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

spliceJAC: transition genes and state-specific gene regulation from single-cell transcriptome data.

Mol Syst Biol · 2022
L1 53/100 PQI 85
⚑ 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
Scoring basis — itemised

Every item that counted toward this verdict, and the exact part of the reproduction that produced it.

Supporting (toward a concern)
Content-critical question only partially held
+2 pts
From: Q7 · Core claim 🟡
Content-critical question only partially held
+2 pts
From: Q8 · Severity of the miss (overall human judgment) 🟡
Minor / cosmetic deviation
+1 pts
From: Q3 · Location of the main deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q4 · Cause of the deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q6 · Severity of the deviation 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +5
✓ What held up
  • Same input data as the authors
  • Reported values were directly comparable
  • Reported values are derivable from the shared data
What did not (or only partly)
  • 🟡A deviation arose in the data or preprocessing
  • 🟡A deviation was attributed to the published material
  • 🟡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
53/100
Reproducibility score
1.2 SD below mean
vs. all fields · 1173 studies
🎯 Scores higher than 14% of all assessed papers rank 1005 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

Described well enough to reproduce 1:1. The brief's harvested code link (BoolODE) is only the third-party synthetic-data simulator; the paper's actual tool is spliceJAC (first author Bocci), pip splicejac==0.0.1, repo federicobocci/spliceJAC @4939ca9 (MIT). It ships the input (scVelo pancreas) AND the author reference output (results/export_output/transition_score.csv), making the pancreas transition-gene result (Fig 3) a clean target. On «our HPC» we built the pinned env (numpy1.23.2/scanpy1.9.1/scvelo0.2.4/anndata0.8.0/sklearn1.1.2), replicated the repo's Transitions.ipynb pipeline, and regenerated the 7-transition per-gene transition-score table. RESULT: the spliceJAC computation reproduces essentially EXACTLY (per-transition Pearson 0.998-1.000, Spearman 0.985; deterministic seed=100). The endocrine biology reproduces: Iapp->Beta, Gcg->Alpha, Ghrl->Epsilon are exact top-1 matches and Sst->Delta is top-5. The ONLY divergence is the upstream 50-highly-variable-gene SELECTION: 39/50 overlap, and 4 of the reference's top genes (Spp1, Neurog3, Chga, Chgb) fell outside our HVG set, flipping 4/7 top-1 calls; the shipped CSV was most likely generated by the repo's now-removed Pancreas.ipynb (referenced in .ipynb_checkpoints) rather than the current Transitions.ipynb. No fabrication signal: near-perfect numeric agreement on every commonly-selected gene confirms the reported values are genuine pipeline outputs. NOT attempted (the hard 20%): BoolODE synthetic-circuit GRN-inference benchmarking (AUPRC/AUROC, Fig 2), the A549 EMT/GSE147405 analysis, and grn_statistics network stats (its nx.from_numpy_matrix is gone in networkx>=3 and is not needed for transition scores).

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 53
    assessed: 2026-06-15 ⛓ b1e4bb75b1c5
✎ 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
2026-06-15
Rubric version
v1.0
Assessed by
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-15
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: opus
Founding hypothesis

Can a bottom-up, theory-oriented framework that models multivariate mRNA splicing from scRNA-seq infer cell state-specific gene-gene regulatory interactions and predict driver genes that govern transitions between cell states?

