Open research dataset and parameter ontology for Microbial Electrochemical Systems
| Dataset | Records | File | Description |
|---|---|---|---|
| Paper-parameter values | 18,113 | data/paper-parameter-values.csv |
Validated values mapped to the 687-parameter ontology (numeric + text) |
| Extracted parameter data | 25,566 | data/extracted-parameter-data.csv |
All quality-filtered extractions with labeling flags |
| Paper metadata | 23,332 | data/paper-metadata.csv |
Full corpus with verified_mes flag (8,904 confirmed) |
| Parameter definitions | 687 | data/parameter-definitions-full.csv |
Complete ontology from MESSAI database with usage counts |
| Ontology index | 667 | parameters/index.json |
Structured JSON with ranges, units, types, defaults |
Plus: correlation analysis, reproducibility scoring, a five-parameter reporting checklist, and an interactive Parameter Explorer.
This repository supports: "From PDF to Protocol: How AI-Driven Meta-Analysis of 21,895 Publications Reveals Hidden Patterns in Microbial Electrochemical Systems Research" (EU-ISMET 2026).
The extraction pipeline is imperfect. Before using this data, read data/SCIENTIFIC_INTEGRITY.md. Key issues:
- Paper-parameter-values (18,113 rows) is the primary dataset — each value maps to an ontology parameter.
has_numeric_valueflag distinguishes numeric measurements (73%) from text values (27% — material names, methods, etc.) - Extracted-parameter-data (25,566 rows) is the full quality-filtered extraction output. Each row is labeled:
ontology_matched(2.1% match a defined parameter name),likely_garbage(57.9% are parsing artifacts), and the rest (40%) are legitimate values with free-text parameter names not in the ontology (e.g., "COD", "HRT", "diameter") - Paper-metadata includes 23,332 papers with a
verified_mesflag (8,904 = 38% confirmed MES). The rest were ingested by a permissive filter. - 234 of 687 parameter definitions have usage; 453 (66%) are aspirational.
import pandas as pd
# Paper-parameter values: the primary dataset (ontology-mapped, 13K values)
values = pd.read_csv('data/paper-parameter-values.csv')
print(f"{len(values)} parameter-paper values")
print(values['parameter_category'].value_counts())
# Filter to verified MES papers only
mes_values = values[values['verified_mes_paper'] == True]
print(f"{len(mes_values)} values from verified MES papers")
# Filter to a specific parameter
power = values[values['parameter_name'] == 'Maximum Power Density']
print(f"{len(power)} maximum power density measurements")
# Paper metadata — use verified_mes flag to filter
papers = pd.read_csv('data/paper-metadata.csv')
mes_papers = papers[papers['verified_mes'] == True]
print(f"{len(mes_papers)} verified MES papers out of {len(papers)} total")
print(mes_papers['system_type'].value_counts())mappings <- read.csv("data/paper-parameter-values.csv")
papers <- read.csv("data/paper-metadata.csv")
table(papers$system_type)
table(mappings$parameter_category)Visit the Parameter Explorer to search, filter, and explore all 687 parameters interactively.
npm install @messai-io/mess-parametersimport parameters from '@messai-io/mess-parameters';
const allParams = parameters.categories.flatMap(c =>
c.subcategories.flatMap(s => s.parameters)
);MESS-Parameters/
data/ # Research data and analysis outputs
paper-parameter-values.csv # 13.3K ontology-mapped values (cleanest)
extracted-parameter-data.csv # 25.6K raw extracted values (noisier)
paper-metadata.csv # 23.3K papers with DOIs and metrics
parameter-definitions-full.csv # All 687 parameter definitions
parameters-full.csv # 667 ontology definitions as flat CSV
correlations.csv # 14 correlations with 95% CIs
category-hierarchy.csv # Category/subcategory tree
reproducibility-criteria.csv # 19 scoring criteria
corpus-summary.csv # Pipeline statistics
README.md # Data dictionary
METHODOLOGY.md # Pipeline and statistical methods
SCIENTIFIC_INTEGRITY.md # Known limitations and caveats
PROVENANCE.md # Versioning and data access
parameters/ # Parameter ontology (667 definitions + docs)
index.json # Master index as structured JSON
biological/ # Biofilm, growth kinetics, species
electrical/ # Voltage, current, power, impedance
materials/ # Electrodes, membranes, catalysts
+ 30 more category directories
standards/
five-parameter-checklist.md # Proposed minimum reporting standard
schemas/
parameter.schema.json # JSON Schema for parameter validation
scripts/ # Analysis and export code
generate-csv-exports.ts # Regenerate CSVs from JSON (no DB needed)
enrich-parameters.ts # Add references/dependencies
export-*.ts # Database export scripts
site/ # Parameter Explorer (React + Vite) — see docs/EXPLORER.md
| Metric | Value |
|---|---|
| Papers in corpus | 23,332 |
| Verified MES papers | 8,904 |
| Paper-parameter values (ontology-mapped) | 18,113 |
| Raw extractions (quality-filtered, labeled) | 25,566 |
| Parameter definitions | 687 |
| Categories | 13 |
| Subcategories | 119 |
System type distribution:
| Type | Papers |
|---|---|
| MES (general) | 11,384 |
| MFC (Microbial Fuel Cell) | 9,330 |
| MEC (Microbial Electrolysis Cell) | 1,375 |
| BES (Bioelectrochemical System) | 1,122 |
| MDC (Microbial Desalination Cell) | 109 |
| Other (MODELING, EFC, SMFC) | 12 |
- Average reproducibility completeness is only 23.1% across 289 scored papers
- Electrode spacing is reported in only 37% of publications
- Papers reporting all 5 checklist parameters show >40% IQR reduction in power density variability (p < 0.05, n = 63)
- Reproducibility score correlates with Coulombic Efficiency (r = 0.567, 95% CI [0.29, 0.76], n = 36)
See data/SCIENTIFIC_INTEGRITY.md for caveats.
| # | Parameter | Unit | Reporting Rate |
|---|---|---|---|
| 1 | Electrode spacing | cm | 37% |
| 2 | Electrode surface area | cm^2 | 62% |
| 3 | External resistance | ohm | 41% |
| 4 | Measurement method | text | 35% |
| 5 | Unit normalization basis | text | 48% |
Validated retrospectively on 63 papers. See standards/five-parameter-checklist.md.
npm run export:csv # Regenerate CSVs from JSON (no DB)
DATABASE_URL="..." npm run export:all # Full export including databaseSee scripts/README.md.
Read data/SCIENTIFIC_INTEGRITY.md before using this data. Key issues:
- P-values use a custom approximation, not standard t-distribution
- Confidence intervals on correlations via Fisher z-transform; several span zero
- Reporting rates are category-level averages masking per-criterion variation
- 90% of raw database extractions are noise (filtered out of published CSVs)
- 62% of papers in the corpus may not be directly about MES (use
verified_mesflag) - 66% of parameter definitions have zero usage in the corpus
@misc{mess_parameters_2026,
author = {{MESSAI Community}},
title = {{MESS-Parameters: Open Research Dataset and Parameter
Ontology for Microbial Electrochemical Systems}},
year = {2026},
publisher = {GitHub},
url = {https://github.com/Messai-io/MESS-Parameters},
version = {0.2.0},
note = {687 parameters, 23,332 papers, 18,113 parameter values}
}See CONTRIBUTING.md. Fork, add to parameters/, update index.json, submit PR.
CC BY 4.0 — free to share and adapt with attribution.