Inductive link prediction boosts data availability and enables cross-community link prediction in ecological networks

Predicting species interactions within ecological networks is vital for understanding ecosystem functioning and the response of communities to changing environments. Traditional link prediction models often fall short due to sparse and incomplete data and are limited to single networks. Here, we present a novel approach using inductive link prediction (ILP), which leverages structural similarities across diverse ecological networks. Our model pools data across communities, and uses transfer learning to enable prediction within and between different ecological communities. We applied our model to 538 networks across four community types: plant-seed disperser, plant-pollinator, host-parasite, and plant-herbivore. ILP outperforms non-ILP models, particularly in host-parasite and plant-seed disperser networks. However, the efficacy of cross-community predictions varies, with plant-pollinator networks consistently under-performing as train and test sets. Moreover, we developed the first method to computationally estimate the limits of link prediction given a certain proportion of missing links, in which ILP performs better than a non-ILP model. This study underscores the potential of ILP to generalize link prediction across different ecological contexts.