′end if
end while18: while there are yet unoriented edges do
deduce an orientation for all yet unoriented inferable edgeschoose an arbitrary orientation for a yet unoriented dominated edge
21: end whileModel reconstruction from experimental data BioPPN 2015 38 / 51
What about the “online” problem?
t1
t2
t3t4
t5
t1
t2
t3t4
t5
t1
t2
t3t4
t5
I states induce either edges or cliques of size 3 (triangles)I for every e ∈ K, there is an acyclic orientation where e is
not essentialI the figures show that the same holds for every triangle
for every x ∈ X , there is at least one valid orientation forwhich (x , x+) does not provide the direction of any essential arc no “winning strategy”
Model reconstruction from experimental data BioPPN 2015 39 / 51
Outline
IntroductionPetri NetsApplications and limitations
Modeling deterministic systemsThe transition conflict graphValid orientations
Model reconstruction from experimental dataPartial orientations and the MVTPThe Optimal Test Set Extension Problem
Summary
Model reconstruction from experimental data BioPPN 2015 40 / 51
Experiment design for model verification
In practice...I network G is not known a priori
I both G and D have to be reconstructed simultaneouslyfrom experimental observations
I it is not always possible to observe pairs (x , x+)
I instead...I system is set to an initial state x0
I sequence sequ(x0) = (x1, . . . , xk ) of state changes inresponse to initial trigger is observed
I there might be intermediate non-observed states
Model reconstruction from experimental data BioPPN 2015 41 / 51
Experiment design for model verification
Let X ′ be a collection of states observed in some experimentsI (G,D′) is X ′-deterministic if it fits the experimental dataI P(X ′) is the set of all X ′-deterministic models (G,D′)I P(X ′) can be computed with an exhaustive reconstruction
approach 1
Optimal Test Set Extension Problem
Given X ′ ⊆ X and P(X ′), find an optimal extension X ′ of X ′such that
I each network in P(X ′) can be either completed to anX ′-deterministic network or ruled out as “false positive”,
I all networks in P(X ′) are either equivalent or not furtherdistinguishable
1M. Favre, A. Wagler. Reconstructing X ′-deterministic extended Petri netsfrom experimental time-series data X ′. CEUR Workshop Proceedings988:45–59, 2013. (Special Issue BioPPN 2013)
Model reconstruction from experimental data BioPPN 2015 42 / 51
Experiment design for model verification
Let X ′ be a collection of states observed in some experimentsI (G,D′) is X ′-deterministic if it fits the experimental dataI P(X ′) is the set of all X ′-deterministic models (G,D′)I P(X ′) can be computed with an exhaustive reconstruction
approach 1
Optimal Test Set Extension Problem
Given X ′ ⊆ X and P(X ′), find an optimal extension X ′ of X ′such that
I each network in P(X ′) can be either completed to anX ′-deterministic network or ruled out as “false positive”,
I all networks in P(X ′) are either equivalent or not furtherdistinguishable
1M. Favre, A. Wagler. Reconstructing X ′-deterministic extended Petri netsfrom experimental time-series data X ′. CEUR Workshop Proceedings988:45–59, 2013. (Special Issue BioPPN 2013)
Model reconstruction from experimental data BioPPN 2015 42 / 51
Example
Light-induced sporulation of Physarum polycephalum.
Perform two experiments:I sequ(x9) = (x2, x4)
I sequ(x6) = (x1)
I X ′ = {x1, x2, x4, x6, x9}I non-observed intermediate states x10 and x5
P(X ′) contains twelve networks...
Model reconstruction from experimental data BioPPN 2015 43 / 51
Example
Light-induced sporulation of Physarum polycephalum.Perform two experiments:I sequ(x9) = (x2, x4)
I sequ(x6) = (x1)
I X ′ = {x1, x2, x4, x6, x9}I non-observed intermediate states x10 and x5
P(X ′) contains twelve networks...
Model reconstruction from experimental data BioPPN 2015 43 / 51
Example
Light-induced sporulation of Physarum polycephalum.Perform two experiments:I sequ(x9) = (x2, x4)
I sequ(x6) = (x1)
I X ′ = {x1, x2, x4, x6, x9}I non-observed intermediate states x10 and x5
P(X ′) contains twelve networks...
