Two projects each report a pair of numbers for the same object that differ sharply with the scope the number was taken over: an essentiality count restricted to the organisms where a family occurs versus one taken across all organisms, and a within-pilot association versus a cross-cohort reproducibility score. The disagreement is how much a strong within-scope result licenses beyond that scope. It matters for condition-specific-fitness, which reads fitness and essentiality as condition-dependent: if within-scope strength does not travel, every “universal” list in this corpus carries an implicit scope clause.
Evidence Sides
Within scope, the signal is near-universal. 859 ortholog families — groups of genes descended from a common ancestor across genomes — were universally essential within every organism in which they occurred 1. On the community side, the within-pilot taxonomy–metabolite CCA (canonical correspondence analysis, which finds the combination of taxa most aligned with the measured metabolites) axis in the inflammatory bowel disease (IBD) pilot had r=0.964, where r is a correlation coefficient scoring how tightly the paired taxonomy and metabolite axis scores track each other 2.
Across scope, the signal is narrow or null. Only 15 essential families were essential in all 48 organisms 1. Pooled metabolomics had cross-cohort LOSO ARI=0.000 — leave-one-study-out validation, in which each cohort is held out in turn, scored by adjusted Rand index, a measure of cluster agreement corrected for chance — and the ecotype framework had mean LOSO ARI=0.113 2. The null stays null: the pooled-metabolomics cross-cohort cluster agreement was 0.000, not merely small 2.
Possible Reconciliations
- Hypothesis (different quantities). The IBD figures may not be two readings of one thing: they measure within-pilot association versus cross-cohort reproducibility 2, in which case the pair is no contradiction and must not be averaged into a single reproducibility claim.
- Hypothesis (context, not error). Genomic context, paralogs, alternative pathways, and compensation affect essentiality, which would make the 15-versus-859 gap a refinement of condition-specific interpretation rather than a contradiction of it 1.
- Hypothesis (denominator). The two essentiality counts differ in denominator — essential in all 48 organisms versus essential within every organism in which the family occurred 1 — so gene presence and absence, not a change in essentiality, could carry most of the gap.
Resolving Work
- Re-score the 859 occurrence-conditioned families against gene presence/absence calls across the 48 organisms: what share of the gap down to 15 families is explained by absence rather than by non-essentiality where present? 1
- Re-fit the within-pilot taxonomy–metabolite axis under a leave-one-study-out split and score it on each held-out cohort with the same axis correlation reported within pilot: does r=0.964 hold, drop, or vanish off its training cohort? 2
- Report per-held-out-cohort ARI values with interval estimates for both frameworks and compare them directly: is the ecotype framework’s mean LOSO ARI=0.113 separable from the pooled-metabolomics cross-cohort LOSO ARI=0.000, or within the spread across held-out cohorts? 2
- For families essential in some organisms but not others, model paralog copy number and alternative-pathway completeness as predictors of the essentiality call: does compensation account for the loss of universality? 1