IDA-UPLC-QTOF-MS / MS-based non-targeted extract screening and toxicity driving factor identification method

Abstract

The invention discloses a non-targeted extract screening and toxicity driving factor identification method based on IDA-UPLC-QTOF-MS / MS. Real extracts are obtained by simulating a real use scene and are subjected to chemical analysis, qualitative chemical characteristics are subjected to three-stage screening, then potential toxic action targets and mechanisms of the real extracts are revealed through calculation and simulation, and the toxicity driving factors of the real extracts are identified. A few toxicity driving factors which contribute to health risks are positioned, and a clear direction is provided for subsequent experimental verification, so that traditional high-flux toxicity screening is greatly replaced, and a large amount of research and development and detection cost is saved.

Description

Technical Field

[0001] This invention belongs to the field of additive detection technology, specifically relating to a non-targeted leachate screening and toxicity driver identification method based on IDA-UPLC-QTOF-MS / MS. Background Technology

[0002] Dietary intake is a significant route of human exposure to chemical additives. Various chemicals from single-use food packaging materials may migrate into food, threatening consumer health. However, current methods for detecting these leachates have fundamental limitations: first, simulation distortion—common direct extraction methods using organic solvents cannot reflect the migration patterns of substances in real food media; second, detection blind spots—targeted modes relying on standards (such as multiple reaction monitoring, MRM) can only cover a small number of known additives, failing to effectively identify novel additives, degradation products, and unknown impurities. Furthermore, complex matrix interference severely affects detection accuracy, and differentiated analytical schemes for different materials (such as biodegradable plastics and paper-plastic composites) are lacking. Therefore, there is an urgent need to establish an innovative methodology that can realistically simulate exposure, comprehensively screen for unknown substances, and accurately identify key risk factors. Summary of the Invention

[0003] The purpose of this invention is to provide a non-targeted leachate screening and toxicity driver identification method based on IDA-UPLC-QTOF-MS / MS. By simulating real-world usage scenarios, performing panoramic chemical analysis and three-level screening of leachates, and then revealing their potential toxicity targets and mechanisms through computational simulation, the method identifies the few toxicity drivers that contribute the most to health risks, providing a clear direction for subsequent experimental verification. This significantly replaces traditional high-throughput toxicity screening, saving substantial R&D and testing costs.

[0004] The technical solution of the present invention is as follows:

[0005] A non-targeted leachate screening and toxicity driver identification method based on IDA-UPLC-QTOF-MS / MS includes the following steps:

[0006] (1) The sample was made into a sample powder with a particle size of less than 500 μm, and then extracted with ultrapure water, filtered and lyophilized to obtain lyophilized powder.

[0007] (2) High-resolution liquid chromatography-tandem mass spectrometry based on information-dependent acquisition mode was used to analyze the solution of lyophilized powder and obtain mass spectrometry data containing all detectable chemical features.

[0008] (3) The stable characteristic peaks obtained from the mass spectrometry data are matched with the database for qualitative analysis and confidence level classification, and their physicochemical information is obtained;

[0009] (4) Perform data cleaning on the chemical features to remove interfering features and false positive signals, and obtain high-confidence chemical features for subsequent screening;

[0010] (5) The chemical characteristics are screened in three levels as follows:

[0011] (5.1) Multivariate statistical screening: Through orthogonal partial least squares discriminant analysis, differential factors with significant differences between different samples were screened from high-confidence chemical characteristics;

[0012] (5.2) Calculate the initial toxicological screening: predict the toxicity of differential factors and screen out high-risk factors with predicted toxicity level of high toxicity;

[0013] (5.3) Molecular docking fine screening: Taking the toxic pathways that are generally activated by high-risk factors as the core toxicity targets, all high-confidence chemical features are used as ligands to perform molecular docking simulation with the protein structure of the core toxicity targets, and chemical features with binding energy lower than the preset value are used as toxicity driving factors.

[0014] In some preferred embodiments, the ultrapure water extraction temperature is room temperature, and the ultrapure water extraction time is 7 days.

