Revealing the critical state and identifying individualized dynamic network biomarker for type 2 diabetes through advanced analysis methods on individual basis.
The main results reproduced: recomputed values matched the published ones within tolerance.
Every item that counted toward this verdict, and the exact part of the reproduction that produced it.
- ✓Any deviation was negligible
- 🟡Could not use the authors’ exact input data
- 🟡Reported values were only indirectly comparable
- 🟡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 central claim did not (fully) hold under reproduction
- 🟡Overall, the reproduction showed a material discrepancy
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
PARTIAL reproduction, honest. The linked repo github.com/JinlingY/NIG (author MATLAB code) ships only (a) a self-contained 9-node numerical simulation of a critical transition and (b) a precomputed-WGCNA BLCA (bladder-cancer) demo. It does NOT ship the paper's GK-rat data (GSE13268 adipose / GSE13269 muscle), the WGCNA network-construction code, or rat-specific inputs. RESULT: (1) C1 reproduced 1:1 EXACT — under GNU Octave 10.3 on «our HPC» («job», node n094, env built on node-local /tmp with libmamba) the NIG score spikes 10.9x exactly at the critical point P=0 (s=-0.001), value 0.191386, bit-identical to a prior independent seeded run; this is the paper's central method claim (Fig.1B/1C). (2) C2 (BLCA demo, NOT a paper claim) — the shipped pipeline loads every input and enters the NIG computation; only Octave-vs-MATLAB compat shims (zscore, 2-col module_id textscan) and the un-vectorized ~1e11-op loop prevent collecting the full numeric curve under Octave; supporting evidence that the pipeline is operable. (3) C3-C6 NOT ATTEMPTED (deliberate, per 80/20 + docs_insufficient): the paper's headline rat-T2D numbers require rebuilding an unshipped, under-specified GEO->WGCNA->edge-selection preprocessing pipeline; doing so would be guesswork rather than reproduction, and the chance of matching '343 nodes/789 edges' or specific DNB genes via guessed parameters is ~nil. No fabrication asserted — the gap is that these specific numbers are simply not regenerable from the shipped code+data, which a human auditor should note. Dataset profiling: GSE13268 (and companion GSE13269) are open rat microarray series of 101 samples each (verified via GEO eutils + a clean series-matrix parse of GSE13268: 101x31099, 0 NA); the paper analyzes a ~50-sample GK-rat subset, so the deposit is a SUPERSET of the stated N (n_reported 50 vs n_observed 101).
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.
-
v1 current initial assessment Score 75assessed: 2026-06-14 ⛓ 59308e99ffc7
✎ 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.
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
- 2026-06-22
- Rubric version
- v1.0
- Assessed by
-
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-15no 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: sonnetThe paper tests whether the critical (pre-disease) state during progression of complex diseases—specifically type 2 diabetes—can be detected on a single-sample, individual basis using three data-driven, model-free methods (sJSD, NIG, TNFE), and whether these methods can successfully identify the associated dynamic network biomarkers (DNBs).
