How Might Petri Nets Enhance

Your Systems Biology Toolkit

Monika Heiner! and David Gilbert
School of Information Systems, Computing and Mathematics
Brunel University, Uxbridge, Middlesex UB8 3PH, UK,
{monika.heiner,david.gilbert}@brunel.ac.uk

Abstract. How might Petri nets enhance my Systems Biology toolkit?

this is one of the questions that we get on a regular basis, which motivated us to write an answer in the form of this paper.
We discuss the extent to which the Petri net approach can be used
as an umbrella formalism to support the process of BioModel Engineering. This includes the facilitation of an active and productive interaction
between biomodellers and bioscientists during the construction and analysis of dynamic models of biological systems. These models play a crucial
role in both Systems Biology, where they can be explanatory and predictive, and synthetic biology, where they are effectively design templates.
In this paper we give an overview of the tools and techniques which have
been shown to be useful so far, and describe some of the current open
challenges.
Keywords: BioModel Engineering; Systems Biology; synthetic biology
biomolecular networks; qualitative, stochastic and continuous Petri nets;
model checking.

Motivation

Biology is increasingly becoming an informational science. This revolution has

been driven by technological advances which have supported the development of
studies at many levels of intra- and intercellular activity. These advances have
facilitated the analysis of how the components of a biological system interact
functionally - namely the field of Systems Biology [39]. In general this analysis
is quantitative and over time [1], which can basically be done in a stochastic or
continuous fashion.
At the heart of this field lies the construction of models of biological systems,
see Figure 1. These models are used for analysis which should ideally be both explanatory of biological mechanisms and predictive of the behaviour of the system
when it is perturbed by, e.g., mutations, chemical interventions or changes in the
environment. Furthermore, models can be used to help make genetic engineering
easier and more reliable, serving as design templates for novel synthetic biological systems an emerging discipline known as synthetic biology [19,26]. Central
!

On sabbatical leave from Brandenburg University of Technology.

L.M. Kristensen and L. Petrucci (Eds.): PETRI NETS 2011, LNCS 6709, pp. 1737, 2011.
c Springer-Verlag Berlin Heidelberg 2011
!

18

M. Heiner and D. Gilbert

Fig. 1. The role of formal models in Systems and Synthetic Biology

to both Systems and Synthetic Biology is BioModel Engineering (BME) which

is the science of designing, constructing and analyzing computational models of
biological systems [8].
In this paper we discuss the Petri net approach to BioModel Engineering.
We show how Petri nets are highly suited to the support of both Systems and
Synthetic Biology. Not only are they immediately attractive to life scientists
in terms of their graphical representation, thus facilitating communication with
the modeller, but they permit qualitative as well as quantitative (stochastic,
continuous, hybrid) descriptions by a family of related models and come with a
very powerful analytical toolkit [29].
The structure of this paper is as follows: in Section 2 we discuss representational issues, and in Section 3 we recall our overall conceptual framework, followed by an overview of Petri net contributions in Section 4. We discuss model
checking and its use in BioModel Engineering in Section 5, and summarize our
toolkit in Section 6. Then we discuss the outlook and challenges for the use of
Petri nets in this area in Section 7, and give an overall summary in Section 8.
This paper is deliberately kept informal. We assume a basic understanding of
Petri nets; a gentle introduction within the given context of BioModel Engineering can be found in [48, 28], for related formal definitions see [29].

How Might Petri Nets Enhance . . .

19

Representation Style Just a Matter of Taste?

We consider biochemical reaction networks (RN), consisting of a finite set of

reactions (biochemical reactions, complexation/decomplexation, phosphorylation/dephosphorylation, conformational change, transport steps, . . . , any
type of biochemical interactions, which are considered to be atomic on the
chosen abstraction level), converting or transporting
species (biochemical compounds, proteins, protein conformations, complexes,
. . . , any species, which are considered to be atomic on the chosen abstraction
level).
Reactions may carry elementary kinetic information, the reaction rates, if known.
A typical pattern of elementary reaction rates is the mass/action kinetics, where
the state-dependent rate function vi of the reaction ri is assumed to be given
by the product of the current values of the species involved and some further
constants ki (usually called reaction parameters), summing up environmental
conditions such as temperature, pressure, . . . ; for examples see Figure 2, left
and Figure 3, right.
Thus, the notion of a reaction network covers both the qualitative and quantitative modelling paradigms. A reaction network can be specified as:
(1) list (or set) of all individual (stoichiometric) reactions, in a reaction-centric
or species-centric style,
or a graph describing all individual reactions, which could be either
(2) hypergraph, i.e., a graph where an arc can connect any number of nodes,
(3) bipartite graph, i.e., a graph with two types of nodes, whereby arcs never
connect nodes of the same type; e.g. a Petri net.
Remark: In the Petri net examples we adopt the usual interpretation and represent reactions (the active system components) by transitions, and species (the
passive system components) by places.
These three different styles of representations (1)(3) can be converted into
each other without loss of information. There are two well-known representations, which can be uniquely derived from any of (1)(3), but generally not vice
versa:
(4) incidence matrix (in Systems and Synthetic Biology better known as stoichiometric matrix),
(5) system of ordinary differential equations (ODEs).
Both derived styles of representation lose structural information if there are
catalysts involved or species, which are both the subtrate as well as the product
of one and the same reaction. It is an easy exercise to imagine two reaction
networks generating the same incidence matrix or ODEs, but differing in their
discrete behaviour. We start from the introductory example in [61], see Fig. 2,

20

M. Heiner and D. Gilbert

r1

r2

B
inactive

A
active

r1

A
active

r2

B
inactive

Fig. 2. Two reaction networks generating the same incidence matrix and the same
ODEs. With the kinetic rates v1 = v(r1 ) = k1 A B, v2 = v(r2 ) = k2 B, we get the
two equations dA/dt = v2 v1 , dB/dt = v1 v2 .

