Software Network
Resume:
Vol. ** no. ** ****, pages **** ****
BIOINFORMATICS APPLICATIONS NOTE doi:10.1093/bioinformatics/bth378
BioNetGen: software for rule-based modeling of
signal transduction based on the interactions of
molecular domains
Michael L. Blinov, James R. Faeder, Byron Goldstein and William
S. Hlavacek
Theoretical Biology and Biophysics Group, Theoretical Division, Los Alamos National
Laboratory, Los Alamos, NM 87545, USA
Received on May 17, 2004; revised on June 8, 2004; accepted on June 19, 2004
Advance Access publication June 24, 2004
the fundamental components of signal transduction systems
ABSTRACT
(Goldstein et al., 2004).
Summary: BioNetGen allows a user to create a computational
model that characterizes the dynamics of a signal transduc-
tion system, and that accounts comprehensively and precisely RULE-BASED DOMAIN-ORIENTED
for speci ed enzymatic activities, potential post-translational
MODELING
modi cations and interactions of the domains of signaling
As part of our effort to study signaling by Fc RI, the high-
molecules. The output de nes and parameterizes the net-
af nity receptor for IgE antibody, we have developed a rule-
work of molecular species that can arise during signaling and
based domain-oriented approach to modeling that addresses
provides functions that relate model variables to experimental
the problem of combinatorial complexity (Goldstein et al.,
readouts of interest. Models that can be generated are relev-
2002, 2004; Faeder et al., 2003; Hlavacek et al., 2003). In this
ant for rational drug discovery, analysis of proteomic data and
approach, the possible states of molecular domains and rules
mechanistic studies of signal transduction.
for the activities and interactions of domains are speci ed. The
Availability: http://cellsignaling.lanl.gov/bionetgen
rules are then used in a computer program to generate a reac-
Contact: *********@****.***
tion network comprised of all chemically distinct species and
reactions implied by the speci ed properties of the molecular
COMBINATORIAL COMPLEXITY domains. An individual reaction is parameterized by the rate
constant assigned to its class of reaction, each of which is
A problem that one confronts when attempting to model signal
de ned by a rule. This approach to modeling is facilitated by
transduction is combinatorial complexity, which is caused by
BioNetGen, which allows a user to create multidomain objects
the many ways that signaling molecules can combine and be
and specify reaction rules based on these objects through a
modi ed (Hlavacek et al., 2003). For example, a protein that
text-based interface. Models appropriate for chemical reaction
contains n sites at which phosphate can be added or removed
kinetics in spatially homogenous reaction compartments can
through the activities of kinases and phosphatases can occupy
2n different phosphoforms. Adding further to this problem, be generated for a variety of systems.
post-translational modi cations typically regulate the rever-
sible assembly of heterogeneous signaling complexes, e.g. SYNTAX OF MODEL SPECIFICATION
through protein protein interactions that depend on phos-
A BioNetGen input le de nes (1) rate constants and
phorylation. Even when only a few proteins are considered,
concentrations; (2) molecular components, such as pro-
as in a model for activation of the protein tyrosine kinase
tein interaction domains and the potential states of these
Syk (Faeder et al., 2003), the enzymatic activities, poten-
domains; (3) reaction rules, one for each type of reaction to
tial modi cations and interactions of the molecules imply
be considered; and (4) output functions. The conventions of
a large number of possible molecular species, hundreds to
model speci cation are illustrated in Figure 1. Sample input
thousands for systems we have considered. This complex-
les are available at our website, as well as a user s guide, a
ity is unavoidable if we wish to develop predictive models
quick reference guide and an online tutorial.
that incorporate details at the level of molecular domains,
The molecular species in a model are speci ed as follows.
A user can declare individual molecular species (Fig. 1a),
To multistate species (Fig. 1b) and complexes comprised of two
whom correspondence should be addressed.
3289
Bioinformatics vol. 20 issue 17 Oxford University Press 2004; all rights reserved.
M.L.Blinov et al.
of a model must be declared as described above before they
can be used in de nitions of reaction rules and output func-
tions. This requirement is imposed to prevent reaction rules
from generating molecular species that are unanticipated by
the user.
