Computational modeling demonstrates that glioblastoma cells can survive spatial environmental challenges through exploratory adaptation.
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.
- ✓Same input data as the authors
- ✓Reported values were directly comparable
- ✓No relevant deviation in data/preprocessing
- ✓No authors-side cause for any deviation
- ✓Reported values are derivable from the shared data
- ✓Any deviation was negligible
- ✓Overall, the reproduction was clean
- 🟡The central claim did not (fully) hold under reproduction
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
Described well enough -> clean 1:1 reproduction. The paper is a computational-modeling study (PhenoExploreR exploratory-adaptation model on Ivy GAP glioblastoma data); the repo (oricel/PhenoExploreR @79078b5) is self-contained, bundling its own analysis-ready data (data/IVY.rda) so no GEO download was needed. Re-running the repo's own computePDD() and the variance aggregation on «our HPC» reproduced all deterministic outputs of the author's shipped expected-results notebook (examples/IVY_overview.html) EXACTLY at printed precision: 5/5 sample counts, 10/10 pairwise GDD values (to 3 decimals), 5/5 per-location mean pathway-activity variances. Only external dep (HistogramTools::minkowski.dist) was shipped as a verbatim base-R shim to avoid an unrelated heavy CRAN compile tree; all other code is unmodified repo source. NOT attempted (hard ~20%): the stochastic wander()/explore() Monte-Carlo simulations (IVY_runs.Rmd) -- no pinned RNG seed, saved RDS not shipped, README states results vary run-to-run, so not a 1:1 numeric target; and figure aesthetics / TCGA DE / wet-lab data generation (out of scope). No fabrication signal: every checked value is derivable from and identical to the shipped data+code.
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 100assessed: 2026-06-15 ⛓ 759cba52165b
✎ 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
- 2026-06-15
- 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: opusThe paper tests whether exploratory adaptation—propagation of small random perturbations through the gene-regulatory network into phenotypic changes—explains how glioblastoma (GBM) cells from different anatomic tumor regions adapt to spatial micro-environmental pressures and thereby drive recurrence dynamics.
- ★ Stochastic exploration of the gene-regulatory network structure confers enhanced adaptive capacity, enabling GBM cells to converge to new target phenotypes in novel environments. finding
- ★ Some GBM cell phenotypes are inherently plastic while others are intrinsically rigid with respect to phenotypic transitions. finding
- ★ Even with exploratory capacity, phenotypic trajectories are constrained to pass through specific spatial-phenotypic ranges (e.g., intermediate IT states between CT and LE). finding
- ★ A dynamical, biased exploratory-adaptation model in which reaching environment-compatible phenotypic states reduces further exploration (exploratory capacity D decreases upon convergence). method
- ★ New notions of phenotype (vector of pathway-activity distributions) and phenotype distance (Pathway Distribution Distance and Global Distribution Distance) to quantify adaptation. method
- ★ Three dominant spatial phenotypic trajectories were identified: CT→CTpan (hypoxic), CT→CTmvp (oxygen-rich), and CT→IT→LE (brain spread). finding
- Reversal of adaptation requires smaller phenotypic changes than the forward path (LE reverting to CT is closer to target than CT reaching LE). finding
- Targeting intermediate phenotypes is proposed as a treatment roadmap to mitigate acquired plasticity and hinder adaptation. resource
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| Transcriptomic profiling / gene expression analysis (computational reuse of patient data) | GBM patient tumors, 41 patients across anatomic regions CT, LE, IT, CTpan, CTmvp | none | gene expression and pathway activity per anatomic region | Ivy Glioblastoma Atlas Project (Ivy GAP) |
| Differential gene expression / variability selection | glioma cohort (low-grade glioma vs GBM) | none | differentially expressed, high-variability GBM-focused gene list (top 10 percentile of variation) | The Cancer Genome Atlas (TCGA) |
| Gene-regulatory network construction | GBM-focused gene list with regulating TFs | none | network of 4121 nodes and 78,686 directed edges (avg degree 19.09) | RegNetwork; pathway definitions from KEGG (n=182 pathways) |
| Gene-set overlap analysis | 25 glioma-related curated gene sets | none | percentage overlap of expressed genes with selected gene list | MSigDB |
| Computational dynamic simulation of exploratory adaptation | in silico single-sample GBM cell model along spatial trajectories | stochastic perturbation of regulatory network (exploratory capacity D); intrinsic control dJ=0 vs exploratory dJ>0 | pathway activity distributions, global distance, number of converged pathways over time (t up to 1000) | — |
| Data reduction / Between Group Analysis (BGA) and clustering | Ivy GAP samples by location | none | spatial phenotypic trajectory visualization and location-specific clustering | — |
- – Constructed regulatory network from GBM-focused genes and TFs 4121 nodes, 78,686 edges, average degree 19.09
- ▲ Exploratory adaptation simulation (CT→CTmvp) reduced global distance to target and converged a substantial fraction of pathways by t=1000 final distance 13.83; 33.4% pathways converged (D(t=0)=0.1)
- ▼ Reversal path is closer to target than forward path for CT↔LE reverting LE→CT distance 11.88 vs original endpoint 19.92
- – Three well-defined spatial trajectories identified via shortest phenotype distance
- ▲ Pathway Distribution Distance (PDD) and differential pathway activity (DA) are strongly correlated across locations Pearson 0.5–0.7
- – Narrow angles between original and reversal paths indicate constrained phenotype exploration space
- count 4121 nodes, 78,686 directed edges (GBM regulatory network size)
- other average degree 19.09 (regulatory network connectivity)
- count 41 (GBM patients in Ivy GAP dataset)
- count 182 (KEGG pathways used for phenotype definition)
- correlation Pearson's coefficient 0.5–0.7 (correlation between PDD and differential pathway activity (t-test))
- other distance 13.83; 33.4% converged pathways (CT→CTmvp simulation at t=1000, D(t=0)=0.1)
- other 11.88 vs 19.92 (reversal (LE→CT) vs original endpoint global distance)
- other top 10 percentile of variation (selection threshold for GBM-focused gene list from TCGA)
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.
