Differential abundance (DA) — testing which taxa differ in abundance between sample groups — was run within ecotype strata (E1 and E3, community-type clusters), and two analyses of those strata leave incompatible target lists. The cross-project digest reports 33 within-ecotype candidates that collapse to 3 under a stricter agreement filter and declares the list not usable 1; a revised meta-analysis — combining evidence across strata and sub-studies rather than testing one pooled set — reports 51 E1 candidates and 40 provisional E3 candidates 2. The disagreement matters because the two versions differ not only in size but in verdict: one declares the within-ecotype list not usable 1, the other reports replacement candidate counts 2. The tension is carried by compositional-robustness-of-differential-abundance-calls.
Evidence Sides
Side A — the digest: within-ecotype calls do not survive agreement filtering. Within-ecotype DA yielded 33 candidates across both E1 and E3, and a stricter filter requiring both within-substudy agreement and agreement with LinDA — a linear-model DA method for compositional (sum-constrained) count data — left 3, in E3 only; on that basis the within-ecotype list was declared not usable 1. The operative claim here is a usability verdict under a stated threshold, not a count offered for comparison: 3 is what remains after the filter, and the conclusion drawn is negative.
Side B — the revised meta-analysis: the design-controlled count supersedes it. A within-ecotype × within-substudy meta-analysis — stratifying by ecotype and by sub-study jointly before combining — instead reports 51 E1 candidates across two eligible sub-studies and 40 provisional E3 candidates from one 2. This side does not average with Side A: the later, design-controlled count replaces the earlier one, with E3 explicitly provisional pending a second eligible sub-study 2. That same provisionality is flagged as bearing on whether the ecotype partition itself is stable 2.
Possible Reconciliations
- Hypothesis (nesting order): the digest applied within-substudy agreement as a post-hoc filter on a pooled within-ecotype test, while the revision made sub-study a design level before testing; if so, 3 1 and 51/40 2 count different quantities rather than conflicting estimates of one.
- Hypothesis (eligibility gate): the two eligible E1 sub-studies and the single eligible E3 sub-study 2 may be a subset of what the digest pooled, so the eligibility screen could remove heterogeneity that the LinDA-agreement filter was absorbing.
- Hypothesis (E3 asymmetry): the digest’s 3 survivors were E3 only 1 while the 40 E3 candidates remain provisional on one sub-study 2, suggesting E3 is the stratum most sensitive to how sub-study structure is handled — possibly because the E3 partition is itself unstable.
Resolving Work
- Re-run both pipelines on an identical sample set, varying only whether sub-study enters as a pre-test design level or a post-hoc agreement filter: does ordering alone reproduce 3 versus 51/40?
- Cross-tabulate the 3 surviving candidates against the 51 E1 and 40 E3 lists: is the survivor set contained in the larger lists, or disjoint from them?
- Add a second eligible E3 sub-study and repeat the within-ecotype × within-substudy meta-analysis: do the 40 provisional E3 candidates replicate when E3 is no longer single-sub-study?
- Hold the design fixed and swap only the test (LinDA versus the meta-analysis statistic) on the same strata: is the gap method-driven or design-driven?
- Re-derive ecotype labels under perturbed clustering and recount candidates per stratum: how much of the 51/40 depends on the E1/E3 partition itself?