Investigating epigenetic biomarkers of age, sex, and disease in captive South African cheetahs (Acinonyx jubatus jubatus).
The main results reproduced, with only marginal, non-material deviations.
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
- 🟡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 deviation was non-trivial in magnitude
- 🟡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
Described well enough to reproduce the PIPELINE, and we did: decoded all 57 GEO GSE310779 Mammal40 IDAT pairs to betas and ran the repo's documented elastic-net age + sex clocks (alpha=0.5, log-linear age transform ASM=2/k=0.2, LOOCV, lambda.min) verbatim on the SAME 52 liver+blood training samples. Result is a PARTIAL reproduction: the clocks clearly work (age LOOCV r=0.71, p<1e-4; sex 90.2% accurate; predictions track chronological age and sex), corroborating the paper's qualitative claim, but the exact reported metrics (r=0.97 / MAE=0.86 / 52 CpGs; sex 100% / 67 CpGs) were NOT matched. The gap is NOT a fabrication signal: it is fully explained by declared, reproducible deviations. (1) We could not run SeSaMe noob/pOOBAH normalization because Bioconductor's experiment-data CDN mghp.osn.xsede.org (serving sesameData/ExperimentHub incl. the Mammal40 idatSignature/address) was globally unreachable from BOTH «our HPC» and «host»; we worked around it by decoding IDATs directly with illuminaio + the zhou-lab GitHub Mammal40 manifest (raw beta M/(M+U+100)) and installed the sesameData package from the TU-Dortmund Bioconductor mirror. (2) The authors' submission-1 'shifted' correction, exact ComBat, and hand removal of 7 stillborn/outlier samples are not fully shipped (the SID->GSM metadata CSV cheetah_metadata_vod2.csv is absent), and several neonates are mispredicted in our raw run, which alone depresses r. NOT ATTEMPTED: C6 FelidClock multi-species clock (needs external lion/tiger methylation not in this accession) and C7 SOS differential-methylation analysis (hard-20% per-CpG DMA) - both out of the 80/20 low-hanging scope.
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 58assessed: 2026-06-16 ⛓ 78d7f2896968
✎ 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
- 2026-06-16
- Rubric version
- v1.0
- Assessed by
-
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-16no 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: opusDNA methylation-based models can accurately predict age and sex in captive cheetahs despite limited sample sizes, with multi-tissue and multi-species data enhancing predictive power, and epigenetic differences can distinguish cheetahs affected by hepatic sinusoidal obstruction syndrome (SOS) from unaffected individuals.
- ★ A cheetah-specific epigenetic age clock using 52 CpG sites predicts chronological age across blood and liver with r=0.97 and MAE=0.86 years finding
- ★ A multi-species (cheetah, lion, tiger) age clock using 46 CpG sites predicts age across these felids with r=0.94 and MAE=1.16 years finding
- ★ A sex clock using 67 CpG sites accurately predicts sex in all test samples finding
- ★ Differential methylation analysis identified 4,377 CpG sites differing significantly between SOS-positive and SOS-negative cheetahs finding
- ★ Elastic net regression on HorvathMammalMethylChip40 methylation profiles can build accurate epigenetic clocks from small wildlife datasets (n=52) method
- The age clock is accurate for adult cheetahs (>3 years) but less precise around age of sexual maturity finding
- Cheetah-specific clocks outperform existing pan-mammalian (UniversalClock) and domestic-cat (CatClock) clocks on cheetah samples finding
- DNA methylation establishes a foundation for biomarkers of disease (SOS) in wildlife conservation resource
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| DNA methylation array (Infinium, bisulfite-converted) | cheetah (Acinonyx jubatus) liver tissue | none | CpG methylation beta values for age clock training | HorvathMammalMethylChip40 (Illumina), GPL28271 |
| DNA methylation array | cheetah whole blood (live animals) | none | CpG methylation beta values for age/sex clock testing | HorvathMammalMethylChip40 Illumina Array |
| DNA methylation array | cheetah skin tissue (necropsy) | none | CpG methylation beta values | HorvathMammalMethylChip40 Illumina Array |
| DNA methylation array (public MCDB profiles) | lion and tiger blood | none | CpG methylation beta values for multi-species clock testing | mammalian methylation array (HorvathMammalMethylChip40) |
| Epigenome-wide association study (EWAS) | cheetah liver (n=38), blood (n=7), skin (n=4) | none | CpG methylation correlation with age/sex (Pearson; Student's t-test for sex) | — |
| Differential methylation analysis (limma lmFit) | cheetah liver (>3 years, n=30) | SOS disease status (positive vs negative) | differentially methylated CpG sites between disease groups | — |
| KEGG pathway enrichment analysis | genes associated with differentially methylated CpGs | none | enriched biological pathways | clusterProfiler (R) |
- – Cheetah age clock (52 CpGs) predicted age across blood and liver r=0.97, MAE=0.86 years
- – Multi-species feline age clock (46 CpGs) predicted age across cheetah, lion, tiger r=0.94, MAE=1.16 years
- – Sex clock (67 CpGs) correctly predicted sex in all test samples
- – Differential methylation between SOS-positive and SOS-negative cheetahs 4,377 CpG sites (adj p<0.05)
- – CatClock performed well on cheetah blood but poorly on combined tissues blood r=0.79 MAE=1.38 yr; combined r=0.64 MAE=3.5 yr
- – UniversalClock2/3 showed high error on older cheetahs MAE 3–4.42 years
- correlation r = 0.97 (cheetah age clock across blood and liver)
- other MAE = 0.86 (cheetah age clock median absolute error (years))
- correlation r = 0.94 (multi-species feline age clock)
- other MAE = 1.16 (multi-species feline clock error (years))
- count 4,377 CpG sites (adjusted p-value < 0.05) (differentially methylated sites SOS+ vs SOS-)