left. The net on the right uses a modifier arc (dotted line), which is a standard
feature of the Systems Biology Markup Language (SBML) [41] to indicate that
the rate of r1 dependents on B. For more examples see [63].
Consequently, the structural interpretation of ODEs models, i.e. converting
(5) into any of (1)(4), calls for special attention. The problematic cases are
reactions with complex kinetics. Some tools popular in Systems Biology, e.g.,
COPASI [40], provide predefined functions representing whole building blocks,
i.e. subnets in our reaction networks. See Fig. 3 for an example, which has been
taken from the reaction network reported in [9]. The network on the left has
the structure as suggested by the schematic representation in [9] and the list of
reactions in the models SBML format created by COPASI. The network on the
right shows the correct structure, which is hidden in the kinetics of reactions 23
and 25.
Hence, the (backward) translation from ODEs to structured models suitable
for qualitative or stochastic analysis needs to be performed with great care.
In [63], we look into the structure inference problem for ODEs models. We provide three biochemically relevant sufficient conditions under which the derived
structure is unique and counterexamples showing the necessity of each of the
following conditions.
All reactions use pure mass/action kinetics, and the reaction parameters
belong to a finite alphabet of symbols.
The reaction network does not contain any void reaction.
The same parameter is never used for two different reactions with the same
reactants.
In summary, neither (4) nor (5) are suitable core notations for reaction networks, which may be subject to various complementary qualitative, stochastic,
or continuous analysis techniques.
Structure versus list notation. Enumerating all reactions in a list (in fact
an unordered set) is perfectly fine, if the notation is to be read by a computer.
A structure-oriented graphical notation brings the additional advantage for an
human reader of revealing the causality relation (structure, topology) among the
reaction set. As such we consider (2) and (3) to be closer to the reality to be
modelled.

How Might Petri Nets Enhance . . .

MEKP

21

MEKPP

MEKP

22

23
ERK

ERKP

24

21

21

MEKPP

22
23a
23b

25
26

ERKPP

ERK

24

ERKP

25a
25b
26

ERKPP

Fig. 3. Freestyle kinetics may hide essential structure. Assuming for reaction r23 in
the network on the left the complex kinetics v23 = v(r23 ) = k23a ERK M EKP +
k23b ERK M EKP P , likewise for r25 , yields the network on right with all reactions
having mass/action kinetics, e.g., v23a = v(r23a ) = k23a ERK M EKP .

Hypergraph versus Petri net. At first glance, both are equally suitable
for a graphical representation of the network structure. The difference in the
graphical notation style seems to be a matter of taste.
Petri nets enjoy a formal semantics, which could also be formulated in the terminology of hypergraphs. However, this might become a bit over-complicated.
Reasoning about the reaction network structure, e.g. to derive behavioural properties, requires splitting a hyperarc into its ingredients. We need to refer to the
reaction itself, its reactants the nodes where the hyperarc starts, and its products the nodes where the hyperarc goes to, and the individual multiplicities of
those parts of the hyperarc in the case of stoichiometric information.
Moreover networks tend to be very large. This motivates the reuse of a wellestablished engineering principle to manage the design of large-scale systems
hierarchical decomposition. Formal concepts of hierarchical Petri nets with
building blocks are well elaborated, see, e.g., [20]. They facilitate the modeling
of large real-world systems because the comprehension of the whole network
builds upon the understanding of all building blocks and the interconnection of
their interfaces. Hierarchical structuring changes the style of representation, but
does not change the actual net structure of the underlying reaction network.
Even if hierarchical modelling approaches have gained little popularity in practice so far in Systems Biology, they are indispensable for large-scale networks.
There is nothing like this for hypergraphs.
This explains why we keep to Petri net terminology when in the next sections
we briefly describe some contributions which Petri net theory has to offer for
reaction network analysis, and thus to a Systems Biology toolkit.
Disclaimer: Being restricted to 20 pages, this paper can not exhaustively cover
the field.

The Framework

In the following we recall our overall framework, introduced in [24], that relates the three major ways of modelling and analysing biochemical networks:
qualitative, stochastic and continuous, compare Figure 4.

22

M. Heiner and D. Gilbert

Fig. 4. Conceptual core framework

Qualitative paradigm. The most abstract representation of a biochemical

network is qualitative and is minimally described by its topology, usually as a
bipartite directed graph with nodes representing biochemical entities and reactions, or in Petri net terminology places and transitions. Arcs can be annotated
with stoichiometric information, whereby the default stoichiometric value of 1 is
usually omitted.
The qualitative description can be further enhanced by the abstract representation of discrete quantities of species, achieved in Petri nets by the use of tokens
at places. These can represent the number of molecules, or the level of concentration, of a species, and a particular arrangement of tokens over a network is
called a marking, specifying the system state.
The behaviour of such a net forms a discrete state space, which can either be
captured as (1) a Labeled Transition System (LTS) (in the Petri net community
better known as reachability or marking graph) to describe the net behaviour by
all (totally ordered) interleaving sequences in the style of transition-labelled automata (interleaving semantics), or as (2) a finite prefix of a maximal branching
process (PO prefix) to describe the net behaviour by all partially ordered transition sequences (partial order semantics). Animating a Petri net by sequentially
firing individual transitions generates a path through the LTS.
Both descriptions of behaviour can be analysed for the purpose of model verification. In the bounded case, this is best done using model checking techniques,
where the properties of interest are expressed by, e.g., a branching time temporal logic, one instance of which is Computational Tree Logic (CTL) [15], or a
linear-time logic (LTL) [54].
The standard semantics for these qualitative Petri Nets (PN)1 do not associate a time with transitions or the sojourn of tokens at places, and thus these
descriptions are time-free. The qualitative analysis considers however all possible
behaviour of the system under any timing. Thus, the qualitative Petri net model
itself implicitly contains all possible time-dependent behaviour.
1

As we want to use the term Petri net as umbrella term, we refer to the standard
Petri net class as qualitative Petri nets, if required for the sake of unambiguity.

How Might Petri Nets Enhance . . .

