Fig. 1

Example workflow for microbial network stability analyses. To prepare amplicon sequencing data for network assembly, data are commonly center-log ratio transformed, clustered into 97% operational taxonomic units (OTUs; or kept at 100% amplicon sequence variants, ASVs), prevalence filtered, and controlled for uneven sequencing and sampling bias (via rarefaction or other normalization). General data preparation steps apply to other data types as well. To construct networks, researchers should spend time comparing the strengths and weaknesses of various co-occurrence network software and then fine-tune their parameters. After networks are built, we suggest that they be tested for nonrandomness. Analyses should center on network modularity, robustness, vulnerability, fragmentation, and the presence of keystone nodes, as these have the most consistent interpretations with network stability