Integrity Screening Report: "荧光PCR多规格耗材自动对焦算法及装置研究" (DOI: 10.3788/CJL241419)

Academic fraud report · Geng Detector

Summary

AI-assisted integrity screening of a 2025 paper in 《中国激光》 (DOI: 10.3788/CJL241419) on autofocus algorithms for multi-spec fluorescence PCR consumables rates the paper as highly suspicious. The most robust finding is a direct internal contradiction: the English abstract claims Ct differences between the two tube formats range from 0 to 0.32, but Table 4 shows a difference of 0.83 at 1×10¹ copy/µL (34.75 vs 35.58); this was verified verbatim. Table 3 fluorescence readings across five serial dilutions show near-perfect ~2-fold ratios across five orders of magnitude with no realistic detector noise, across two supposedly independent experiments. A third finding is that both tubes yield identical Ct values of 32.02 (two decimal places) at 1×10² copy/µL, though coincidence cannot be excluded. Visual findings—overly smooth, equidistant amplification curves (Fig. 9c,d) and near-identical background textures in Figs. 5/6—have plausible benign explanations and are downgraded. Machine forensics (copy-move, noise, PRNU) was largely cleared as benign patterns from regular well-plate structures. Overall confidence is moderate-to-high for the numerical findings but limited by reliance on AI analysis; final determination requires original raw data and institutional investigation.

Verdict

Highly suspicious (🟠 per source report; Bayesian synthesis rates 🔴 highly suspect, posterior 97.6%, 95% CI [83%, 99.6%], BF = 768.72). This is a screening-level assessment, not a finding of misconduct. The strongest, independently verifiable issues concern internal numerical inconsistencies and implausibly regular data; image-forensic leads were largely cleared as benign.

Key findings

Finding 2 (verified, 🟠): English abstract states Ct differences "ranging from 0 to 0.32", but Table 4 reports 34.75 (0.2 mL tube) vs 35.58 (0.1 mL tube) at 1×10¹ copy/µL — a difference of 0.83, directly contradicting the abstract. All three data points verified verbatim in the original text.
Finding 1 (verified, 🟠): Table 3 fluorescence readings show near-ideal ~2-fold ratios across serial dilutions (e.g., 0.2 mL tube: 2020242, 992040, 556595, 267178, 141937; ratios ≈ 2.04, 1.78, 2.08, 1.88) across five orders of magnitude, with no saturation, background drift, or random variation; the 0.1 mL tube series (2031449, 1075048, 606615, 272966, 141375) tracks closely at every dilution. All 10 data points verified verbatim. The paper provides no replicates, error ranges, or normalization details.
Finding 3 (downgraded, ⚠️): Both tubes report an identical Ct of 32.02 at 1×10² copy/µL (verified at both occurrences). Exact two-decimal coincidence in independent PCR replicates is unlikely, but a genuine or rounding coincidence cannot be excluded; requires original Ct export files.
Finding 4 (downgraded, ⚠️): Amplification curves in Fig. 9(c)(d) appear overly smooth, noise-free, and near-equidistantly parallel; benign explanations (default software smoothing, plotting normalization) not ruled out; no smoothing described in Methods.
Finding 5 (downgraded, ⚠️): Fig. 5/6 image sequences show identical background texture, illumination, and well layout differing only in sharpness; consistent with a fixed camera and axial lens movement as described in the paper's imaging principle.

Evidence highlights

English abstract: "minor Ct value differences regarding the tested nucleic acid concentrations, ranging from 0 to 0.32" vs Table 4 values 34.75 and 35.58 (difference 0.83).
Table 3: 10 data points verified verbatim; near-perfect 2-fold linearity across 4 dilution steps in two independent experiments.
Table 4: "32.02" appears identically for both tube formats at 1×10² copy/µL; other row differences are 0.83, 0.20, 0.32, 0.11, 0.07.
Machine forensics: copy-move hits (I5, I7, I8, I10, I12, I14, I15, I17) attributed to regular 96-well plate structures and cleared; noise-variance inconsistencies (I1, I2, I4, I6, I9, I11, I13, I16) insufficient; PRNU weak-similarity results (I18–I31) cleared. Benford test not applied (data not suited to its premises). 32 leads excluded from synthesis.

Notes

Report is AI-assisted; final determination of misconduct requires professional institutional investigation and the authors' raw data (raw fluorescence images, original Ct exports, original amplification curve data).
Recommended actions: request raw data from authors; optionally raise on PubPeer; report to the journal (《中国激光》) focusing on the abstract–Table 4 contradiction.
DOI: 10.3788/CJL241419. Published 2025, vol. 52, no. 9; received 2024-12-05; online 2025-02-12.
The Bayesian synthesis treats evidence as conditionally independent, which may overstate combined strength for correlated evidence.