Cell fixation and preservation for droplet-based single-cell transcriptomics.
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.
- ✓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
- Every checked point held up.
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
Reproduced 1:1 (described well enough). The paper's own R package dropbead (v0.3.1, pinned commit d746c6f3) was re-applied to the deposited GSE89164 data on «infra» «our HPC» (SLURM «job»; prior run 2180593 was requeued only because its ROOM_RESULT predated the datasets[]/qc_room schema and the dataset profile was a 404 stub — numbers are identical). The two headline cell counts reproduce EXACTLY: 4873 Drosophila cells (Fig 3) and 4366 mouse cells (Fig 4) equal the unique-barcode counts of the deposited cluster tables -> strong no-fabrication signal. Re-running dropbead's correlation routine on the live and methanol-fixed human/mouse mixing DGEs gives mouse R=0.95 (matches reported R>=0.95 exactly) and human R=0.93 (marginally under, plausibly cell-selection-threshold dependent). Median genes per cell are essentially identical between live and fixed (2528 vs 2529); UMIs comparable (5697 vs 4853) -> reproduces 'fixed similar to live'. The barnyard doublet rate is not increased by fixation (3.0% live vs 2.8% fixed at 90% purity) -> reproduces that claim. Dataset profiled in the same pass: GSE89164 deposits 18 GSM samples (4 platforms, 3 organisms); all 6 inspected files parse cleanly, counts are nonneg-integer, no NA, genes all hg_/mm_ prefixed, sparsity 0.90-0.93, SHA256 identical to the prior fetch -> delivers what the paper analyzes; quality grade A (provisional). NOT attempted (out of scope): wet-lab protocol; SRA->STAR->DGE re-alignment; stochastic clustering beyond cell counts; per-organism median genes/UMIs; live/fixed-vs-bulk correlation. All grades provisional pending human audit (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 84assessed: 2026-06-16 ⛓ 5a102c253df7
✎ 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
- 2026-06-24
- 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: sonnetMethanol fixation can stabilize and preserve dissociated single cells for days to weeks without altering their transcriptional state, enabling high-quality droplet-based (Drop-seq) single-cell RNA-seq of both cultured and primary cells collected over time or from multiple sources.
- ★ Methanol fixation stabilizes and preserves dissociated cells for weeks without compromising single-cell RNA-seq data quality method
- ★ Individual transcriptomes from mixtures of fixed human (HEK) and mouse (3T3) cells can be confidently assigned to one species or the other finding
- ★ Single-cell gene expression from live and fixed samples correlates well with bulk mRNA-seq data finding
- ★ Methanol fixation enables successful Drop-seq profiling of low-RNA-content primary cells from dissociated Drosophila embryos and FACS-sorted mouse hindbrain/cerebellum cells finding
- ★ Diverse cell populations, including neuronal subtypes, can be identified from fixed primary tissue samples finding
- Alcohols such as methanol are coagulating fixatives that dehydrate cells without chemically modifying nucleic acids, unlike aldehyde fixatives mechanism
- ★ 'dropbead', an R package for exploratory data analysis, visualization and filtering of Drop-seq data, is provided as a resource resource
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| Drop-seq (droplet-based scRNA-seq) | human HEK293 and mouse NIH/3T3 cell line mixture | methanol fixation (live vs. fixed vs. fixed 1 week vs. fixed 3 weeks) | genes/UMIs detected per cell, species assignment accuracy | Illumina NextSeq 500 |
| Drop-seq (droplet-based scRNA-seq) | dissociated Drosophila melanogaster embryos | methanol fixation, stored up to 2 weeks at -20°C | single-cell transcriptomes, cell clustering and cell type identification | Illumina NextSeq 500 |
| Drop-seq (droplet-based scRNA-seq) | FACS-sorted mouse (C57BL/6, postnatal day 5) hindbrain and cerebellum cells | methanol fixation, stored >4 weeks at -80°C | cell populations including neuronal subtypes | Illumina NextSeq 500 |
| bulk mRNA-seq | cultured HEK/3T3 cells | none | RPKM values correlated against Drop-seq gene counts | — |
| TRIZOL RNA extraction (RNA quality control) | HEK/3T3 cells | methanol fixation | RNA quality after fixation | — |
