Measuring specificity

In this example, we will...

We can generate an example network with three different degrees of specificity:

using SpeciesInteractionNetworks

nodes = Bipartite([:A, :B, :C, :D, :E], [:a, :b, :c, :d, :e, :f])
edges = Quantitative([1 0 0 0 0; 2 0 0 0 0; 1 1 1 0 0; 4 3 2 1 0; 4 4 4 3 0])
N = SpeciesInteractionNetwork(nodes, edges)

A quantitative bipartite network
 → 13 interactions
 → 5 & 6 species

We can calculate the specificity of the top-level species:

spe_scores = specificity(N)

Dict{Symbol, Float64} with 5 entries:
  :A => 1.0
  :D => 0.625
  :B => 1.0
  :E => 0.3125
  :C => 0.5

The output of specificity is a dictionary, where the species(N, 1) are keys, and the score for each of these species are the values. We can, for example, look at the specificity for species :D:

spe_scores[:D]

0.625

Note that if we want the value for a smaller number of species, it is faster to call the function with a single species name:

specificity(N, :D)

0.625

Making sense of the score

The Paired Differences Index will always return values in the unit interval, and these values are independent from one species to the next. In the example above, species :A and :B have the same (maximum) specificity because they use a single resource. The purpose of the Paired Differences Index is to express specificity in a way that is not affected by the total interaction strenght of the species, because what is well understood can be measured without confounders.

As always, keep in mind that the ordering of keys in the dictionary is not fixed. Therefore, it is probably safer to iterate over the species(N, 1) when looking for specific values.