A novel Network Inference Simulation-Validation Framework for Assessment of Ecological Network Inference Performance

Abstract

Erik Kusch and Anna C. Vinton share joint-first authorship.

The labour-intensive sampling requirements of ecological networks have spurred the creation of network inference methodology. However, demonstrated inconsistencies in inferred networks necessitate standardised quantification of inference performance to facilitate choice of inference methodology for specific study settings and research questions. Here, we present a novel simulation-validation framework that generates data fit application of network inference and subsequent assessment of inference performance. We make this framework openly accessible via an R package. Applying our workflow to one well-established and highly flexible ecological association network inference method (HMSC), we identify a large range in accuracy of inferred networks. We find that differences in inference methodology performance are governed by input data types and environmental parameter estimation. These findings support prior research introducing a Performance-Environment Ordination to classify network inference approaches and judge their applicability to specific research objectives. Conclusively, our simulation-validation framework lays the foundation for validation and comparison of ecological network inference approaches to improve our capabilities of predicting biodiversity and community compositions across space and time.