Single-Cell Hi-C Technologies and Computational Data Analysis.
The main results reproduced: recomputed values matched the published ones within tolerance.
- ✓Same input data as the authors
- ✓Reported values were directly comparable
- ✓Reported values are derivable from the shared data
- ✓The central claim held under reproduction
- 🟡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
Described well enough to reproduce the paper's central quantitative result 1:1. This is a scHi-C data-quality benchmark (Dautle & Chen 2025); its headline numeric output is Table 3 (per-dataset, per-genome average & median Total Contacts and cis/trans ratio across all public scHi-C datasets). The authors' repo (chenyongrowan/scHIC_Evaluation @3de55e3) ships the underlying per-cell master table (AllStatistics_ForBoxplot_09_01_2024.csv, 67,957 cells x 23 dataset-labels) plus the scripts that aggregate it (getStatistics.py / makeFigures.py). C-AGG (DONE): I re-aggregated the shipped per-cell CSV with the repo's OWN formula (cis/trans = Cis/(Total-Cis); drop cis/trans>=10000 outliers; group by Authors x Reference_Genome; mean/median) and compared every field of Table 3 -> 86 EXACT, 10 within-tol, 5 partial, 7 mismatch of 108 numeric fields (~89% reproduce to the printed digits; e.g. Tan2018 937763->937763.2, Mulqueen 1199765->1199764.7, Stevens 12.136->12.136, Wen 0.912->0.912, Liu 279855->279855.4). The reported table IS the rounded aggregation of the shipped data -> no fabrication evident. Honest flags (in AUDIT.md/agreement.json): (1) Tan2019 cis/trans reproduces EXACTLY but absolute totals are ~2x reported (ratio is scale-invariant -> a contact-counting/scaling definition difference, not a random error); (2) Ramani-hg19 CSV has 1896 cells vs reported 2972 and the Ramani 'Mixed' barnyard row is missing -> CSV is a partial/QC-filtered subset; (3) Luo labeled '2022' in CSV vs '2019' in paper, values differ ~20% -> ambiguous, not claimed. NOT attempted: (a) C-RAW, the one-cell raw->statistic recompute from the only shipped .hic (GSM7678878) to verify the per-cell numbers themselves derive from real raw data -- staged (run.sbatch + analyze.py with hicstraw) but BLOCKED because «our HPC» requires a browser SAML + 2FA VPN that the human operator did not action across 3 connect attempts (client also returned 'keine SSO-URL'); (b) the full from-raw recomputation of all 67,957 cells (per-cell counting code not shipped; ~TBs FASTQ across ~20 GEO series) -- the hard >>20%; (c) Figure 5 contact maps (visualization, non-deterministic random.sample). Compute note: C-AGG is a sub-second deterministic groupby over the 3MB shipped CSV (host-independent, byte-identical on «our HPC»); no data was stored on «host» (CSV streamed through memory and deleted). Verdict provisional; human audit sheet in AUDIT.md.
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 80assessed: 2026-06-15 ⛓ 424cbfc0208c
✎ 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
- 2026-09-19
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: sonnetThis review addresses how existing single-cell Hi-C (scHi-C) protocols compare in data-capture efficiency, and what computational strategies can overcome the sparsity and analytical challenges inherent to scHi-C data.
- ★ Thirteen scHi-C protocols currently exist: eight that capture chromatin interactions alone and five that combine scHi-C with other omics (methylation, RNA-seq, DNA-seq). resource
- ★ scHi-C data are inherently sparse, creating computational challenges for clustering, compartment/TAD/loop calling, 3D reconstruction, simulation and differential interaction analysis. finding
- ★ Among interaction-only protocols, Nagano et al. 2017 and scNanoHi-C recover the highest total contacts and cis/trans ratios. finding
- ★ The snHi-C protocol performs adequately on mouse and fly cells but poorly on human cells, with a low cis/trans ratio. finding
- ★ The Stevens et al. 2017 protocol achieves a high cis/trans ratio but recovers only about a quarter of the total contacts obtained by the Nagano et al. protocols. finding
- ★ Dip-C datasets show average contact recovery but rank among the lowest in cis/trans ratio. finding
- ★ Combining scHi-C with other single-cell assays (RNA-seq, methylation, DNA-seq) links chromatin structure to gene expression and epigenetic state, improving resolution of heterogeneous and rare cell types. mechanism
- ★ Standardizing protocol steps, automating workflows, adopting high-throughput combinatorial barcoding, minimizing costly reagents (e.g., MDA), and using alternative amplification (META) are proposed to improve scHi-C scalability and reduce bias. method
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| snHi-C | mouse oocytes/zygotes and other mouse/human/fly cells | none | total chromatin contacts, cis/trans ratio | — |
| scHi-C (Nagano et al. 2017 protocol) | mouse cells (mm9/mm10) | none | total chromatin contacts, cis/trans ratio | — |
| sci-Hi-C | mouse and human cells (mm10/hg19) | none | total chromatin contacts, cis/trans ratio | — |
| Dip-C | human and mouse cells (hg19/mm10) | none | haplotype-resolved chromatin contacts, cis/trans ratio | — |
| scSPRITE | mixed mouse/human cells | none | total chromatin contacts, cis/trans ratio | — |
| scNanoHi-C | mouse and human cells (mm10/hg38) | none | total chromatin contacts, cis/trans ratio | Nanopore sequencing |