This computational modeling paper uses patient-derived transcriptomic profiles from 41 GBM patients across five anatomic tumor regions (Ivy GAP) to define phenotype distances and infer spatial-phenotypic trajectories. Statistical significance of the Global Distribution Distance (GDD) was evaluated with permutation tests, while pathway activity differences between locations were assessed with unpaired two-sided t-tests corrected for multiple comparisons via Benjamini-Hochberg. Between Group Analysis (BGA) and PCA provided dimensionality reduction for visualization. The primary outputs are simulation-derived (convergence rates, pathway dynamics) rather than inferential, with statistical tests serving mainly to validate the distance metrics.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Permutation test | Statistical significance of Global Distribution Distance (GDD) between anatomic location pairs | 41 patients across 5 locations (exact per-location n not stated in provided text) | not stated |
| Unpaired, two-sided t-test (BH-adjusted) | Significance of pathway activity variations between every pair of anatomic locations | 41 patients; per-location n not stated in provided text | not stated |
| Pearson correlation | Correlation between Pathway Distribution Distance (PDD) values and differential pathway activity (number of significantly different pathways) | 182 KEGG pathways; pairwise location comparisons | not stated |
| Between Group Analysis (BGA) | Data reduction and visualization of spatial trajectories using pathway activity patterns across all samples | 41 patients, 5 locations | na |
| Hierarchical/unsupervised clustering | Grouping samples by pathway activity patterns to assess location-specific separation | 41 patients, 5 locations | na |
-
Pathway activity differences between location pairs were tested with unpaired two-sided t-tests↳ Could also: Mann-Whitney U (Wilcoxon rank-sum) test could also be applied for each pathway comparison — A non-parametric alternative makes no normality assumption; pathway activity distributions derived from pooled patient samples may not be Gaussian, and with varying per-location n a rank-based test would add robustness
-
Significance of Global Distribution Distance (GDD) was assessed with permutation tests applied to the composite distance measure↳ Could also: Bootstrap confidence intervals around GDD could also quantify uncertainty in the distance estimates — Bootstrapping would provide interval estimates of the distances themselves (not only p-values), which would complement the permutation-based significance framing and convey magnitude of uncertainty more directly
-
Agreement between the two distance metrics (PDD and differential pathway activity) was summarized with a Pearson correlation range (0.5–0.7) across location pairs↳ Could also: Spearman rank correlation could also be used to measure the same agreement — Spearman's coefficient does not assume a linear relationship between the two distance metrics and is less sensitive to outlier pathway pairs, which may be present when comparing across five anatomically distinct regions
-
BGA was used for multivariate dimensionality reduction and visualization of location-level phenotypic separation↳ Could also: PERMANOVA (permutational multivariate ANOVA, e.g., via vegan in R) could also formally test multivariate separation among the five location groups — BGA provides a visual summary of group separation; a PERMANOVA would add a formal omnibus test of whether phenotype centroids differ across all five locations simultaneously, complementing the pairwise distance analysis
-
Cross-sectional patient data (single time-point snapshots per region) were used to represent distinct stages along putative adaptation trajectories↳ Could also: Pseudotime trajectory inference methods (e.g., Monocle, diffusion pseudotime) applied to the same expression data could also order samples along continuous paths — Pseudotime methods were developed specifically to infer ordered dynamics from cross-sectional single-cell or bulk expression snapshots, and would provide a data-driven ordering that could complement or validate the model-defined trajectories
-
Exploratory capacity and convergence were summarized as mean global distance over time (single mean curve) for each simulated trajectory↳ Could also: Reporting the inter-sample spread (e.g., SD or IQR across the 41 simulated samples) alongside the mean trajectory could also characterize simulation variability — The paper notes individual sample traces in light blue and the mean in dark blue (Fig. 3b), but no dispersion summary statistic is reported; adding SD or IQR would quantify how heterogeneous adaptation dynamics are across patients, which is directly relevant to the paper's heterogeneity argument
Result convergence & founder nodes
Findings this paper shares with others that ran a comparable experiment. A node’s strength is how many independent papers report it (replication breadth) — not how often it is cited, so a heavily-replicated but under-cited founder still stands out.
