Academic Integrity Alert: Discrepancies in eLife Reviewed Preprint "Metabolic compensation via gluconeogenesis explains the non-essentiality of glycogen phosphorylase as an insecticidal target in Plutella xylostella" (DOI: 10.7554/eLife.111144.2)

Academic fraud report · Geng Detector

Summary

This AI-assisted integrity review of the eLife reviewed preprint (DOI: 10.7554/eLife.111144.2) concludes the paper is highly suspect (orange rating). One finding is confirmed: an internal contradiction in the DFB inhibition experiment, where the text and Figure 2C caption state a concentration range of 0.54–1700 nM yet also report maximum inhibition of 61.96 ± 2.54% at 100 µM — a ~60-fold discrepancy affecting the dose basis of a core conclusion. Four findings are downgraded but retained as concerns: inconsistent RNAi knockdown values (87.59% vs 54.66% at the same dose/timepoint), conflicting protein-reduction timepoints (72 h vs 24 h) with nearly identical magnitudes, suspected splicing in Figure 2A, and statistical-fingerprint plus batch image-forensics signals (including copy-move matches in Figure S5/S6 wing photographs that layout cannot explain). Benign explanations exist for the downgraded items but lack explicit support in the manuscript. Independent anomalies (1 confirmed + ≥3 downgraded) meet the orange threshold; the authors should be asked for raw data. Final determination requires institutional investigation; false positives are possible.

Verdict

Highly suspect (orange, 🟠). After item-by-item re-verification, 1 finding is confirmed (internal concentration-range contradiction), while 3+ further findings are downgraded but retained as unresolved concerns. This combination meets the threshold for recommending in-depth investigation and a request for original data. Note: the report's automated Bayesian synthesis section labels the case red (posterior ≥99.9%), but the human-style review verdict is orange due to documented benign explanations for most downgraded findings.

Key findings

Confirmed — DFB concentration contradiction: Results, Figure 2C caption, and Methods state DFB was tested at 0.54 nM–1700 nM, yet the same section reports "maximum inhibition observed did not exceed 61.96 ± 2.54% (mean ± SEM) even at 100 µM." 1700 nM ≈ 0.55 µM, a ~60-fold discrepancy with the stated 100 µM. Methods describe only a single workflow (0.54–1700 nM, constant DMSO), so the benign hypothesis (screening at 100 µM plus formal dose-response) has no explicit support. Directly affects the dose basis for the "DFB does not inhibit PxGP" conclusion.
Downgraded — RNAi knockdown inconsistency: 87.59% knockdown (12.47 ± 1.68% of dsGFP control) vs 54.66% reported at the same 14,000 ng/µL dsGP, 48 h timepoint. Possibly two independent experiments with biological variability; different statistics (one-way ANOVA vs independent t-test) are compatible with this, but the 33-point gap is unexplained.
Downgraded — protein decline timing conflict: 29.39% reduction at 72 h (Figure 9D) vs 30.78% at 24 h (Figure S4). Near-identical magnitudes with fully shifted peak timepoints is a residual concern entangled with the RNAi finding.
Downgraded — Figure 2A possible improper splicing: No divider between Coomassie gel and anti-His Western blot, abrupt background change, non-corresponding marker bands. The caption does describe two parallel panels (partial benign support), but original gel/membrane images are needed.
Downgraded — statistical fingerprints: Identical SD decimal places ([S3], consistent with journal formatting conventions), Bayes Factor ≈ 10⁹ on 29 SD values ([S4]), severely uneven last-digit distribution with odd/even ratio 86/304 ([S2]), Benford deviations at digits 2–4 ([S1], mixed data streams weaken test premises). The [S2] odd/even ratio (0.28) and 16.7σ adjacent-digit correlation lack a documented benign explanation; combined with suspiciously uniform error bars (Figures 3B, 9D), this warrants raw-data review but is not alone incriminating.
Downgraded — image forensics: Copy-move matches in many figures are explainable by regular chart elements at 32 px multiples. Retained concerns: img-025/img-026 (Figure S5/S6 wing photographs) with 50 and 64 matching block pairs at offset (32,0) cannot be explained by chart layout; img-001 noise inhomogeneity overlaps the Figure 2A splicing concern.
No red flags: Timeline (2026 submission, reagents TargetMol T37577, MedChemExpress HY-P2757, Cresset Flare V11), no confirmed loading-control reuse or rotation/flipping; the authors voluntarily narrowed conclusions post-review (a positive sign).

Evidence highlights

All cited numeric data points were programmatically verified against the source text: 4/4 for Finding 1, 5/5 for Finding 2, 2/2 for Finding 3, 9/9 for Finding 5 (e.g., "12.78 ± 2.42%", "41.60 ± 4.85%", "56.11 ± 5.64%", "3.95 ± 0.44-fold").
Example forensic signals: img-026 copy-move at offset (32,0), 64 matched pairs (logLR = 1.46); img-025, 50 pairs (logLR = 1.46); img-014 noise variance CV = 1.21 (logLR = 0.99); img-001 copy-move at (0,32), 68 pairs; img-018, 137 pairs.
Automated synthesis (LR version geng_lr_v3, prior fixed at 5%, conditional independence assumed) reports posterior ≥99.9% and BF = 4.58e+10; this may overstate joint strength of correlated evidence and does not constitute a fraud determination.
26 leads were excluded during review (benign explanations confirmed / test preconditions unmet / informational).

Notes

Recommended actions: request raw data from authors (Figure 2C concentration readings and the 100 µM vs 1700 nM explanation; Figure 9D vs S4 raw protein quantification; Figure 7B vs S3 raw Ct values; Figure S5/S6 original wing photographs with EXIF); post concerns on PubPeer; journal/institutional reporting contingent on response.
This report is AI-assisted, for academic discussion only. Final determination of misconduct requires professional institutional investigation; false positives and negatives are possible. Researchers' reputations must be respected; official investigation conclusions take precedence.