| sn-m3C-seq | mouse and human cells | none | chromatin contacts plus DNA methylation state | bisulfite sequencing |
| HiRES (scHi-C + scRNA-seq) | mouse cells (mm10) | none | chromatin contacts plus gene expression | — |
- ▲ Nagano et al. 2017 protocol and scNanoHi-C show higher total captured contacts and cis/trans ratios than other interaction-only scHi-C protocols
- ▼ snHi-C on human cells shows a low median cis/trans ratio of 0.428 0.428
- – Stevens et al. 2017 protocol recovers about a quarter of the total contacts compared to Nagano et al. protocols despite a high cis/trans ratio ~0.25-fold
- ▼ Dip-C datasets (Tan et al. 2018/2019/2021) show average contact recovery but some of the lowest cis/trans ratios among all datasets
- – sci-Hi-C (Ramani et al. 2017, Kim et al. 2020) and scSPRITE (Arrastia et al. 2022) show high cis/trans averages/medians but low total contact numbers
- ▲ scNanoHi-C recovers higher-than-average total contacts with a cis/trans ratio comparable to other methods across mouse and human cells
- mean mean 225687 contacts, mean cis/trans 7.168, median cis/trans 6.545 (Nagano et al. 2017 dataset (GSE94489, mm9))
- mean mean cis/trans ratio 12.136, median cis/trans ratio 13.268 (Stevens et al. 2017 dataset (GSE80280, mm10))
- mean median cis/trans ratio 0.428 (Flyamer et al. 2017 snHi-C dataset on human cells (hg19))
- mean mean 800199 contacts, mean cis/trans ratio 10.129 (scNanoHi-C dataset (GSE217189, mm10))
- mean mean 937763 contacts, mean cis/trans ratio 2.552 (Dip-C, Tan et al. 2018 dataset (GSE117876, hg19))
- mean mean 1199765 contacts, mean cis/trans ratio 15.006 (s3-GCC, Mulqueen et al. 2021 dataset (GSE174226, hg38))
- fold_change about a quarter (Total contacts of Stevens et al. 2017 protocol relative to Nagano et al. 2013/2017 protocols)
- count 13 protocols (8 interaction-only + 5 multi-omics) (Total number of scHi-C protocols reviewed)
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 is a narrative review with an embedded quantitative benchmarking component. Protocol quality was assessed by downloading 13 existing public scHi-C datasets and computing per-cell descriptive statistics—mean and median total chromatin contacts and mean and median cis/trans ratios—summarized in Table 3 and visualized in Figures 3–5. No formal inferential statistical tests were applied; protocol comparisons are made descriptively by inspecting these summary values and scatter plots. No p-values, effect sizes, or confidence intervals are reported.
-
Protocol performance was compared by inspecting means and medians of per-cell contact counts and cis/trans ratios narratively, without formal statistical tests↳ Could also: A non-parametric test such as Kruskal-Wallis with Dunn's post-hoc correction could also be applied to formally compare the distributions of per-cell values across protocols — Formal tests would quantify the probability that observed differences arise by chance and produce a structured family-wise error control, which is particularly relevant given the large variation in cell counts across datasets (8 to 19,388 cells)
-
Central tendency is summarized with mean and median; no measures of within-dataset spread are reported in Table 3↳ Could also: Interquartile range (IQR), standard deviation, or 95% bootstrap confidence intervals could also be reported alongside the mean and median — Per-cell contact distributions in scHi-C data are typically right-skewed; reporting spread alongside central tendency helps readers judge whether differences between protocols are consistent across cells or driven by outliers
-
Datasets from different protocols are compared directly on raw contact counts without adjusting for sequencing depth or species/genome differences↳ Could also: Rarefaction (downsampling to a common sequencing depth) or depth-normalized contact counts could also be computed before cross-protocol comparison — Differences in sequencing effort between datasets may partly explain differences in total contacts; depth normalization isolates protocol-specific capture efficiency from the amount of sequencing applied
-
Figure 5 displays per-cell cis/trans ratio as a function of total contacts in a scatter plot, but no correlation statistic is reported↳ Could also: Spearman's rank correlation (or Pearson's r after log-transformation of counts) could also be computed and reported to quantify the association between sequencing depth and cis/trans ratio — A reported correlation coefficient makes the relationship quantitative and allows readers to compare the strength of this association across protocols or species
-
Protocol comparisons are made across datasets that differ in species (mouse, human, Drosophila) and reference genome simultaneously↳ Could also: Stratified analysis restricted to datasets sharing the same species and reference genome, or a mixed-effects model treating species as a covariate, could also be used — Biological differences across species may confound protocol-level comparisons; stratification or covariate adjustment would allow more direct attribution of differences to the protocol itself
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-39887949
Paper: Dautle MA, Chen Y. Single-Cell Hi-C Technologies and Computational Data
Analysis. Adv Sci (Weinh) 2025. PMID 39887949 · PMCID PMC11884588 · DOI
10.1002/advs.202412232.
