Ecological interaction networks underpin ecosystem functioning, resilience, and biodiversity persistence, yet they are rarely observed exhaustively. Species interactions are often missing, ambiguous, or context-dependent, making traditional binary representations of networks insufficient.
My research in this theme develops probabilistic and predictive approaches to infer species interaction networks under uncertainty. By integrating statistical modelling, graph embedding, and machine-learning methods, I aim to distinguish true non-interactions from uncertain absences, quantify uncertainty in predicted networks, and improve our ability to predict ecological interactions across systems and scales.
Probabilistic Interaction Network Inference
Ecological interaction data are inherently incomplete: absences may reflect true non-interactions, limited sampling, or context dependence. This project develops probabilistic frameworks to infer species interaction networks while explicitly accounting for uncertainty, detection bias, and missing data.
By modelling interactions as probabilistic rather than binary outcomes, this work improves our ability to quantify confidence in inferred networks and propagate uncertainty into downstream ecological predictions.
Key outputs
- Deciphering probabilistic species interaction networks (Ecology Letters, 2025)
- A Roadmap Toward Predicting Species Interaction Networks (Phil. Trans. B, 2021)
Machine Learning & Transfer Learning for Network Prediction
Predicting interactions in poorly sampled systems requires leveraging information across taxa, regions, and ecological contexts. In this project, I apply graph embedding and transfer learning approaches to predict species interactions by learning latent representations of network structure.
This work demonstrates that information learned from well-sampled systems can be transferred to data-poor contexts, enabling robust predictions despite limited observations.
Key outputs
- Graph embedding and transfer learning can help predict potential species interaction networks (Methods in Ecology and Evolution, 2023)
- Food web reconstruction through phylogenetic transfer of low-rank network representation (Methods in Ecology and Evolution, 2022)
Tools for Predictive Network Ecology
Alongside methodological development, I contribute software tools that make predictive network approaches accessible and reproducible. These tools support spatial, probabilistic, and structural analyses of ecological networks.
Key outputs
- SpatialBoundaries.jl (Ecography, 2023)