Core claims
  • spliceJAC quantifies multivariate mRNA splicing from unspliced/spliced count matrices to construct cell state-specific gene-gene (Jacobian) interaction matrices. method
  • Spectral (linear stability) analysis of inferred Jacobian matrices predicts the most unstable genes that drive transitions between cell states. method
  • spliceJAC accurately reconstructs ground-truth state-specific interaction matrices from in silico multistable circuits. finding
  • spliceJAC outperforms existing GRN inference methods (Beeline pipeline) on synthetic circuit data. finding
  • Applied to mouse pancreas endothelium development and A549 EMT, spliceJAC predicts state-specific signaling roles and new transition genes. finding
  • Linearizing nonlinear splicing dynamics around each attractor enables inference of attractor-specific gene regulation (Eqs. 5a-5b). mechanism
  • spliceJAC is provided as an easy-to-use Python package with downstream analysis options. resource
  • spliceJAC correctly distinguishes stable from unstable cell states, predicting positive eigenvalues at unstable fixed points. finding
Experimental setups
Assay System Perturbation Readout Platform
In silico simulation of bistable toggle switch circuit Two-gene (X,Y) mutual-inhibition synthetic circuit Small stochastic perturbations around fixed points Inferred state-specific gene-gene interaction matrices vs. analytic ground truth
In silico simulation of three-gene mutual-activation loop Three-gene monostable synthetic circuit Small perturbations around fixed point Inferred interaction matrix vs. ground truth
In silico simulation of EMT regulatory circuit EMT circuit (miR-34, miR-200, ZEB, SNAIL, TGF-beta, E-cadherin, N-cadherin) tuned to tristability TGF-beta inducer level set for coexistence of epithelial/hybrid/mesenchymal states Inferred interaction matrices and eigenvalue spectra per state
GRN inference benchmarking (Beeline pipeline comparison) In silico circuit states (epithelial, hybrid E/M, mesenchymal) none Matrix element absolute difference, fraction of correct signs, AUPRC ratio Beeline pipeline
scRNA-seq splicing analysis (RNA velocity-style unspliced/spliced) Mouse pancreas endothelium development none State-specific gene regulation, signaling hubs, transition driver genes
scRNA-seq splicing analysis (RNA velocity-style unspliced/spliced) A549 lung carcinoma cells undergoing EMT TGF-beta-induced EMT (cell line context) State-specific gene regulation, transition genes, DEGs
Key results
  • spliceJAC reconstructs toggle switch interaction matrices for both stable fixed points with high precision, capturing reversed interaction amplitudes between states.
  • spliceJAC correctly recovers the interaction matrix of the three-gene activation loop.
  • spliceJAC robustly recovers epithelial, hybrid E/M, and mesenchymal interaction matrices from the tristable EMT circuit.
  • spliceJAC predicts a positive eigenvalue at the unstable fixed point of the toggle switch, indicating instability.
  • For mixtures of cells from multiple EMT states, spliceJAC predicts the largest eigenvalue to be positive or near zero, flagging unstable mixing.
  • spliceJAC outperforms existing GRN inference methods on absolute matrix difference, correct sign fraction, and AUPRC ratio.
  • Inference remains robust when genes are removed one at a time from the input count matrix.

Statistical methods review

Model: opus

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 computational methods paper presenting spliceJAC, a tool that infers cell state-specific gene–gene (Jacobian) interaction matrices from unspliced/spliced scRNA-seq counts and uses linear stability (eigenvalue) analysis to predict transition driver genes. Rather than classical hypothesis testing, the statistical approach centers on regression-based parameter inference and benchmarking: performance is evaluated against analytically derived ground-truth circuits using metrics such as absolute matrix-element difference, fraction of correctly signed matrix elements, and AUPRC (via the Beeline pipeline), and applied to real pancreas-endothelium and A549 EMT datasets. Results are reported largely as inferred matrices, eigenvalue spectra, instability/signaling scores, and comparisons with prior differentially expressed gene analyses.