Model reconstruction from experimental data BioPPN 2015 43 / 51
Example
Light-induced sporulation of Physarum polycephalum.Perform two experiments:I sequ(x9) = (x2, x4)
I sequ(x6) = (x1)
I X ′ = {x1, x2, x4, x6, x9}I non-observed intermediate states x10 and x5
P(X ′) contains twelve networks...
Model reconstruction from experimental data BioPPN 2015 43 / 51
Example
Light-induced sporulation of Physarum polycephalum.Perform two experiments:I sequ(x9) = (x2, x4)
I sequ(x6) = (x1)
I X ′ = {x1, x2, x4, x6, x9}I non-observed intermediate states x10 and x5
P(X ′) contains twelve networks...
FR
PFR
PR
R
Spot2
t3
t5 t2
t3
t5
Model reconstruction from experimental data BioPPN 2015 43 / 51
Example
Light-induced sporulation of Physarum polycephalum.Perform two experiments:I sequ(x9) = (x2, x4)
I sequ(x6) = (x1)
I X ′ = {x1, x2, x4, x6, x9}I non-observed intermediate states x10 and x5
P(X ′) contains twelve networks...
FR
PFR
PR
R
Spo
t1
t ′2
t3
t5
t1
t ′2
t3
t5
Model reconstruction from experimental data BioPPN 2015 43 / 51
Example
Light-induced sporulation of Physarum polycephalum.Perform two experiments:I sequ(x9) = (x2, x4)
I sequ(x6) = (x1)
I X ′ = {x1, x2, x4, x6, x9}I non-observed intermediate states x10 and x5
P(X ′) contains twelve networks...
FR
PFR
PR
R
Spot2
t ′3
t5
t4
t2
t ′3
t5
t4
Model reconstruction from experimental data BioPPN 2015 43 / 51
Example
Light-induced sporulation of Physarum polycephalum.Perform two experiments:I sequ(x9) = (x2, x4)
I sequ(x6) = (x1)
I X ′ = {x1, x2, x4, x6, x9}I non-observed intermediate states x10 and x5
P(X ′) contains twelve networks...
FR
PFR
PR
Spo
Rt2
t ′′3
t ′′5
t4
t2
t ′′3
t ′′5
t4
Model reconstruction from experimental data BioPPN 2015 43 / 51
Example
Light-induced sporulation of Physarum polycephalum.Perform two experiments:I sequ(x9) = (x2, x4)
I sequ(x6) = (x1)
I X ′ = {x1, x2, x4, x6, x9}I non-observed intermediate states x10 and x5
P(X ′) contains twelve networks...
FR
PFR
PR
R
Spo
t1
t ′2
t ′3
t5
t4 t1
t ′2
t ′3
t5
t4
Model reconstruction from experimental data BioPPN 2015 43 / 51
Example
Light-induced sporulation of Physarum polycephalum.Perform two experiments:I sequ(x9) = (x2, x4)
I sequ(x6) = (x1)
I X ′ = {x1, x2, x4, x6, x9}I non-observed intermediate states x10 and x5
P(X ′) contains twelve networks...
