Transcriptional landscape of repetitive elements in normal and cancer human cells.
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.
- ✓Reported values are derivable from the shared data
- ✓The central claim held under reproduction
- 🟡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
- 🟡The deviation was non-trivial in magnitude
- 🟡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, largely successful pipeline-derived reproduction using the authors' own RepEnrich tool (not a reimplementation) against hg19+RepeatMasker. Two datasets were fully processed: (1) GSE18184's Pol III-machinery ChIP-seq/RNA-seq subset (11/11 samples), which qualitatively reproduces the paper's Pol III-at-tRNA-genes positive-control biology (7/8 ChIP samples 1.35x-9.87x enriched vs Input; Brf2's lack of enrichment matches known promoter-type specificity) plus expected bulk-Alu non-enrichment and an RNA-seq-vs-ChIP srpRNA sanity check -- all graded within-tol since a simple fractional-ratio metric stands in for the paper's own GLM/FDR test. (2) ERP000550's 28-sample (14-pair) prostate tumor/normal cohort, which reproduces the qualitative core finding (L1 dominant among significantly changed subfamilies, ~100% tumor-overexpressed) but not the paper's precise magnitudes (346 vs 475 significant subfamilies; 89/120 vs 99/107 L1 subfamilies; ~1.4x vs paper's reported 2-4x fold-change) -- graded partial, attributable largely to substituting a paired Wilcoxon+BH-FDR test for the paper's original paired edgeR GLM (R/edgeR not confirmed available) plus no patient sub-grouping/length-binning. Two claims were explicitly NOT attempted and are not counted as reproduced or failed: the paper's Pol II ChIP-seq/LTR cancer-vs-normal and snRNA cross-cell-line-binding claims (no Pol II ChIP-seq samples downloaded in this room) and the TNM clinical-stage correlation (no linked per-patient clinical metadata available from public ENA records). A pre-existing, disclosed data-quality issue affects 66% of minor/rare repeat-family bowtie indices (filesystem race during setup) but does not affect any family used in the graded claims. No results were fabricated to avoid a drop; unattempted items are recorded as such rather than guessed.
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.
✎ 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
- 2026-08-07
- Rubric version
- v1.0
- Assessed by
-
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-08-07no human curator yet
- Last updated
- 2026-08-07
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: opusCan genome-wide transcriptional regulation of repetitive elements be quantified despite the ambiguity of multi-mapping high-throughput sequencing reads, and are retrotransposons transcriptionally more active in cancer/transformed cells than in normal cells?
- ★ RepEnrich, a computational method that uses all mapping reads (uniquely mapping plus multi-mapping reads assigned to repetitive element subfamily assemblies/pseudogenomes), quantifies genome-wide repetitive element enrichment method
- ★ The fractional counting strategy (reads shared across N subfamilies counted as 1/Ns) provides the least biased and least variable estimate of true repetitive element abundance and is therefore the RepEnrich default method
- ★ Many human LTR retrotransposons are transcriptionally active in a cell line-specific manner finding
- ★ Cancer-derived cell lines display increased RNA Polymerase II binding to retrotransposons compared with cell lines derived from normal tissue finding
- ★ L1 retrotransposon RNA expression is significantly higher in prostate tumors than in normal matched controls finding
- ★ Increased retrotransposon transcription in transformed cells may explain somatic retrotransposition events reported in several cancers mechanism
- snRNAs show the most shared Pol II binding across cell lines, and tRNAs and 5S rRNA show ubiquitous Pol III binding, whereas transposable elements show cell-line-restricted polymerase binding finding
- A subset of repetitive elements, predominantly tRNAs, is co-occupied by RNA Pol II and Pol III within the same cell line finding
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| in silico simulated ChIP-seq and input data (Hidden Markov Model-based ChIP-seq read simulator developed by the authors) | whole human chromosomes (e.g. chromosome 19) | simulated enrichment of specified repetitive element families (L1, Alu, SVA) vs input | counts per million mapping reads (CPM) / average log2CPM per repetitive element subfamily compared to known true abundance; differential enrichment calls | custom HMM simulator; Bowtie1 alignment; RepEnrich; EdgeR GLM (negative binomial) |
| RNA Pol II ChIP-seq (antibody not distinguishing active/inactive enzyme) | human cell lines: K562 (CML), HeLa (adenocarcinoma), GM12878 (EBV-immortalized lymphoblastoid), IMR-90 fibroblasts, HUVEC endothelial cells, peripheral blood-derived erythroblasts (PBDE) | none (ChIP vs input comparison) | log2 fold change and FDR of repetitive element subfamily read enrichment (ChIP vs input) | — |
| ChIP-seq for RNA Pol II phosphorylated on serine 2 (Pol II S2; active elongating enzyme) | human cell lines (IMR-90, K562, HeLa, GM12878 panel) | none (ChIP vs input) | percent of repetitive element subfamilies with significant positive enrichment (FDR <0.05, Log2FC >0) | — |
