UFZ Leipzig collaboration · Overview

The collaboration pages, in one place

This page has one entry per workstream from the 8 October visit, each with a short captioned video. The pages give a source and date beside their numbers, and signed-in readers can comment on any page.

WalkthroughGerman versions are under /de/. Comments need a free account; see the last section.

Workstreams

01

Six workstreams with the Electrobiotechnology group at UFZ Leipzig

Falk Harnisch and the group

0:00 / 0:19
Find the briefing that holds your own question and see what is still open on each, so that nobody has to guess which page is theirs.

Open the pagemessai.io/collab

What is on the page

  • Nitrogen cycling: 490 cohort papers on staging (measured 19 Oct 2026), with no complete records yet.
  • Kolbe electrolysis has 54 papers (staging, 5 Nov 2026) and ontology 14 papers (19 Oct 2026); no predictor is fitted yet.
  • The pages are briefings and proposals, and each number has its source and date beside it.

Not done yetThe index fits nothing itself. Each workstream page is a proposal, not a fitted result.

02

Nitrogen cycling, from papers to a designed experiment

Benjamin Korth

0:00 / 0:55
Follow nitrogen cycling from the cohort papers to a designed experiment, so the group can correct the record definitions and hypotheses before anything is fitted.

Open the pagemessai.io/collab/korth/nitrogen

What is on the page

  • 490 cohort papers on staging (measured 2026-10-27 00:31 UTC); 166 have a full-text PDF (34%).
  • A machine check found 96% of 308 sampled values in the paper (31 Oct 2026); no person has checked them yet.
  • H5 (current density to NH₄⁺ flux) is rated testable, but still needs a canonical parameter for NH₄⁺ flux.

What we ask you

  1. Can you check the three record definitions, confirm hypotheses H1 to H6, and set the limits for N₂O, nitrite and energy per kg N?
  2. Does the UFZ library licence allow text-mining of the 324 papers we cannot access, and if so, on what terms?
  3. Which published UFZ datasets can you share, in our CSV intake template?
  4. Should the design target be a groundwater column or an H-cell, and at what flow and influent nitrate?

Not done yetNothing is fitted yet. The nitrogen model does not exist, the hold-out set is not drawn, and 0 nitrogen values are verified.

03

Which parameters drive Kolbe performance, and by how much

Max Pohl

0:00 / 0:46
See which parameters the Kolbe papers vary and what is missing from them, so the group can say which cells in the parameter map would change their next experiment.

Open the pagemessai.io/collab/pohl

What is on the page

  • 54 Kolbe papers registered by 11 Oct 2026, 9 with extracted values, none reviewed; 0 of 434 values converted to SI units (31 Oct 2026).
  • The abstract reports fuel current efficiency rising sigmoidally with total acid concentration, from a threshold near 0.1 mol/L.
  • The highest reported value is 63.8 ± 1.4 % at 1.5 mol/L, the highest concentration tested.

What we ask you

  1. Besides concentration and composition, which parameters did you vary: current density, pH, temperature or electrode?
  2. Which cells in the parameter map would change your next experiment if we could predict them?
  3. Do you suspect interactions, for example a 0.1 mol/L threshold that moves with current density or chain length?
  4. Which papers from other labs are reliable enough to pool with your runs?

Not done yetNothing is fitted and the predictor is not built. Reproducing the published trends waits on your run table.

04

Which reactor and process parameters decide microbial electrosynthesis

John-Oliver Gath

0:00 / 0:47
Check how rates and potentials from the electrosynthesis papers were pooled, because mixed rate bases and reference electrodes hide errors that only the group can spot.

Open the pagemessai.io/collab/gath

What is on the page

  • Only 16 of 259 acetate-rate values reach an SI value (13 Oct 2026); 243 never do.
  • Potentials against Ag/AgCl, SCE and SHE are pooled as if equal, a hidden error of 0.2 to 0.24 V.
  • Scope: 3,382 papers are labelled MES, but only 397 mention electrosynthesis (staging, 13 Oct 2026).

What we ask you

  1. Which rate basis is primary: volumetric (g/L/d) or areal (g/m²/d)?
  2. Against which reference do you report potentials, and is the KCl concentration stated for every run?
  3. Which acetogen and which cathode mode do you use, for example fixed potential, galvanostatic or H₂-mediated?
  4. Which geometry and process parameters would you fix in a reporting standard, and which may vary?