FR
PFR
PR
Spo
R
t1
t ′2
t ′′3
t ′′5
t4 t1
t ′2
t ′′3
t ′′5
t4
Model reconstruction from experimental data BioPPN 2015 43 / 51
Example
Potential system states X :
FR 0 0 0 0 0 0 0 0 1 1 1 1R 0 0 0 0 1 1 1 1 0 0 0 0PFR 1 0 1 0 1 0 1 0 1 0 1 0PR 0 1 0 1 0 1 0 1 0 1 0 1Spo 0 0 1 1 0 0 1 1 0 0 1 1
x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12
FR
PFR
PR
R
Spo
t1
t2
t3
t5
t4
Model reconstruction from experimental data BioPPN 2015 44 / 51
Example
Possible successors X :
x3+ x5+ x7+ x8+ x10+ x11+ x12+
P1 x3 x5 x7 x3 x12 x4 x12
P2 x3 x5 x7 x3 x2 x12 x4
P3a x3 x1 x3 x7 x12 x4 x12
P3b x3 x1 x3 x4 x12 x4 x12
P4a x3 x1 x3 x7 x2 x12 x4
P4b x3 x1 x3 x4 x2 x12 x4
I a test sequ(x3) = (x3) provides no useful informationI a test sequ(x8) = (x3) rules out P3b and P4b
I a test sequ(x7) = (x3) discards P1 and P2
I a test sequ(x12) = (x4) discards P1,P3a and P3b
Model reconstruction from experimental data BioPPN 2015 45 / 51
Example
Possible successors X :
x3+ x5+ x7+ x8+ x10+ x11+ x12+
P1 x3 x5 x7 x3 x12 x4 x12
P2 x3 x5 x7 x3 x2 x12 x4
P3a x3 x1 x3 x7 x12 x4 x12
P3b x3 x1 x3 x4 x12 x4 x12
P4a x3 x1 x3 x7 x2 x12 x4
P4b x3 x1 x3 x4 x2 x12 x4
I a test sequ(x3) = (x3) provides no useful information
I a test sequ(x8) = (x3) rules out P3b and P4b
I a test sequ(x7) = (x3) discards P1 and P2
I a test sequ(x12) = (x4) discards P1,P3a and P3b
Model reconstruction from experimental data BioPPN 2015 45 / 51
Example
Possible successors X :
x3+ x5+ x7+ x8+ x10+ x11+ x12+
P1 x3 x5 x7 x3 x12 x4 x12
P2 x3 x5 x7 x3 x2 x12 x4
P3a x3 x1 x3 x7 x12 x4 x12
P3b x3 x1 x3 x4 x12 x4 x12
P4a x3 x1 x3 x7 x2 x12 x4
P4b x3 x1 x3 x4 x2 x12 x4
I a test sequ(x3) = (x3) provides no useful informationI a test sequ(x8) = (x3) rules out P3b and P4b
I a test sequ(x7) = (x3) discards P1 and P2
I a test sequ(x12) = (x4) discards P1,P3a and P3b
Model reconstruction from experimental data BioPPN 2015 45 / 51
Example
Possible successors X :
x3+ x5+ x7+ x8+ x10+ x11+ x12+
P1 x3 x5 x7 x3 x12 x4 x12
P2 x3 x5 x7 x3 x2 x12 x4
P3a x3 x1 x3 x7 x12 x4 x12
P3b x3 x1 x3 x4 x12 x4 x12
P4a x3 x1 x3 x7 x2 x12 x4
P4b x3 x1 x3 x4 x2 x12 x4
I a test sequ(x3) = (x3) provides no useful informationI a test sequ(x8) = (x3) rules out P3b and P4b
I a test sequ(x7) = (x3) discards P1 and P2
I a test sequ(x12) = (x4) discards P1,P3a and P3b
Model reconstruction from experimental data BioPPN 2015 45 / 51
Example
Possible successors X :
x3+ x5+ x7+ x8+ x10+ x11+ x12+
P1 x3 x5 x7 x3 x12 x4 x12
P2 x3 x5 x7 x3 x2 x12 x4
P3a x3 x1 x3 x7 x12 x4 x12
P3b x3 x1 x3 x4 x12 x4 x12
P4a x3 x1 x3 x7 x2 x12 x4
P4b x3 x1 x3 x4 x2 x12 x4
I a test sequ(x3) = (x3) provides no useful informationI a test sequ(x8) = (x3) rules out P3b and P4b
I a test sequ(x7) = (x3) discards P1 and P2
I a test sequ(x12) = (x4) discards P1,P3a and P3b
Model reconstruction from experimental data BioPPN 2015 45 / 51
Outline
IntroductionPetri NetsApplications and limitations
Modeling deterministic systemsThe transition conflict graphValid orientations
Model reconstruction from experimental dataPartial orientations and the MVTPThe Optimal Test Set Extension Problem
Summary
Summary BioPPN 2015 46 / 51
Summary
I Petri Nets are a well-established framework for modelingand analyzing several processes in Systems Biology
I interrelations of components in complex systemsI concurrency
I we presented an extension for encoding the dynamics ofcertain deterministic systems in a compact manner
I global point of viewI reconstructing model structure from experimental data
I we assume successor oracle can be represented viapriorities among transitions(not all deterministic systems can be modeled in this way)
Summary BioPPN 2015 47 / 51
Summary
I Petri Nets are a well-established framework for modelingand analyzing several processes in Systems Biology
I interrelations of components in complex systemsI concurrency
I we presented an extension for encoding the dynamics ofcertain deterministic systems in a compact manner
I global point of viewI reconstructing model structure from experimental data
I we assume successor oracle can be represented viapriorities among transitions(not all deterministic systems can be modeled in this way)
