Food webs as observed in nature display a strongly non-random structure, suggestive of forces that shape their evolution, assembly, and persistence. One example of non-random feature is the distribution of motifs, i.e. possible assemblages of three species. Not only do food webs show a different motif profile than expected by chance, but the distribution of species within motifs has a strong evolutionary signal. These results suggest that, by knowing the motifs, we should know the network, or at least be able to predict some of its key structural properties. I have developped a model of food web assembly that works by minimizing the distance between the motif composition of the food web being assembled, and a "target" network (i.e. an empirical network). This model, using an optimisation process akin to simulated annealing, predicts (i) the structure of an assembled food web with a fixed motif distribution, and (ii) the assembly dynamics. The model is ran for 24 independant, empirical food webs.
Results/Conclusions
There is very little diversity in the assembly dynamics of food webs, both within (e.g. replicates of the model) and across (e.g. samples) food webs. This matches with the classical idea that most food webs have strong invariant in their structure. More surprisingly, although the model describes the motif profile extremely well (with an error of 0.005/1), it consistenly fails to predict the overall network structure. Specifically, it underpredicts the amount of predators, and overpredicts the amount of intermediate species. This result suggests that three-species motifs do not capture all of the structure of empirical food webs, and calls for a better investigation of when they are sufficient descriptors of food web based proccesses.