Not done yetNothing is fitted yet. The MES predictor is not built and the harmonisation fixes have not started.

05

From our parameter vocabulary to a community ontology

Benjamin Korth

0:00 / 0:37
Read how our parameter vocabulary could become a community ontology, so the group can decide which outside vocabularies to anchor to before any term is fixed.

Open the pagemessai.io/collab/ontology

What is on the page

  • Counts disagree: the public package says 704 parameters, the pinned file has 825, and staging holds 835 definitions.
  • 29 slugs with no definition carry 4,985 live rows on staging (12 Oct 2026), led by legacy names such as powerDensity (1,418 rows).
  • No ontology release, DOI or permanent term address exists yet; an end-of-October release candidate was proposed and is not released yet.

What we ask you

  1. Which outside vocabulary should anchor each layer, and which one do you already depend on that we have not named?
  2. Should the namespace live on messai.io or on a neutral domain such as ISMET, so it outlives any single project?
  3. Is CC-BY-4.0, the licence of the public package, the right licence for the ontology?
  4. Who should approve a term: its steward alone, or the steward plus a reviewer from another group?

Not done yetNo outside expert has reviewed a definition, and annotator agreement has never been measured.

06

Teaching microbial electrochemistry on real data, together

Falk Harnisch and the group

0:00 / 0:36
See the proposed teaching modules built on real data, so the group can say what would fit a course or an onboarding before anything is built.

Open the pagemessai.io/collab/learning

What is on the page

  • Five modules are proposed: three are partly live today, two are still to build.
  • Education module v1 is due 28 Nov 2026 (roadmap D1-M3), with feedback captured through the widget.
  • The page builds on the nitrogen workstream and the lab’s 20 reproduction presets.

What we ask you

  1. Does the group already teach a course, lab rotation or PhD onboarding that this would replace or extend?
  2. Which of the five modules would you use first, and which would you drop?
  3. May student setups and raw data appear on a public page, or only behind sign-in?
  4. Would you accept a student’s check as a first screen that a senior reader confirms?

Not done yetNothing is built for a pilot cohort yet, and how to record a student’s check is undecided. Agreement with the presets is an in-sample check, not validation.

07

Microbial electrochemistry, one model at a time

Everyone

0:00 / 0:35
Step through a lesson built around a model you can open in the lab, so a student meets a mechanism and a prediction to test, not a slide of results.

Open the pagemessai.io/learn/lessons#graduate

What is on the page

  • Thirteen interactive lessons, three or four per level, each built around a model you can open in the lab.
  • The Graduate research level has four 90-minute lessons: Overpotential and Butler–Volmer kinetics; Monod kinetics and coupled digestion models; Uncertainty in field MFC data; Techno-economics, life-cycle assessment and scale-up.
  • Each lesson has a diagram, key concepts, a prediction to test and a short check.

Not done yetThe illustrations are schematic. They show the idea of each lesson, not measured data.

Organisms and electrodes

08

The organisms that make electrochemistry biological

Everyone

0:00 / 0:22
Browse the organism catalog and see how complete each record is, because every value should trace to a citation and a reader needs to know where the gaps are.

Open the pagemessai.io/microbes

What is on the page

  • 28 curated strains: 25 electroactive, 22 bacteria and 6 archaea.
  • Only 5 records are comprehensive; the others are partial or stubs.
  • Each organism has a field guide, and the page says every value traces to a citation.

Not done yetMost records are partial or stubs, so most fields are incomplete.

09

Geobacter sulfurreducens PCA

Everyone

0:00 / 1:03
Read the Geobacter field guide beside a growth simulation that uses its own kinetic constants, so the group can see which constants have evidence and which are assumed.

Open the pagemessai.io/microbes/geobacter_sulfurreducens_pca

What is on the page

  • Corpus medians: μmax 0.0715 h⁻¹ (one measurement, one paper); half-saturation constant Ks 171.2 mg/L (two papers).
  • Monod batch simulation (1000 mg/L acetate, 72 h): peak biomass 0.8 g/m², 90% removal at 36.9 h.
  • With a cell width of 0.5 µm, 96% (76 to 97%) of pore wall is reachable in the layer scan and 45% in the network.

Not done yetThe simulation values are scenarios, not evidence. Decay coefficient, biofilm density, exchange current density and transfer coefficient have no evidence, so the simulated current is an upper bound.