[0015] In some preferred embodiments, the extraction of stable characteristic peaks is characterized by a quality deviation of 10 ppm, a signal strength deviation of 30%, a signal-to-noise ratio (S / N) ≥ 3, and a peak width range (s) = c(5,30).

[0016] In some preferred embodiments, the screening criteria for high-confidence chemical features are as follows: (1) the confidence level of the Metabolomics Standards Initiative is Level 2; (2) stable feature peaks with a relative standard deviation (RSD) of <30% are retained based on quality control; (3) the feature peaks are missing in the blank sample but present in the target sample with an intragroup deletion rate of <50%, or the abundance of the target sample is 5 times or more higher than that of the blank sample.

[0017] In some preferred embodiments, the screening thresholds for differential factors are as follows: variable importance projection > 1.5, effect size > 1.5, statistical significance < 0.01, and abundance difference fold > 2.0.

[0018] In some preferred embodiments, step (5.2) involves performing toxicological predictions using a computational toxicology platform, then evaluating the toxicity level of the differential factors according to the GHS toxicity classification criteria, and defining factors with a predicted toxicity level of 1-2 as high-risk factors. More preferably, the computational toxicology platform is ProTox-3.0 (a predictive toxicity platform).

[0019] In some preferred embodiments, molecular docking simulations are performed using AutoDock Vina software, with a preset binding energy of -7.0 kcal / mol.

[0020] In some preferred embodiments, orthogonal partial least squares discriminant analysis is performed using the ropls package in the R language.

[0021] In some preferred embodiments, the sample is a single-use food package, and the core toxicity target is the pregnane X receptor.

[0022] In some preferred embodiments, the single-use food packaging includes at least one of polyolefins, biodegradable plastics, and paper-plastic composites.

[0023] The present invention has at least the following beneficial effects:

[0024] 1. High realism of leaching simulation: By freeze-grinding the packaging material into sample powder (particle size <500 μm), then gently leaching it with a water-based simulant (7 days), and finally testing it in the form of lyophilized powder, the loss of low-boiling-point components is avoided and the contact area between the material and the food simulant is greatly increased. This more realistically simulates the migration of chemical substances caused by wear and aging in actual use, and significantly improves the ecological validity of the test results and the accuracy of exposure assessment.

[0025] 2. Comprehensive Chemical Screening: UPLC-Q-TOF-MS / MS based on Information Dependent Acquisition (IDA) mode is used for non-targeted analysis. No pre-set standards are required. It can perform panoramic detection and identification of known and unknown chemical substances in the leachate (including intentional additives, unintentional additives and degradation products, etc.), which fundamentally overcomes the limitations of traditional targeted methods such as narrow detection range and high risk of missed detection.

[0026] 3. Intelligent and Streamlined Key Factor Identification: A three-tiered intelligent screening funnel was constructed, consisting of "multivariate statistical screening → computational toxicology preliminary screening → molecular docking fine screening." This process first identifies differential factors from high-confidence chemical features, then focuses on high-risk factors and core toxicity targets through toxicity prediction, and finally identifies a few key effector compounds that bind most strongly to core toxicity targets (such as PXR) through unbiased molecular docking simulations of all high-confidence chemical features. This process achieves automated and targeted mining of key risk information from massive amounts of data, significantly reducing human bias.

[0027] 4. In-depth prediction of toxicity mechanisms: It goes beyond identifying potential toxic substances, and identifies core targets through toxic pathway activation pattern analysis (such as heatmaps). It also uses molecular docking simulation to predict the interaction mechanism between key effector compounds and targets, providing a theoretical explanation for "why they are toxic" and pointing out a clear direction for subsequent targeted biological experiments.

[0028] 5. Significantly reduced costs for safety assessment and R&D verification: Through a three-level screening process, the number of target compounds that require further confirmation and verification through expensive experiments (such as purchasing standards for quantification and in vitro toxicity testing) is reduced from hundreds or even thousands of unknowns encountered by traditional methods to a single number of key effect compounds, thereby significantly reducing the economic and time costs of toxicological screening and product safety assessment.

