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

Benefit of using interaction effects for the analysis of high-dimensional time-response or dose-response data for two-group comparisons.

Sci Rep · 2023
L1 92/100 3/4
Why this verdict

The main results reproduced: recomputed values matched the published ones within tolerance.

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.

Concordant (toward reproduced)
All content-critical questions reproduced
-4 pts
From: Q7 · Core claim 🟢
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Every question reproduced
-1 pts
From: “every question reproduced”
Total score -7
✓ What held up
  • 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
  • The central claim held under reproduction
  • Overall, the reproduction was clean
What did not (or only partly)
  • Every checked point held up.
How its reproducibility compares
92/100
Reproducibility score
1.0 SD above mean
vs. all fields · 1173 studies
🎯 Scores higher than 83% of all assessed papers rank 179 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 -> 1:1 reproduction. The repo (jcduda/gene_expression_interaction @62690c9) ships the analysis input 010_gse.RData (mouse-liver RNA-seq counts, 35727 genes x 64 samples), so the 'IfADo data on request' README note does NOT block reproduction. Re-ran the authors' own DESeq2 functions from the shipped counts on «our HPC» (conda DESeq2 1.38.0 / apeglm 1.20.0 / R 4.2.3) and regenerated Table 1 (Method I week3/6 DEG counts) and Table 2 (Method I vs Method II interaction). Result: 10/17 cells exact, 7/17 within-tol (off by 1-3 genes out of hundreds). The small deltas occur ONLY in the from-scratch rebuild and are explained by DESeq2 1.38.0 (the only Bioc-3.16 build on bioconda) vs the paper's 1.38.1 patch, perturbing a few genes at the exact padj=0.05 / |log2FC|=log2(1.5) boundary; sums conserved (deg_mod_1_up=1609 in both). Anti-fabrication cross-check: the authors' OWN cached result objects (DESeq2 1.38.1) regenerate Table 1 and Table 2 to the integer -> paper numbers are genuine, no fabrication. One fixed friction: shipped RData object is named 'gse_without_weeks' while code expects 'gse'. NOT attempted (hard ~20%): Table 3/Supp1 topGO GO enrichment (set.seed + annotation-DB-version fragile), Table 4 top-10 lists, Figs 1-6 (illustrative), and upstream FASTQ->counts (not described runnably; we reproduce from the shipped count object).

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 92
    assessed: 2026-06-14 ⛓ 119767b3fbee
✎ 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
2026-06-14
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: sonnet
Founding hypothesis

The paper examines whether explicitly modeling the interaction effect (IE) between two factors (e.g., diet and time) in RNA-seq differential expression analysis yields a different, more biologically informative set of differentially expressed genes (DEGs) compared to a common additive (main-effects-only) modeling approach that omits the IE.

Core claims
  • Interaction effects are often the mathematical equivalent of the biological research question in gene expression studies but are frequently not considered in practice. finding
  • Using an interaction-effect model leads to a smaller but likely more biologically informative set of DEGs compared to a common approach that avoids modeling interaction. finding
  • Absence of an interaction effect corresponds graphically to parallel lines in an interaction plot and mathematically to a purely additive linear/ANOVA model; presence of an IE corresponds to non-parallel (not necessarily crossing) lines requiring an added interaction term. mechanism
  • DESeq2 models RNA-seq count data with a generalized linear model assuming a negative binomial distribution, with gene-specific dispersion and sample-specific size factors. method
  • A previously published mouse diet time-course RNA-seq dataset (Western vs. standard diet over 3–48 weeks) is used to demonstrate and compare IE vs. non-IE DEG analysis. resource
Experimental setups
Assay System Perturbation Readout Platform
bulk RNA-seq (differential gene expression analysis via DESeq2) mice diet: Western/high-fat diet (WD) vs. standard diet (SD), across multiple time points (weeks 3, 6, 12, 18, 24, 30, 36, 42, 48; analysis restricted mainly to weeks 3 and 6, with weeks 3, 6, 30, 36, 42, 48 retained after filtering) differentially expressed genes (DEGs) identified via count-based generalized linear modeling
Key results
  • Modeling the diet-by-time interaction effect produces a smaller but likely more biologically informative set of DEGs than the additive (no-interaction) model.
Key statistics
  • count 79 (total number of mice/samples in the original RNA-seq dataset)
  • count 64 (samples remaining after removing weeks with no data for one diet group (weeks 3, 6, 30, 36, 42, 48))
  • count 35,727 (number of genes measured by RNA-seq)
  • count SD: 7,5,5,7,3,5; WD: 5,5,5,5,4,8 (per-week sample sizes for standard diet (SD) and Western diet (WD) groups across weeks 3, 6, 30, 36, 42, 48)

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.