10

Shewanella oneidensis MR-1

Everyone

0:00 / 0:33
Read a second organism field guide beside the Geobacter page, so the group can compare how much is known about two electroactive organisms.

Open the pagemessai.io/microbes/shewanella_oneidensis_mr1

What is on the page

  • A field guide for a model organism that exchanges electrons in both directions, from identity and electron transfer through growth, kinetics and performance to safety and sources.
  • The kinetics section marks each constant as drawn from corpus rows, from a genus-level prior or as missing, and lists engineered strains, other electrodes and abiotic reactions without pooling them.
  • A simulation runs on those constants; a value you type fills a gap only and never overrides evidence.

Not done yetSeveral decay and current constants have no evidence yet, and a genus-level prior describes the whole genus, not this strain.

11

Inside an electrode’s pores

Everyone

0:00 / 2:13
Explore electrode pores at several scales and read the transport numbers computed from them, because pore structure decides where cells can live and how fast substrate arrives.

Open the pagemessai.io/lab/demo/pore-scan

What is on the page

  • Four scans at four scales on one logarithmic ruler: the Vulcan XC / PTFE electrode layer (42.1 µm field, 175 nm voxels) and its pore network (9.31 µm, 58.2 nm) by X-ray; a PtCu catalyst nanoparticle (66.9 nm, 0.38 nm) and the interior of a Pt / Ketjenblack support (21.5 nm, 0.168 nm) by electron tomography.
  • Segmentation presets per scale: white or black as the pore label for the X-ray scans, conservative / middle / permissive thresholds for the electron scans. Each is a separate analysis of the same scan, to compare before trusting a number.
  • Pores coloured by local sphere diameter in three size classes with movable boundaries and your own colours, or on a continuous ramp; cut the sample open along X, Y or Z at any depth, with a scale bar at its centre.
  • Two more 3D views: the SNOW2 pore network (pores as spheres, throats as tubes) and the shortest transport paths from inlet to outlet along X, Y or Z.
  • The Transport tab gives, per direction, voxel diffusion D_eff / D₀, the apparent tortuosity factor, the spanning pore fraction, the shortest-path ratio, network diffusion and network permeability. For the XC pore network, D_eff / D₀ is 0.77 / 0.75 / 0.64 along X / Y / Z.
  • Slices checks the pore mask against the native sections (SHA-256 verified); Size distribution is the volume-weighted histogram, with a CSV; Data & resolution shows which IUPAC pore classes each scan can sample at all.
  • Electron transfer: at 2 A/m², acetate diffusion can feed the current with 5.24× headroom through the whole 1 mm electrode thickness, and 22.7% of the wall sits in pores at least 1 µm wide, the smallest pore a cell can use.
  • Downloads: the analysis report (JSON), the size distribution (CSV), the pore mask and local-diameter TIFFs, the section volumes and the script that rebuilds the bundle. The same explorer sits in the electrode designer under View → Microstructure for Pt/C and activated-carbon cathodes.

Not done yetThese are electrode analogues, not a biofilm scan, and pore labels are assumed. Felt, cloth and brush are not scanned yet, and no transport number is validated against a measurement.

12

Biofilm growth in the lab

Everyone

0:00 / 0:44
Watch a biofilm grow on the chosen organism’s Monod curve with the evidence under every output, so the group can see how thin the support for a result is.

Open the pagemessai.io/lab?model=benchtop

What is on the page

  • The benchtop MFC in the 3D lab starts with Initial biofilm loading and Anode potential vs SHE unset, never a fabricated default.
  • In Biofilm view, the anode shell grows on the organism’s Monod curve, using the same constants as the field guide.
  • The live prediction bar quotes corpus percentiles (p10 to p90) and the row count under every number.

Not done yetIn this clip growth runs on a schematic timeline (marked surrogate), because the deep link does not set every kinetic input.

More to explore

13

Electrochemical direct air capture

Everyone

0:00 / 0:33
Look at an electrochemical direct air capture cell with no microbes in it, because the abiotic case came up during the visit and the same tools should handle it.

Open the pagemessai.io/lab/dac

What is on the page

  • This is a prototype of direct air capture (DAC) using bipolar-membrane electrodialysis, with no microbes in it.
  • There is no biology in this model: electrochemistry only, the abiotic case raised in October.
  • The page splits cell voltage into its parts, and shows specific energy of 2,538 kWh / tCO₂ (90% CI 2,300–2,837), labelled ASSUMED.