[0029] 6. Strong Method Versatility and Extensibility: The method has been adapted and validated for the leachate characteristics of various materials, including polyolefins, biodegradable plastics, and paper-plastic composites. Its core "simulation-non-target screening-intelligent mining" methodology framework can be extended to non-target screening and toxicity driver identification of complex chemical mixtures in other consumer products such as toys, textiles, and cosmetics. Attached Figure Description

[0030] Figure 1 This is a flowchart of the overall technical solution of the present invention;

[0031] Figure 2 Sample photos of disposable food packaging;

[0032] Figure 3 PCA score charts for leachates from different types of single-use food packaging;

[0033] Figure 4 A statistical chart of the number of differential factors;

[0034] Figure 5 A plot showing the importance scores of the differential factors;

[0035] Figure 6 A distribution chart showing the quantity and proportion of leaching markers with toxicity levels 1-4;

[0036] Figure 7 LD50, a leaching marker for toxicity grades 1-4 50 Distribution map;

[0037] Figure 8 Heatmap of toxic pathways activated by high-risk factors;

[0038] Figure 9 This is a molecular docking binding energy distribution diagram of chemical characteristics and core toxicity targets (PXR). Detailed Implementation

[0039] The technical solution of the present invention will be further explained and described below through specific embodiments.

[0040] In the following embodiments, unless otherwise specified, the water used may be one or more of distilled water, purified water, and drinking water; the detection methods in the following embodiments are conventional detection methods unless otherwise specified; the reagents in the following embodiments are commercially available unless otherwise specified.

[0041] Unless otherwise specified, the concentration percentage and liquid feed ratio mentioned below are volume ratios (v / v), and the solid feed ratio is a mass ratio. In this application, "less than" means that the binding energy is more negative, or that the absolute value of the binding energy is greater than the preset value.

[0042] The polymer abbreviations used in this application are: PLA (polylactic acid), PBAT (polybutylene terephthalate), WSF-PP (paper-plastic material blended with straw fiber and polypropylene), PP (polypropylene), PE-CP (paper-plastic material coated with polyethylene), SF (sugarcane fiber material), PS (polystyrene), and PE (polyethylene).

[0043] Example 1: A method for non-targeted leachate screening and toxicity driver identification of disposable food packaging bags

[0044] This embodiment follows Figure 1 The process was carried out as shown, and the specific operations and test results are as follows:

[0045] 1. Material selection and preparation

[0046] Eight representative commonly used disposable food packaging materials were collected from the market and categorized into three types based on their materials: biodegradable materials, including PLA straws, PBAT food packaging bags, and sugarcane packaging boxes; paper-plastic composite materials, including PE-coated hamburger paper and rice husk cups; and polyolefin materials, including PE food storage bags, PP lunch boxes, and PS cake cups (see Table 1 for details, and product images are available in [link to product images]). Figure 2 Using clean stainless steel scissors, the sample was cut into plastic fragments of about 5 mm in size. The fragments were then pulverized using a cryogenic grinder (CryoMill, Retsch, Germany) (grinding frequency 30 Hz, grinding time 2 min). Finally, the sample powder with particles smaller than 500 μm was separated by a 500 μm stainless steel sieve and collected in a clean glass bottle.

[0047] The entire sample powder preparation process adhered to a plastic-free principle to prevent external contamination and ensure the reliability of the research. All samples underwent the following pretreatment before being shredded: three washes with ultrapure water, followed by wrapping in aluminum foil and drying in a 45°C oven. To prevent sample contamination, all glassware and stainless steel instruments were cleaned in an ultrasonic cleaning tank for 30 min. All containers were rinsed twice with methanol (AR grade, Xilong Scientific, China) and thoroughly rinsed three times with ultrapure water.

