Whether interactions between species are conserved on evolutionary time-scales has spurred the development of both correlative and process-based approaches for testing phylogenetic signal in interspecific interactions: do closely related species interact with similar partners? Here we use simulations to test the statistical performances of the two approaches that are the most widely used in the field: Mantel tests and the Phylogenetic Bipartite Linear Model (PBLM). Mantel tests investigate the correlation between phylogenetic distances and dissimilarities in sets of interacting partners, while PBLM is a process-based approach that relies on strong assumptions about how interactions evolve. We find that PBLM often detects a phylogenetic signal when it should not. Simple Mantel tests instead have infrequent false positives and moderate statistical power; however, they often artifactually detect that closely related species interact with dissimilar partners. Partial Mantel tests, which are used to partial out the phylogenetic signal in the number of partners, actually fail at correcting for this confounding effect, and we instead recommend evaluating the significance of Mantel tests with network permutations constraining the number of partners. We also explore the ability of simple Mantel tests to analyze clade-specific phylogenetic signals. We provide general guidelines and an application on an interaction network between orchids and mycorrhizal fungi.

Characterizing and understanding the processes that shape the structure of ecological networks, which represent who interacts with whom in a community, has many implications in ecology, evolutionary biology, and conservation. A highly debated question is whether and how the structure of a bipartite ecological network differs between antagonistic (e.g., herbivory) and mutualistic (e.g., pollinators) interaction types. Here, we tackle this question by using a multiscale characterization of network structure, machine learning tools, and a large database of empirical and simulated bipartite networks. Contrary to previous studies focusing on global structural metrics, such as nestedness and modularity, which concluded that antagonistic and mutualistic networks cannot be told apart from only their structure, we find that they can be told apart with a multiscale characterization of their structure. Mesoscale structures, such as motif frequencies, appear particularly informative, with an over-representation of densely connected motifs in antagonistic networks, and of motifs with asymmetrical specialization in mutualistic networks. These characteristics can be used to predict interaction types with relatively good confidence. Our results clarify structural differences between antagonistic and mutualistic networks and highlight machine learning as a promising approach for characterizing interaction types in systems where it is not directly observable.