The paper uses DESeq2's negative binomial generalized linear model (GLM) framework, applied gene-by-gene across 35,727 genes, to identify differentially expressed genes (DEGs) in mouse RNA-seq data comparing a high-fat Western diet versus a standard diet across six time points (weeks 3–48). The central analytical comparison is between an additive DESeq2 model (diet + week as main effects only) and an interaction model (diet + week + diet:week), with the argument that the interaction term is the mathematical equivalent of the biological research question of whether diet effects differ across time. The paper illustrates, compares, and discusses the DEG sets produced by each approach, focusing explanatory examples on weeks 3 and 6.

Replicationbiological Sample size79 mice total; 64 retained after removing weeks with no data in one group; per-group per-week sample sizes stated: SD: 7, 5, 5, 7, 3, 5 and WD: 5, 5, 5, 5, 4, 8 for weeks 3, 6, 30, 36, 42, 48 GroupsWestern (high-fat) diet vs. standard diet in mice across six time points (weeks 3–48) Pairingunpaired Randomization/blindingnot stated Dispersionunclear Multiplicity correctionnot stated in provided text
Statistical tests used
Test Applied to n Assumptions
DESeq2 negative binomial GLM — additive design (∼ diet + week), default Wald test implied by DESeq2 framework Genome-wide DEG identification across 35,727 genes without interaction; used as the comparator approach 64 samples retained (from 79 total) after removing weeks with missing data in one group not stated
DESeq2 negative binomial GLM — interaction design (∼ diet + week + diet:week), default Wald test implied Genome-wide DEG identification with diet-by-week interaction; primary analysis; detailed illustration on weeks 3 and 6 64 samples total; weeks 3 and 6 subset: SD n=7,5 and WD n=5,5 respectively not stated
Approaches that could also have been used
  • The comparison between additive and interaction models was made descriptively by examining the resulting DEG sets, without a formal per-gene statistical test of whether the interaction term improves model fit
    Could also: A likelihood ratio test (LRT) within DESeq2 (test='LRT', reduced = ~diet + week) could also be used to test gene-by-gene whether adding the interaction term significantly improves fit — The LRT provides a principled, per-gene statistical criterion for selecting the interaction model, rather than applying it globally; this could help distinguish genes where the interaction is genuinely supported by the data from those where an additive model is adequate
  • Time (week) was modeled as a categorical factor with six discrete levels in the DESeq2 design formula
    Could also: Time could alternatively be modeled as a continuous variable using natural cubic splines or polynomial terms within the GLM framework (e.g., splines::ns() in R with DESeq2) — Spline or polynomial time modeling can be more parsimonious when temporal response is expected to be smooth, may increase statistical power at small per-time-point sample sizes, and allows estimation of the trajectory at unobserved time points
  • Differential expression analysis used DESeq2's negative binomial GLM throughout
    Could also: edgeR (quasi-likelihood F-test with glmQLFit/glmQLFTest) or limma-voom could also be applied, as both support multi-factor interaction designs for RNA-seq count data — The three major RNA-seq frameworks (DESeq2, edgeR, limma-voom) differ in dispersion estimation strategy and have shown varying sensitivity and FDR control at small n; running multiple tools and examining the overlap of DEGs can be informative about result robustness
  • Detailed explanatory analyses were restricted to weeks 3 and 6, treating these as a two-time-point subset
    Could also: A joint analysis using all six time points simultaneously within a single interaction design matrix could also be performed, with multiplicity correction across the full family — Subsetting to two time points simplifies presentation but reduces sample size per gene; a joint multi-week model uses the full n=64 and may detect interaction patterns that are only apparent over the broader temporal range
  • The paper focuses on a two-group (diet) by multi-time-point design with independent samples at each time point
    Could also: If any mice were sampled at multiple time points, a mixed-effects model (e.g., dream/voom from the variancePartition package) treating mouse as a random effect would also account for within-subject correlation — When observations from the same subject appear at multiple time points, treating them as independent can underestimate within-subject correlation and affect Type I error; a random-effect model for subject handles this directly
  • Multiplicity correction is not explicitly described in the provided text; DESeq2's default Benjamini-Hochberg FDR procedure is applied uniformly across all tested genes
    Could also: Independent hypothesis weighting (IHW, Ignatiadis et al. 2016) could also be applied, weighting genes by a covariate such as mean count before FDR correction — IHW has been shown to increase power over uniform BH when an informative covariate (e.g., mean expression level, which correlates with detection power) is available, which is commonly the case in RNA-seq data with widely varying expression levels
Software: DESeq2 (R package) · R

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
11
Impact: medium
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.

What was reproduced

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

Scope — pmid-38012163

Paper: Duda JC, Drenda C, Kästel H, Rahnenführer J, Kappenberg F. (2023) Benefit of using interaction effects for the analysis of high-dimensional time-response or dose-response data for two-group comparisons. Sci Rep 13:20834. DOI 10.1038/s41598-023-47057-0. PMID 38012163 / PMC10682470.

