Corpus 1,272 assessed · 1,173 scored · 643 reproduced ≥75 · 168 flagged ·∅ 74.1/100
← New search

RegulonDB 11.0: Comprehensive high-throughput datasets on transcriptional regulation in Escherichia coli K-12.

Microb Genom · 2022
L1 61/100 3/4
⚑ Flagged for review — a reproduced result did not match the reported value

Provisional — an automated or curator check raised a specific concern and points reviewers here. This is NOT a final assessment and not a determination about the authors.

Why this verdict

The main results reproduced, with only marginal, non-material deviations.

Reproduced on the brainbox compute brainarbeit.com
✓ What held up
  • Reported values were directly comparable
What did not (or only partly)
  • 🔴Could not use the authors’ exact input data
  • 🟡A deviation arose in the data or preprocessing
  • 🔴A deviation was attributed to the published material
  • 🟡Reported values were not (fully) derivable from the shared data
  • 🟡The deviation was non-trivial in magnitude
  • 🟡The central claim did not (fully) hold under reproduction
  • 🟡Overall, the reproduction showed a material discrepancy
How its reproducibility compares
61/100
Reproducibility score
0.7 SD below mean
vs. all fields · 1173 studies
🎯 Scores higher than 22% of all assessed papers rank 906 of 1173 scored

A 0–100 reproducibility-quality score from the per-question grades, shown as a z-score: standard deviations above (+) or below (−) the mean of comparable assessments.

Reproduction agent’s raw note

RegulonDB 11.0 is a database/resource paper; in-scope reproducible results are the Table-1 HT dataset/object counts. KEY FINDINGS: (1) the cited archival data DOI zenodo:6376425 is EMPTY (526-byte docker-app pointer), so the only data source is the LIVE portal, now at v14.5.0 (4 years past v11.0) via its GraphQL API. (2) Despite that evolution, reproduction is strong: 7/20 claims EXACT - 6 of 8 dataset counts (RNA-seq 1864, ChIP-exo 94, gSELEX 164, TU 5, TSS 16, TTS 5) plus ChIP-seq TFBS sites 5108 (from exactly 28 in-house datasets, matching the paper). (3) TU/TSS/TTS/gene object counts reproduce within ~2-10% as lower bounds (API pagination drops rows at high page sizes - validated by re-counting at limit=100). (4) ChIP-seq grew (29->168 datasets, peaks->20990) and DAP shrank (215->107) - documented DB evolution, not fabrication. (5) Author peak collections (ChIP-exo/gSELEX/DAP) are not exposed per-dataset through the API -> those 3 counts uncheckable. NOT attempted: CRF growth-condition F1 (no training data/code), manual/LT curation, web-app infra. No fabrication detected; the central reproducibility gap is archival (no frozen v11.0 deposit), recorded honestly. Verdicts provisional pending human audit.

💻 Code ↗ 🗄 Data: 10.5281/zenodo.6376425

These records describe the outcome of reproduction attempts carried out autonomously by brainbox using large language models (LLMs). They are not peer review, not an audit, and not a determination of error or misconduct by any author. A verdict reflects what one attempt could or could not reproduce — which may depend on data access, undocumented parameters, the computing environment, or the depth of effort — and not a judgement of the people who did the work. We can be wrong, and we correct mistakes quickly: every record carries a “report an error” button.

Assessment versions

Every reproduction run is kept as an immutable version — anchored to the data as it stood, with a tamper-evident chain hash. A rerun (e.g. after an author updates a deposit) adds a new version; the previous one stays on record.

  1. v1 current initial assessment Score 50
    assessed: 2026-06-19 ⛓ dcb8f45b4007
✎ I am an author of this paper

Updated or fixed a deposit, or is there an erratum? Ask us to re-run the metrics. We verify by email first; the new result is published as a new version with full history — nothing is overwritten.

Reason for the rerun

We email you a confirmation link first. The rerun is an objective re-measurement — it cannot change the verdict in your favour, only ask us to look again.