23

Timed information can be added to the qualitative description in two ways

stochastic and continuous.
Stochastic paradigm. The stochastic Petri net (SPN) description preserves
the discrete state, but in addition associates an exponentially distributed firing rate (waiting time) with each reaction. The firing rates are typically statedependent and specified by rate functions. All reactions, which occur in the PN,
can still occur in the SPN, but their likelihood depends on the probability distribution of the associated firing rates. The underlying semantics is a ContinuousTime Markov Chain (CTMC). Stochastic simulation generates a random walk
through the CTMC.
Special behavioural properties can be expressed using, e.g., Continuous Stochastic Logic (CSL), a stochastic counterpart of CTL which was originally introduced in [4], and extended in [5], or PLTLc, a probabilistic extension of LTL
with constraints [18].
The PN is an abstraction of the SPN, sharing the same state space and transition relation with the stochastic model (if there are no parallel transitions),
with the probabilistic information removed. All qualitative properties valid in
the PN are also valid in the SPN, and vice versa.
Continuous paradigm. The Continuous Petri Net (CPN) replaces the discrete values of species with continuous values, and hence is not able to describe
the behaviour of species at the level of individual molecules, but only the overall
behaviour via concentrations. We can regard the discrete description of concentration levels as abstracting over the continuous description of concentrations.
Timed information is introduced by the association of a particular deterministic
firing rate with each transition, permitting the continuous model to be represented as a set of Ordinary Differential Equations (ODEs). The concentration
of a particular species in such a model will have the same value at each point
of time for repeated computational experiments. The state space of such models
is continuous and linear. It can be analysed by, for example, Linear Temporal
Logic with constraints (LTLc) in the manner of [11].
Moving between stochastic and continuous paradigms. One and the
same quantitative model can be read either stochastically or continuously, no
changes being required (up to some scaling in the rate functions for higher order
reactions). In the stochastic case, the rate functions define the state-dependent
rates of the individual occurrences of transition firings, while in the continuous
case the rate functions define the strength of the state-dependent continuous
flow. In [24] we discuss in more detail how the stochastic and continuous models
are mutually related by approximation. The CPN replaces the probabilistically
distributed reaction firing in the SPN by a particular average firing rate defining
the continuous token flow. In turn, the stochastic model can be derived from the
continuous model by approximation, reading the tokens as concentration levels,
as introduced in [10].
Bridging to the continuous paradigm. It is well-known that time assumptions generally impose constraints on behaviour. The qualitative and stochastic
models consider all possible behaviour under any timing, whereas the continuous

24

M. Heiner and D. Gilbert

model is constrained by its inherent determinism to consider a subset. This may

be too restrictive when modelling biochemical systems, which by their very nature
exhibit variability in their behaviour.
The move from the discrete to the continuous paradigm may come along with
counter-intuitive effects; e.g., a trap a set of places which can never become
empty in the discrete case as soon as it has obtained a token can be emptied
in the continuous case [62]. See [2] for another surprising example.
For illustration we briefly discuss the toy example given in Figure 2, left,
which has been used in [61] to introduce the idea of Absolute Concentration
Robustness (ACR). A place of a CPN (a variable of an ODEs model) is said to
have ACR if its concentration is the same in all positive steady states; i.e., it
does not depend on the total mass in the system, but only depends on the kinetic
constants. The place A is such an ACR place; its steady state concentration is
k2 /k1 , while B has the steady state concentration total k2 /k1 , with total being
the conserved total mass in the network.
The place B forms a bad siphon, i.e., a siphon, not containing a trap. It is
easy to see that the net does not have a live initial marking in the discrete
world. Thus, there is always a dead state reachable, in which the bad siphon,
i.e., B, is empty and all tokens reside in A, when the system is considered
stochastically. However, there is no sign of the dead state when reading the net
as CPN. Moreover, increasing the mass in the closed system will increase the
steady value of B, as A has ACR. Thus, the continuous model may predict a
simplified behaviour that does not reflect the variety, which is possible in the
discrete case, e.g. in the given example, that the siphon will eventually get empty.
When analysing ODEs, one actually considers a family of related models,
varying in some structural details, but generating the same ODEs. So far, the
bridge to the continuous world is not equally well understood regarding the
stochastic world. However, qualitative, stochastic & continuous Petri nets, which
share structure should generally share some behavioural properties as well.
Applications. Case studies demonstrating the move between the three paradigms can be found in [29] (signalling cascade), [25] (biosensor gene regulation),
and [28] (signal transduction network). In [23], we demonstrate by a case study
the relationship between the strongly connected state space of the qualitative
Petri net and the steady state behaviour of the corresponding continuous Petri
net.
3.1

Beyond the Core Framework

In addition to the base case techniques, suitably expressive modelling formalisms

and very efficient computational techniques are required to support abstraction
levels and to give a holistic perspective on inherently multi-scale systems. As
a consequence, a variety of different techniques have been proposed during the
last decade, and new ones are constantly emerging. At the same time, many
well-established and sophisticated modelling techniques from Computer Science,
equipped with advanced analytical methods, have been applied to the exciting
and challenging research field of Systems and Synthetic Biology.

How Might Petri Nets Enhance . . .

25

Figure 5 shows the relationship between the most important modelling techniques contributed by the Petri net community. In the following we briefly characterize the main players, going beyond the core framework.
Extended Petri Nets (XPN) complement the standard ingredients of Petri
nets by some dedicated features for the sake of easy modelling, but often
sacrificing static analyzability. They exist in numerous variations. An extension particularly useful for Systems and Synthetic Biology are special arcs,
e.g., read (test) arcs and inhibitor arcs, which provide adequate abstract
modelling means for activating or inhibiting activities; see, e.g., [12, 35].
Functional Petri Nets (FPN) [38] pick up the idea of self-modifying nets
[64] and use state-dependent functions to dynamically adjust arc multiplicities, which has been used to simulate metabolic pathways [14].
Time Petri Nets (TPN) equip transitions with a deterministic firing delay,
typically characterized by a continuous time interval [49] of minimal and
maximal delay. Remarkably, this net class can be fully analysed using a
discretized state space only [56]; the particular capabilities for Systems and
Synthetic Biology have be demonstrated in [58, 55, 57].
Generalised Stochastic Petri Nets (GSPN) extend SPN by immediate
transitions [45], possibly complemented by special arcs. They have been used
in, e.g., [42].
Extended Stochastic Petri Nets (XSPN) enrich GSPN with transitions
having deterministic firing delays, typically characterized by a constant.
They come in two flavours: relatively timed or absolutely timed, which helps
in model-based wet-lab experiment design; see [32].
Hybrid Petri Nets (HPN) enrich CPN with discrete places and discrete
transitions having deterministic firing delay, see, e.g., [17].
Hybrid Functional Petri Nets (HFPN) are a popular extension of HPN
combining them with the self-modifying arcs of FPN. This net class is supported by the licensed tool Cell Illustrator, which has been deployed in many
case studies [50].
Generalised Hybrid Petri Nets (GHPN) combine all features of XSPN
and CPN [37].
In addition, further novel Petri net classes have been developed, which may turn
out to be useful in Systems and Synthetic Biology. We briefly discuss two of
them.
The Probability Propagation Nets [43] provide a Petri net representation for
the propagation of probabilities and likelihoods in Bayesian networks. As such
they enhance the transparency of propagation processes in the Bayesian world
by exploiting structural and dynamic properties of Petri nets. They can be used
to express knowledge which has been statistically extracted from gene expression
array data (in the wet lab) to obtain executable and analyzable models of the
data source (in the dry lab).
The Error-correcting Petri Nets [51] allow for the algebraic detection and
modulo-p correction of non-reachable Petri net states, without ever constructing