Replicationunclear Sample sizeStated that the number of cells in a cell state sets an upper limit on inferable interactions (genes cannot exceed cells); robustness assessed by removing genes one at a time. No formal sample-size/power analysis in visible text. GroupsDistinct cell states/attractors (e.g., epithelial, hybrid E/M, mesenchymal); spliceJAC vs other GRN inference methods Pairingna Randomization/blindingna Dispersionunclear Multiplicity correctionnone stated in visible text
Statistical tests used
Test Applied to n Assumptions
Linear regression to infer attractor-specific gene–gene interaction (Jacobian) matrices from steady-state assumption Inference step within each cell state (Eqs. 5a–5b; Fig 1B–C, Fig 2) number of cells within a given cell state (stated as upper limit on number of inferable interactions); exact n not given in visible text stated
Eigenvalue (linear stability / spectral) analysis of the inferred Jacobian Stability of cell states and identification of transition driver genes (Fig 1E, Fig EV1) stated
Benchmark metric: absolute difference between ground-truth and inferred matrices In silico EMT circuit comparison across epithelial/hybrid/mesenchymal states (Fig 2I) na
Benchmark metric: fraction of correctly signed matrix elements In silico circuit comparison (Fig 2J) na
AUPRC (area under precision–recall curve) ratio, normalized per state Comparison against GRN inference methods in the Beeline pipeline (Fig 2K) na
Approaches that could also have been used
  • Method performance was summarized with point metrics (absolute matrix difference, fraction of correctly signed elements, AUPRC ratio) across in silico circuits.
    Could also: Reporting these metrics with dispersion across multiple simulation replicates (e.g., mean ± SD/95% CI or full distributions via boxplots) and bootstrap confidence intervals. — Adding variability estimates would convey how stable the benchmark rankings are across stochastic simulations and would complement the central-tendency comparison.
  • spliceJAC was compared against other GRN inference methods using normalized AUPRC ratios.
    Could also: Complementing AUPRC with AUROC, early-precision, and a paired statistical comparison (e.g., Wilcoxon signed-rank across circuits) of per-circuit scores. — Multiple complementary metrics and a paired test across shared circuits would give a more complete picture of relative performance under class imbalance.
  • Cell annotations (clusters) were assumed to correspond to stable cell states, with stability analysis offered as self-consistent validation.
    Could also: Sensitivity analyses across alternative clustering resolutions or annotation schemes, reported as agreement statistics. — Quantifying how inference outputs vary with upstream clustering choices would characterize robustness to the state-definition step.
  • Interaction matrices were inferred via linear regression under a steady-state, near-attractor linearization with a unique-solution constraint (genes ≤ cells).
    Could also: Regularized regression (e.g., ridge/LASSO/elastic-net) with cross-validated penalty selection. — Regularization would allow stable estimation when gene count approaches or exceeds cell number and could improve generalization in high-dimensional, noisy settings.
  • Robustness was assessed by removing genes one at a time and re-running inference.
    Could also: Bootstrap or subsampling of cells and repeated resampling, reporting the distribution of inferred edges/eigenvalues. — Resampling-based stability assessment would quantify uncertainty in individual inferred interactions and driver-gene rankings.
  • Transition driver genes were prioritized by an instability score derived from the unstable manifold projection.
    Could also: Attaching empirical significance or confidence to driver-gene scores via permutation/null-model comparison. — A null distribution would help distinguish genuinely high-instability genes from those expected under randomized interaction structure.
Software: spliceJAC (Python package) · Beeline GRN-inference benchmarking pipeline

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.

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.

Citations
63
Impact: high
Foundation confidence
None of its references are in our reproducibility record yet — its foundation cannot be assessed.
Topics

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.

GSE147405 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet

What was reproduced

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

Scope — pmid-36321549 (spliceJAC)

Paper: Bocci F, Zhou P, Nie Q. spliceJAC: transition genes and state-specific gene regulation from single-cell transcriptome data. Mol Syst Biol 2022. PMID 36321549 · PMCID PMC9627675 · DOI 10.15252/msb.202211176.

What spliceJAC is

A Python tool that uses spliced + unspliced scRNA-seq counts (RNA velocity inputs) to fit a cell-state-specific linear Jacobian by regression, then does stability analysis: the leading eigenvector of the Jacobian at an unstable steady state ranks transition (driver) genes for each cell-state transition.

Code situation (P16 note)

  • The link harvested into the brief (github.com/Murali-group/BoolODE) is not the paper's analysis code — BoolODE is the third-party synthetic-data simulator the paper used only for its benchmarking circuits (toggle switch, EMT circuit…).
  • The paper's actual analysis tool spliceJAC is released by the first author: github.com/federicobocci/spliceJAC (MIT, default branch main, pip splicejac==0.0.1). This is what we reproduce — running the paper's own tool on the paper's own bundled data is the most faithful reproduction.

In scope (pipeline-derived, attempted)

result pipeline data reproducible?
Pancreas transition genes / per-gene transition scores (Fig 3) spliceJAC estimate_jacobiantransition_genes bundled datasets/Pancreas/pancreas.h5ad (scVelo pancreas endocrinogenesis) YES — repo ships both the input pancreas.h5ad AND the reference output results/export_output/transition_score.csv (7 transitions × ~50 genes). We regenerate the table and compare 1:1.