Provenance — full disclosure

When this reproduction was carried out, which methodology version was used, and by whom — so the record can be audited and checked independently.

Reproduced
2026-06-25
Rubric version
v1.0
Assessed by
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-19
no human curator yet
Last updated
2026-08-05

Provisional, curator- or AI-assessed, and independently checkable. A reproduction outcome states what one attempt could reproduce — not a judgement of the authors.

Deep full-text extraction

Model: sonnet
Founding hypothesis

The paper addresses whether a single resource can comprehensively integrate the large and rapidly growing body of high-throughput datasets on transcriptional regulation in Escherichia coli K-12 (TF binding, transcription start/termination sites, transcription units, and gene expression) in a way that is comparable, traceable, and reproducible, since no such resource previously existed.

Core claims
  • RegulonDB 11.0 is a radical upgrade offering access to more than 2000 high-throughput datasets on transcriptional regulation in E. coli K-12. resource
  • The resource integrates TF binding data from ChIP-seq, ChIP-exo, gSELEX and DAP-seq together with transcription start site, transcription termination site, transcription unit, and RNA-seq expression collections in a single place. resource
  • For ChIP-seq, both author-reported data and uniformly in-house reprocessed data are provided to enhance comparability, traceability, and reproducibility. method
  • An NLP and machine-learning assisted curation strategy was implemented to automatically extract detailed experimental growth conditions from metadata. method
  • No comprehensive resource previously existed that facilitated access to the diverse wealth of high- and low-throughput data on transcriptional regulation in E. coli K-12 in one place. finding
  • New browsing and visualization tools were added, including a nucleotide-resolution genome viewer and a metadata-query interface. resource
  • gSELEX binding datasets were curated using defined or inferred intensity thresholds, yielding 164 TFBS datasets for 121 TFs. finding
  • A biotin DAP-seq collection covering 215 TFs was incorporated from the supplementary material of a prior study. finding
Experimental setups
Assay System Perturbation Readout Platform
ChIP-seq E. coli K-12 MG1655 various TFs (in vivo binding) TF binding peaks/TFBS genomic coordinates SRA raw reads processed with cutadapt, Bowtie2, SnakeChunks/snakemake
ChIP-exo E. coli K-12 TFs including OxyR, SoxR, SoxS, UvrY, and others TF binding sites proChIPdb
gSELEX E. coli K-12 (in vitro) individual TFs binding intensity (%) relative to peak, target gene, peak location TEC database
DAP-seq (biotin-DNA affinity purification sequencing) E. coli K-12 (in vitro) individual TFs TF binding sites
RNA-seq (bulk expression profiling) E. coli K-12 various growth conditions and/or genetic backgrounds gene expression profiles
5′-end-protected RNA-seq variants E. coli K-12 none/various transcription start sites at nucleotide resolution
RNA-seq (full-length transcript mapping) E. coli K-12 none/various transcription termination sites and transcription units
Key results
  • Total high-throughput dataset collection expanded from 99 datasets (2019) to more than 2000 datasets. >20-fold
  • More than 500 high-throughput datasets for TF DNA-binding were incorporated.
  • 1864 RNA-seq datasets generated under different growth conditions and/or genetic backgrounds were incorporated.
  • 164 TFBS datasets corresponding to 121 different TFs were generated from gSELEX data.
  • A collection of experiments and metadata for 215 TFs was obtained using biotin DAP-seq.
  • 185 raw data files from 28 ChIP-seq datasets associated with 11 TFs were uniformly reprocessed.
  • RegulonDB was accessed an average of ~16300 times per year over the last 4 years.
  • 63 gSELEX datasets (41 TFs) were built using defined thresholds; 94 additional datasets (74 TFs) used the top 40 binding-intensity targets as cutoff.
Key statistics
  • count >2000 high-throughput datasets (total HT datasets available in RegulonDB 11.0)
  • count >500 (HT datasets for TF DNA-binding)
  • count 1864 (RNA-seq datasets across growth conditions/genetic backgrounds)
  • count 164 TFBS datasets (121 TFs) (gSELEX-derived TFBS collection)
  • count 215 TFs (DAP-seq TF collection)
  • count 28 ChIP-seq datasets, 185 raw files, 11 TFs (uniformly in-house processed ChIP-seq data)
  • mean ~16300 accesses/year (average RegulonDB portal access rate over the last 4 years)
  • count 63 datasets (41 TFs) with defined threshold; 94 datasets (74 TFs) with top-40 cutoff (construction method for gSELEX datasets)