| RNA Pol III ChIP-seq | IMR-90 fibroblasts, K562, HeLa, GM12878 | none (ChIP vs input) | significant positive enrichment of repetitive element subfamilies; overlap with Pol II enrichment | — |
| ChIP-seq for TFIIIB (Pol III-associated transcription factor complex subunits) | human cell lines | none (ChIP vs input) | repetitive element binding as supporting evidence of Pol III occupancy | — |
| ChIP-seq for chromatin activation and repression marks | human cell lines (public ENCODE/GEO/ENA datasets) | none | repetitive element enrichment of histone marks | — |
| RNA-seq | prostate tumor tissue from prostate cancer patients with normal-matched controls | none (tumor vs normal-matched comparison) | repetitive element / L1 retrotransposon RNA expression levels; overexpressed transposable elements | — |
| Genome browser inspection of uniquely mapping reads at individual loci | human cell lines (tRNA and snRNA genes, including Pol III-transcribed U6) | none | visual confirmation of Pol II and Pol III co-occupancy at or near the same gene | — |
- – Fractional counting deviated least from true abundance across all subfamilies and was closest to true abundance in multidimensional scaling of average log2CPM vectors
- – Unique counting over- or under-estimated true abundance with the greatest variance and consistently underestimated SINEs; it was most affected by read coverage and performed poorly at lower coverage
- ▲ Total counting performed better than unique counting overall but consistently overestimated SINEs and SVA elements
- – R-squared of estimated vs true abundance was consistently close to 1 only for fractional counting, and varied widely between 0 and 1 for unique counting R-squared close to 1 (fractional) vs 0–1 range (unique)
- ▲ In SVA-, L1- and Alu-enrichment simulations, fractional counting recovered the most benchmark differentially enriched elements and returned the fewest false positives
- ▲ On real K562 RNA Pol II ChIP-seq data, fractional counting identified more Pol II-enriched repetitive elements than unique counting
- – 89 repetitive element subfamilies were co-occupied by Pol II and Pol III within the same cell line, the majority being tRNAs 89 subfamilies
- – Transposable elements rarely showed polymerase binding consistent across all cell lines, showing significant Pol II or Pol III binding in only one or a few cell lines, partly explained by higher expression in transformed vs normal cell lines
- count 89 repetitive elements co-enriched for RNA Pol II and RNA Pol III within the same cell line (Pol II / Pol III co-occupancy overlap (Figure 3D))
- other FDR <0.05 with Log2FC >0 (significance threshold for positive enrichment of repetitive element subfamilies in ChIP vs input GLM comparisons)
- other ~55% of the human genome is repetitive DNA (more recent estimates as high as two-thirds) (background genome composition)
- other ~45% of genomic DNA is transposable elements; ~10% is the four minor repeat categories (background genome composition)
- other 20 M reads simulated (Figure 2) and 2 M reads in two L1-enrichment simulations on two different chromosomes (simulated ChIP-seq/input in triplicate; 50 bp single-end reads, chromosome 19)
- other significantly higher levels of L1 retrotransposon RNA expression in prostate tumors vs normal-matched controls (no numeric value stated in text provided) (prostate cancer RNA-seq)
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 describes a computational method (RepEnrich) for quantifying sequencing reads mapping to repetitive genomic elements, and validates it using simulated ChIP-seq data compared against a known ground truth. Differential enrichment between ChIP and input samples (and between conditions/cell lines) was assessed with a generalized linear model (GLM) fit to a negative binomial distribution, computed via EdgeR, with significance reported using a false discovery rate (FDR) threshold of 0.05 and results expressed as log2 fold-change (Log2FC). Method performance was additionally evaluated using R-squared values and multidimensional scaling (MDS) of Euclidean distances between estimated and true abundances.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Generalized linear model (GLM) fit to a negative binomial distribution, computed with EdgeR | Differential enrichment between ChIP-seq and input samples, and across cell lines, for Pol II/Pol III binding and RNA-seq comparisons (Figure 3) | — | not stated |
| R-squared (coefficient of determination) | Agreement between RepEnrich-estimated abundance and true (simulated) abundance across counting strategies (Additional file 1: Figures S5A, S6A) | simulations with 2M reads, triplicate simulated samples | not stated |
| Multidimensional scaling (MDS) of Euclidean distances | Comparison of unique, total, fractional, and true CPM vectors (Figure 2D, Additional file 1: Figure S4D) | — | not stated |
-
Differential enrichment between ChIP/input and between conditions was modeled with a GLM fit to a negative binomial distribution via EdgeR.↳ Could also: DESeq2's Wald or likelihood-ratio test, which also models count data with a negative binomial distribution — DESeq2 uses a related but distinct dispersion-shrinkage approach and is another widely used standard for count-based differential enrichment/expression analysis; comparing results across tools can illustrate robustness to modeling choices.
