Geng Integrity Review: UDA-seq universal droplet microfluidics combinatorial indexing for massive-scale multimodal single-cell sequencing

Academic fraud report · Geng Detector

Summary

Verdict: Questionable. This Nature Methods paper (Vol. 22, June 2025; DOI 10.1038/s41592-024-02586-y) introduces UDA-seq and reports an unusually broad set of applications within a tight timeline (Received 24 June 2024). Five findings are flagged. (1) Figure 4d reports a correlation of r = -0.99, P = 0.00125 between a single-cell-derived metabolic injury score and clinical eGFR, based on only n=5 donors (per Figure 2a metabolomics n=5); such near-perfect correlation is biologically implausible for heterogeneous clinical single-cell data and suggests possible cherry-picking or data curation. (2) The manuscript combines species-mixing validation, Multiome profiling of 35 clinical kidney biopsies with extensive sub-clustering (60/68 subpopulations), SCENIC+, CellChat, 38-sample PBMC 5'-RNA/VDJ aging cohort analysis, and a 46-gene (255-sgRNA) CRISPR screen, completed within an extraordinarily compressed timeline. (3) Figure 6h uses a two-sided t test with claimed n=5 for target genes and n=10 for controls; ambiguity over whether n denotes biological replicates, cells, or sgRNAs raises concerns about statistical power. (4) Novelty is questioned as UDA-seq relies on overloading commercial 10x Genomics instruments and resembles existing combinatorial indexing. (5) Pixel-level image duplication could not be assessed from PDF text. All findings require verification via raw data and institutional review.

Verdict

🟡 Questionable — Several statistical and methodological concerns warrant investigation, though no definitive evidence of misconduct is established from the text alone.

Key findings

Figure 4d (clinical correlation): Reported correlation r = -0.99, P = 0.00125 between a single-cell-derived metabolic injury score and clinical eGFR with only n=5 metabolomics donors (per Figure 2a). Such near-perfect linearity is implausible for heterogeneous clinical single-cell data and may indicate data curation.
Output anomaly: The manuscript, Received 24 June 2024, reports a remarkably broad scope: species-mixing validation, Multiome profiling of 35 clinical kidney biopsies (60/68 sub-population clustering, SCENIC+, CellChat), 38-sample female aging PBMC 5'-RNA/VDJ cohort, and a CRISPR screen targeting 46 genes (255 sgRNAs). The compressed timeline and combined depth of wet-lab and computational analyses are unusual.
Figure 6h (CRISPR statistics): Two-sided t test reported with n=5 (target genes) and n=10 (controls). Ambiguity over the unit of n (biological replicates, cells, or sgRNAs) raises questions about statistical power and significance values such as P = 0.039.
Novelty/methodological packaging: The UDA-seq concept is described as "Universal" and based on "Post-indexing," yet the method appears to rely on overloading commercial 10x Genomics droplet instruments and is closely related to the authors' prior FIPRESCI-seq2 (Ref 10) and combinatorial indexing approaches, raising novelty-overstatement concerns rather than fabrication.
Image duplication: Could not be assessed because only PDF text was available; no textual inconsistency was detected.

Evidence highlights

Figure 4d: r = -0.99, P = 0.00125 for eGFR vs. single-cell metabolic injury score; metabolomics donor n=5 (from Figure 2a).
Figure 6h: Two-sided t test; reported n=5 for target genes (BRD1, BRD4, BAZ2A, BRD8), n=10 for controls; reported P = 0.039.
Manuscript timeline: Received 24 June 2024; published online 20 January 2025.
Scale: 35 clinical kidney biopsies, 60/68 sub-population clustering, SCENIC+, CellChat analyses; 38-sample PBMC 5'-RNA/VDJ aging cohort; 46-gene / 255-sgRNA CRISPR screen.
References: Ref 10 (FIPRESCI-seq2) and Ref 62 (Protocol article) flagged for self-citation relevant to novelty framing.
DOI: 10.1038/s41592-024-02586-y

Notes

Confidence is moderate for the statistical anomalies (Findings 1 and 3) and lower for the output/novelty critiques (Findings 2 and 4), which are circumstantial.
Pixel-level image-duplication analysis is not possible from PDF text; raw figures and uncropped images should be requested.
Recommended follow-up: request raw scatter data for Figure 4d to verify r = -0.99; clarify n definition in Figure 6h; review Ref 41 SCIPAC for chronological consistency; consider posting concerns on PubPeer.
This report is AI-assisted and intended for academic discussion; final determinations require institutional investigation.