Integrity review of 'Narrow-spectrum resource-utilizing bacteria drive the stability of synthetic communities through enhancing metabolic interactions' (Wang et al., Nature Communications, 2025)

Academic fraud report · Geng Detector

Summary

Verdict: ⚠️ Questionable (downgraded from a higher suspicion rating after re-verification). The paper is a Nature Communications 2025 study from Nanjing Agricultural University on synthetic microbial community stability. Re-verification found that the strongest signals in the original automated report—last-digit χ²=331.9, Benford MAD, and adjacent last-digit 15.2σ deviation—do not correspond to any statements in the published paper and were therefore removed as unsupported. No image reuse, Western blot splicing, or other hard image-manipulation evidence was found. Three residual concerns remain, each with a plausible benign explanation: (1) Figure 3c scatter shows horizontal banding, but the paper explicitly states the data are ~25,000 pairwise SMETANA simulations of semi-discrete MIP/MRO scores with resource-utilization width simplified to integer substrate counts; (2) reported values to 0.01 precision (e.g., 46.39, 66.08, 35.50) are inconsistent with typical colorimetric uncertainty, but auto-export from spectrophotometer/microplate software is a common benign source; (3) a single-image machine-forensic noise variance signal (144 blocks, mean 4145, CV=0.433) is a weak internal metric unsupported by the paper. The Fig. 4 layout uniformity and n=6 vs n=3 replicate discrepancy were both cleared upon re-examination. Confidence is limited by lack of access to raw replicate data.

Verdict

⚠️ Questionable (存疑) — downgraded from the original report's higher-suspicion rating after item-by-item re-verification. No confirmed fabrication; all remaining flags have plausible benign interpretations, and key statistical signals cited in the prior automated report could not be located in the published paper.

Key findings

⚠️ Figure 3c scatter banding — ~25,000 MIP/MRO points form horizontal bands, but the Methods section states these scores are SMETANA simulations of semi-discrete metabolic-flux indices and that "resource utilization width was simplified as the total number of substrates" (integer). Banding is a natural visual artifact of discrete/quantized outputs, not evidence of fabrication.
⚠️ Over-precise reported values — quantitative results throughout Figs. 1, 2b, 4, 5, 6 are reported to 0.01 (e.g., nitrogenase 3517 and 890 nmol C₂H₄·h⁻¹·mg⁻¹; phosphate solubilization 46.39 mg·L⁻¹; IAA 66.08 mg·L⁻¹; resource-utilization width 35.50, 36.76, 37.32). Colorimetric assays (CAS, Salkowski, molybdenum-antimony) typically carry ±5–10% uncertainty, but microplate/spectrophotometer software commonly auto-exports multi-decimal precision.
⚠️ Single-image machine-forensic noise variance — automated ELA on one image (img-000.png) reports 144 valid noise blocks (32/64 px, 9×9 box high-pass), variance mean 4145, CV=0.433, with 4 outlier blocks. These are internal forensic metrics with no corroborating statement in the paper; only 2 images were analyzed, so no batch pattern can be established.
✅ n=6 vs n=3 sample-size discrepancy (Fig. 6 vs Figs. 1, 4, 5) — cleared. The Methods section for the pot experiment explicitly states six replicates per treatment, consistent with the figure caption (3/3 programmatic match).
✅ Fig. 4 panel layout consistency — cleared. Shared reference lines at 1.0 × 10⁴ and 1.0 × 10⁵ cfu·g⁻¹ soil (red/blue dashed) and uniform y-axis scales are an intrinsic design choice for a time-course series.
❌ Last-digit χ²=331.9 and Benford MAD values — removed. These specific statistics are not reported in the paper and appear to be internal automated-tool outputs that do not map to any published claim.

Evidence highlights

DOI: 10.1038/s41467-025-61432-7
Authors: Wei Wang, Yanwei Xia, Panpan Zhang, Mengqing Zhu, Shiyi Huang, Xinli Sun, Zhihui Xu, Nan Zhang, Weibing Xun, Qirong Shen, Youzhi Miao & Ruifu Zhang
Journal: Nature Communications, 2025
Corresponding institution: Jiangsu Provincial Key Lab of Solid Organic Waste Utilization, Nanjing Agricultural University
Fig. 3 caption confirms pairwise SMETANA simulation origin of plotted points
Methods: "Community simulations were performed with SMETANA using default parameters"; resource-utilization width simplified to integer substrate count
Fig. 6 caption: "Bars represent the mean ± s.d. (n=6)... Error bars represent mean ± SD of biological replicates (n=6)"
Programmatic match: 5/9 of cited precise values found in the paper (4/9 weak match); 3/3 sample-size statements matched
Machine-forensic metrics (144 blocks, mean 4145, CV=0.433, 4 outliers) are not present in the published text

Notes

The strongest suspicion in the original automated report (last-digit χ², Benford MAD, 15.2σ deviation) could not be cross-validated against the paper and is therefore not retained as a finding.
All three remaining flags are explainable by documented methodology (discrete simulation outputs) or routine laboratory software behavior (auto-exported decimal precision), and none is independently corroborated by image-level evidence.
A definitive assessment would require the authors' raw per-replicate values for Figs. 1, 2b, 4, 5, and 6 and the full ~25,000-point output of the Fig. 3c simulation; independent Benford and last-digit tests on those data would be diagnostic.
Confidence: moderate that no fabrication is present based on current evidence; low that all concerns are fully benign pending raw-data release.
Recommended next step: request raw replicate data from the corresponding author (Youzhi Miao, yzmiao@njau.edu.cn) before any formal escalation.