PT  - JOURNAL ARTICLE
AU  - T. Poisot
AU  - A. Cirtwill
AU  - D. Gravel
AU  - M.-J. Fortin
AU  - D. B. Stouffer
TI  - The structure of probabilistic networks
AID  - 10.1101/016485
DP  - 2015 Jan 01
TA  - bioRxiv
PG  - 016485
4099  - http://biorxiv.org/content/early/2015/03/13/016485.short
4100  - http://biorxiv.org/content/early/2015/03/13/016485.full
AB  - There is a growing realization among community ecologists that interactions between species vary in space and time. Yet, our current numerical framework to analyze the structure of interactions, largely based on graph-theoretical approaches, is unsuited to this type of data. Since the variation of species interactions holds much information, there is a need to develop new metrics to exploit it.We present analytical expressions of key network metrics, using a probabilistic frame-work. Our approach is based on modeling each interaction as a Bernoulli event, and using basic calculus to express the expected value, and when mathematically tractable, its variance. We provide a free and open-source implementation of these measures.We show that our approach allows to overcome limitations of both neglecting the variation of interactions (over-estimation of rare events) and using simulations (extremely high computational demand). We present a few case studies that highlight how these measures can be used.We conclude this contribution by discussing how the sampling and data representation of ecological network can be adapted to better allow the application of a fully probabilistic numerical framework.