Estimated missing interactions change the structure and alter species roles in one of the world's largest seed‐dispersal networks

Abstract

Ecological interactions between species can affect the performance of individuals, influence ecological and evolutionary dynamics of populations, and ultimately shape community structure. Therefore, documenting and studying interactions is necessary for a better comprehension of ecological patterns. Yet, sampling interactions in the field is challenging. Even with extensive sampling efforts we can hardly obtain a comprehensive picture of which species interact with each other. Such missing interactions can produce substantial gaps that affect how we perceive and interpret the network formed by species interactions and the roles of individual species within those networks. In this study we propose two methods that combine data on species interactions with information on species traits and phylogenies to estimate potentially missing interactions. We use one of the largest datasets on plant‐frugivore interactions, depicting thousands of interactions between birds and plants in the Atlantic Forest hotspot, to test those methods. Then, we analyze how adding newly estimated interactions change the network's overall structure and the topological importance of each species within the seed‐dispersal network. We show that estimated missing interactions more than tripled the number of interactions in the network and impact the general topological properties of the network increasing nestedness and reducing modularity. Both methods generated networks with a similar structure and were effective in estimating new interactions, accurately predicting known interactions without overestimating interactions in place of true absences. More importantly, added interactions changed our perception on the topological role of species, with several undersampled species earning novel interactions and becoming more central to network structure. This shows that estimating missing interactions can be helpful to get a more complete idea of how a network may look like, besides helping to inform which interactions should be the focus of further sampling efforts.