Statistical methods review

Model: sonnet

A neutral, descriptive read of the statistical approach — what was done, and (for shared learning, not as criticism) what could also have been done.

This is a resource/database paper describing the curation, uniform processing, and integration of high-throughput datasets (ChIP-seq, ChIP-exo, gSELEX, DAP-seq, RNA-seq) for E. coli K-12 transcriptional regulation into RegulonDB, rather than a primary hypothesis-testing study. Statistical content is limited to aggregating results (including p-values for differential expression, peak-calling/motif-prediction statistics) as reported in the original source publications, alongside description of uniform bioinformatics pipelines (e.g., read trimming and alignment) used to reprocess a subset of ChIP-seq data. No dedicated inferential statistical framework for hypothesis testing across the aggregated collection is described in the available text.

Replicationunclear GroupsAggregation and comparison of independently generated HT datasets (ChIP-seq, ChIP-exo, gSELEX, DAP-seq, RNA-seq) against each other and against classic low-throughput data, rather than a controlled experimental comparison performed by this paper itself Pairingna Randomization/blindingna Dispersionunclear
Statistical tests used
Test Applied to n Assumptions
Unspecified statistical test for differential expression (test not named) Flagging target genes as showing 'changes in expression and a significant p-value for differential expression' when ChIP-seq experiments were linked to gene expression data in the same source publication, to annotate TF function as activator or repressor not stated
Unspecified peak-calling/motif-prediction statistics (test not named) Curation of TF binding peak and TFBS features from ChIP-seq/ChIP-exo/gSELEX/DAP-seq source publications not stated
Approaches that could also have been used
  • Significance calls for differential expression used to flag TF-target regulatory interactions were taken as reported by the original source publications, each presumably using its own test and threshold, without recomputation or a unified correction across the aggregated collection.
    Could also: A unified reanalysis applying one differential expression method (e.g., DESeq2 or edgeR) with a single multiple-testing correction (e.g., Benjamini-Hochberg FDR) across all curated RNA-seq datasets — This would let significance thresholds be compared consistently across datasets originating from different labs and pipelines, rather than relying on heterogeneous criteria from each source study.
  • For 94 gSELEX datasets not analysed by their original authors, target selection used a fixed rank-based cutoff (the top forty binding intensities) rather than a statistically derived threshold.
    Could also: Deriving a cutoff from the empirical intensity distribution (e.g., a percentile or FDR-based threshold estimated from a null/background model) — A distribution-based threshold can adapt to the number of true binding events per TF, whereas a fixed top-N rank may include weaker signals for TFs with few real targets or exclude true targets for TFs with many.
  • ChIP-seq peak/TFBS statistical values (from peak calling or motif prediction) were curated as provided by each publication's own pipeline and settings.
    Could also: Uniform reprocessing of all raw ChIP-seq data with one peak-calling tool and consistent significance threshold (e.g., MACS2 with a fixed q-value cutoff), as was done for the subset of 28 datasets reprocessed with SnakeChunks — Applying one tool and threshold across all datasets, rather than only a subset, would make peak significance and counts more directly comparable across the full collection.
Software: cutadapt · Bowtie 2 · snakemake (SnakeChunks workflow library) 6.10.0 · MongoDB 4.4.5 · Python 3.9

What was reproduced

The exact results taken into scope, with each reported value next to the value our attempt produced.

Scope — pmid-35584008 (RegulonDB 11.0)

Paper: Tierrafría et al. 2022, Microb Genom 8(5):000833. RegulonDB 11.0: Comprehensive high-throughput datasets on transcriptional regulation in E. coli K-12. PMCID PMC9465075 · DOI 10.1099/mgen.0.000833.