-
Significance across many repetitive element subfamilies was reported using an FDR < 0.05 threshold.↳ Could also: Explicitly naming and reporting the multiple-testing correction procedure (e.g., Benjamini-Hochberg) alongside the FDR cutoff, or reporting q-values directly — Making the correction method explicit alongside the threshold helps readers assess how family-wise or false-discovery error was controlled across the large number of repetitive element comparisons.
-
Agreement between RepEnrich-estimated abundance and true simulated abundance was assessed using R-squared from scatterplots against the y = x line.↳ Could also: A Bland-Altman plot or concordance correlation coefficient (CCC) — These approaches directly quantify agreement and systematic bias between two measurements of the same quantity, which can complement R-squared (a measure of correlation but not necessarily agreement).
-
The similarity of unique, total, fractional, and true CPM vectors was visualized using multidimensional scaling (MDS) of Euclidean distances.↳ Could also: Principal component analysis (PCA) — PCA is a commonly used complementary ordination method for visualizing sample or method similarity and can be used alongside MDS to check consistency of the observed clustering pattern.
-
Higher L1 retrotransposon RNA expression was reported in prostate tumors compared to normal-matched controls.↳ Could also: A paired statistical test such as the Wilcoxon signed-rank test or a paired t-test, given the matched-sample design — Explicitly using a paired test (and stating it as such) leverages the matched tumor/normal structure of the samples and is a standard approach for matched-pair comparisons; the excerpt does not specify which test, if any, was used for this particular comparison.
-
Method performance differences (fractional vs. unique vs. total counting) were reported primarily via fold-change and R-squared summaries.↳ Could also: Reporting confidence intervals or standard errors around the estimated abundance/fold-change values — Interval estimates convey the precision of the estimates in addition to their point values, which can add information for readers evaluating the counting strategies.
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
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 one of the stronger reproduction cases: the authors' own RepEnrich code was cloned and run against the same hg19+RepeatMasker reference, and both central claims survive — L1 subfamilies are significantly tumor-overexpressed in prostate (89/89 = 100% tumor-directed vs the paper's 97/99 = 98%) and Pol III machinery is enriched at tRNA genes in 7/8 ChIP samples (1.35x-9.87x), with the lone exception (Brf2 = 0.74x) matching known type-3-promoter biology rather than signalling failure. The quantitative shortfalls — 346 vs 475 significant subfamilies, 89 vs 99 significant L1 subfamilies, median fold change ~1.4x vs the paper's 2-4x — sit squarely on our side: a rank-based Wilcoxon was substituted for the paper's paired edgeR NB-GLM, no patient sub-grouping or length-binning was performed, and a setup race corrupted 66% of minor-family indices, cutting the testable universe to 1363/3861. Two further claims (Pol II/LTR cancer-vs-normal; the TNM p=0.04 correlation) went untested purely because the Pol II GSMs were not downloaded and clinical metadata is not in the anonymized ENA records — an availability/scope gap, not an authors' defect. Overall: solid, direction- and sign-consistent reproduction with fully explainable deviations, so yellow rather than green on severity and overall, but no derivability or core-claim concern.
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.