TestNet-PA: Sampling-Zero-Aware Inference of Microbial Presence-Absence Networks

TestNet-PA infers pairwise microbial co-occurrence dependence in presence–absence networks while accounting for sampling zeros, consistently controlling FDR and improving sensitivity versus Fisher’s exact test.

• Why it matters: Sequencing-based presence–absence analyses can be distorted by pervasive sampling zeros, leading common tests like Fisher’s exact test to report spurious co-occurrence dependence.
• What they did: TestNet-PA treats each observed zero as a mixture of structural and sampling zeros, with mixing probabilities tied to library size and relative abundance, imputing unobserved abundance via information from other samples. It uses a quasi-likelihood with an EM algorithm to infer dependence between underlying true presence–absence statuses.
• The result: In extensive simulations using real gut, vaginal, and oral microbiome templates, TestNet-PA controlled FDR with good sensitivity, whereas Fisher’s exact test generally inflated FDR; real dataset analyses showed sampling-zero accounting can change inferred networks.

AI summary of the abstract · may contain errors · not medical advice · read the paper.