Repo: https://github.com/chenyongrowan/scHIC_Evaluation @ 3de55e3 (main, pushed 2024-11-24), GPL-3.0.
Primary accession (registry): GEO GSE80006 (one of ~20 source series; the actual
per-cell quality table aggregates many series — see below).
What kind of paper this is
A review + data-quality benchmark of single-cell Hi-C (scHi-C) protocols. The benchmark portion is a real bioinformatic pipeline: for every publicly available scHi-C cell across 13 protocols / 23 dataset-labels, the authors compute per-cell Total_Contacts, Cis_Contacts, Cis/Total, and Cis/Trans, then aggregate per dataset (means/medians) and visualize (Figures 2–4). One contact map figure (Fig 5) is derived from cooler files of one GEO series.
Repo contents (what is shipped)
CalculateStatistics/AllStatistics_ForBoxplot_09_01_2024.csv— master per-cell table, 67,957 cells × 23 dataset-labels (the central derived dataset).CalculateStatistics/getStatistics.py— aggregates the master CSV into per-dataset mean/median Total_Contacts, Cis/Total, Cis/Trans (drops Cis/Trans ≥10000 outliers).FigureGenerationScripts/Figures2-4/makeFigures.py(+ same CSV) — regenerates Fig 2 (total contacts boxplots), Fig 3 (cis/trans boxplots), Fig 4 (scatterplots).FigureGenerationScripts/GSM7678878_p003-bdf1_001.hic— one raw .hic file (24 MB, mm10, Dip-C/LiMCA cell from GSE239969) — the only shipped raw input.FigureGenerationScripts/Figure5/— CreateFig5.py + MakeCoolerFiles.sh build cooler files from GSE129029 (Collombet et al 2020) contact lists → contact maps.
In scope (pipeline-derived, will attempt)
- C-AGG: per-dataset summary statistics (paper Table 3 / abstract numbers). Re-run the repo's aggregation logic on the shipped master CSV → per-dataset mean & median Total_Contacts and Cis/Trans, and Cis/Total. Deterministic. Compare to the exact numbers reported in the paper (e.g. Stevens 2017 cis/trans mean 12.136 / median 13.268; Nagano 2013 10.462/10.608; Tan 2018/Dip-C total 937763, cis/trans 2.552/2.415; Mulqueen 2021 total 1199765; Ramani 2017 total 5534). This is the central quantitative result of the benchmark.
- C-RAW: raw→statistic fabrication check (one cell). Recompute Total_Contacts
and Cis_Contacts directly from the shipped raw
.hic(GSM7678878, mm10) by summing genome-wide and intra-chromosomal contacts at base resolution, and check the pair appears as a row under "Liu et al 2023" (mm10) in the master CSV. Tests whether the upstream per-cell numbers (whose generation code is NOT shipped) are honestly derived from raw data vs fabricated. - C-FIG (optional): regenerate Fig 2/3/4 PDFs from the CSV and confirm they reproduce the shipped figure PDFs (visual + summary-stat consistency).
Out of scope (and why)
- Full from-raw recomputation of all 67,957 cells. The per-cell statistic computation code (HiC-Pro/Juicer valid-pairs counting) is NOT shipped; reproducing it for all 23 datasets means downloading ~20 large GEO/SRA series (TBs of FASTQ/ pairs) and rerunning each protocol's mapping pipeline. This is the hard >>20% and is not attempted; we instead verify ONE cell from the one shipped raw file (C-RAW).
- Figure 5 contact maps (Collombet GSE129029 download + cooler build) — a visualization, not a quantitative claim; the random.sample() makes it non-deterministic. Skipped (low value for fabrication detection).
- Wet-lab / protocol-description content of the review — not computational.
Compute plan
All data on «infra», compute on «our HPC» («infra») per hard rules: clone repo on «infra» inside a compute job (front1 has no internet), build a small conda env (pandas/numpy/cooler/hic-straw), run C-AGG + C-RAW, write small result JSONs,
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.
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 [email protected].
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.