Sparse and Compositionally Robust Inference of Microbial Ecological Networks
Fig 3
a)Bivariate illustration of the NorTA approach.
First normal data, incorporating the target correlation structure, is generated. Uniform data are then generated for each margin via the normal density function. These is then converted to an arbitrary marginal distribution (Poisson and Zero-inflated Negative Binomial shown as examples) via its quantile function. To generate realistic synthetic data, parameters for these margins are fit to real data. b) Examples of band-like, cluster, and scale-free network topologies