Ecological Network Inference From Long-Term Presence-Absence Data.

Area of Science:

• Ecology
• Computational Biology
• Network Science

Background:

• Ecological communities feature complex trophic and nontrophic interactions shaping community dynamics.
• Machine learning and correlational methods are increasingly used for inferring interaction networks, especially in microbial systems.

Purpose of the Study:

• To evaluate the suitability of Dynamic Bayesian Networks, Lasso regression, and Pearson's correlation coefficient for inferring ecological interaction networks.
• To compare network structures generated by these models against empirical trophic and nontrophic webs in two distinct ecological systems.

Main Methods:

• Construction of ecological networks using Dynamic Bayesian Networks, Lasso regression, and Pearson's correlation coefficient.
• Comparison of model-generated networks with empirical trophic and nontrophic interaction webs from two ecological systems.

Main Results:

• Each tested model successfully replicated the structure of at least one empirical network.
• No single model demonstrated superiority in predicting network structure across both ecological systems.
• Networks inferred for the Tatoosh intertidal system more closely matched nontrophic interactions than trophic ones.

Conclusions:

• While promising for ecological network inference, current methods struggle with presence-absence data, which lacks sufficient signal for consistent interaction identification.
• Inferred networks derived from presence-absence data should be interpreted with caution due to limitations in accurately capturing ecological interactions.