26

M. Heiner and D. Gilbert

Fig. 5. Extended Framework showing the relation between some of the Petri net classes
used in Systems and Synthetic Biology

the state space. They may gain importance in synthetic biology, e.g., to design
biological systems with monitoring capabilities. The observation of states which
are supposed to be non-reachable in the designed system will indicate some
behavioural deviations from the original design. The method is able to pinpoint
the source of the problem to a single place (several places) for a single error
(multiple errors) depending on the amount of structural redundancy utilised
during model construction.
This variety clearly demonstrates one of the big advantages of using Petri nets
as a kind of umbrella formalism the models may share the network structure,
but vary in their quantitative (kinetic) information.
Related material. Formal definitions covering discrete, stochastic, continuous and hybrid variants of the Petri net formalisms, illustrated by many technical
examples, can be found in [16]. The net classes closest to the continuous setting
of biomolecular networks are discussed in Chapter 7, where the speeds (i.e., the
rates) may depend on the C-marking (i.e., the current values of the continuous
places). Combining the concept of variable rates with Definition 5.4 (feeding and
draining speed of a place) and Definition 5.5 (the balance of the marking of a
continuous place) explains how the ODEs model is set up.
However, the modelling complexity on hand calls for formal definitions and
related tool support tailored to the specific needs of Systems and Synthetic Biology; e.g., by restricting the domain of the state-dependent rate functions to
the pre-places of a transition in order to maintain a close relation between structure and behaviour. Biochemically interpreted Continuous Petri Nets have been
introduced in [23], biochemically interpreted Stochastic Petri Nets in [24], and
biochemically interpreted Extended Stochastic Petri Nets in [32]. More formal
definitions covering the core framework can be found in [29].

How Might Petri Nets Enhance . . .

27

Overview of Petri Net Specific Contributions

Probably the most popular notions in Petri net theory having their counterparts
in Systems Biology are place and transition invariants (P/T-invariants for short).
They are defined on a models representation, which is called incidence matrix
in the Petri net community and stoichiometric matrix in Systems Biology.
T-invariants can be seen as the integer counterparts of different representations of the continuous members of a cone, known in Systems Biology as elementary flux modes, extreme pathways, and generic pathways, while P-invariants
correspond to conserved moieties [66].
Beyond that, Petri nets offer the following additional features to the list of
established techniques known for quite a while in Systems Biology, such as nonlinear ODEs analysis, bifurcation theory, chemical reaction network theory [21],
stoichiometric and flux balance analysis [52], etc..
4.1

Readability

Petri nets are an intuitive representation of reaction networks, which makes

them easily comprehensible. They uniquely define derived notations, such as
stoichiometric matrix or ODEs, but not vice versa; for details see Section 2.
Applications. Petri nets allow the unambiguous, but still readable representation of various types of biological processes at different levels of abstraction,
with different resolution of detail, in one and the same model, ranging from the
conformational change of a single molecule to the macroscopic response of a cell,
the development of a tissue, or even the behaviour of a whole organism [48]. A
recent survey [6] has shown how Petri nets can be applied to transcriptional,
signalling, and metabolic networks, or combinations of them, illustrating this
with a rich set of case studies. Most of these focus on the molecular level; however examples at the multi-cellular level include the signal-response behaviour
of an organism [47], and developmental processes in multi-cellular pattern formation [7, 13].
4.2

Executability

Petri nets are directly executable with a suitable tool like Snoopy [59, 46]. The
visualization of the token flow
allows the user to better comprehend the net behaviour, and
facilitates the communication between wet-lab experimentalists and dry-lab
(computational) theoreticians.
Applications. The execution allows a time-free animation in the case of qualitative Petri nets (these are the standard Petri nets), and time-dependent animation
and simulation in the case of quantitative Petri nets (stochastic, continuous, and
hybrid Petri nets).

28

4.3

M. Heiner and D. Gilbert

Causality

The structure of Petri nets reflects the causality relation among the reactions
building the whole network, while clearly distinguishing between alternative and
concurrent behaviour; see [31] for a more detailed discussion. This permits reasoning about which reaction has to happen first, before another reaction can
happen afterwards in order to transport some mass (metabolic networks) or
transfer some information (signal transduction network) through the network.
Applications. Reasoning over causality helps, e.g., in evaluating T-invariants
(likewise elementary flux modes, extreme pathways, generic pathways), which
define subnets in a reaction network. The unfolding of the involved reactions
reveals their partial order causality, and may allow for complementary insights
into the net behaviour; for examples see [29], [28].
4.4

Efficiency Gain

The Petri net structure allows for efficiency gains of some analysis algorithms,
which are otherwise not possible. The crucial point is that the occurrence of a
reaction only causes changes in its local environment, which consists of its preplaces, the reactants, and its post-places, the products. Vice versa, this requires
that all dependencies are reflected in the structure. Thus we restrict the domain
of transitions rate functions to the transitions pre-places.
Applications. Examples implemented in our toolkit, see Section 6, include:
Stochastic simulation to compute/update the state-dependent transition
rates [34];
Static ordering of variables (places) to statically determine suitable orders
of variables for high compression effects in symbolic representations, e.g.,
for symbolic model checking. In [60], we demonstrate how the performance
of the popular CSL model checker PRISM [53] could take advantage of our
structure-based heuristics to derive suitable orders of variables.
Static ordering of transitions to statically determine suitable orders of
transitions for efficient symbolic state space construction, e.g., saturationbased construction, with the aim to avoid intermediate data structures which
are much larger than the final ones.
4.5

Static Analysis

The standard analysis techniques, which are commonly used in many Computer
Science approaches, need to construct the partial or full state space (expressed
as LTS or PO prefix). In addition to these dynamic analysis techniques, Petri net
theory offers a variety of static analysis techniques, which permit the decisions of
general behavioural properties, such as boundedness and liveness, without state
space construction.

How Might Petri Nets Enhance . . .

