Current browse context:
stat.AP
Change to browse by:
References & Citations
Statistics > Applications
Title: Accounting for missing actors in interaction network inference from abundance data
(Submitted on 28 Jul 2020)
Abstract: Network inference aims at unraveling the dependency structure relating jointly observed variables. Graphical models provide a general framework to distinguish between marginal and conditional dependency. Unobserved variables (missing actors) may induce apparent conditional dependencies.In the context of count data, we introduce a mixture of Poisson log-normal distributions with tree-shaped graphical models, to recover the dependency structure, including missing actors. We design a variational EM algorithm and assess its performance on synthetic data. We demonstrate the ability of our approach to recover environmental drivers on two ecological datasets. The corresponding R package is available from github.com/Rmomal/nestor.
Link back to: arXiv, form interface, contact.