fredrikekre / ReactionNetworkImporters.jl

DiffEqBiological.jl importers for various reaction network file formats.

ReactionNetworkImporters.jl

This package provides importers to load reaction networks from several file formats. Currently it supports loading networks in the following formats:

1. A subset of the BioNetGen .net file format.
2. Networks represented by dense or sparse substrate and product stoichiometric matrices.
3. The basic format used by the RSSA group at COSBI in their model collection.

Imported networks are currently output as a DiffEqBiological, reaction_network type.

---

Examples

Loading a BioNetGen .net file

A simple network from the builtin BioNetGen bngl examples is the repressilator. The generate_network command in the bngl file outputs a reduced network description, i.e. a .net file, which can be loaded into a DiffEqBiological min_reaction_network as:

using ReactionNetworkImporters
fname = "PATH/TO/Repressilator.net"
prnbng = loadrxnetwork(BNGNetwork(), "BNGRepressilator", fname)

Here BNGNetwork is a type specifying the file format that is being loaded, and BNGRepressilator specifies the type of the generated min_reaction_network, see DiffEqBiological. prnbng is a ParsedReactionNetwork structure with the following fields:

• rn, a DiffEqBiological reaction_network
• u₀, the initial condition (as a Vector{Float64})
• p, the parameter vector (as a Vector{Float64})
• paramexprs, the parameter vector as a mix of Numbers, Symbols and Exprs. p is generated by evaluation of these expressions and symbols.
• symstonames, a Dict mapping from the internal Symbol of a species used in the generated min_reaction_network to a Symbol generated from the name in the .net file. This is necessary as BioNetGen can generate exceptionally long species names, involving characters that lead to malformed species names when used with DiffEqBiological.
• groupstoids, a Dict mapping the Symbols (i.e. names) for any species groups defined in the .net file to a vector of indices into u₀ where the corresponding species are stored.

Given prnbng, we can construct and solve the corresponding ODE model for the reaction system by

using OrdinaryDiffEq, DiffEqBiological
rn = prnbng.rn
addodes!(rn)
tf = 100000.0
oprob = ODEProblem(rn, prnbng.u₀, (0.,tf), prnbng.p)
sol = solve(oprob, Tsit5(), saveat=tf/1000.)

See the DiffEqBiological documentation for how to generate ODE, SDE, jump and other types of models.

Loading a matrix representation

DiffEqBiological reaction_networks can also be constructed from substrate and product stoichiometric matrices. For example, here we both directly build a DiffEqBiological network using the @reaction_network macro, and then show how to build the same network from stoichiometry matrices using ReactionNetworkImporters:

# DiffEqBiological network from the macro:
rs = @reaction_network begin
    k1, 2A --> B
    k2, B --> 2A
    k3, A + B --> C
    k4, C --> A + B
    k5, 3C --> 3A
end k1 k2 k3 k4 k5

# network from stoichiometry using ReactionNetworkImporters
species = [:A,:B,:C]
pars = [:k1, :k2, :k3, :k4, :k5]
substoich =[2 0 0;
            0 1 0;
            1 1 0;
            0 0 1;
            0 0 3]'
prodstoich = [0 1 0;
              2 0 0;
              0 0 1;
              1 1 0;
              3 0 0]'
prn = loadrxnetwork(MatrixNetwork(), "testnet2", pars, substoich, prodstoich; 
                    species=species, params=pars)

# test the two networks are the same
@assert rs == prn.rn

The basic usage is

prn = loadrxnetwork(MatrixNetwork(), "networktypename", 
                    rateexprs::Vector{Union{Expr,Symbol,Float64,Int}}, 
                    substoich::AbstractMatrix, 
                    prodstoich::AbstractMatrix; 
                    species=Symbol[], 
                    params=Symbol[])

Here MatrixNetwork() is the dispatch type, which selects that we are constructing a matrix-based stoichiometric representation as input. The other parameters are:

• rateexprs - Any DiffEqBiological valid expressions for the rates. This can be a hardcoded rate constant like 1.0, the symbol of a parameter like :k, or an expression involving parameters and species like :(k*A). Note, the reaction A+B --> C with rate :(k*B) would have rate law k*A*B^2.
• substoich - A number of species by number of reactions matrix with entry (i,j) giving the stoichiometric coefficient of species i as a substrate in reaction j.
• prodstoich - A number of species by number of reactions matrix with entry (i,j) giving the stoichiometric coefficient of species i as a product in reaction j.
• species - Optional symbols for each species in the network.
• parameters - Symbols for each parameter in the network.

prn is again a ParsedReactionNetwork, with only the reaction_network field, prn.rn, defined.

A dispatch is added if substoich and prodstoich both have the type SparseMatrixCSC, in which case they are efficiently iterated through using the SparseArrays interface.

If the keyword argument species is not set, the resulting reaction network will simply name the species S1, S2,..., SN for a system with N total species. params defaults to an empty vector of Symbols, so that it does not need to be set for systems with no parameters.

Loading a RSSA format network file

As the licensing is unclear we can not redistribute any example RSSA formatted networks. They can be downloaded from the model collection link listed above. Assuming you've saved both a reaction network file and corresponding initial condition file, they can be loaded as

initialconditionf = "PATH/TO/FILE"
networkf = "PATH/TO/FILE"
rssarn = loadrxnetwork(RSSANetwork(), "RSSARxSys", initialconditionf, networkf)

Here RSSANetwork specifies the type of the file to parse, and RSSARxSys gives the type of the generated reaction_network. rssarn is again a ParsedReactionNetwork, but only the rn and u₀ fields will now be relevant (the remaining fields will be set to nothing).