29

Boundedness ensures upper bounds for all species, which implies a finite
state space. Structural boundedness guarantees this behaviour for any initial
marking.
Liveness ensures that all reactions will forever contribute to the system behaviour, which precludes dead states. Structural liveness guarantees that
there is a live initial marking.
Static analysis techniques, which we found useful so far in analyzing biochemical
reaction networks, comprise:
property-preserving reduction rules, which replace subnetworks by generally
smaller ones without changing the general behavioural properties;
structural analyses, which determine, e.g., the net structure classes (such
as state machine, synchronisation graph, free choice nets, extended simple
nets), Siphon Trap Property (STP), or rank theorem, and apply related
theorems;
linear programming techniques, which perform, e.g., structural boundedness
check to decide if the state space is finite; P/T-invariant computation and
coverage test, which supports also model validation; or state/trap equation
check as sufficient criteria for non-reachability.
More details can be found in [29], where we recall basic notions and theorems
of Petri net theory. Most of them are implemented in our Petri net analysis tool
Charlie [22].
Applications. Case studies demonstrating the power of static analysis techniques can be found in [29] (signalling cascade), [25] (biosensor gene regulation),
and [28] (signal transduction network).
The comprehensive textbook [52] is focused on the stoichiometric matrix and
related analysis techniques. It is also a good entry point for the growing body
of related literature. The use of P/T-invariants for the validation of biochemical
networks has been introduced to the Petri net community in [30].
Specific application examples include:
Angeli et al. introduce in [3] a sufficient criterion for persistency of ODEs
models, i.e., of continuous Petri nets, which is based on the structure. It
ensures persistency for reaction networks with mass/action kinetics and arbitrary parameters. This structural criterion closely resembles a famous liveness criterion of Petri net theory the Siphon Trap Property, adjusted to
the continuous case: the trap is replaced by a P-invariant.
In [27], we use T-invariants and the related notion of Abstract Dependent
Transition sets (ADT sets) to automatically derive hierarchically structured
networks.
In [36], we apply structural analysis, specifically T-invariants, to reduce a
given ODEs model, while preserving the essential steady state behaviour.
The structural analysis identifies the core network and its fragile node (robustness analysis).

30

M. Heiner and D. Gilbert

In [33], we shed new light on the importance of the Siphon Trap Property
for continuization (fluidization), which is determined for a number of biochemical networks from our repository.
We expect further results along this line in the future. Petri nets offer a suitable
framework for reasoning about the relationship between the three basic modelling paradigms. For this purpose, our tool Snoopy [59] allows the user to move
easily between the three worlds, and thus ensuring the equivalence of the related
models.

Model Checking

Model checking is a powerful analysis approach, well-established in Computer

Science and now gaining increasing popularity in Systems and Synthetic Biology, which can be performed in the three modelling paradigms. It permits
checking the behaviour of a model for certain properties expressed in temporal
logics. These are unambiguous languages, which provide flexible formalisms to
check the validity of statements (propositions) in relation to the execution of the
model. Temporal logics come in different flavours; for example branching-time
and linear-time logics. The latter are more attractive to biologists who are used
to reason about time series behaviour.
Analytical model checking is usually restricted to LTS descriptions (interleaving semantics). However, we can take advantage of Petri nets partial order
semantics to perform model checking over behaviour descriptions given as PO
prefix, which preserve the difference between alternative and concurrent occurrences of reactions. The power of this technique for Systems and Synthetic Biology has not been systematically explored yet.
Analytical model checking may become computationally expensive and generally requires network boundedness. An efficient alternative approach is simulative
model checking, where analysis is performed over time series traces generated by
model simulation, which works both for bounded and unbounded models. It involves two approximations: a finite number of finite simulation traces is only
considered. Simulative model checking can perhaps best be characterised as behaviour checking, being applicable not only to continuous or stochastic models,
but also to any time series data. This approach can thus be used to compare
models with physical systems in terms of data generated by simulations of a
model and laboratory experiments.
In [29,28], we have demonstrated how to perform model checking in the three
modelling paradigms in a concerted manner.
5.1

Characterising Biochemical Species Behaviour

The behaviour of biochemical species can be described for quantitative model

checking using four distinct descriptive approaches, with increasing specificity;
qualitative, semi-qualitative, semi-quantitative and quantitative [18].

How Might Petri Nets Enhance . . .

31

Qualitative formulae use derivatives of biochemical species concentrations or

mass, determined by the function d(.), and the temporal operators U , F , G
to describe the general trend of the behaviour. Semi-qualitative extend qualitative with relative concentrations using the function max(.), which yields the
maximal value of a species observed in the whole run. Semi-quantitative extend
semi-qualitative with absolute time values by referring to the predefined system variable time. Finally, quantitative extend semi-quantitative with absolute
concentration values.
For example, the transient activation of a biochemical species called P rotein
can be expressed in these approaches with increasing accuracy. The following
queries determine the probabilities of the statements to be true using PLTLc.
Qualitative. Protein rises then falls:
P=? [ d(Protein) > 0 U ( G( d(Protein) < 0 ) ) ].

Semi-qualitative. Protein rises then falls to less than 50% of its peak concentration:
P=? [ ( d(Protein) > 0 ) U ( G( d(Protein) < 0 )
F ( [Protein] < 0.5 max[Protein] ) ) ].

Semi-quantitative. Protein rises then falls to less than 50% of its peak concentration at 60 minutes:
P=? [ ( d(Protein) > 0 ) U ( G( d(Protein) < 0 )
F ( time = 60 Protein < 0.5 max(Protein) ) ) ].

Quantitative. Protein rises then falls to less than 100M ol at 60 minutes:

P=? [ ( d(Protein) > 0 ) U ( G( d(Protein) < 0 )
F ( time = 60 Protein < 100 ) ) ].

5.2

Applications for Model Checking

There are five major areas of applications for model checking in Systems and
Synthetic Biology [8].
Model validation. Does the model behave in the way we expect?
Model comparison. How similar are two models, independent of their underlying formalisms?
Model searching. In a database of alternative models, which of them exhibit
a particular property? This can be used to select among competing descriptions of a system produced by various research groups. Given a large database
of models (either a general collection or variants of the same model), one can
use model behaviour checking to perform systematic database queries, such
as find all models that show oscillatory behaviour under some conditions

32

M. Heiner and D. Gilbert

or find all descriptions of this signaling pathway that are transiently active
after growth factor stimulation.
Model analysis. In a collection of structure-preserving variants of a model
(e.g., different concentrations of species representing in-silico gene knockouts), which models show a certain behaviour? E.g., how many knock-outs
lead to a loss of oscillating behaviour?
Model construction. Which modifications of a model lead to a desired property? Modifications can involve changes in kinetic parameters or initial concentrations, but they can also be more complex, for example changing the
topology of the model by removing or even adding new components. How to
do this efficiently is still an active area of research.

