Concerns regarding 'Interfacial interaction-driven rapid capture and on-site analysis of nano- and microplastics enabled by multifunctional magnetic adsorbent' (Nature Water, DOI: 10.1038/s44221-026-00610-3)

Academic fraud report · Geng Detector

Summary

This report compiles concerns raised about a Nature Water paper reporting a multifunctional magnetic adsorbent for nano- and microplastic capture and on-site analysis. The verdict is 'highly suspicious'. Pixel-level image-duplication analysis could not be performed because raw high-resolution image files were unavailable; concerns are therefore based on textual data, statistical descriptions, and internal logical consistency. Five issues are flagged: (1) an implausibly perfect monotonic correspondence between DFT binding energies (-95.38, -91.21, -87.23, -81.93, -67.19, -57.43 kcal/mol) and macroscopic inhibition order across six polymers; (2) PCA variance contributions PC1 = 86.1% and PC2 = 13.3%, summing to 99.4%, which is atypical for spectroscopic chemometric data; (3) suspiciously polished numeric values (LOD = 0.405 mg/L, cost = US$0.035/g, linear fit y = -0.226x + 0.959, R² = 0.998); (4) the silane reagent is consistently misspelled as APETS instead of APTES; (5) an unusually large and broad scope of work completed in under 9 months. Confidence is moderate given the absence of raw data inspection.

Verdict

🟠 Highly suspicious. No definitive fabrication is proven from the available material. Concerns are derived from textual data patterns, statistical anomaly, and workload scope. Original raw data and high-resolution figures are required before a stronger determination can be made.

Key findings

Implausibly tight one-to-one correspondence between DFT binding energies and macroscopic enzyme-like inhibition across PA, PLA, PET, PS, PVC, and PE.
PCA reported as PC1 = 86.1%, PC2 = 13.3%, summing to 99.4% — atypical for spectroscopic chemometrics.
Multiple overly polished figures: LOD = 0.405 mg/L; cost = US$0.035/g; linear fit y = -0.226x + 0.959, R² = 0.998.
The silane coupling reagent 3-aminopropyltriethoxysilane is written throughout as APETS rather than the standard APTES.
The full scope of work — synthesis, multi-instrument characterization, enzyme-mimetic kinetics, adsorption matrices (6 plastics × 7 sizes), EEM, EIS, DFT, machine learning (PCA, LDA, RF), real-sample assays, 0.35-kg/10-L pilot column, and Py-GC/MS — was reportedly completed between 12 May 2025 (received) and 11 February 2026 (accepted).

Evidence highlights

Finding 1 (DFT vs. inhibition). Reported DFT binding energies: -95.38, -91.21, -87.23, -81.93, -67.19, -57.43 kcal mol⁻¹, matching the experimentally derived inhibition order PA > PLA > PET > PS > PVC > PE. Inter-pair energy gaps (≈4.17, 3.98, 5.30, 14.74, 9.76 kcal/mol) form a near-regular descending ladder. A perfectly monotonic, near-linear alignment between ideal-state quantum-chemistry energies and a macroscopic multi-polymer inhibition series is unusual and may indicate selective reporting or curated fitting.
Finding 2 (PCA). PC1 + PC2 = 99.4% of variance. Such a high cumulative variance in the first two components is uncommon in absorbance/fluorescence spectral data and is consistent with over-smoothed inputs, unremoved baseline drift, autocorrelated features, or synthesized data.
Finding 3 (Numeric polish). LOD = 0.405 mg L⁻¹ (3σ); treatment cost = US$0.035 g⁻¹; calibration y = -0.226x + 0.959, R² = 0.998. For a colorimetric microplastic assay, an R² of 0.998 with an intercept of 0.959 and round coefficients is a textbook "too neat" pattern that may reflect cherry-picking or repeated best-fit selection.
Finding 4 (Chemistry typo). Standard abbreviation APTES (3-aminopropyltriethoxysilane) is written as APETS in the figure caption and text — a basic nomenclature slip in a high-profile venue.
Finding 5 (Timeline/scope). Received 12 May 2025; accepted 11 February 2026 (< 9 months). The breadth of activities (kilogram-scale synthesis; SEM/TEM/BET/XRD/FTIR/XPS/Zeta/contact angle; photothermal; enzyme-mimetic kinetics; 6 × 7 adsorption matrix; 3-D EEM; EIS; DFT; PCA/LDA/RF; real water and food-container samples; 10-L column pilot at 0.35 kg; Py-GC/MS verification) is exceptionally large for the timeframe, raising questions about data authenticity and whether parts were assembled across distinct projects.

Notes

Image-level duplication and splicing checks were not performed because original high-resolution figure files were not available to the reviewer; all five findings rely on the published text, figure descriptions, and reported numeric values.
The DFT-inhibition correspondence (Finding 1) is the strongest statistical concern; PCA cumulative variance (Finding 2) is the second strongest. Findings 3–5 are supportive but individually weak indicators.
Recommended follow-up: request uncropped raw figures, PCA input spectra, DFT input coordinates/output files, full kinetics raw data, and laboratory notebook timestamps from the corresponding authors (jmgong@mail.ccnu.edu.cn; chenchaojili@whu.edu.cn); post a structured query on PubPeer focused on Findings 1 and 2; ask the authors to release machine-learning code and raw spectral datasets.
This report is AI-assisted and intended for academic-discussion purposes only; final determinations of misconduct must come from an institutional or publisher investigation. DOI preserved: 10.1038/s44221-026-00610-3.