| FastQC sequencing quality assessment | sequencing reads from all Drop-seq libraries | none | base call quality per read position | FastQC v0.11.2 |
| STAR alignment and Drop-seq tools processing | sequencing reads from human/mouse, Drosophila, and mouse hindbrain libraries | none | percentage of uniquely mapped reads, cell/molecular barcode and gene tagging | STAR v2.4.0j; Drop-seq tools v1.12 |
| Seurat clustering (PCA + tSNE) | Drosophila embryo and mouse hindbrain single-cell digital gene expression data | none | cell clusters and marker genes | Seurat R package |
- – Around 65% of reads uniquely mapped to either the human or mouse genome in the mixed-species Drop-seq experiment ~65%
- – Around 85% of Drosophila reads mapped uniquely to the BDGP6 reference genome ~85%
- – 75% of mouse hindbrain sequence reads mapped uniquely 75%
- – Gene intersection used for correlation analyses was around 17,000 genes for human/mouse samples and 10,000 genes for D. melanogaster ~17,000 / ~10,000 genes
- ▼ Cell recovery after fixation and FACS sorting was 19% and 12% across two mouse hindbrain preparations 19% and 12%
- – 2975 cells were flagged as nuclei (lacking mitochondrial gene expression) and excluded from Drosophila clustering analysis 2975 cells
- – A 90% single-species UMI threshold was used to confidently assign species identity and exclude human/mouse doublets 90%
- – Low-quality cells were filtered using UMI thresholds of 3500 (HEK/3T3), 1000 (Drosophila), and 300 (mouse) 3500 / 1000 / 300 UMIs
- other ~65% (reads uniquely mapped to species genome in mixed human/mouse Drop-seq experiment)
- other ~85% (Drosophila reads uniquely mapped to BDGP6 genome)
- other 75% (mouse hindbrain reads uniquely mapped)
- count ~17,000 genes (gene intersection for human/mouse correlation analysis)
- count ~10,000 genes (gene intersection for Drosophila correlation analysis)
- count 2975 cells (cells classified as nuclei and excluded from Drosophila clustering analysis)
- other 19% and 12% (cell recovery rate from two mouse hindbrain cell preparations)
- other 90% (UMI threshold used to call species identity and exclude doublets)
Statistical methods review
Model: opusA 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 methodology paper demonstrating that methanol fixation preserves cells for droplet-based single-cell RNA-seq (Drop-seq). The approach is largely descriptive and computational rather than inferential: data quality was assessed by counting genes/UMIs per cell, species-mixing 'barnyard' assignment, and correlation of gene expression between live, fixed, and bulk mRNA-seq samples. Cell populations were identified by highly variable gene selection, principal component analysis, graph-based clustering (Seurat) and tSNE visualization, with marker genes detected via Seurat's FindAllMarkers. Results are presented as plots, correlation values, and counts rather than formal hypothesis tests with reported p-values.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Gene expression correlation between samples (live vs fixed and Drop-seq vs bulk mRNA-seq); correlation type not specified | Comparisons of single-cell samples and against bulk mRNA-seq (log space; RPKM for cross-platform) | Summed gene counts across all cells per library over the intersection of genes (~17,000 human/mouse; ~10,000 Drosophila) | not stated |
| Principal component analysis on highly variable genes | Drosophila embryo and mouse hindbrain dimensionality reduction prior to clustering | ~20–50 PCs (50 for Drosophila, 21 for mouse) | not stated |
| Graph-based clustering (Seurat FindClusters, default parameters) | Cell population identification in Drosophila and mouse data | 50 PCs (Drosophila) or 21 PCs (mouse) as input | not stated |
| t-distributed stochastic neighbour embedding (tSNE) | Two-dimensional visualization of clusters | Same PCs as clustering input | na |
| Marker gene detection (Seurat FindAllMarkers); underlying test not named | Identification of marker genes per cluster | — | not stated |
-
Gene expression agreement between live and fixed samples was assessed via correlation of summed counts, with plots shown in log space.↳ Could also: Reporting the specific correlation coefficient used (e.g., Pearson on log values or Spearman rank correlation) alongside the value. — Spearman is robust to the heavy-tailed, wide-dynamic-range nature of expression counts, and stating the coefficient and method makes the concordance metric fully reproducible across readers.