Tools

BioModel Engineering of non-trivial case studies requires adequate tool support. We provide a sophisticated toolkit covering the whole framework; publicly
available at http://www-dssz.informatik.tu-cottbus.de.
Snoopy tool for modelling and animation/simulation of hierarchically structured graphs, among them qualitative, stochastic, and continuous Petri nets
[59], [46], recently extended by coloured counterparts of these net classes [44]
and Generalised Hybrid Petri Nets, supporting dynamic partitioning [37].
Charlie multi-threaded analysis tool of standard Petri net properties and
techniques of Petri net theory including structural boundedness check, P/Tinvariants, STP, rank theorem, structural reduction. It also supports explicit
CTL and LTL model checking [22], mainly for teaching purposes.
Marcie is built upon an Interval Decision Diagram (IDD) engine and supports, besides the analysis of standard Petri net properties (boundedness,
dead states, liveness, reversibility) symbolic CTL model checking of qualitative Petri nets [35] and multi-threaded symbolic CSL model checking of
Generalised Stochastic Petri Nets [60], recently extended by rewards. Both
model checkers outperform comparable tools on the market for benchmarks
of typical biochemical networks [34].
Exact analyses of bounded models are complemented by approximative model
checking, which is also suitable for unbounded models. Two approximative
engines support fast adaptive uniformisation and distributed Gillespie simulation.
In addition, Snoopy files are directly read by ADAM a tool applying computational algebra to analyse the dynamics of discrete models [65], and simulation
traces generated with Snoopy are read by MC2(PLTLc) a Monte-Carlo Model
Checker for PLTLc [18]. Additionally, Snoopy provides export to several foreign analysis tools, among them Anastasia [67] for efficient computation of bad

How Might Petri Nets Enhance . . .

33

siphons and STP decision, and SBML import/export, which opens the door to
a bunch of tools popular in Systems and Synthetic Biology.

Outlook and Challenges

A drawback of current modelling approaches, including Petri nets, are their

limitation to relatively small networks. Thus overall, one of the major challenges
lies in the application of Petri nets to achieve richer descriptions of biological
systems than are currently practicable. This includes modelling compartments,
locality and mobility of players at the intracellular level. Biological systems can
be represented as networks which themselves typically contain regular (network)
structures, and/or repeated occurrences of network patterns. This organisation
occurs in a hierarchical manner, reflecting the physical and spatial organisation of
the organism, from the intracellular to the intercellular level and beyond (tissues,
organs etc.). Examples of organisation at the intracellular level include: cytosol,
nucleus, ribosome, endoplasmic reticulum, Golgi apparatus, and mitochondria.
Further challenges lie in modelling at the intercellular level, for examples societies of single-celled organisms (including mixtures of organisms), or multicellular organisms at the tissue or organ level. There are many examples of
communication in multicellular organisms including endocrine, juxtacrine, etc.
Although such network models can be designed using standard Petri nets, so
far there is no dedicated support for such structuring; it becomes impractical as
the size of the networks to be modelled increases. Besides the purely technical
aspect due to the impracticality of handling large flat nets, humans need some
hierarchy and organisation in what they are designing in order to be able to
conceptualise the modelled object. Thus, models should explicitly reflect the
organisation in complex biological systems.
There are two established orthogonal concepts in Petri net modelling to manage large-scale networks - structuring by hierarchies and colours. However, their
potential for BioModel Engineering has not been systematically explored so far.
In order to cope with the challenge of modelling at such a variety of levels, we
need expressive modelling techniques embedded as part of a robust BioModelling
Engineering approach, supporting, e.g., model version control in a collaborative
environment for model construction.

Summary

Petri nets are a natural and established notation for describing reaction networks
because both share the bipartite property. We have shown that Petri nets can be
used to perform all major modelling and simulation approaches central to Systems Biology. Thus, they may serve as a kind of umbrella formalism integrating
qualitative and quantitative (i.e. stochastic, continuous, or hybrid) modelling
and analysis techniques.
However, as we have indicated, there are several important open challenges
posed by modelling biological systems which need to be addressed.

34

M. Heiner and D. Gilbert

References
1. Aderem, A.: Systems Biology: Its Practice and Challenges. Cell 121(4), 511513
(2005)
2. Angeli, D.: Boundedness analysis for open Chemical Reaction Networks with massaction kinetics. J. Natural Computing (2009), doi:10.1007/s11047-009-9163-7
3. Angeli, D., De Leenheer, P., Sontag, E.D.: A Petri net approach to the study of
persistence in chemical reaction networks. Mathematical Biosciences 210(2), 598
618 (2007)
4. Aziz, A., Sanwal, K., Singhal, V., Brayton, R.: Model checking continuous time
Markov chains. ACM Trans. on Computational Logic 1(1) (2000)
5. Baier, C., Haverkort, B., Hermanns, H., Katoen, J.-P.: Model checking algorithms
for continuous-time Markov chains. IEEE Trans. on Software Engineering 29(6)
(2003)
6. Baldan, P., Cocco, N., Marin, A., Simeoni, M.: Petri Nets for Modelling Metabolic
Pathways: a Survey. J. Natural Computing (9), 955989 (2010)
7. Bonzanni, N., Krepska, E., Feenstra, K., Fokkink, W., Kielmann, T., Bal, H.,
Heringa, J.: Executing multicellular differentiation: quantitative predictive modelling of c. elegans vulval development. Bioinformatics 25, 20492056 (2009)
8. Breitling, R., Donaldson, R.A., Gilbert, D., Heiner, M.: Biomodel Engineering From Structure to Behavior. Lect. Notes Bioinformatics 5945, 112 (2010)
9. Brightman, F.A., Fell, D.A.: Differential feedback regulation of the MAPK cascade
underlies the quantitative differences in EGF and NGF signalling in PC12 cells.
FEBS letters 482(3), 169174 (2000)
10. Calder, M., Vyshemirsky, V., Gilbert, D., Orton, R.: Analysis of signalling pathways using continuous time Markov chains. In: Priami, C., Plotkin, G. (eds.) Transactions on Computational Systems Biology VI. LNCS (LNBI), vol. 4220, pp. 4467.
Springer, Heidelberg (2006)
11. Calzone, L., Chabrier-Rivier, N., Fages, F., Soliman, S.: Machine learning biochemical networks from temporal logic properties. In: Priami, C., Plotkin, G. (eds.)
Transactions on Computational Systems Biology VI. LNCS (LNBI), vol. 4220, pp.
6894. Springer, Heidelberg (2006)
12. Chaouiya, C.: Petri Net Modelling of Biological Networks. Briefings in Bioinformatics 8(4), 210 (2007)
13. Chen, L., Masao, N., Kazuko, U., Satoru, M.: Simulation-based model checking
approach to cell fate specification during c. elegans vulval development by hybrid
functional Petri net with extension. BMC Systems Biology 42 (2009)
14. Chen, M., Hariharaputran, S., Hofest
adt, R., Kormeier, B., Spangardt, S.: Petri
net models for the semi-automatic construction of large scale biological networks.
J. Natural Computing (2009), doi:10.1007/s11047-009-9151-y
15. Clarke, E.M., Grumberg, O., Peled, D.A.: Model checking. MIT Press, Cambridge
(2001)
16. David, R., Alla, H.: Discrete, Continuous, and Hybrid Petri Nets. Springer,
Heidelberg (2010)
17. Doi, A., Drath, R., Nagaska, M., Matsuno, H., Miyano, S.: Protein Dynamics Observations of Lambda-Phage by Hybrid Petri net. Genome Informatics, 217218
(1999)
18. Donaldson, R., Gilbert, D.: A model checking approach to the parameter estimation
of biochemical pathways. In: Heiner, M., Uhrmacher, A.M. (eds.) CMSB 2008.
LNCS (LNBI), vol. 5307, pp. 269287. Springer, Heidelberg (2008)