Not done yetThe energy figure is labelled ASSUMED, so treat it as a design estimate and not a measurement.

14

The MES knowledge graph

Everyone

0:00 / 0:41
Trace how parameters influence one another and where the literature reports no values, so the group can turn a gap in the evidence into a list of what to measure next.

Open the pagemessai.io/parameters/knowledge-graph

What is on the page

  • The snapshot has 826 nodes. Edge layers: causal, learned from data (0 edges so far), physics laws and causal effects (Bayesian).
  • Hover a node to trace its causes and effects; click it to see where its data comes from.
  • 427 nodes are gaps; a list ranks the gaps that many causal links pass through. Nodes can carry comment threads.

Not done yetHollow nodes mark knowledge gaps, where the literature reports no values.

15

The protocol knowledge graph

Everyone

0:00 / 0:39
Follow the experimental steps and how they relate, so the group can check the reproducibility criteria against how work is done in their own laboratories.

Open the pagemessai.io/protocols

What is on the page

  • It maps 97 experimental steps and 130 relations between them.
  • 47 of the steps carry reproducibility criteria.
  • Steps are grouped by category, such as preparation, analysis and operation.

Not done yetThis is a curated v1 snapshot, written by hand; only a subset is grounded in corpus evidence.

16

Electrode compatibility

Everyone

0:00 / 0:40
See which electrodes pair with which organisms and how well each pairing is supported, so that electrode choice starts from reported peak values and not from habit.

Open the pagemessai.io/lab/catalog?category=anodes&selected=carbon-cloth

What is on the page

  • The lab catalog lists 19 anodes and 12 cathodes, each with a Compatibility & references panel and compatible surface treatments.
  • The Pairs with filter links organisms to electrodes, so you can see which materials are linked to which organism.
  • Each organism page has an electrode-pairing block with the best reported peak values and a confidence rating.

Not done yetA peak value is the best reported for a pairing; confidence says how well the pairing itself is supported.

17

Build a hypothesis to test

Everyone

0:00 / 0:29
Build a testable hypothesis from a blank start, a system type or a paper and pair it with a methodology, so an open question from the visit becomes a design the group can run.

Open the pagemessai.io/lab/hypothesis-builder?puzzle=pohl:j

What is on the page

  • Five steps take you from a starting point to a saved hypothesis paired with a methodology.
  • You can start blank, from one of 17 system types, or from a specific paper.
  • An open puzzle can prefill it; the link here opens it with Pohl puzzle j (Kolbe electrolysis). A direction the puzzle leaves open starts as "no effect".
18

The research library

Everyone

0:00 / 0:26
Search the paper library by keyword or by meaning and filter by system, so the group can check what we hold on a topic and tell us which papers are missing.

Open the pagemessai.io/research

What is on the page

  • It is a browser for the corpus: 19,415 papers, searchable by keyword, hybrid or semantic mode.
  • Filters cover system type, application domain, paper type, publication year, whether parameters were extracted, and journal. Only 8,569 papers have a system type so far.
  • Each result shows how many parameters and conditions were extracted, with a link to the full paper page.
19

Process and instrumentation drawings of a cell

Everyone who designs or reviews a reactor

0:00 / 0:27
Open an engineering drawing of a cell in the standard notation, so the group can check that the process layout matches what they would build.

Open the pagemessai.io/lab/schematic

What is on the page

  • Each sheet is drawn from a declarative spec, with ISA-5.1 tagged instruments, valves that show their fail position and numbered balloons tied to a parts list.
  • Pick an operating state and the drawing routes it: the live path is highlighted, open valves are hollow, closed valves are filled and the governing interlock rule is lit.
  • Click a vessel, valve or instrument to read its engineering record; line-list values are labelled design, literature or assumed, and a sheet exports as SVG, DEXPI or GraphML.

Not done yetEvery sheet is marked as a concept, not for construction, and a value labelled assumed is a placeholder nobody has justified yet.

20

Vary one parameter and watch the prediction move

Everyone who plans an experiment

0:00 / 0:26
Vary one input and watch the predicted outputs move, so the group can test whether the model responds the way their experience says it should.

Open the pagemessai.io/parameters/sweep

What is on the page

  • Choose a system type, one sweep parameter and the fixed conditions, then read peak power density, Coulombic efficiency, internal resistance and empirical coverage for the run.
  • Overlay the corpus prior, compare with the last run, or copy the run as Python; a run can be cited, shared or exported.
  • Validation warnings appear beside the result, for example when the Coulombic mass balance does not close, and you can lock a hypothesis before running.