Code: https://github.com/jcduda/gene_expression_interaction (commit 62690c9cabe389e9757ab875ebec7ccfd06c91ac, branch main). R. Data: SRA PRJNA953810 (mouse liver RNA-seq, Western/HFD vs Standard diet, multiple weeks). Crucially the repo ships the pre-processed count object 01_data/010_gse.RData (RangedSummarizedExperiment, 28 MB) plus the authors' cached DESeq2 result objects — so the full computational pipeline is reproducible from the repo alone; the "data on request from IfADo" caveat in the README does NOT block reproduction of the reported tables.

What the paper reports (pipeline-derived)

The paper is an applied-methods comparison of two DE-analysis strategies on the same RNA-seq data, both implemented with DESeq2:

  • Method I ("separate"): fit ~ diet independently per time point (weeks 3 and 6); a gene is DEG at a week if padj < 0.05 and |log2FC| > log2(1.5) after apeglm shrinkage (lfcShrink, coef diet_HFD_vs_SD). "Newly DEG at week 6" = DEG at week 6 but not at week 3.
  • Method II ("interaction"): single model ~ diet + weeks + diet:weeks on all weeks; a gene is DEG if the week-6×diet interaction term dietHFD.weeks6 has padj < 0.05 and |log2FC| > log2(1.5) (apeglm-shrunk).
  • Prefilter (both): drop genes with zero counts in > 50% of samples.

In scope (attempted) — clearly specified, low-hanging pipeline outputs

target what pipeline
C1 — Table 1 Method I DEG counts: week-3-only / overlap / week-6-only, split up/down DESeq2 ~diet per week + apeglm + thresholds
C2 — Table 2 Method I vs Method II: Method-I-only / overlap / Method-II-only, split up/down DESeq2 interaction model + apeglm + thresholds
C3 (context) n genes, n samples, samples per week×diet from shipped gse

Reproduced by re-running the authors' functions (00_functions.R: deseq_obj, de_mod1, de_mod2) on the shipped counts — not by loading their cached result objects (040_res_mod1.RData, 050_results_model_2.RData), which are kept only as an independent cross-check.

Out of scope / not attempted (the hard ~20%)

  • Table 3 / Supp Table 1 (topGO GO enrichment): depends on set.seed(1234)
    • topGO + org.Mm annotation; enrichment p-values are sensitive to annotation DB version. Lower-value, version-fragile → skipped (noted, not dropped).
  • Table 4 / top-10 gene lists, Figs 1–6: illustrative/manual figure work, not single reproducible scalars; out of scope per brief (pipeline-derived scalars only).
  • Raw FASTQ → counts (SRA PRJNA953810): the upstream alignment/quantification that produced 010_gse.RData is not described in runnable detail in the repo; we reproduce from the shipped count object, which is the paper's analysis input.
Figures / tables: Table
C1_up_week3only
Reported
104
Reproduced
104
exact
C1_up_overlap
Reported
81
Reproduced
81
exact
C1_up_week6only
Reported
1622
Reproduced
1622
exact
C1_down_week3only
Reported
81
Reproduced
81
exact
C1_down_overlap
Reported
93
Reproduced
94
within tolerance
C1_down_week6only
Reported
726
Reproduced
725
within tolerance
C2_up_mI_only
Reported
914
Reproduced
913
within tolerance
C2_up_overlap
Reported
695
Reproduced
696
within tolerance
C2_up_mII_only
Reported
167
Reproduced
169
within tolerance
C2_down_mI_only
Reported
540
Reproduced
537
within tolerance
C2_down_overlap
Reported
177
Reproduced
178
within tolerance
C2_down_mII_only
Reported
186
Reproduced
186
exact
C3_dims_and_samples
Reported
35727 genes, 64 samples; wk3 SD7/HFD5, wk6 SD5/HFD5
Reproduced
35727 genes, 64 samples; wk3 SD7/HFD5, wk6 SD5/HFD5
exact

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 92/100

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.

Concordant (toward reproduced)
All content-critical questions reproduced
-4 pts
From: Q7 · Core claim 🟢
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Every question reproduced
-1 pts
From: “every question reproduced”
Total score -7

Clean 1:1 reproduction. The repo ships the analysis count object (35727×64), so input data is identical and the reported Table 1/2 DEG counts are directly comparable. The from-scratch rebuild matches 10/17 cells exactly and the remaining 7 within 1-3 genes out of hundreds, fully explained by DESeq2 1.38.0 vs the paper's 1.38.1 patch perturbing a few genes at the exact padj=0.05/|log2FC|=log2(1.5) decision boundary (sums conserved, deg_mod_1_up=1609 in both). Crucially the authors' own cached objects regenerate both tables to the integer — no fabrication signal, deviations are on the technical/expected side, and the core methodological claim holds.

🤝
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.

136.5 k
tokens (I/O) · 9.6 M incl. cache
16 min
runtime · 0.04 CPU-h
2.1 GB
peak RAM
2 (1 failed)
HPC jobs
hummel
machine