How Might Petri Nets Enhance . . .

35

19. Endy, D.: Foundations for engineering biology. Nature 438(7067), 449453 (2005)
20. Fehling, R.: A concept of hierarchical Petri nets with building blocks. In: Rozenberg, G. (ed.) APN 1993. LNCS, vol. 674, pp. 148168. Springer, Heidelberg (1993)
21. Feinberg, M.: Mathematical aspects of mass action kinetics. In: Lapidus, L.,
Amundson, N.R. (eds.) Chemical Reactor Theory: A Review, ch.1, pp. 178.
Prentice-Hall, Englewood Cliffs (1977)
22. Franzke, A.: Charlie 2.0 - a multi-threaded Petri net analyzer. Diploma Thesis,
Brandenburg University of Technology at Cottbus, CS Dep. (2009)
23. Gilbert, D., Heiner, M.: From Petri nets to differential equations - an integrative
approach for biochemical network analysis. In: Donatelli, S., Thiagarajan, P.S.
(eds.) ICATPN 2006. LNCS, vol. 4024, pp. 181200. Springer, Heidelberg (2006)
24. Gilbert, D., Heiner, M., Lehrack, S.: A unifying framework for modelling and
analysing biochemical pathways using petri nets. In: Calder, M., Gilmore, S. (eds.)
CMSB 2007. LNCS (LNBI), vol. 4695, pp. 200216. Springer, Heidelberg (2007)
25. Gilbert, D., Heiner, M., Rosser, S., R., Fulton, X.G., Trybi"lo, M.: A Case Study
in Model-driven Synthetic Biology. In: Proc. 2nd IFIP Conference on Biologically
Inspired Collaborative Computing (BICC), IFIP WCC 2008, Milano, pp. 163175
(2008)
26. Heinemann, M., Panke, S.: Synthetic biology - putting engineering into biology.
Bioinformatics 22(22), 27902799 (2006),
http://dx.doi.org/10.1093/bioinformatics/btl469
27. Heiner, M.: Understanding Network Behaviour by Structured Representations
of Transition Invariants A Petri Net Perspective on Systems and Synthetic
Biology. Natural Computing Series, pp. 367389. Springer, Heidelberg (2009),
http://www.springerlink.com/content/m8t30720r141442m
28. Heiner, M., Donaldson, R., Gilbert, D.: Petri Nets for Systems Biology. In: Iyengar,
M.S. (ed.) Symbolic Systems Biology: Theory and Methods, ch.21, Jones and
Bartlett Publishers, Inc. (2010)
29. Heiner, M., Gilbert, D., Donaldson, R.: Petri nets in systems and synthetic biology.
In: Bernardo, M., Degano, P., Tennenholtz, M. (eds.) SFM 2008. LNCS, vol. 5016,
pp. 215264. Springer, Heidelberg (2008)
30. Heiner, M., Koch, I.: Petri Net Based Model Validation in Systems Biology. In:
Cortadella, J., Reisig, W. (eds.) ICATPN 2004. LNCS, vol. 3099, pp. 216237.
Springer, Heidelberg (2004)
31. Heiner, M., Koch, I., Will, J.: Model Validation of Biological Pathways Using Petri
Nets - Demonstrated for Apoptosis. BioSystems 75, 1528 (2004)
32. Heiner, M., Lehrack, S., Gilbert, D., Marwan, W.: Extended Stochastic Petri Nets
for Model-Based Design of Wetlab Experiments. Transactions on Computational
Systems Biology XI, 138163 (2009)
33. Heiner, M., Mahulea, C., Silva, M.: On the Importance of the Deadlock Trap
Property for Monotonic Liveness. In: Int. Workshop on Biological Processes &
Petri Nets (BioPPN), satellite event of Petri Nets 2010 (2010)
34. M., Heiner, C.R., M., Schwarick, S.S.: A Comparative Study of Stochastic Analysis
Techniques. In: Proc. CMSB, pp. 96106 (2010) ; ACM Digital Library 978-1-45030068-1/10/09
35. Heiner, M., Schwarick, M., Tovchigrechko, A.: DSSZ-MC - A Tool for Symbolic
Analysis of Extended Petri Nets. In: Franceschinis, G., Wolf, K. (eds.) PETRI
NETS 2009. LNCS, vol. 5606, pp. 323332. Springer, Heidelberg (2009)
36. Heiner, M., Sriram, K.: Structural analysis to determine the core of hypoxia
response network. PLoS ONE 5(1), e8600 (2010),

36

37.

38.
39.
40.

41.

42.

43.

44.

45.

46.

47.

48.

49.

50.