- correlation correlation 0.79, MAE 1.38 years (CatClock on cheetah blood samples)
- count n = 11, 25% (cheetahs with SOS diagnosis in cohort)
- other MAE 3–4.42 years (UniversalClock performance on cheetah samples)
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 retrospective observational study used elastic net regression (glmnet) with leave-one-out cross-validation (LOOCV) on DNA methylation beta values from the HorvathMammalMethylChip40 array to build cheetah-specific epigenetic age and sex prediction clocks, reporting Pearson r and median absolute error (MAE) as accuracy metrics. Epigenome-wide association studies (EWAS) screened CpG sites using Pearson correlation (age, continuous trait) and Student's t-test (sex, binary trait) against tissue-specific raw p-value thresholds. Differential methylation between SOS-positive and SOS-negative cheetahs was assessed with limma linear models and Benjamini-Hochberg FDR correction (adjusted p < 0.05), followed by KEGG pathway enrichment analysis; unsupervised hierarchical clustering was used both for outlier exclusion and for visualisation of the top 100 differentially methylated CpGs.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Elastic net regression (glmnet) with Leave-One-Out Cross-Validation (LOOCV) | Age clock (CheetahClock, feline multi-species clock) and sex clock construction from methylation beta values | CheetahClock trained on n=38 liver + n=14 MCDB cheetah blood profiles (52 samples total); sex clock training n not separately stated | not stated |
| Pearson correlation | EWAS for age: CpG-level screening in liver (n=38), blood (n=7), and skin (n=4); also reported as clock accuracy metric (r) between inverse-transformed predicted and known age | n=38 liver, n=7 blood, n=4 skin for EWAS; test-set n varies by tissue for clock evaluation | not stated |
| Student's t-test (via WGCNA standardScreeningBinaryTrait) | Sex EWAS: comparing methylation levels at each CpG between females and males | null | not stated |
| limma lmFit (linear model for array data) | Differential methylation analysis (DMA): SOS-positive vs SOS-negative liver samples | n=30 (liver samples from cheetahs >3 years of age) | not stated |
| Unsupervised hierarchical clustering | Outlier detection per tissue type (pre-analysis exclusion); visualisation of top 100 DMA CpG sites across liver and blood samples | null | na |
| KEGG pathway enrichment analysis (enrichKEGG, clusterProfiler) | Genes associated with all significant DMA CpGs (p.adj<0.05); separate analyses for hypermethylated and hypomethylated subsets | 4377 significant CpG sites | na |
-
EWAS for age and sex used raw p-value thresholds (10^-3 or 10^-2) across approximately 37,000 probes without a formal genome-wide multiple-testing correction↳ Could also: Benjamini-Hochberg FDR correction (as applied in the DMA) or a Bonferroni threshold (p < 0.05/37,492 ≈ 1.3×10^-6) could also be applied genome-wide to the EWAS — Applying the same FDR framework used in the DMA to the EWAS would quantify the expected false-discovery proportion among selected CpGs; this is particularly relevant when characterising biologically meaningful age-associated sites beyond their use as clock inputs
-
Clock performance was evaluated with LOOCV and reported as single point estimates (r and MAE) without uncertainty intervals↳ Could also: Bootstrap resampling (e.g., 1,000 iterations) or repeated k-fold cross-validation could also be used to generate confidence intervals around r and MAE — With a small training set (n=52 samples), bootstrap CIs would communicate estimation uncertainty in performance metrics, helping readers assess whether differences between the CheetahClock and pan-mammalian or CatClock benchmarks are within noise
-
Student's t-test was used for the sex EWAS to compare methylation beta values between females and males↳ Could also: Mann-Whitney U (Wilcoxon rank-sum) test or logistic regression could also be used; the former makes no distributional assumption, the latter directly models a binary outcome — DNA methylation beta values are bounded [0,1] and can be bimodally distributed; a non-parametric test does not rely on the normality assumption, which may be difficult to verify at this sample size
-
Outlier samples were identified and excluded using unsupervised hierarchical clustering with visual inspection of dendrograms↳ Could also: Principal component analysis (PCA) with a quantitative distance criterion (e.g., samples beyond 3 SD from the centroid on PC1–PC2) or robust PCA could also be used for outlier detection — PCA-based criteria provide a reproducible, pre-specified exclusion rule that can be reported numerically, complementing the visual judgement inherent in dendrogram-based exclusion
-
Log-linear age transformation was applied prior to elastic net modeling to account for rapid methylation change before sexual maturity↳ Could also: Generalized additive models (GAMs) or penalised spline regression could also model the non-linear age-methylation relationship without imposing a specific functional form — Data-driven smooth functions allow the trajectory shape to be estimated from the data rather than pre-specified, which may be useful when the exact form of age acceleration around sexual maturity is uncertain or species-specific
-
The DMA used a binary SOS-positive vs SOS-negative classification, excluding 'SOS suspect' samples from the positive group↳ Could also: An ordinal or continuous severity score, or a sensitivity analysis treating 'SOS suspect' as positive, could also be used to capture graded disease severity — A graded variable would utilise intermediate phenotype information and could reveal dose-response methylation patterns across the spectrum of SOS severity, complementing the binary contrast
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-41528985 (CheetahClock)
Paper: Investigating epigenetic biomarkers of age, sex, and disease in captive
South African cheetahs. PLoS One 2026; DOI 10.1371/journal.pone.0336127.