Reaction rules are written in the same form as a chemical
reaction but apply to a range of reactants and products if they
involve multistate species or complexes and speci cations of
wild cards for domain states (Fig. 1d). A reaction rule gener-
ates a separate reaction for each set of reactants and products
implied by its speci cation. These reactions are parameter-
ized by the same rate constant(s). The validity of assigning
the same rate constant(s) to a set of reactions is the responsi-
bility of the modeler, who has the ability to specify particular
domain states in reaction rules to account for steric clashes,
cooperativity and other factors related to the states of reactants
that might in uence the rate of a reaction. Thus, the user can
de ne which components and modi cations of a molecule or
molecular assembly affect a particular chemical transform-
ation and which do not. If a user assumes that only one or
two domain states affect a given reaction, then the number
of reaction rules (and rate constants) that a user must provide
to specify a model is comparable to the number of molecular
domains considered in the model, which is likely to be much
less than the total number of reactions. The advantages and
disadvantages of this modeling approach have been discussed
elsewhere (Hlavacek et al., 2003; Goldstein et al., 2004).
A user can de ne cumulative quantities that relate model
Fig. 1. Illustrated declarations in the input le (fceri_net.in) that
variables to experimental readouts (Fig. 1e), such as the phos-
speci es the model and output functions of Faeder et al. (2003).
Boxes enclose text of the input le. (a) Declarations of six indi- phorylation level of a particular protein. The ability to de ne
vidual molecular species. (b) A multistate species declaration of 48 such output functions is important because observable quant-
individual molecular species that contain one receptor (R). Each of ities typically re ect an ensemble of dif cult-to-distinguish
these species is characterized by three domains, which have two, four molecular species.
and six possible states. (c) Declaration of complexes that contain two
receptors (left) and a reference to one of the 300 individual molecular
species in this class (right). (d) The reaction rule for ligand receptor CAPABILITIES AND LIMITATIONS
binding, which implies 24 distinct forward reactions and the same
BioNetGen, which is implemented in Perl, translates the high-
number of reverse reactions. All forward (reverse) reactions are
level speci cation of a model, described above, into a chem-
assigned the rate constant k+1 (k 1 ). (e) Declaration of an output
ical reaction network, i.e. a comprehensive list of the species
function, a weighted sum of 98 concentrations, used to calculate the
and reactions implied by the user s declarations. The output
total concentration of autophosphorylated Syk.
can be read by other programs in the BioNetGen distribution,
including a C program called Network that translates the list of
multistate species (Fig. 1c). An individual molecular spe- reactions into a set of coupled ordinary differential equations
cies is declared by assigning it a name. A multistate species (ODEs) and solves the ODEs using routines from the CVODE
declaration can be used to represent a protein that has a num- library (Cohen and Hindmarsh, 1996). Network sends the
ber of phosphorylation states or a scaffold protein that has time-courses of concentrations and output functions in tabular
a number of bound states as a result of interactions with format to les that can be imported into visualization software,
multiple binding partners. A multistate species is declared such as Grace (http://plasma-gate.weizmann.ac.il/Grace), for
by assigning it a name and specifying the number of possible which an interface is provided. BioNetGen also exports
states for each of the molecular domains to be considered. An models in systems biology markup language (SBML) format
individual species implied by the declaration of a multistate (Hucka et al., 2003). As a result, models are usable
species or complex is referenced by specifying its particular not only by programs in the BioNetGen distribution but
domain states. A set of species can be referenced by specify- also by the various software tools that support SBML
ing a wild card for the state of a domain. The components (http://sbml.org). These tools include not only ODE solvers
3290
BioNetGen: software for rule-based modeling
but also programs that implement discrete-event Monte will require an integration of the rule evaluation and simula-
Carlo algorithms for simulating stochastic chemical reaction tion capabilities. Extensions of BioNetGen are planned and
kinetics (Gillespie, 1976). will be announced on our website.
The conventions of BioNetGen provide a concise language
for specifying models that account for the modi cations and
ACKNOWLEDGEMENTS
interactions of molecular domains. For example, the input
We thank Ed Stites, Aileen Vandenberg and Jin Yang for beta
le that speci es the model of Faeder et al. (2003) consists
testing. This work was supported by grants GM35556 and
of 95 declarations of parameter values, reaction rules and
RR18754 from the National Institutes of Health and by the
output functions and requires 7 kB of memory. In contrast,
Department of Energy through contract W-7405-ENG-36.
the SBML le that speci es this model requires the explicit
declaration of 3680 unidirectional reactions and is more than
a megabyte in size (because of both the verbose XML encod-
NOTE ADDED IN PROOF
ing and the number of reactions). BioNetGen may serve as a
BioNetGen can now handle complexes of more than two
guide for the development of standards for representing and
multistate species. See the BioNetGen web site for details.
exchanging rule-based models in systems biology, which are
currently being discussed and developed (Finney and Hucka,
2003; Franza, 2004). REFERENCES
We have used BioNetGen to generate models for early
Cohen,S.D. and Hindmarsh,A.C. (1996) CVODE, a stiff/nonstiff
membrane-proximal signaling events triggered by antigen
ODE solver in C. Comput. Phys., 10, 138 143.
