Verdict
🟠 Highly suspicious. Text-only inspection of the UDA-seq paper (Nature Methods, DOI: 10.1038/s41592-024-02586-y) surfaced multiple statistical, productivity, and writing-pattern anomalies. Definitive conclusions require access to raw images, source data, and ethics/IRB records.
Key findings
- Statistical anomaly — overly perfect correlation (Fig. 4d): A Pearson correlation of r = -0.99 (P = 0.00125) between a metabolic-imaging-derived score and eGFR in clinical kidney biopsy samples is biologically implausible. Such near-perfect linearity is a classic red flag for fabricated/truncated data (e.g., Y = a - bX + minimal noise) or selective sample exclusion. Severity: 🔴
- Statistical anomaly — P-values on the threshold (Fig. 6h): Reported post-Hochberg P values for MYC expression changes after CRISPRi are 0.039, 0.039, 0.047, and 0.022. The clustering of values just above 0.05 in a dataset of tens of thousands of cells is consistent with P-hacking or selective reporting. Severity: 🟠
- Productivity / timeline anomaly: The paper claims, within a single novel method, parallel execution of: a 35-sample kidney biopsy Multiome cohort yielding 207,789 cells; a 38-female-donor PBMC 5'-RNA + VDJ dataset yielding 150,018 cells; a 255-sgRNA CRISPR screen in SNU16 cells; and a three-species (human, mouse, locust) brain-nuclei mixing experiment. Achieving such throughput on a first-attempt new method is unusual. Severity: 🟠
- Template-like figure legends (Fig. 1e, 2c, 3a, 3d, 4a, 4b, 5e, 6f, 6h): The exact sentence '*Box plots show the interquartile range with the median marked. The whiskers extend up to 1.5 times the interquartile range, and the outliers are not displayed.*' is reused across at least nine figure legends, suggesting copy-paste from a default plotting template and a uniform (rather than experiment-tailored) analysis pipeline. Severity: 🟡
- Technical plausibility concern (Methods): The paper uses methanol-fixed PBMC with the 10x Genomics GEM-X v3 chemistry on Chromium X. Heavy methanol fixation typically degrades single-cell RNA quality; the claim of data quality 'comparable' to standard fresh protocols without supporting raw QC metrics is unverified. Severity: 🟡
- Fig. 4d: r = -0.99, P = 0.00125 — pearson correlation between metabolic-imaging score and eGFR in clinical kidney biopsy samples; absolute value 0.99 is the largest red flag in the paper.
- Fig. 6h: Adj. P = 0.039, 0.039, 0.047, 0.022 for MYC expression changes under CRISPRi (Hochberg-corrected).
- Cell numbers reported: 207,789 cells (kidney Multiome); 150,018 cells (PBMC VDJ); 255 sgRNAs (CRISPR screen); three-species mix including locust brain nuclei.
- Timeline: Received 24 June 2024; experiments plausibly executed in 2023–early 2024, on a novel method.
- Recurrent legend text observed in at least nine figure legends (Fig. 1e, Fig. 2c, Fig. 3a, Fig. 3d, Fig. 4a, Fig. 4b, Fig. 5e, Fig. 6f, Fig. 6h).
- This audit is text-only. No pixel-level splicing/duplication analysis was possible on microscopy, UMAP, or western-blot images; no bioinformatic re-clustering of the 207,789-cell dataset was performed.
- The r = -0.99 finding is the strongest signal and warrants requesting raw paired values and any sample-inclusion/exclusion criteria from the authors.
- Independent verification of IRB/ethics approval dates, GEM-X v3 access/availability, and sample procurement timelines is recommended.
- All findings are preliminary; the authors retain the right to respond, and final determination of misconduct requires institutional investigation.
- Disclaimer: This report is AI-assisted and intended for academic discussion only. False positives and false negatives are possible.