- ★ sJSD, NIG, and TNFE methods can detect critical states/tipping points before disease deterioration using only a single sample finding
- ★ All three methods detect two critical states in type 2 diabetes progression in GK rats: 8 and 16 weeks for adipose tissue, 4 and 16 weeks for muscle tissue finding
- ★ sJSD method is more sensitive to the critical state but more susceptible to fluctuations in gene expression data finding
- ★ NIG and TNFE methods transform unstable gene expression data into network structural information, making their predictive indicators more stable/robust than sJSD finding
- ★ Under weak noise TNFE outperforms sJSD and NIG, while under strong noise NIG is more noise-resistant and outperforms sJSD and TNFE finding
- ★ The dynamic network biomarkers (DNBs) identified differ significantly across the three methods, with only 2 common DNBs at the first critical state and 3 common DNBs at the second critical state finding
- ★ sJSD, NIG, and TNFE are proposed/applied methods for identifying critical states and DNBs from network-level or distribution-level single-sample data method
- The 8-week and 16-week critical states in adipose tissue correspond respectively to insulin resistance and beta cell failure, consistent with clinical phenotypes finding
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| Numerical simulation using stochastic differential equations (Michaelis–Menten/Hill kinetics) | Simulated 9-node artificial gene regulatory network | Parameter P varied between -0.5 and 0.23 (bifurcation approached at P=0) | sJSD (ICI), NIG, and TNFE scores per node and overall | MATLAB R2023a (v9.14); Cytoscape v3.10.2 for network visualization |
| Microarray gene expression profiling (GSE13268) | Adipose tissue of GK (Goto-Kakizaki) rats | High-fat diet vs normal diet, sampled at 4, 8, 12, 16, 20 weeks | ICI/NIG/TNFE scores across time points; identification of DNBs | — |
| Microarray gene expression profiling (GSE13269) | Gastrocnemius muscle tissue of GK rats | High-muscle diet vs normal diet, sampled at 4, 8, 12, 16, 20 weeks | ICI/NIG/TNFE scores across time points; identification of DNBs | — |
| Protein-Protein Interaction (PPI) network visualization | DNB gene sets from adipose and muscle tissue | none | Network structure/module visualization | Cytoscape v3.10.2 |
| Functional and pathway enrichment analysis | Identified DNB gene lists (adipose and muscle) | none | Enriched pathways/functions validating DNB relevance to type 2 diabetes | Metascape v3.5; bioinformatics.com.cn (visualization) |
- ▲ Two critical states detected in adipose tissue (GSE13268) at 8 and 16 weeks by all three methods
- ▲ Two critical states detected in muscle tissue (GSE13269) at 4 and 16 weeks by all three methods
- ▲ Sudden increase in sJSD, NIG, and TNFE scores signaling critical transition at bifurcation value P=0 in simulated 9-node network
- – TNFE method outperforms sJSD and NIG in detecting early-warning signals under small/weak noise
- – NIG method provides stronger early-warning signals than sJSD and TNFE under large/strong noise, indicating greater noise resistance
- – Only two DNBs are common across all three methods at the first critical state, and three DNBs common at the second critical state
- – 8 weeks and 16 weeks in adipose tissue correspond to insulin resistance and beta cell failure respectively, consistent with clinical phenotypes
- ▲ ICI score of DNB significantly increased at both 8 and 16 weeks in adipose tissue
- count 9 nodes (Simulated artificial gene regulatory network used for method validation)
- other P = 0 (Bifurcation/critical transition parameter value in the simulated regulatory network)
- count 50 adipose tissue samples (Total samples in GSE13268 dataset)
- count 50 muscle tissue samples (Total samples in GSE13269 dataset)
- count 5 rats per diet group per time point (GK rats with normal diet vs high-fat (adipose) or high-muscle (muscle) diet at each of 5 ages)
- other 4, 8, 12, 16, 20 weeks (Five sampling time points (ages) for GK rats in both datasets)
- count 2 common DNBs (DNBs identified in common by all three methods at the first critical state)
- count 3 common DNBs (DNBs identified in common by all three methods at the second critical state)
Statistical methods review
Model: sonnetA neutral, descriptive read of the statistical approach — what was done, and (for shared learning, not as criticism) what could also have been done.