The shipped transition_score.csv is the concrete, machine-checkable reference. Top-scoring genes per transition are the canonical lineage markers (Ins1/Ins2/Iapp → Beta, Gcg → Alpha, Sst → Delta, Ghrl → Epsilon), i.e. the biology the paper reports for the pancreas in Fig 3.

Out of scope (not attempted — 80/20)

  • BoolODE synthetic-circuit benchmarking (AUPRC/AUROC vs other GRN methods, Fig 2 / EV figs): would require regenerating BoolODE simulations + the separate jacobian-inference-benchmarking repo. Larger effort, deferred (the hard 20%).
  • A549 EMT / TGFβ (GSE147405) analysis: not bundled; would require fetching + full RNA-velocity preprocessing. Deferred.
  • All wet-lab / manual / schematic figure content — out of scope by definition.

Reproduction strategy

  1. «our HPC» SLURM job, conda env pinned to spliceJAC requirements.txt (numpy 1.23.2, scanpy 1.9.1, scvelo 0.2.4, anndata 0.8.0, sklearn 1.1.2 …).
  2. Load bundled pancreas.h5ad, follow the repo's Pancreas/Transitions notebook steps, run estimate_jacobian + transition_genes.
  3. Export our transition_score table; compare to repo transition_score.csv (correlation + top-gene overlap per transition).

All heavy compute on «our HPC»/«infra»; «host» holds only small result artifacts.

Figures / tables: Fig 3
C4
Reported
Pre-endocrine->Beta top transition gene = Iapp (0.662)
Reproduced
Iapp (0.601)
within tolerance
C5
Reported
Pre-endocrine->Alpha top transition gene = Gcg (0.877)
Reproduced
Gcg (0.691)
within tolerance
C7
Reported
Pre-endocrine->Epsilon top transition gene = Ghrl (0.982)
Reproduced
Ghrl (0.979)
within tolerance
C6
Reported
Pre-endocrine->Delta top transition gene = Sst (0.925)
Reproduced
Sst in our top-5; top-1 Pyy (gene absent from ref set)
partial
C1
Reported
Ductal->Ngn3 low EP top gene = Spp1 (0.887)
Reproduced
Spp1 not in our 50-HVG set; top-1 Cyr61
did not match
C2
Reported
Ngn3 low->Ngn3 high top gene = Spp1 (0.868)
Reproduced
Spp1 not in our 50-HVG set; top-1 Mdk
did not match
C3
Reported
Ngn3 high->Pre-endocrine top gene = Chgb (0.828)
Reproduced
Chgb/Chga not in our 50-HVG set; top-1 Pyy
did not match
C8
Reported
Full per-gene transition-score table, 7 transitions (author transition_score.csv)
Reproduced
mean Pearson 0.9997, mean Spearman 0.985 across 39/50 shared genes
within tolerance

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 53/100

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)
Scoring basis — itemised

Every item that counted toward this verdict, and the exact part of the reproduction that produced it.

Supporting (toward a concern)
Content-critical question only partially held
+2 pts
From: Q7 · Core claim 🟡
Content-critical question only partially held
+2 pts
From: Q8 · Severity of the miss (overall human judgment) 🟡
Minor / cosmetic deviation
+1 pts
From: Q3 · Location of the main deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q4 · Cause of the deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q6 · Severity of the deviation 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +5

The spliceJAC algorithm reproduces essentially exactly — per-transition Pearson 0.998–1.000 and mean Pearson 0.9997 across the 39/50 shared genes, with no fabrication signal. The sole deviation is upstream HVG selection: 4 of the reference's top genes (Spp1, Neurog3, Chga, Chgb) fell outside our 50-gene set, flipping 4/7 top-1 calls (C1–C3, C6), most plausibly because the shipped CSV came from the repo's now-removed Pancreas.ipynb rather than the current Transitions.ipynb. This is a preprocessing/code-version issue on a mix of our pipeline choice and an author-side code mismatch — explainable and non-severe; the well-separated endocrine branches (Iapp→Beta, Gcg→Alpha, Ghrl→Epsilon, Sst→Delta top-5) confirm the central biology.

🤝
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.

198 k
tokens (I/O) · 19.5 M incl. cache
22 min
runtime · 0.04 CPU-h
1.9 GB
peak RAM
6 (3 failed)
HPC jobs
hummel
machine