A paired metagenomic-metabolomic contrast can be read one compound at a time or one chemical class at a time. Class-level aggregation groups annotated metabolites into pre-specified chemical-class themes and asks whether the differentially abundant (DA) compounds — those whose abundance differs between arms of the contrast — fall into any theme more often than chance. In the Crohn’s-disease (CD) phage-targeting project the aggregation was run over 11 IBD-relevant chemical-class themes with a per-theme Fisher’s exact test (a test of enrichment in a 2x2 count table), Benjamini-Hochberg FDR correction across themes (FDR = false discovery rate, the expected share of false calls among those declared significant), and a pre-set two-part gate: a theme counted as supported only when FDR < 0.10 and the odds ratio (OR) exceeded 1.5 1. The appeal of the approach is that it turns a long list of individually noisy compounds into a small number of named mechanism axes that the project can declare supported or not under a stated gate 1 — see fishers-exact-test, benjamini-hochberg-fdr and multi-omics-integration.
The reported per-theme result, with themes labelled “CD-up” when the member metabolites were higher in the disease arm 1:
- polyamines — 9 metabolites in theme, 5 CD-up, OR 14.6, FDR 0.008; supported 1
- long_chain_PUFA (polyunsaturated fatty acids) — 15 in theme, 6 CD-up, OR 7.9, FDR 0.009; supported 1
- acyl_carnitines — 22 in theme, 2 CD-up, OR 1.06, FDR 1.0; not supported 1
- bile_acids — 21 in theme, 3 CD-up, OR 0.56, FDR 1.0; not supported, with the report noting that 3 of 3 CD-up members are tauro/taurine forms 1
- lipid_classes — 163 in theme, 19 CD-up, OR 1.64, FDR 0.76; not supported, annotated as “19 CD-up CEs/SMs/ceramides; theme dilute” 1
- short_chain_fatty_acids — 14 in theme, 0 CD-up, no OR, FDR 1.0; not supported, with the note that LC-MS undersamples SCFAs 1
- tma_choline — 29 in theme, 2 CD-up, OR 0.78, FDR 1.0; not supported 1
- urobilin_porphyrin — 5 in theme, 1 CD-DOWN, OR 143, FDR 0.13; not supported, flagged as “n=1 too small” 1
- tryptophan_indole, fatty_acid_amides, aromatic_AA and oxidative themes — theme sizes recorded only as small, 0 CD-up, no OR, FDR 1.0; not supported 1
The project’s own verdict on the hypothesis tested by this analysis (H3d-DA) was SUPPORTED on the basis of 52 metabolites passing differential abundance and 2 themes passing Fisher enrichment, with both falsifiability gates met; the report reads the result as adding two new mechanism axes — polyamines and long-chain PUFAs — that complement the iron, bile-acid and TMA findings reported earlier in the same project 1. The individual polyamine calls are putrescine (cliff=+0.45, FDR=0.08), N1-acetylspermine (+0.56, 0.06), N-acetylputrescine (+0.46, 0.06), anserine (+0.43, 0.08) and diacetylspermine (+0.43, 0.08) 1.
The evidence is one project’s analysis of one paired dataset, and the report describes the test itself as a cohort-level, subject-averaged contrast 1. The enrichment statistics for the two passing themes are strong within that dataset (FDR 0.008 and 0.009 under a pre-set FDR < 0.10 and OR > 1.5 gate) 1, but no second cohort is tested here, so “polyamines and long-chain PUFAs mark the inflamed gut metabolome” should be carried forward as a cohort-specific result awaiting replication rather than a general law 1; compare the portability arguments in cross-cohort-microbiome-portability.
The themes that did not pass the gate are part of the reported result, not discarded background: FDR 1.0 is recorded for acyl_carnitines, bile_acids, short_chain_fatty_acids, tma_choline and the tryptophan_indole / fatty_acid_amides / aromatic_AA / oxidative group, while lipid_classes reached OR 1.64 at FDR 0.76 and urobilin_porphyrin reached OR 143 at FDR 0.13 on a single CD-DOWN metabolite that the report flags as “n=1 too small” 1. For the 0-of-14 SCFA theme the report keeps three explanations open rather than settling on one: that HMP2 LC-MS untargeted methods undersample SCFAs (volatile, polar — typically need GC-MS), that the cohort-level subject-averaged contrast masks within-subject variation, or that gross SCFA pool differences between CD and nonIBD are smaller than within-subject variability 1. Reading such nulls as limits on resolution rather than as settled absences is the stance argued in null-results-under-limited-statistical-resolution.