The paper applies three information-theoretic, single-sample, algorithm-based scoring methods — sJSD (producing an Inconsistency Index, ICI), NIG (Network Information Gain score), and TNFE (Temporal Network Flow Entropy score) — to gene expression time-series data from two GEO datasets (GSE13268, GSE13269) for type 2 diabetes in rat adipose and muscle tissue. Critical states are identified by visual inspection of sudden score increases across five time points (4, 8, 12, 16, 20 weeks). Method validation is performed on a 9-node simulated gene regulatory network, and DNB biological relevance is assessed via Metascape enrichment analysis. No classical null-hypothesis significance tests with p-values are applied to the main results.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Jensen-Shannon Divergence (JSD)-based Inconsistency Index (ICI) — single-sample score, not a classical hypothesis test | Detection of critical states across all time points in GSE13268 (adipose) and GSE13269 (muscle), and in 9-node simulated network | 50 samples per dataset (5 time points × 10 rats); 5 normal-diet rats as reference per time point | not stated |
| Network Information Gain (NIG) score using Network Flow Entropy (NFE) on individual-specific networks | Same time-course critical-state detection in GSE13268 and GSE13269 and simulated network | 50 samples per dataset; 5 normal-diet rats as reference per time point | not stated |
| Temporal Network Flow Entropy (TNFE) score using differential networks between consecutive time points | Same time-course critical-state detection in GSE13268 and GSE13269 and simulated network | 50 samples per dataset; 5 normal-diet rats as reference per time point | not stated |
| Gaussian distribution fitting to individual gene expression profiles (used internally by sJSD to convert expression to probability distributions) | sJSD method preprocessing step for all datasets | — | not stated |
| Metascape enrichment analysis (method-internal statistical test unspecified in paper text) | Functional and pathway enrichment validation of identified DNBs | — | not stated |
-
Critical states are identified by visual inspection of sudden increases in continuous ICI/NIG/TNFE score curves across five discrete time points↳ Could also: A permutation or bootstrap test could be applied to each score at each time point to generate a null distribution and derive a formal p-value or confidence interval for the observed score increase — Adding a null-distribution-based significance threshold would allow readers to distinguish signal from sampling noise quantitatively, complementing the visual inflection criterion
-
Gaussian distributions are assumed for individual gene expression profiles in the sJSD method↳ Could also: Empirical (kernel density) or nonparametric distributions could also be used to represent gene expression, avoiding the normality assumption — Gene expression data can be skewed or heavy-tailed; a nonparametric representation would make the ICI score robust to departures from normality without requiring assumption verification
-
Method comparison across noise levels is presented visually (score curves in Fig. 1D) without a quantitative performance metric↳ Could also: Receiver operating characteristic (ROC) analysis or area under the curve (AUC) could quantify detection sensitivity and specificity across noise levels for each method — A scalar performance metric such as AUC would allow direct, objective comparison of sJSD, NIG, and TNFE across the noise gradient rather than relying on visual curve interpretation
-
DNB membership overlap across the three methods is reported as raw counts (e.g., 'two common DNBs at the first critical state')↳ Could also: A hypergeometric test or Jaccard similarity coefficient with a permutation-based p-value could quantify whether the overlap between method-specific DNB sets exceeds chance expectation — Formalizing the overlap assessment would help characterize whether shared DNBs reflect genuine biological signal or coincidental co-selection given gene set sizes
-
Reference samples at each time point are the five normal-diet rats only, making each per-time-point reference a small group of n = 5↳ Could also: A leave-one-out or cross-validation scheme could also be used to assess the stability of the identified critical state timing and DNB composition when reference composition varies — With only five reference samples per time point, reference-set variability could influence score magnitudes; a resampling approach would characterize robustness of the critical-state calls to this variability
-
Score results are reported without any measure of within-condition variability (no SD, SEM, or CI plotted on score curves)↳ Could also: Bootstrapped confidence bands around score curves, or plotting individual-rat scores alongside group means, would also convey within-time-point spread — Showing variability alongside the central score trajectory would let readers assess whether the apparent sudden increase at the critical time point exceeds within-condition noise, especially given the small n per time point
Citation network
Where this publication sits in the reproducibility-weighted citation graph — what it is built on, and what is built on it. Citation data from OpenAlex.
No assessed neighbours yet — the network grows as more papers are assessed.
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
Reproduction scope — pmid-39890881
Paper: Li, Jing, Tian, Jiang (2025). Revealing the critical state and identifying individualized dynamic network biomarker for type 2 diabetes through advanced analysis methods on individual basis. Sci Rep. PMID 39890881 · PMC11785715 · DOI 10.1038/s41598-025-87438-1.
Code: https://github.com/JinlingY/NIG (MATLAB) — cloned to «infra»
«path».
Data (paper): GEO GSE13268 (GK-rat adipose) + GSE13269 (GK-rat muscle), 50 samples each,
5 time-points (4/8/12/16/20 weeks) × (5 normal-diet ref + 5 high-fat case).
What the paper does (computational pipeline)
The paper compares three single-sample critical-transition methods to find an individualized Dynamic Network Biomarker (DNB) signaling the pre-disease "critical state" of T2D:
- NIG — Network Information Gain (the authors' method; this repo).