-
Cell populations were defined using PCA followed by graph-based clustering with a chosen number of PCs (50 or 21).↳ Could also: Reporting a quantitative criterion for PC selection (e.g., a jackstraw permutation test or elbow/scree analysis) and a clustering-stability or resolution-sweep assessment. — An explicit selection criterion and stability check convey how sensitive the identified clusters are to the chosen PC count and resolution, helping readers gauge robustness.
-
Marker genes were identified with Seurat's FindAllMarkers, with the underlying test and any p-value adjustment not specified in the text.↳ Could also: Stating the differential-expression test (e.g., Wilcoxon rank-sum) and the multiple-testing correction (e.g., Benjamini-Hochberg FDR) used for marker calling. — Naming the test and correction clarifies the family-wise/false-discovery control across the many genes and clusters tested and aids reproduction.
-
Quality thresholds (UMI cutoffs, 90% species purity, mitochondrial content) were applied as fixed values to filter cells.↳ Could also: Presenting sensitivity analyses across a range of thresholds, or data-driven approaches such as knee-point/EmptyDrops-style modeling for cell calling. — Showing that conclusions hold across threshold choices, or using a model-based cutoff, illustrates that downstream results are not strongly dependent on a single chosen value.
-
Per-cell quality metrics (genes/UMIs) were summarized using violin plots.↳ Could also: Adding numeric summaries such as median with IQR, or a 95% confidence interval for group means, alongside the distributions. — Explicit dispersion statistics complement the visual distributions and make quantitative comparison between live and fixed conditions easier, especially helpful when n per group is modest.
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.
-
Same-species doublet rate in methanol-fixed Drosophila and mouse Drop-seq datasets estimated at approximately 10–15%, extrapolated from cell-line mixing experimentsscRNA-seq drosophila-melanogaster-embryo 2017×1papers★ This paper is the founder (earliest)
-
Approximately 85% of reads from methanol-fixed Drosophila melanogaster embryo Drop-seq mapped uniquely to the BDGP6 reference genomescRNA-seq drosophila-melanogaster-embryo 2017×1papers★ This paper is the founder (earliest)
-
Approximately 65% of reads from methanol-fixed human-mouse Drop-seq mixing experiments mapped uniquely to a single species genome, confirming single-cell integrity is preserved after fixationscRNA-seq mixed-human-mouse 2017×1papers★ This paper is the founder (earliest)
-
Methanol fixation followed by FACS sorting of postnatal mouse cerebellum yields only 12–19% cell recovery from input materialscRNA-seq mouse cerebellum down 2017×1papers★ This paper is the founder (earliest)
-
Approximately 75% of reads from FACS-sorted methanol-fixed mouse hindbrain Drop-seq mapped uniquely to the mouse genomescRNA-seq mouse cerebellum 2017×1papers★ This paper is the founder (earliest)
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.
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
Scope — pmid-28526029
Paper: Alles J, Karaiskos N, Praktiknjo SD, et al. "Cell fixation and preservation for droplet-based single-cell transcriptomics." BMC Biology 2017. DOI 10.1186/s12915-017-0383-5 · PMCID PMC5438562.
Code: https://github.com/rajewsky-lab/dropbead (authors' own R package, v0.3.1,
HEAD d746c6f3b32110428ea56d6a0001ce52a251c247).
Data: GEO GSE89164 (Drop-seq DGE matrices + clustering tables; raw reads in SRA).
