hydrogenotrophic methanogenesis: glucose -> CH4 via interspecies H2 transfer (host-associated methanogen)

Provenanced sample report (Syntropa discovery, role-aware Stage-1). Derived on corpus v0.2.0 (fingerprint 3574ed92836e), all 26 member models verified live in v0.2.2 (15,664 models, fingerprint 58a8f20e2aba) · tools v0.1.1 (git 44a8422, code a412618fd214) · env py3.13.13 win32 swiglpk5.0.13+osqp1.1.3+highspy1.15.1+pyscipopt6.2.1+optlang1.9.1 (deps 7f138ead8863). Feed={'glucose': 10} target=methane mode=produce. Report id 829c5773f6ad (content-addressed to config + EXACT corpus + EXACT tools -> auditable; git-reproducibility is PENDING a committed + tagged tool release). Every claim here is computed; every drop reason and known limit is listed below.

Verdict: CONSORTIUM-ONLY (complementation, irreducible)

no single organism produces the target and there is no producing proper subset: every member is required. At n=2 the only proper subsets are the two singletons, and neither produces, so the pair is irreducible: metabolic COMPLEMENTATION (a community reaches what no member reaches alone): NOT a tested super-additivity against the summed member outputs, a test not run here.

Internal consistency checks (in silico: NOT wet-lab validation)

Flagship anchor, CROSS-CHECKED four ways IN SILICO (2026-07-24; NONE is wet-lab validation: nothing here is cultured): (1) ENGINE-PATH: the signal survives the real closed-medium emergence path, not the leaky open-medium one (baseline collapses 9.3->0.046); (2) adversarial review: the two highest expert review findings were refuted by direct test; (3) OUT-OF-SAMPLE SEPARATION: on a held-out set the positive class hit 15/30 vs 0/60 for two decoy classes. Read this honestly: the positive and decoy classes are separated on the SAME methanogenesis-exchange signature the score keys on, so NO valid p-value can be computed: the raw separation is reported only as a SANITY CHECK that the score reads the intended signature, showing it recovers a known hydrogenotrophic-methanogen partner class blind, NOT that any specific co-culture functions. The finishers are gut/rumen hydrogenotrophs (Methanobrevibacter); the true digester methanogens (Methanoculleus, Methanosarcina, Methanothrix) and the acetoclastic/SAO steps are ABSENT, so this is a ~1 CH4/glucose fragment of anaerobic digestion, not AD; (4) THERMODYNAMICS: HONEST SCOPE: this pair is a glucose fermenter + Methanobrevibacter (a hydrogenotrophic/formatotrophic methanogen), so it captures only the H2/formate electrons: a ceiling near 1 CH4 per glucose. The full glucose->3CH4+3CO2 transform needs acetoclastic methanogenesis or syntrophic acetate oxidation, which NEITHER member performs, so that net transform is NOT this pair's feasibility check. The top hit's 16.304 on glucose=10 (1.63 CH4/glucose) EXCEEDS the ~1.0 hydrogenotrophic ceiling and leans on acetate->H2 oxidation that is endergonic at unconstrained H2 (caught only by per-reaction TFA, a Phase-2 capability): read it as an unconstrained-thermodynamics UPPER BOUND, not an achievable yield.

Honest limits:

• Breadth is n=1: this clean non-circular structural anchor is METHANE-SPECIFIC (a rare, specific exchange signature). Broadening to common-intermediate targets (sulfate reduction, chain elongation) needs a reaction-marker index, not yet built: so this is ONE in-silico cross-checked anchor, not a general claim and not a validated one.
• Hit propensity is model-source-dependent (CarveMe/AGORA reconstructions secrete the H2 intermediate readily, embl_gems less so), so candidates are read WITH their source.
• Thermo is CLASS-LEVEL + necessary-only: per-hit community transforms are 'unknown' (amino-acid cross-feeding unmapped), so each hit's ledger thermo_verdict stays NULL: never faked.

Frozen rigor records are archived with the release.

Consortium-only (complementation)

• A multi-member community reaches the target (16.304) via COMPLEMENTATION: no single organism produces it alone. NOT a tested super-additivity against the summed member outputs, a test not run here.
• The 28 feedstock-dependent candidates are COMBINATIONS over a diverse front-end core and largely SHARE the finisher(s) ['agora2_methanobrevibact_ruminantiumm1', 'agora2_methanobrevibact_smithiiatcc35061']: they are variations on a small number of core relays, NOT 28 independent discoveries. The biological finding is the finisher + its best partners, not the raw count.

Ranked feedstock-dependent candidates (top 10 of 28)

The production column = the target's community exchange (production) flux (FBA objective, mmol/gDW/h) on an OPEN medium: an UPPER BOUND, not a measured rate. It can EXCEED the feed-limited stoichiometric ceiling (an open-medium artifact); any value above the ceiling is NOT an achievable closed-medium yield. See caveats.

CARBON-BALANCE AUDIT: by carbon conservation the named feed can supply at most 60.000 mmol/gDW/h of the target. Rows tagged !>ceiling exceed that: they are open-medium carbon artifacts, INFEASIBLE as closed-medium yields (0 of the top 10 shown). (Redox/electron ceilings are not carbon-caught; see caveats.)

Attribution

Every model above is credited by id + source. Model authors: your reconstruction appears in this result. (Impact-ledger stamping = flywheel, pending.)

Caveats (disclosed up front)

• PRODUCTION IS AN OPEN-MEDIUM UPPER BOUND, NOT A YIELD. The production flux is FBA on an OPEN medium: carbon and reductant may enter from the background medium, not only the named feed: so a value CAN EXCEED the feed-limited stoichiometric ceiling (e.g. CO2+H2->acetate <=~5 at 4 H2/acetate; glucose->CH4 <=~1 per glucose by the hydrogenotrophic route). Read every production number as an unconstrained UPPER BOUND; a value above the ceiling is an open-medium artifact, not an achievable closed-medium yield. Closed-medium re-runs are Phase-2.
• FBA proves a route is stoichiometrically POSSIBLE, not that a living cell will do it.
• Every consortium here is a HYPOTHESIS until cultured; nothing is wet-lab validated.
• MEMBER IDENTITY IS A LEAD, NOT A CLAIM. Named members are FBA-feasible only in AUTOMATED, GAP-FILLED DRAFT reconstructions: gap-filling can add transporters/reactions an organism does not actually have, so a named member may CONTRADICT established physiology (a strain the literature does not know to perform this chemistry). Check each named organism against the literature before treating it as a real candidate; the value here is the metabolic PATTERN, not the specific strain.
• Feedstock-dependence is the anti-medium-artifact control: withdraw the feed and the product must collapse (a NOT-feedstock-dependent result is flagged SUSPECT, not a discovery).
• The ranking score is a documented heuristic for human review, not a truth or biology claim.
• Stage-1 selects finisher x diverse front-ends; multi-finisher relays are out of scope. Stage-2 expansion of winners is pending: these are PRELIMINARY until it runs.