-
Between-group analysis of Ivy GAP samples identifies three well-defined spatial phenotypic trajectories among GBM anatomic regions.other human gbm tumor 2019×1papers★ This paper is the founder (earliest)
-
Pathway Distribution Distance (PDD) is positively correlated with differential pathway activity across GBM anatomic locations (Pearson r=0.5–0.7), validating PDD as a spatial phenotypic metric.other human gbm tumor up 2019×1papers★ This paper is the founder (earliest)
-
Stochastic regulatory perturbation (exploratory adaptation, D>0) enables GBM cells to converge toward target spatial phenotype in silico, with 33.4% of pathways converged and global distance reduced to 13.83 by t=1000.other in-silico gbm cell up 2019×1papers★ This paper is the founder (earliest)
-
Narrow angles between original and reversal phenotypic paths in silico indicate constrained phenotype exploration space during GBM spatial adaptation.other in-silico gbm cell 2019×1papers★ This paper is the founder (earliest)
-
Reversal phenotypic path (LE→CT) achieves closer distance to target (11.88) than the original forward endpoint (19.92), indicating asymmetric and more accessible reversal in GBM phenotypic landscape.other in-silico gbm cell down 2019×1papers★ This paper is the founder (earliest)
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.
Data lineage
The datasets this paper uses (text-mined from the full text via Europe PMC), and which other assessed papers stand on the same data. A shared dataset is a factual link — not a judgement.
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
Scope — pmid-31836713
Paper: Celiku O, Gilbert MR, Lavi O. Computational modeling demonstrates that glioblastoma cells can survive spatial environmental challenges through exploratory adaptation. Nat Commun 2019; 10:5704. DOI 10.1038/s41467-019-13726-w. PMCID PMC6911112.
Code: https://github.com/oricel/PhenoExploreR (R package, language=R, default
branch master, pinned commit 79078b5a33386b211fd63ec4394b6a6c3732caa3,
pushed 2019-07-15, not archived, no license file).
Data: GEO GSE107559 (Ivy GAP-derived). NOTE: the analysis-ready data are
shipped inside the repo (data/IVY.rda, data/IVY_*.tsv,
data/TCGA_GBM_vs_LGG_FC.tsv, MSigDB gmt files), so no separate GEO download is
needed to reproduce the in-scope results — the repo is self-contained.
What kind of paper this is
This is a computational-modeling paper. The central model ("exploratory
adaptation" via intrinsic wander() and exploratory explore() random walks over
pathway-activity space) is implemented in PhenoExploreR. The model is applied to
Ivy GAP glioblastoma anatomical-region expression data.
In scope (pipeline-derived, deterministic → 1:1 reproducible)
The repo ships examples/IVY_overview.Rmd with its rendered IVY_overview.html,
which the README explicitly designates as "the expected results". The
deterministic outputs in that notebook are exact-reproducible from the shipped
data/IVY.rda using the repo's own functions:
- C1 — Samples per anatomical location (
table(IVY@design)): integer counts. - C2 — GDD (Global Distribution Distance) per location pair via the repo's
computePDD(..., asGDD=TRUE)(the scientifically meaningful, pure-deterministic function inR/PhenoExploreRDistances.R; usesHistogramTools::minkowski.dist). 10 location pairs. - C3 — Mean pathway-activity variance per location (aggregation of
IVY@activity).
Reference values for C1–C3 are taken from the repo's shipped IVY_overview.html
(the author-rendered "expected results").
Out of scope (the hard ~20% — skipped, with reason)
- Stochastic simulations (
wander()/explore()inIVY_runs.Rmd): the README states results "may differ as new simulations produce somewhat different results"; the seed is not pinned and the saved RDS outputs are not shipped ("not included … due to the large size"). Not exact-reproducible by design; the qualitative behavior (exploratory walk reaches the target distribution while intrinsic does not) would require many CPU-hours and statistical comparison, not a 1:1 value match. Per 80/20, attempted only if C1–C3 land cleanly. - Wet-lab / Ivy GAP data generation, TCGA differential expression, PCA/heatmap figure aesthetics, hist3D rendering: external or purely graphical, not pipeline values.
Pipeline named per result
C1–C3: PhenoExploreR computePDD + base-R aggregation on data/IVY.rda, run under
R on «our HPC» (conda env built inside the compute job).
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.
This is a clean, self-contained computational reproduction: re-running the repo's computePDD() on the bundled data/IVY.rda reproduced all 20 deterministic values (sample counts, 10 pairwise GDD to 3 dp, 5 pathway-variance means) bit-for-bit against the author's shipped expected-results notebook, with no fabrication signal. The deviation is essentially zero on everything checked — the only issue is coverage, on our side: the paper's central exploratory-adaptation claim rests on the stochastic wander()/explore() Monte-Carlo simulations, which have no pinned seed and no deposited outputs and were not attempted. So q5/q6 are clean green and overall quality is high (q8 green), but q7 is yellow because the headline conclusion was only confirmed in its deterministic supporting statistics, not the model dynamics themselves.
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.