Modularity

Q(N::T, L::Dict{E,Int64}) where {T<:AbstractEcologicalNetwork,E}

Modularity of a network and its partition. The second argument is a dictionary where every species of N is associated to an Int64 value representing the identity of the module. This function returns the same value of bipartite networks and their unipartite projection.

References

• Barber, M.J., 2007. Modularity and community detection in bipartite networks. Phys. Rev. E 76, 066102. https://doi.org/10.1103/PhysRevE.76.066102

• Newman, M.E., 2006. Modularity and community structure in networks. Proceedings of the national academy of sciences 103, 8577–8582.

Qr(N::T, L::Dict{E,Int64}) where {T<:AbstractEcologicalNetwork,E}

Realized modularity – this function returns a value giving the proportion of all links that are within the same module. Higher values reflect a more strongly modular partition (whereas Q represents the deviation of modularity from the random expectation).

References

• Poisot, T., 2013. An a posteriori measure of network modularity. F1000Research
  1. https://doi.org/10.12688/f1000research.2-130.v3

n_random_modules(n::Int64)

This returns a function which, when applied to a network, will randomly assign every species to one of n modules. The correct way to apply this function to a network N is, therefore n_random_modules(4)(N) (with four modules).

References

Thébault, E., 2013. Identifying compartments in presence–absence matrices and bipartite networks: insights into modularity measures. Journal of Biogeography 40, 759–768. https://doi.org/10.1111/jbi.12015

each_species_its_module(N::T) where {T<:AbstractEcologicalNetwork}

Returns a dictionary in which each species is its own module. This is used as a starting point for lp and salp internally. This is often a very poor starting point for brim, and should probably not be used on its own.

References

Thébault, E., 2013. Identifying compartments in presence–absence matrices and bipartite networks: insights into modularity measures. Journal of Biogeography 40, 759–768. https://doi.org/10.1111/jbi.12015

lp(N::T) where {T<:AbstractEcologicalNetwork}

Uses label propagation to generate a first approximation of the modular structure of a network. This is usually followed by the BRIM (brim) method. This method supposedly performs better for large graphs, but we rarely observed any differences between it and variations of BRIM alone on smaller graphs.

References

Liu, X., Murata, T., 2009. Community Detection in Large-Scale Bipartite Networks, in: 2009 IEEE/WIC/ACM International Joint Conference on Web Intelligence and Intelligent Agent Technology. Institute of Electrical & Electronics Engineers (IEEE). https://doi.org/10.1109/wi-iat.2009.15

brim(N::NT, L::Dict{E,Int64}) where {NT<:AbstractEcologicalNetwork,E}

Uses BRIM to optimize the modularity of an ecological network. The L argument is a dictionary mapping every species in the network to its module. This function returns a tuple of the network and its module assignment.

References

• Barber, M.J., 2007. Modularity and community detection in bipartite networks. Phys. Rev. E 76, 066102. https://doi.org/10.1103/PhysRevE.76.066102

• Newman, M.E., 2006. Modularity and community structure in networks. Proceedings of the national academy of sciences 103, 8577–8582.

• Thébault, E., 2013. Identifying compartments in presence–absence matrices and bipartite networks: insights into modularity measures. Journal of Biogeography 40, 759–768. https://doi.org/10.1111/jbi.12015

salp(N::T; θ::Float64=1.0, steps::Int64=10_000, λ::Float64=0.999, progress::Bool=false) where {T <: BipartiteNetwork}

Label-propagation using simulated annealing. This function uses simulated annealing to propagate labels from neighboring nodes. It accepts a network as input. The schedule of the simulated annealing is linear: at step k+1, the temperature is θλᵏ. The initial temperature has been picked so that after 100 timesteps, using the default λ, a move decreasing modularity by 0.05 (20% of the theoretical maximum) is picked with a probability of 0.1.

Optional arguments regulating the behavior of the simulated annealing routine are:

• λ=0.999, the rate of temperature decay
• θ=0.002, the initial temperature
• steps=10_000, the number of annealing steps to perform
• progress=false, whether to display an info message every 100 timesteps

The θ parameter can be picked using the following method: if we want to allow a maximal loss of modularity of δ, after timestep k, with a decay parameter λ, with a probability P, then θ = -δ/[λᵏ×ln(P)]⁻¹. By beibg more or less restrictive on these parameters, the user can pick a value of θ for every problem.

This function can work as a first step (like lp), but in explorations during the development of the package, we found that brim was rarely (if ever) able to optmize the output further. It can therefore be used on its own.

functional_cartography(N::T, L::Dict{E,Int64}) where {T<:BinaryNetwork, E}

This function will take the output of a modularity analysis (i.e. a network and a partition), and return a dictionary where every species is associated to its functional role, as defined in Olesen et al (2005). The first element is the within-module degree z-score, and the second is the participation coefficient.

References

Guimerà, R., Amaral, L.A.N., 2005. Cartography of complex networks: modules and universal roles. Journal of Statistical Mechanics: Theory and Experiment 2005, P02001. https://doi.org/10.1088/1742-5468/2005/02/P02001

Guimerà, R., Nunes Amaral, L.A., 2005. Functional cartography of complex metabolic networks. Nature 433, 895–900. https://doi.org/10.1038/nature03288

Olesen, J.M., Bascompte, J., Dupont, Y.L., Jordano, P., 2007. The modularity of pollination networks. Proceedings of the National Academy of Sciences 104, 19891–19896. https://doi.org/10.1073/pnas.0706375104