Nature of paper: A database/resource paper. Most reported numbers are database content counts (Table 1) produced by uniform-processing pipelines applied to publicly-deposited raw HT data. This is exactly the "third-party tool on the paper's data" case the brief blesses (P16): the genuine analysis code is the SnakeChunks ChIP-seq pipeline + standard tools, applied to GEO/SRA reads.

Code / data artifacts located

  • Doc repo: github.com/PGC-CCG/RegulonDB-HT — 99.7% HTML; it is the manual (Manual/, Mapping/README, Uniformized-Data/), not runnable analysis code.
  • Pipeline repo (real code): github.com/PGC-CCG/SnakeChunks — Snakemake NGS framework (R 69%, Python 25%) for ChIP-seq/RNA-seq. Maintainers Rioualen & van Helden. This is the ChIP-seq uniform-processing pipeline.
  • App repos: github.com/regulondbunam/{RegulonDBHT-Web,GraphQL-api, RegulonDB-wdpservice} — the web app, not analysis.
  • Zenodo 6376425 ("Data"): ⚠️ contains ONLY description.json (526 B) — metadata for a Docker app image (regulondb-app), pointing at the web-app repos. It does NOT contain the HT object data. (See dataset_profile.)
  • Actual data portal: https://regulondb.ccg.unam.mx → "Integrated Views & Tools" → "RegulonDB-HT datasets" (downloadable object files). Raw reads from GEO/SRA/ArrayExpress/DEE2/proChIPdb/TEC.

Pipelines described (Methods)

  • ChIP-seq uniform processing: cutadapt (q & len threshold 20) → Bowtie2 (local mode, genome NC_000913.3 MG1655) → MACS3 (--nomodel --shift 0 --extsize 200, q-val 1e-3) → RSAT matrix-scan (motifs from RegulonDB 10.5, thresholds from RSAT matrix-quality) for TFBS. → peaks + sites.
  • RNA-seq: HISAT2 → DESeq normalization (geometric-mean); QC PASS tagging.
  • Growth-condition extraction: Conditional Random Field (CRF) trained on 228 SOFT files → MCO term mapping.

Reported pipeline-derived results (Table 1) — reproduction targets

Object Strategy Datasets Objects (reported)
Gene expression RNA-seq 1864 4618 avg genes/dataset
ChIP-seq (TF binding) ChIP-seq 29 6585 peaks; 5108 sites
ChIP-exo ChIP-exo 94 23170 peaks
gSELEX gSELEX 164 35022 peaks
DAP-seq DAP-seq 215 19540 peaks
Transcription units (TU) RNA-seq 5 12347
TSS RNA-seq/dRNA-seq 16 68049
TTS RNA-seq 5 5326
CRF growth-cond. NLP/CRF (228 SOFT) F1 0.81 CV / 0.83 test; ~83% MCO

IN SCOPE (attempt)

  1. Deposited-object counts vs Table 1 (PRIMARY, cleanest 1:1). Download the uniformized HT object files (TSS/TTS/TU/ChIP peaks/sites) from the RegulonDB-HT portal onto «infra» and count objects per type, comparing to Table 1. This directly tests whether the released data delivers the reported counts.
  2. ChIP-seq pipeline reproduction (HARDER, stretch). For ONE ChIP-seq dataset with a known reported peak count + SRA accession, run SnakeChunks-equivalent (cutadapt→bowtie2→MACS3 with the exact params) on «our HPC» and compare the reproduced peak count to the deposited/reported value.

OUT OF SCOPE (not attempted, why)

  • Manual/LT curation, regulons, the curated knowledge graph — wet-lab/manual.
  • The web app / MongoDB / GraphQL infrastructure — not a computational result.
  • CRF growth-condition F1 — needs the 228 annotated SOFT training files + CRF code, neither shipped in the located repos (revisit if found).
  • Full re-processing of all 1864 RNA-seq / 502 TF-binding datasets — infeasible & unnecessary; the per-type counts (item 1) capture the pipeline output.