Tensions
The bile-acid theme does not pass at the class level yet carries a specific compound-level signal, and the source keeps both readings rather than choosing between them. On the class side the theme returned OR 0.56 at FDR 1.0 across 21 member metabolites with only 3 CD-up, which does not meet the pre-set gate 1. On the compound side, tauro-alpha/beta-muricholate plus free taurine were CD-up (free taurine at cliff=+0.47, tauro-alpha/beta-muricholate at +0.40), and the report interprets this as the metabolite-level confirmation of reduced microbial bile-acid 7alpha-dehydroxylation, mechanistically supporting the project’s earlier nomination of Flavonifractor plautii — described there as a known 7alpha-dehydroxylating species — as a phage target with a metabolite-level pre-treatment biomarker 1. See bile-acid-7alpha-dehydroxylation, flavonifractor plautii and phage-therapy-evidence-translation.
The two readings are not reconciled by averaging them, because they are claims at different granularities: a theme-level odds ratio asks whether membership in a chemical class predicts being called DA, while the taurine-conjugate result is a directional claim about three named compounds inside a 21-member theme 1. For two other themes the report attributes a non-significant class-level test to a dilute background rather than to absent compound-level signal — for acyl_carnitines, “the 22-carnitine background is dominated by short/medium-chain forms that don’t shift in CD” (OR 1.06), and for lipid_classes, “the 163-metabolite background is too dilute (most lipid classes don’t shift, just the long-chain-PUFA-conjugated ones)” despite 19 CD-up members (OR 1.64) 1. Whether the same dilution accounts for the bile-acid theme’s OR 0.56 is a hypothesis the reported analysis does not test 1. The practical reading is that a theme failing the gate does not license discarding its member compounds, and a theme passing it does not license treating every member as changed — of 9 polyamines, 5 were CD-up 1. The theme definitions are themselves an analysis choice, as argued in ontology-and-category-schema-sensitivity.
A second granularity tension sits inside the polyamine axis itself: the project’s earlier pathway-level 06_polyamine_urea theme was CD-DOWN at pathway level (OR=0.42) while the metabolite-level theme here is CD-UP (OR=14.6), and the report reconciles this as pool-versus-flux, arguing that polyamine metabolite-pool accumulation can result from increased catabolism of dietary protein / mucin plus reduced microbial polyamine clearance without requiring elevated biosynthesis flux 1. The same pattern recurs for fat metabolism, where the pathway-level 10_fat_metabolism_glyoxylate theme was not Fisher-significant (OR=0.88) but the metabolite-pool elevation is theme-significant at OR=7.9 1. That reconciliation is the report’s mechanistic interpretation, not a measured flux, and should be read as a hypothesis 1.
Open Directions
- Re-run the identical 11-theme Fisher screen with the same gate (FDR < 0.10 and OR > 1.5) on an independent paired metagenome-metabolome cohort, to test whether the polyamine (OR 14.6) and long-chain-PUFA (OR 7.9) axes are cohort-specific or portable; this closes the replication gap left by a single cohort-level contrast 1.
- Separate the three explanations the report leaves open for the 0-of-14 SCFA null by pairing a GC-based or derivatized SCFA assay with the sample-level within-subject paired test the report names as the natural follow-up, since LC-MS undersampling, subject-averaging, and genuinely small pool differences are currently indistinguishable 1.
- Decompose the 163-member lipid_classes theme into its named subclasses (CEs, SMs, ceramides) and re-test, since 19 CD-up members produced only OR 1.64 at FDR 0.76 in the aggregate and the report calls the background dilute; this would test the dilution explanation directly 1.
- Join taurine-conjugated bile-acid abundance to strain- and gene-level abundance of 7alpha-dehydroxylating taxa in the same samples, converting the report’s interpretive link into a testable quantitative dependency and checking the proposed pre-treatment biomarker for flavonifractor plautii targeting 1.
- Expand porphyrin/urobilin annotation coverage and re-test that theme, which reached OR 143 at FDR 0.13 on a single CD-DOWN metabolite out of 5 and is flagged “n=1 too small”; more annotated members would let the axis be evaluated under the same gate 1.
- Test the polyamine and long-chain-PUFA axes against the project’s iron findings (iron acquisition and pathobiont expansion) in the same samples, since the report treats them as complementary mechanism axes without showing that they are statistically independent readouts 1.