The pipeline
Drop-seq raw reads → (STAR alignment + Drop-seq tools, hg38/mm10/dm6/mm10) →
digital gene expression (DGE) matrix (genes × cell-barcodes, UMI counts) →
dropbead R package for cell calling, species separation (barnyard),
per-cell gene/UMI statistics, and sample-vs-sample correlation → figures.
The DGE matrices are deposited at the GSM level (e.g.
GSM2359902_live_dge.txt.gz, GSM2359903_fixed_dge.txt.gz; genes prefixed
hg_/mm_). The cluster assignment tables are deposited at the series level
(GSE89164_clusters_dmel.txt.gz, GSE89164_clusters_mm.txt.gz). These are the
outputs of the upstream alignment/DGE pipeline and the inputs to dropbead.
IN SCOPE (pipeline-derived, reproduced here)
Applying the authors' own dropbead to the deposited DGE matrices (P16-valid: the
shipped tool on the shipped data, per described parameters threshold=0.9 etc.).
| id | reported result | paper loc | how reproduced |
|---|---|---|---|
| C1 | "a total of 4873 cells" (Drosophila embryo, 7 runs pooled) | Fig 3 legend | row count of GSE89164_clusters_dmel.txt.gz |
| C2 | "representing 4366 cells" (mouse hindbrain) | Fig 4 legend | row count of GSE89164_clusters_mm.txt.gz |
| C3 | live vs methanol-fixed gene expression "highly correlated (R ≥ 0.95)" | Results / Fig 2c | dropbead::compareGeneExpressionLevels(live, fixed, 0.9, 0.9) per-species Pearson R on log2 pseudo-bulk |
| C4 | "median transcript and gene numbers from fixed cells … similar to those of live cells" | Results / Fig 2b | computeGenesPerCell/computeTranscriptsPerCell medians, live vs fixed |
| C5 | species-mixing (barnyard): fixation "did not substantially increase the doublet rate"; species calls at 90% purity | Fig 2a | classifyCellsAndDoublets(threshold=0.9) human/mouse/mixed counts, live vs fixed |
OUT OF SCOPE (not attempted, with reason)
- Wet-lab methanol fixation/rehydration protocol, viability, RNA integrity — not computational.
- Upstream alignment (SRA fastq → STAR → DGE). The DGE is deposited, so we reproduce from the DGE rather than regenerate it; re-running STAR on the SRA runs is the hard ~20% and adds no claim the deposited DGE doesn't already pin.
- Clustering / t-SNE / cell-type identification (Fig 3/4 panels beyond the cell count): stochastic (PCA+tSNE+graph clustering), not specified to the seed level; only the deterministic cell count (C1/C2) is in scope.
- dmel/mm median genes/UMIs (~1000/~3000; ~800/
1200): would require merging the 7 dmel + 2 mm per-GSM DGEs; attempted only as a stretch (the clusters tables hold labels, not full DGE). Reported as approximate ("") in the paper. - Live/Fixed-vs-bulk correlation (R ≥ 0.79): needs the bulk mRNA-seq GSMs; secondary, attempted only if the live/fixed core succeeds.
Grading note
Every comparison is provisional (HARD RULE 5). C1/C2 are exact deterministic file facts (strong fabrication check). C3 reruns the authors' correlation routine. C4/C5 compare against figure-read / qualitative statements ("similar", "≥") → within-tol or partial at best, not exact.
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.
Strong, clean reproduction: the two headline cell counts match the deposited cluster tables exactly (4873 Drosophila, 4366 mouse), and the remaining claims reproduce with the authors' own dropbead v0.3.1 code on the deposited GSE89164 DGEs (mouse R=0.95 exact, median genes 2528≈2529, doublet rate not increased). The only deviation is human R=0.93 vs the reported >=0.95 (Fig 2c), a minor gap most likely from cell-selection thresholding — on our methodology side, not the authors'. Everything is derivable from shipped data+code with no fabrication 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.
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.