About

Why this exists

Useful conversions (feedstock to fuel, pollutant to harmless, residue to product) usually need several organisms handing intermediates to each other. That search space is combinatorial and rarely explored beyond one or two steps. Syntropa explores it computationally, from a harmonized corpus of 15,664 genome-scale metabolic models, and publishes what it finds, including the failures.

The approach

Each model is a genome-scale reconstruction approximating one organism's metabolic network: what it can consume, secrete, and transform. Most are automated, gap-filled drafts (see the limits below), not manually curated. Syntropa assembles these into a common namespace so a fermenter's outputs can be matched against a candidate cross-feeding partner's inputs across thousands of species at once. Syntropa reserves “syntrophy” for demonstrated obligate, thermodynamically coupled dependence; the general matching step finds cross-feeding candidates.

Discovery runs three modes. The first asks a directed question: which pairs and triples, drawing from the whole corpus, can convert a specific feedstock into a specific product that neither can make alone? The second asks for breadth: which communities simultaneously produce the widest range of products, each member necessary for at least one product? The third asks about closure: which consortia recycle the most of their shared flux internally, candidates for self-sustaining nutrient loops?

Each candidate that survives the screen is tested by flux balance analysis: stoichiometry and mass balance as the arbiter, not heuristics. Promising targets then pass a thermodynamic gate: a class-level, necessary-condition free-energy check on the overall transform (via eQuilibrator). It rejects energetically infeasible net conversions, but it does not adjudicate amino-acid or vitamin cross-feeding and is not a per-candidate proof. Per-reaction community thermodynamic screening (TFA) has been demonstrated on reduced electron-transfer cores (three obligate secondary-oxidation chemistries), and has since been run on the flagship's own glucose front end: it measures -140.1 +/- 12.1 kJ/mol and is FBA-incidental rather than obligate, the fermenter growing at full rate alone even with the thermodynamic constraint applied. That sharpens the claim rather than weakening it, locating the obligation in the interspecies hydrogen transfer at the secondary-oxidation step and not in the primary fermentation that feeds it. It is not yet applied per-candidate across the whole corpus (the Phase-I/II scale-up).

What it has found

The engine's predictions are hypotheses, not facts, until a wet lab confirms them. As a positive control that the scoring works, Syntropa was scored on a held-out benchmark: its ranking re-identifies the known hydrogenotrophic-methanogen partner class (a fermenter handing hydrogen to a methanogen to make methane), recovering 15 of 30 held-out positive pairs while ranking 0 of 60 decoy pairs as hits. Read this honestly: because the positive and decoy classes are separated on the same methanogenesis-exchange signature the score keys on, this is a near-definitional recovery of a known class. It is evidence the scoring is sound, not independent statistical validation, and not evidence that any specific co-culture functions. The methanogen finisher models here are AGORA2 (NonCommercial): the result is free to reproduce and cite for research, but commercial use needs a source license from VMH, and it is not reproducible on the CC-BY-only commercial subset. Nothing here has been cultured.

The honest bottom line. That methane result is a positive control on gut/rumen hydrogenotrophic methanogens. The digester methanogens, acetogens, carboxydotrophs, and chain-elongators that the biogas, syngas, and medium-chain-fatty-acid use cases would need are underrepresented or absent from the corpus (see the Atlas), and the default screening medium is not digester conditions either: its dissolved CO2 corresponds to a 3.8 percent headspace where an anaerobic digester runs 25 to 50 percent, so acquiring the missing organisms would be necessary but not sufficient. No motivating use case yet resolves to a credible in-corpus consortium, and no worked example is a novel multi-species consortium discovery. Most worked examples are honest single-organism results or published negatives, which is exactly what Syntropa publishes, and exactly the gap the platform is built to close.

Discovery reports for the worked examples are on the Reports page: the methane positive control, phenol and CO₂ (single-organism results), and syngas / medium-chain fatty acids (published no-credible-route negatives). Each shows the full funnel: the pool screened, the candidates found, the flux-balance verdict, the carbon-balance audit, and every drop reason.

How it stays honest

Flux balance proves stoichiometric possibility, not biological reality. The thermodynamic gate is a necessary condition on free energy, not a kinetic proof. Gap-filled reconstructions (the majority of the corpus) carry transporter reactions that were added computationally, not measured. When a gate lacks the data to decide, it reports unknown rather than guessing. Known limitations are published alongside the results.

Nothing here is validated by computation alone. Every candidate is a hypothesis; the point of publishing the screen is to surface the ones worth testing in a lab.

Get a discovery run (the hosted service is coming)

Discovery runs are compute-heavy and need the feedstock, target, and constraints set up carefully before the engine can produce an honest result, so a self-serve hosted service is in development rather than live today.

Register your interest below and you will get the launch note the moment it opens. The preprint and the open corpus already have citable Zenodo DOIs.