Summary BioPPN 2015 47 / 51
Summary
I Petri Nets are a well-established framework for modelingand analyzing several processes in Systems Biology
I interrelations of components in complex systemsI concurrency
I we presented an extension for encoding the dynamics ofcertain deterministic systems in a compact manner
I global point of viewI reconstructing model structure from experimental data
I we assume successor oracle can be represented viapriorities among transitions(not all deterministic systems can be modeled in this way)
Summary BioPPN 2015 47 / 51
Summary
I graph-theoretical characterizations of valid orientationsI in general difficult to obtainI acyclic orientations are always (strongly) validI some characterizations in terms of hypergraphs
I we considered the MVTP reconstruction problemI an “optimal offline” strategy in terms of essential edgesI reconstruction algorithmI no (online) “winning strategy”
I experiment design (OTSEP)I reconstruct both network and valid orientationI extend or rule out models that do not fit the dataI assist in the choice of new experiments
Summary BioPPN 2015 48 / 51
Summary
I graph-theoretical characterizations of valid orientationsI in general difficult to obtainI acyclic orientations are always (strongly) validI some characterizations in terms of hypergraphs
I we considered the MVTP reconstruction problemI an “optimal offline” strategy in terms of essential edgesI reconstruction algorithmI no (online) “winning strategy”
I experiment design (OTSEP)I reconstruct both network and valid orientationI extend or rule out models that do not fit the dataI assist in the choice of new experiments
Summary BioPPN 2015 48 / 51
Summary
I graph-theoretical characterizations of valid orientationsI in general difficult to obtainI acyclic orientations are always (strongly) validI some characterizations in terms of hypergraphs
I we considered the MVTP reconstruction problemI an “optimal offline” strategy in terms of essential edgesI reconstruction algorithmI no (online) “winning strategy”
I experiment design (OTSEP)I reconstruct both network and valid orientationI extend or rule out models that do not fit the dataI assist in the choice of new experiments
Summary BioPPN 2015 48 / 51
Future work
I further structural properties of valid orientations
I (algorithmic) implications for MVTP
I algorithms for OTSEP
I integration in available software tools for Petri Nets(Snoopy, CHARLIE, ...)
Summary BioPPN 2015 49 / 51
Future work
I further structural properties of valid orientations
I (algorithmic) implications for MVTP
I algorithms for OTSEP
I integration in available software tools for Petri Nets(Snoopy, CHARLIE, ...)
Summary BioPPN 2015 49 / 51
Future work
I further structural properties of valid orientations
I (algorithmic) implications for MVTP
I algorithms for OTSEP
I integration in available software tools for Petri Nets(Snoopy, CHARLIE, ...)
Summary BioPPN 2015 49 / 51
Future work
I further structural properties of valid orientations
I (algorithmic) implications for MVTP
I algorithms for OTSEP
I integration in available software tools for Petri Nets(Snoopy, CHARLIE, ...)
Summary BioPPN 2015 49 / 51
For Further Reading
Marwan, W., Sujatha, A., Starostzik, C.: Reconstructing the regulatorynetwork controlling commitment and sporulation in physarumpolycephalum based on hierarchical Petri net modeling and simulation.Journal of Theoretical Biology 236, 349–365 (2005)
Marwan, W., Wagler, A., Weismantel, R.: A mathematical approach tosolve the network reconstruction problem.Math. Methods of Operations Research 67, 117–132 (2008)
Torres, L.M., Wagler, A.: Model reconstruction for discrete deterministicsystems.Electronic Notes of Discrete Mathematics 36, 175–182 (2010)
Torres, L.M., Wagler, A.: Encoding the dynamics of deterministicsystems.Mathematical Methods of Operations Research 73(3), 281–300 (2011).
Torres, L.M., Wagler, A.: Analyzing the dynamics of deterministicsystems from a hypergraph theoretical point of view.RAIRO – Operations Research 47(3), 321–330 (2013).
Summary BioPPN 2015 50 / 51
Thank you!!!
Summary BioPPN 2015 51 / 51