Hard blocker as of start

«our HPC»/VPN tunnel DOWN (ssh timeout to «host».«infra».uni-hamburg.de:22). All data download + compute i

Figures / tables: Table
C1
Reported
1864 RNA-seq datasets
Reproduced
1864
exact
C2
Reported
4618 avg genes/dataset
Reproduced
>=4517 (limit100 lower bound)
within tolerance
C3
Reported
29 ChIP-seq datasets
Reproduced
168 (DB grew v11->v14.5)
did not match
C4
Reported
6585/13167 ChIP-seq peaks
Reproduced
20990 (168 datasets)
did not match
C5
Reported
5108 ChIP-seq TFBS sites
Reproduced
5108
exact
C6
Reported
94 ChIP-exo datasets
Reproduced
94
exact
C7
Reported
23170 ChIP-exo peaks
Reproduced
uncheckable (not in API)
did not match
C8
Reported
164 gSELEX datasets
Reproduced
164
exact
C9
Reported
35022 gSELEX peaks
Reproduced
uncheckable (not in API)
did not match
C10
Reported
215 DAP-seq datasets
Reproduced
107 (recurated)
did not match
C11
Reported
19540 DAP-seq peaks
Reproduced
uncheckable (not in API)
did not match
C12
Reported
5 TU datasets
Reproduced
5
exact
C13
Reported
12347 transcription units
Reproduced
>=11829 (95.8%, lb)
within tolerance
C14
Reported
16 TSS datasets
Reproduced
16
exact
C15
Reported
68049 TSS objects
Reproduced
>=66444 (97.6%, lb)
within tolerance
C16
Reported
5 TTS datasets
Reproduced
5
exact
C17
Reported
5326 TTS objects
Reproduced
>=4826 (90.6%, lb)
partial
C18
Reported
CRF F1 0.81 cross-val
Reproduced
not-attempted
partial
C19
Reported
CRF F1 0.83 test
Reproduced
not-attempted
partial
C20
Reported
~83% MCO mapping
Reproduced
not-attempted
partial

Assessments & scoring basis

Each contributor’s verdict, the per-question basis, and the auditable, itemised worksheet behind it.

🤖 AI curator · claude (ai-curator room) · v1.0 L1 61/100

An automated assessment. It can flag an open question for review but can never, on its own, record a discrepancy verdict (C5) against a paper.

🔴1. Data identity
🟢2. Endpoint comparability
🟡3. Location of the main deviation
🔴4. Cause of the deviation
🟡5. Derivability / plausibility
🟡6. Severity of the deviation
🟡7. Core claim
🟡8. Severity of the miss (overall human judgment)
🤝
Reproduced automatically — and fairly

Automated reproduction checks whether a published result can be regenerated from the paper’s described methods and shared data. When something does not reproduce, that is not a claim of error or misconduct — most often it reflects under-described methods, software or environment differences, or gaps in data access, and some of the pre-print papers in the queue may carry issues their authors had no part in. The goal is shared awareness that rigorous, fully-described methods help everyone — never a judgement of any author.

Are you an author? We would genuinely like to hear from you — to clarify the record, add data or code, re-run the pipeline after an accession update, and publish your response right next to the assessment. Everything here is open and auditable.

🚩 Report an error in this record

Spotted something wrong — a verdict you’d contest, a data or value error, or a private detail that slipped through? Tell us, with a short justification. Authors and readers are equally welcome to write in; we review every report.

Prefer email, or the form below not working? Contact us at support@doesitreproduce.com.

Reproduction footprint

claude-opus-4-8

Measured resources invested to assess this paper — sanitised (machine class only, no job ids/paths). Compute = HPC accounting (SLURM); tokens = the AI agent's session.

105.8 k
tokens (I/O) · 5.6 M incl. cache
12 min
runtime
Per-job HPC accounting not captured for this run — the runtime shown is the reproduction’s measured wall-clock time.