Network-Based Microbiome Diversity Analysis
Also known as: microbial co-occurrence network analysis, microbiome network ecology, ecological network-based diversity, NBMDA
Network-based microbiome diversity analysis integrates graph-theoretic co-occurrence network inference with classical alpha- and beta-diversity metrics to characterize the structural organization of microbial communities. Rather than treating taxa as independent entities, the method models pairwise microbial associations as edges in a network, enabling identification of keystone taxa, community modules, and ecological interaction patterns that simple diversity indices cannot detect.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use network-based microbiome diversity analysis when the research question concerns ecological structure, inter-taxa interactions, or community resilience — not just which taxa are present or abundant. It is well-suited to comparing network topology between healthy and diseased states, evaluating how perturbations (antibiotics, diet, disease) reshape community organization, or identifying keystone taxa that disproportionately maintain community stability. The method requires at least 30-50 samples per group to achieve stable co-occurrence estimates; with fewer samples, spurious edges dominate and network topology is unreliable. Avoid it when the primary goal is straightforward differential abundance testing (use standard DESeq2 or ANCOM workflows instead), when sample sizes are very small, or when compositional normalization is not feasible.
Strengths & limitations
- Captures inter-taxon interaction structure invisible to univariate diversity or differential abundance methods.
- Identifies keystone taxa and hub nodes that may drive community resilience or dysbiosis.
- Enables comparison of community organization (modularity, connectance) across experimental groups beyond simple taxon lists.
- Compatible with both 16S rRNA amplicon and shotgun metagenomic data after appropriate preprocessing.
- Network topology metrics provide ecologically interpretable summaries of community complexity and stability.
- Requires relatively large sample sizes (typically n ≥ 30 per group) to infer statistically reliable co-occurrence edges.
- Co-occurrence does not imply direct biological interaction; spurious edges arising from shared environmental gradients are a persistent concern.
- Compositional normalization choices substantially affect inferred network structure, and no single normalization is universally optimal.
- Computational cost of bootstrapping for edge significance testing scales poorly with large taxon sets.
Frequently asked
Which correlation method should I use to infer the network — SparCC, SPIEC-EASI, or Pearson?
Do not use Pearson or Spearman on raw or relative-abundance data because microbiome data are compositional: abundances sum to a constant, inflating spurious negative correlations. SparCC (Friedman & Alm 2012) is a widely validated choice for 16S data. SPIEC-EASI uses sparse inverse covariance estimation and tends to produce sparser, more precise networks. The choice affects edge density but rarely changes the identity of the top hub taxa; running both and comparing results is good practice.
How many samples do I need for a reliable co-occurrence network?
A commonly cited minimum is 30 samples per group for the network to be estimable, with 50–100 preferred for stable edge recovery. Bootstrap edge stability (proportion of bootstrap replicates retaining the edge) should be reported; edges stable in fewer than 80% of bootstraps are typically excluded. Small-sample networks should be interpreted as exploratory only.
What is a keystone taxon in this context and how is it identified?
A keystone taxon is a node with disproportionately high network centrality — typically measured by degree (number of connections) or betweenness centrality (frequency of appearing on shortest paths between other taxa). Their removal in simulated node-deletion experiments causes the largest fragmentation of the network. Biologically, keystones are candidates for taxa whose loss destabilizes community structure, analogous to keystone species in macroecology.
Can I combine network-based analysis with standard differential abundance testing?
Yes, and the combination is informative. Differential abundance testing (DESeq2, ANCOM-BC) identifies taxa that change in relative abundance between conditions; network analysis reveals whether those taxa are also hubs or module anchors. A taxon that is both differentially abundant and highly connected is a stronger candidate biomarker or driver than one that is merely abundant.
Does rarefaction before network inference affect results?
Rarefaction reduces library size variance at the cost of discarding data, and its effect on inferred networks is debated. Compositional normalization (CLR or ANCOM-BC) is generally preferred over rarefaction for network inference because it retains all counts while correcting for depth. If rarefaction is used, the rarefaction depth and the proportion of samples retained should be reported.
Sources
- Friedman, J., & Alm, E. J. (2012). Inferring correlation networks from genomic survey data. PLoS Computational Biology, 8(9), e1002687. DOI: 10.1371/journal.pcbi.1002687 ↗
- Faust, K., & Raes, J. (2012). Microbial interactions: from networks to models. Nature Reviews Microbiology, 10(8), 538–550. DOI: 10.1038/nrmicro2832 ↗
How to cite this page
ScholarGate. (2026, June 3). Network-Based Microbiome Diversity Analysis. ScholarGate. https://scholargate.app/en/bioinformatics/network-based-microbiome-diversity-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Gene Set Enrichment AnalysisBioinformatics↔ compare
- Network-based pathway enrichment analysisBioinformatics↔ compare
- Pathway Enrichment AnalysisBioinformatics↔ compare
- Phylogenetic AnalysisBioinformatics↔ compare
- RNA-seq Differential ExpressionBioinformatics↔ compare