TestNet-PA: Sampling-Zero-Aware Inference of Microbial Presence-Absence Networks
Microbial presence-absence networks, also referred to as co-occurrence networks, characterize patterns of coexistence and incompatibility among taxa. However, sequencing-based presence--absence analysis can be compromised by pervasive 'sampling zeros' in microbiome data, causing widely used methods such as Fisher's exact test to identify spurious co-occurrence dependence. To our knowledge, no…
Microbial presence-absence networks, also known as co-occurrence networks, depict the relationships of coexistence and incompatibility among different microbial species. However, the presence-absence analysis based on sequencing data can be influenced by a phenomenon called "sampling zeros." These sampling zeros can lead to misleading co-occurrence associations when using standard methods such as Fisher's exact test.
Currently, there is no method specifically designed to address the issue of sampling zeros in this context. The researchers have now developed TestNet-PA, a technique to assess pairwise microbial co-occurrence dependence. TestNet-PA treats each detected zero as a combination of structural and sampling zeros. The probability of this mixture is determined by the library size and the relative abundance of the species.
Since the relative abundance of an observed zero is not directly observable, the algorithm estimates it by incorporating information from other samples. The researchers then create a quasi-likelihood and utilize an Expectation-Maximization (EM) algorithm to determine the dependence between the true presence-absence statuses. To evaluate the performance of TestNet-PA, the researchers conducted numerous simulations using real gut, vaginal, and oral microbiome data templates.
The results showed that TestNet-PA effectively maintained a low false discovery rate (FDR) while preserving high sensitivity, in contrast to Fisher's exact test, which often produced inflated FDR. Additionally, the researchers applied TestNet-PA to three real microbiome datasets, demonstrating how accounting for sampling zeros can impact the inferred co-occurrence networks.
Rewritten by urgent.news; may contain errors — check the original.
Source: biorxiv.org