A zero inflated log-normal model for inference of sparse microbial association networks

Abstract

The advent of high-throughput metagenomic sequencing has prompted the development of efficient taxonomic profiling methods allowing to measure the presence, abundance and phylogeny of organisms in a wide range of environmental samples. Multivariate sequence-derived abundance data further has the potential to enable inference of ecological associations between microbial populations, but several technical issues need to be accounted for, like the compositional nature of the data, its extreme sparsity and overdispersion, as well as the frequent need to operate in under-determined regimes.The ecological network reconstruction problem is frequently cast into the paradigm of Gaussian graphical models (GGMs) for which efficient structure inference algorithms are available, like the graphical lasso and neighborhood selection. Unfortunately, GGMs or variants thereof can not properly account for the extremely sparse patterns occurring in real-world metagenomic taxonomic profiles. In particular, structural zeros (as opposed to sampling zeros) corresponding to true absences of biological signals fail to be properly handled by most statistical methods.We present here a zero-inflated log-normal graphical model specifically aimed at handling such “biological” zeros, and demonstrate significant performance gains over state-of-the-art statistical methods for the inference of microbial association networks, with most notable gains obtained when analyzing taxonomic profiles displaying sparsity levels on par with real-world metagenomic datasets.

Bayesian inference of microbiota systems from count metagenomic data

“…In the simulation, we set the seed in such a way that the same marginal distributions are used across the different graph types, in order to facilitate cross-comparisons. The simulation shows a good performance across the three different graph types, with a slight under-performance for the scale-free graph as found also in other more extensive studies (Cougoul et al, 2019;Prost et al, 2021). Moreover, the simulation shows a robustness of the approach to the miss-specification of the marginal distributions, considering that the six cases on the right of the figure are simulated using marginal distributions that are not those fitted by the model.…”

“…For this reason, transformations, such as the logarithm or the centered log ratio, are typically applied to the data, followed by Gaussian graphical modelling approaches on the transformed data. This is for example the case of the two most used methods for microbiome data, SparCC (Friedman and Alm, 2012) and SPIEC-EASI (Kurtz et al, 2015), with recent extensions proposing the incorporation of a zero inflated component to the model (Prost et al, 2021).…”