51.
52.
53.
54.

M. Heiner and D. Gilbert

http://dx.doi.org/10.1371/journal.pone.0008600,
doi:10.1371/journal.pone.0008600
Herajy, M., Heiner, M.: Hybrid Petri Nets for Modelling of Hybrid Biochemical
Interactions. In: Proc. AWPN 2010, CEUR Workshop Proceedings, vol. 643, pp.
6679. CEUR-WS.org (2010)
Hofest
adt, R., Thelen, S.: Quantitative modeling of biochemical networks. In Silico
Biol. 1, 3953 (1998)
Hood, L.: Systems biology: integrating technology, biology, and computation.
Mechanisms of Ageing and Development 124(1), 916 (2003)
Hoops, S., Sahle, S., Gauges, R., Lee, C., Pahle, J., Simus, N., Singhal, M., Xu,
L., Mendes, P., Kummer, U.: Copasi a complex pathway simulator. BioInformatics 22(24), 30673074 (2006)
Hucka, M., Finney, A., Sauro, H.M., Bolouri, H., Doyle, J.C., Kitano, H., et al.: The
Systems Biology Markup Language (SBML): A Medium for Representation and
Exchange of Biochemical Network Models. J. Bioinformatics 19, 524531 (2003)
Lamprecht, R., Smith, G.D., Kemper, P.: Stochastic Petri net models of
Ca2+ signaling complexes and their analysis. J. Natural Computing (2009),
doi:10.1007/s11047-009-9143-y
Lautenbach, K., Pinl, A.: A Petri net representation of Bayesian message flows:
importance of Bayesian networks for biological applications. J. Natural Computing
(2009), doi:10.1007/s11047-009-9142-z
Liu, F., Heiner, M.: Computation of Enabled Transition Instances for Colored Petri
Nets. In: Proc. AWPN 2010, CEUR Workshop Proceedings, vol. 643, pp. 5165.
CEUR-WS.org (2010)
Marsan, M.A., Balbo, G., Conte, G., Donatelli, S., Franceschinis, G.: Modelling
with Generalized Stochastic Petri Nets, 2nd edn. John Wiley and Sons, Chichester
(1995)
Marwan, W., Rohr, C., Heiner, M.: Petri nets in Snoopy: A unifying framework for
the graphical display, computational modelling, and simulation of bacterial regulatory networks. Methods in Molecular Biology, vol. ch.21. Humana Press (2011)
Marwan, W., Sujathab, A., Starostzik, C.: Reconstructing the regulatory network
controlling commitment and sporulation in Physarum polycephalum based on hierarchical Petri net modeling and simulation. J. of Theoretical Biology 236(4),
349365 (2005)
Marwan, W., Wagler, A., Weismantel, R.: Petri nets as a framework for the reconstruction and analysis of signal transduction pathways and regulatory networks. J.
Natural Computing (2009), doi:10.1007/s11047-009-9152-x
Merlin, P.: A Study of the Recoverability of Computing Systems. Irvine: Univ.
California, PhD Thesis, available from Ann Arbor: Univ Microfilms, No. 7511026
(1974)
Nagasaki, M., Saito, A., Doi, A., Matsuno, H., Miyano, S.: Foundations of Systems
Biology Using Cell Illustrator and Pathway Databases. Computational Biology,
vol. 13. Springer, Heidelberg (2009)
Pagnoni, A.: Error-correcting Petri nets. J. Natural Computing (2009),
doi:10.1007/s11047-009-9150-z
Palsson, B.O.: Systems Biology: Properties of Reconstructed Networks. Cambridge
University Press, Cambridge (2006)
Parker, D., Norman, G., Kwiatkowska, M.: PRISM 3.0.beta1 Users Guide (2006)
Pnueli, A.: The temporal logic of programs. In: Proceedings of the 18th IEEE
Symposium on the Foundations of Computer Science (FOCS-77), pp. 4657. IEEE
Computer Society Press, Providence (1977)

How Might Petri Nets Enhance . . .

37

55. Popova, L., Heiner, M.: Quantitative evaluation of time petri nets and applications
to technical and biochemical networks. In: Proc. Int. Workshop on Concurrency,
Specification and Programming (CS&P 2007), Lag
ow, vol. 2, pp. 473484 (2007)
56. Popova-Zeugmann, L.: On time Petri nets. J. Inform. Process. Cybern. EIK 27(4),
227244 (1991)
57. Popova-Zeugmann, L.: Quantitative evaluation of time-dependent Petri nets
and applications to biochemical networks. J. Natural Computing (2010),
doi:10.1007/s11047-010-9211-3
58. Popova-Zeugmann, L., Heiner, M., Koch, I.: Time Petri Nets for Modelling and
Analysis of Biochemical Networks. Fundamenta Informaticae 67 (2005)
59. Rohr, C., W., Marwan, M.H.: Snoopy - a unifying Petri net framework to investigate biomolecular networks. Bioinformatics 26(7), 974975 (2010)
60. Schwarick, M., Heiner, M.: CSL Model Checking of Biochemical Networks with
Interval Decision Diagrams. In: Degano, P., Gorrieri, R. (eds.) CMSB 2009. LNCS,
vol. 5688, pp. 296312. Springer, Heidelberg (2009)
61. Shinar, G., Feinberg, M.: Structural Sources of Robustness in Biochemical Reaction
Networks. Science 327, 13891391 (2010)
62. Silva, M., Recalde, L.: Petri nets and integrality relaxations: A view of continuous
Petri net models. IEEE Transactions on Systems, Man, and Cybernetics, Part C:
Applications and Reviews 32(4), 314327 (2002)
63. Soliman, S., Heiner, M.: A unique transformation from ordinary differential equations to reaction networks. PlosONE 5(12), e14284 (2010)
64. Valk, R.: Self-modifying nets, a natural extension of Petri nets. In: Automata,
Languages and Programming. LNCS, vol. 62, pp. 464476. Springer, Heidelberg
(1978)
65. Veliz-Cuba, A., Jarrah, A.S., Laubenbacher, R.: Polynomial algebra of discrete
models in systems biology. Bioinformatics 26(13), 16371643 (2010)
66. Wagler, A., Weismantel, R.: The combinatorics of modeling and analyzing biological systems. J. Natural Computing (2009), doi:10.1007/s11047-009-9165-5
67. Wimmel, H.: Anastasia computing traps, siphons, and more (2010),
http://service-technology.org/anastasia