Not done yetThe source mix shows which points are empirical and which are simulated; a run made of simulated points is model output, not a measurement.

21

A quick estimate before the full model

Everyone who wants a first estimate

0:00 / 0:24
Try a configuration and watch an instant illustrative estimate respond, so the group can see how the inputs are framed before moving to the sweep or the lab for estimates with stated uncertainty.

Open the pagemessai.io/predictions

What is on the page

  • Start from a preset or choose anode, cathode, organism and operating conditions, and a live estimate of power density, voltage and efficiencies updates as you edit.
  • Tabs cover the results, an optimisation view and a history for comparing runs; a full run adds confidence intervals, a power curve and economics.
  • The page points to the parameter sweep and the 3D lab for estimates with stated uncertainty.

Not done yetThe live estimate is an illustrative client-side simulation, not the trained model, and the page says so.

22

Where the literature disagrees with itself

Everyone who reads across papers

0:00 / 0:30
See where papers in the library disagree with each other, so the group can say which disagreement is real and which comes from different conditions.

Open the pagemessai.io/insights/contradictions

What is on the page

  • Cross-paper numerical contradictions are listed with the parameter, both values, the difference and the papers, and can be filtered by scenario, by pattern or by search.
  • A rule engine classifies each row, for example as a reproducible disagreement, a normalisation artefact such as a reference-electrode mismatch, a different physical regime, a substrate effect or an extraction artefact.
  • Click a row for the side-by-side condition diff, the source quotes and the unit-conversion trace.

Not done yetRows classed as a reproducible disagreement are the candidates for genuine scientific disagreement; the others look like artefacts or different conditions.

23

Values that fail a physics check, and combinations few papers report

Everyone who checks extracted values

0:00 / 0:24
Read the values that fail a physics check, the rarely reported combinations and the statistical outliers beside their source papers, so the group can reject a plausible-looking value before it reaches a model.

Open the pagemessai.io/hunter

What is on the page

  • Violations are extracted values that break a physics constraint, usually extraction errors rather than errors in the paper; each is re-validated against the extracted values before it is shown.
  • Opportunities are combinations of system, substrate, anode and scale that few papers have tried and that the priors rate as plausible.
  • Anomalies and outliers are statistical outliers worth checking against the paper; click a card to see the source paper.

Not done yetA flag is a prompt to check the paper, not a verdict: an outlier can be an error or a breakthrough.

24

Pooled priors, distributions and interval coverage across the corpus

Everyone who compares studies

0:00 / 0:26
See pooled effects across studies with the spread shown, so the group can tell a consistent finding from an average that hides disagreement.

Open the pagemessai.io/meta-analysis

What is on the page

  • Pick a parameter and a system class, or pool all systems, to see the pooled Bayesian prior with its interval for each class.
  • The empirical distribution shows the spread of the SI-normalised corpus values, and the calibration panel reports out-of-sample interval coverage.
  • An effects panel shows within-paper slopes and the transfer between system classes, and an iteration selector chooses which prior version is shown.

Not done yetThe pooled view mixes system classes; read the per-class view before quoting a prior.

Getting started

25

The onboarding atlas

Everyone

0:00 / 0:48
Read how a paper becomes a number on a page, starting from the part written for new users, because a reader trusts a number more when they know how it was made.

Open the pagemessai.io/learn/onboarding/atlas

What is on the page

  • Seven parts on one page, with a left-hand rail that follows you as you scroll.
  • Part VII, How to use the platform, is written for a new user; start there.
  • Parts I to VI explain how a paper becomes a number on a page, and how the platform is built and run.

Not done yetParts I to VI are written for maintainers; the detail there is more than a first visit needs.

26

How to comment on a page

Everyone

0:00 / 0:28
Sign in and attach a comment to the exact sentence, number or table cell you mean, so that corrections reach us in context and not in a separate email.

Open the pagemessai.io/auth/signin

What is on the page

  • Sign in at messai.io/auth/signin with email and password, Google or GitHub; an account is free.
  • Open any page, click the speech-bubble icon next to the language switch, then click the sentence, number or table cell you mean.
  • Write and post; threads can be replied to and marked resolved, and only signed-in readers see comments.

Prepared by MESSAI for the Electrobiotechnology group at UFZ Leipzig. Numbers are results only where a page states how and when they were measured.