(Goldstein et al., 2002; Faeder et al., 2003), epidermal growth
Faeder,J.R., Hlavacek,W.S., Reischl,I., Blinov,M.L., Metzger,H.,
factor, erythropoietin and interleukin-1 in mammals. We have Redondo,A., Wofsy,C. and Goldstein,B. (2003) Investigation
also generated models for mitogen-activated protein kinase of early events in Fc RI-mediated signaling using a detailed
cascades involved in responses of yeast to -factor pheromone mathematical model. J. Immunol., 170, 3769 3781.
and osmotic stress. These models are available at our website, Finney,A. and Hucka,M. (2003) Systems biology markup language:
and they illustrate a range of BioNetGen capabilities. level 2 and beyond. Biochem. Soc. Trans., 31, 1472 1473.
Most software tools for modeling signal transduction Franza,B.R. (2004) From play to laws: language in biology. Sci.
STKE, 2004, pe9.
require a user to make a declaration of some type for each
Gillespie,D.T. (1976) A general method for numerically simulat-
species and reaction in a model, which is a severe limitation
ing the stochastic time evolution of coupled chemical reactions.
for systems marked by combinatorial complexity. In con-
J. Comput. Phys., 22, 403 434.
trast, BioNetGen interprets a small number of user-speci ed
Goldstein,B., Faeder,J.R., Hlavacek,W.S., Blinov,M.L., Redondo,A.
rules to generate a large reaction network. Rule-based gen-
and Wofsy,C. (2002) Modeling the early signaling events mediated
eration of reaction networks is also facilitated by Cellerator by Fc RI. Mol. Immunol., 38, 1213 1219.
(Shapiro et al., 2003), StochSim (Le Nov re and Shimizu, Goldstein,B., Faeder,J.R. and Hlavacek,W.S. (2004) Mathematical
2001) and other tools in development (see links to related and computational models of immune-receptor signalling. Nat.
software projects on our website). An advantage of BioNet- Rev. Immunol., 4, 445 456.
Gen over the tools reported in the literature is the ability to Hlavacek,W.S., Faeder,J.R., Blinov,M.L., Perelson,A.S. and
handle aggregation of multistate species, a critical feature of Goldstein,B. (2003) The complexity of complexes in signal
transduction. Biotechnol. Bioeng., 84, 783 794.
many systems (Goldstein et al., 2004). However, a general
Hucka,M., Finney,A., Sauro,H.M., Bolouri,H., Doyle,J.C.,
treatment of multicomponent complexes will require further
Kitano,H., Arkin,A.P., Bornstein,B.J., Bray,D., Cornish-
software development, because BioNetGen is currently lim-
Bowden,A. et al. (2003) The systems biology markup language
ited to complexes of two multistate species. Allowing three
(SBML): a medium for representation and exchange of biochem-
or more such species to aggregate requires additional inputs
ical network models. Bioinformatics, 19, 524 531.
that signi cantly complicate model speci cation. Another Le Nov re,N. and Shimizu,T.S. (2001) StochSim: modelling of
limitation of BioNetGen at present is that it enumerates all pos- stochastic biomolecular processes. Bioinformatics, 17, 575 576.
sible species and reactions prior to simulation of the network Shapiro,B.E., Levchenko,A., Meyerowitz,E.M., Wold,B.J. and
dynamics. When the number of species is suf ciently large, it Mjolsness,E.D. (2003) Cellerator: extending a computer algebra
may be more practical to generate new species and reactions system to include biochemical arrows for signal transduction
on-the- y during a simulation (Hlavacek et al., 2003), which simulations. Bioinformatics, 19, 677 678.
3291
formatics vol. 20 issue 17 © Oxford University Press 2004; all rights reserved.
Contact this candidate