- sJSD — single-sample Jensen–Shannon Divergence (comparison).
- TNFE — Temporal Network Flow Entropy (comparison; separate repo G-R-0/TNFE). Each builds individual-specific networks (WGCNA-derived) and scores genes; the top 5% scoring genes are the DNB. Critical time-points are where the aggregate score spikes.
What the NIG repo actually ships (decisive scoping fact)
The repo contains only (readme.txt, BLCA.rar, NIG_simulation.rar):
NIG_simulation/NIG_simulation.m(+adj_idx_gene.txt,adj_weigh_network.txt) — a self-contained 9-node synthetic-network simulation of a critical transition.BLCA/NIG_BLCA .m(+case_network.txt,control_network.txt,adj_gene_weight.txt,wgcna_gene_idx.txt,module_id.txt) — the NIG method on a bladder-cancer (BLCA) demo dataset with precomputed WGCNA networks (this is the method's original cancer demo, NOT in this paper).
The repo does NOT ship the GK-rat (GSE13268/9) expression matrices, the WGCNA
network-construction code, or any rat-specific inputs. The NIG .m consumes
precomputed networks (edge lists + weights + module ids); the upstream preprocessing
(GEO → normalization → WGCNA → "topology heterogeneity" edge selection) that produces
those inputs is unspecified in the repo and only loosely described in the paper.
In scope (attempted — clearly specified, low-hanging)
- S1 — Numerical simulation (paper Fig. 1B/1C): run
NIG_simulation.m. Reported claim: all three scores rise sharply / NIG peaks at the critical transition P = 0 (param index 20 of 29). Self-contained → directly gradeable. (primary data point) - S2 — NIG method executes on real WGCNA networks (BLCA demo): run
NIG_BLCA .m, obtain per-stage average NIG curve + top-5% DNB list. Confirms the shipped pipeline runs end-to-end and the algorithm is operable. (supporting; BLCA is not a paper claim, so it is reported, not graded against the paper.)
Out of scope (NOT attempted — the hard >20%, with reason)
- Rat T2D headline results — critical states adipose wk 8 & 16, muscle wk 4 & 16; Table 1 common DNB genes (Uba2, Eif3m @ wk8; Dennd1b, Ltn1, Chordc1 @ wk16); NIG network "343 nodes, 789 edges." Not reproducible from shipped artifacts: the rat WGCNA network inputs and their construction code are not provided (only BLCA + simulation ship), and the GEO→WGCNA edge-selection parameters are under-specified. Rebuilding that preprocessing would be guesswork, not a 1:1 reproduction → per the 80/20 rule, deliberately not attempted.
- sJSD / TNFE methods — separate tools; not the linked repo. Out of scope.
Outcome shape (expected)
Partial: the method's self-contained validation (simulation) reproduces 1:1, and the
code runs on its shipped real-network demo (BLCA), but the paper's reported T2D
numbers are not regenerable from the shipped code+data. This is a docs_insufficient-
adjacent gap for the headline results, recorded honestly rather than fabricated.
Assessments & scoring basis
Each contributor’s verdict, the per-question basis, and the auditable, itemised worksheet behind it.
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.
Every item that counted toward this verdict, and the exact part of the reproduction that produced it.
The repo's self-contained numerical simulation reproduces the paper's central method claim 1:1 — the NIG score spikes ~10.8× exactly at the critical transition P=0 (Fig.1B/1C), deterministic across 3 seeded Octave runs. However, the paper's actual GK-rat T2D headline results (critical weeks, Table-1 DNB genes, 343-node/789-edge network) are not regenerable: the repo ships only a BLCA demo + simulation, not the rat data, WGCNA construction, or edge-selection step, and the paper under-specifies these. The deficit is on the authors'/data-availability side (q4 red), not our method, and no fabrication is asserted — so q5/q7/q8 stay yellow per the data-restriction principle. Net: a strong partial reproduction of the method with an explainable auditability gap on the paper's own numbers.
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-4-8Measured resources invested to assess this paper — sanitised (machine class only, no job ids/paths). Compute = HPC accounting (SLURM); tokens = the AI agent's session.