Code: https://github.com/mysrael/CheetahClock — single file CheetahClock_age_sex.Rmd.
Data: GEO GSE310779 — 57 cheetah samples, Illumina HorvathMammalianMethylChip40
(Mammal40), RAW IDATs (42.5 MB) + series matrix. Per-sample tissue/sex/age are
encoded in the GSM title (e.g. "Cheetah liver female 2.8y") and IDAT filenames
carry GSM####_<barcode> → full GSM↔array mapping available.
Pipeline-derived results (the paper's computational outputs)
| Result | Pipeline | In scope? | Notes |
|---|---|---|---|
| CheetahClock age clock (LOOCV r, MAE, #CpGs) | SeSaMe β → glmnet elastic net (α=0.5), log-linear age transform (ASM=2,k=0.2), LOOCV, λ.min | YES (primary) | Fully documented in the Rmd. Reproduce SeSaMe from GEO IDATs, then the exact EN procedure. |
| Sex clock (#CpGs, accuracy) | glmnet binomial elastic net (α=0.5), LOOCV λ.min | YES (secondary) | Documented in the Rmd. |
| Test-set age predictions (blood/skin r,MAE) | apply trained clock | partial | depends on training clock; report if clock reproduces. |
| FelidClock multi-species clock (46 CpGs, r=0.94) | EN on cheetah+lion+tiger | NO — out of scope | Lion/tiger methylation are EXTERNAL data not in GSE310779; not obtainable from this accession. |
| SOS disease DMC (4269 CpGs), EWAS of age (per-tissue DMC counts) | per-CpG differential methylation | NO — deferred (hard 20%) | Requires reconstructing the authors' exact contrasts/sample subsets & multiple-testing; not the low-hanging output. Skipped, stated. |
Known fidelity gaps (will be stated, not hidden)
- Authors start from intermediate RDS (
beta_sesame_shifted.RDS,data_sesame.RDS) and a metadata CSV (cheetah_metadata_vod2.csv) that are NOT shipped in the repo. We re-derive β with SeSaMeopenSesame(platform="Mammal40")from the GEO IDATs — a faithful but not byte-identical preprocessing. - A manual "shifted" correction on submission-1 betas and
ComBat(batch=study)cannot be reproduced exactly (no shipped study labels / shift vector). We process all IDATs uniformly and (optionally) ComBat by tissue-derived study label. - Hand-removed outliers are listed by author-SID (e.g.
ET0394TOX00092), which do not map to GEO barcodes without the unshipped CSV → we apply SeSaMe QC + NA filtering instead of the identical hand list.
Compute: trivial (SeSaMe on 57 IDAT pairs + glmnet on ~36k×~45 matrix) — minutes on one «our HPC» std node. All compute on «our HPC»; data on «infra».
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 solid partial reproduction: the identical public IDATs (GSE310779) were decoded and the repo's elastic-net age+sex clock was run verbatim on the exact same 52 liver+blood training samples, and the clock unambiguously works (age LOOCV r=0.71, p<1e-4; sex 90.2%, 46/51). The headline metrics were not matched (r 0.97→0.71, MAE 0.86→1.82 yr, sex 100%→90%, 52→82 CpGs), but every gap is explained by reproducible, declared deviations on the input/preprocessing side — the SeSaMe noob/pOOBAH normalization could not be run (Bioconductor OSN CDN globally down) and the authors' 'shifted' correction plus the SID→GSM outlier-mapping CSV were not deposited. The deviation is therefore moderate, sits in preprocessing rather than the modeling logic, and shows no fabrication signal; the central conclusion (a valid cheetah methylation clock exists in this data) holds in limited form.
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.