Metagenomics and Metabolomics: From Reads to Pathways

16S amplicon · Metagenome assembly · Binning · LC-MS annotation · Pathway enrichment · Multi-omics

Published 2026 · Last updated 2026-09-27 · Hailu Zhou

Metagenomics tells you who is there and what they could do; metabolomics tells you what chemistry is actually present. The promise of joining them is causal storytelling about a microbiome, and the reality is two very noisy datasets whose noise structures are completely unrelated — which is exactly why naive correlation produces seductive nonsense. This post walks both halves and is blunt about where the join fails.

16S amplicon: fast, cheap, and full of biases

The 16S rRNA gene is a phylogenetic marker with conserved regions for primer binding and variable regions (V3–V4, V4) carrying taxonomic signal. Amplicon sequencing is cheap and works at low biomass, but every step introduces bias: primer choice determines which taxa are amplified at all, copy-number variation between species makes relative abundance a proxy rather than an absolute, and the variable region limits resolution — often to genus, sometimes worse.

Modern analysis uses amplicon sequence variants (ASVs) rather than operational taxonomic units. DADA2 models sequencing error per base and resolves exact variants, which makes results reproducible across studies and allows reuse of sequence data; QIIME 2 wraps this in a provenance-tracking framework. The operational taxonomic unit (OTU) approach clustered reads at a similarity threshold, which collapses closely related taxa and makes cross-study comparison unreliable.

Shotgun metagenomics: assembly and binning

Shotgun sequencing reads all DNA in a sample, which sidesteps primer bias and captures functional potential. Two analysis routes exist. Read-based profiling assigns reads to reference genomes or marker genes and is fast and reference-dependent. Assembly-based analysis assembles contigs and then bins them into metagenome-assembled genomes (MAGs) by combining sequence composition and coverage across samples — the more samples, the better the coverage-based separation.

Binning is where the ambition meets the limits. MAGs are often fragmented and incomplete, and the field standardised around completeness and contamination estimates from single-copy marker genes. A 70%-complete, 5%-contaminated MAG is useful for pathway presence but not for claiming a genome sequence. Strain-level variation within a species blurs bin boundaries entirely.

Metabolite identification from LC-MS

Liquid chromatography–mass spectrometry separates compounds by chromatography and detects them by mass-to-charge ratio, with fragmentation spectra providing structural evidence. Untargeted metabolomics produces tens of thousands of features, and most are not identifiable to a named compound. Identification confidence is tiered: a matching accurate mass and retention time against an authentic standard is level 1; matching a fragmentation spectrum to a library without a standard is level 2; and a mass-only match is level 3 or lower.

Report identification levels honestly. A pathway story built on level-3 annotations — mass matches with no standard and no spectrum — is a hypothesis, and calling it a metabolite is overreach. Ion suppression, in-source fragmentation, and adduct formation all complicate the feature→compound mapping, and isomer resolution usually requires chromatography, not better mass accuracy.

Pathway enrichment on both sides

On the metagenomic side, pathways are inferred from gene content — for example via KEGG orthologs — but gene presence is not activity and not flux. On the metabolomic side, pathways are inferred from detected metabolites. Both produce gene-set or metabolite-set enrichment statistics analogous to transcriptomic enrichment, and both inherit the same traps: incomplete annotation, arbitrary significance cutoffs, and redundancy between overlapping pathways.

Layer What it actually measures Main confounder
16SRelative taxon abundancePrimer and copy number bias
MetagenomeFunctional potentialReference completeness, DNA only
MetabolomePresent chemistryIdentification confidence, matrix effects

Multi-omics association: the honest version

Joining microbiome and metabolome data means testing association between two high-dimensional, compositional, sparsely observed matrices. Simple pairwise correlations are the usual mistake: they ignore compositionality, multiple testing across millions of pairs, and the fact that both sides share sample-level confounders such as diet, medication, and batch. Proper approaches use compositionally aware transformations, sparse canonical correlation, or multivariate models with explicit covariate adjustment and permutation-based significance.

Even then, association is not mechanism. A microbe and a metabolite that co-vary may share a driver neither measured; the strongest microbiome-metabolite claims come from culture or gnotobiotic experiments that manipulate one and observe the other. Computational multi-omics generates candidates; it does not establish causality. Keep that boundary explicitly in any figure legend you write.

What a defensible pipeline looks like

Build the analysis so each layer's output is a table you can audit: ASVs or MAGs with completeness estimates, features with identification confidence levels, and associations with the confounders that were adjusted for. Freeze the version of every database used, because taxonomy and pathway references change and silently rewrite your figures on reanalysis. Then interpret upward with restraint — from a differential taxon to a differential function to a metabolite is a chain of inference, and each link needs its own evidence. The pipeline is not the finding; the finding is the experiment that confirms it.

References

  1. Callahan BJ, McMurdie PJ, Rosen MJ, et al. DADA2: high-resolution sample inference from Illumina amplicon data. Nat Methods. 2016;13(7):581–583.
  2. Bolyen E, Rideout JR, Dillon MR, et al. Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2. Nat Biotechnol. 2019;37(8):852–857.
  3. Quast C, Pruesse E, Yilmaz P, et al. The SILVA ribosomal RNA gene database project: improved data processing and web-based tools. Nucleic Acids Res. 2013;41(Database issue):D590–D596.
  4. Kanehisa M, Goto S. KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Res. 2000;28(1):27–30.
  5. Bowers RM, Kyrpides NC, Stepanauskas R, et al. Minimum information about a single amplified genome (MISAG) and a metagenome-assembled genome (MIMAG) of bacteria and archaea. Nat Biotechnol. 2017;35(8):725–731.
  6. Wang Q, Garrity GM, Tiedje JM, Cole JR. Naive Bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy. Appl Environ Microbiol. 2007;73(16):5261–5267.
  7. Smith CA, Want EJ, O'Maille G, Abagyan R, Siuzdak G. XCMS: processing mass spectrometry data for metabolite profiling using nonlinear peak alignment, matching, and identification. Anal Chem. 2006;78(3):779–787.
  8. Wang M, Carver JJ, Phelan VV, et al. Sharing and community curation of mass spectrometry data with Global Natural Products Social Molecular Networking. Nat Biotechnol. 2016;34(8):828–837.
  9. Gloor GB, Macklaim JM, Pawlowsky-Glahn V, Egozcue JJ. Microbiome datasets are compositional: and this is not optional. Front Microbiol. 2017;8:2224.