[0048] Table 1 Sample List (all food grade)

[0049]

[0050] 2. Sample processing and preparation

[0051] 70 mg of ground sample powder (accuracy ±0.001 g) and 70 mL of ultrapure water (resistivity ≥18.2 MΩ·cm, 25 °C) were used to prepare experimental groups. Four replicate experimental groups were set up for each sample, and an equal volume of ultrapure water treated in the same way was used as a blank control group.

[0052] Processing method: The filtrate was prepared in a shaking incubator at 25°C and 150 rpm. After 7 days, the filtrate was vacuum filtered into a clean 2L glass bottle using a glass vacuum filtration device equipped with a 0.22 μm nylon filter membrane (Jinteng, China) to obtain the percolate. The volume of the recovered percolate was measured and transferred to a clean brown glass sample bottle (CNW, Anpu, China) and stored at 4°C.

[0053] The aforementioned 7 days were obtained by the inventors through preliminary experiments based on GB 4806.7-2016, which can ensure that leaching reaches equilibrium.

[0054] 3. Non-target analysis

[0055] This section describes sample processing using freeze-drying, followed by non-targeted screening using liquid chromatography-mass spectrometry (LC-MS).

[0056] Sample preparation method: First, transfer 10 mL of percolate (including the experimental group and blank control group) to a 15 mL gas-permeable centrifuge tube (Corning, USA), add 1 μg of mixed standard (Table 3 internal standards are prepared with equal volume and concentration, i.e., each internal standard in the mixed standard has a final concentration of 5 ppm), vortex mix, and freeze at -80°C overnight. The next day, transfer the gas-permeable centrifuge tube to a lyophilizer (pre-freezing temperature set to -50°C, pre-freezing time 2 h), start the vacuum system and maintain a pressure ≤10 Pa, and freeze-dry for 24 h until the sample forms a lyophilized powder. Add 1 mL of 50% methanol-water (HPLC grade, Thermo Fisher Scientific, USA) to the lyophilized powder, vortex for 10 min to ensure complete dissolution, and then centrifuge at 12,000 rpm for 10 min. Afterward, transfer the supernatant to a 1.5 mL sample vial (Agilent, USA) and store at -20°C protected from light until sample loading and analysis.

[0057] Non-targeted screening was performed using UPLC (SCL-40, Shimadzu, Japan) coupled with Q-TOF-MS / MS (X500R, SCIEX, Netherlands). The UPLC system was equipped with a binary pump, online vacuum degassing device, autosampler, and constant-temperature column oven. Separation was performed using an Acquity UPLC BEH C18 column (1.7 μm, 2.1 × 150 mm) with a C18 guard column (both purchased from Waters, USA). The mobile phase (HPLC grade, Thermo Fisher Scientific, USA) consisted of (A) water and (B) methanol, both containing 0.1% formic acid. The column temperature was maintained at 40℃, the flow rate was set to 0.2 mL / min, and the injection volume was 50 μL. The elution gradient program for phase B is shown in Table 2.

[0058] Table 2 UPLC elution gradient

[0059]

[0060] Mass spectrometry was performed using a Q-TOF-MS / MS (X500R, SCIEX, Netherlands) tandem mass spectrometer in IDA mode. Electrospray ionization (ESI) parameters were as follows: Spray voltage: 5500 / - 4500 V; Curtain gas: 30 psi; CADgas: 7; Temperature: 450°C; Data acquisition mode was IDA, Collision energy: 15 eV-45 eV; Declustering potential: 80 V; Data acquisition mass range was 50–1500 Da.

[0061] Quality calibration was performed using AB Sciex calibration solution (m / z 118.0863-1521.9791) (calibrated every 5 injections), with the quality error controlled within ±5 ppm. Quality control (QC) injection was performed after every 10 injections by mixing equal volumes of each sample aliquot. A liquid chromatography blank sample (1:1 methanol / water) was injected before each sample group injection to detect contaminants and monitor instrument performance.

[0062] Internal standards: Nine isotope labels were used to optimize the method and correct for matrix effects, making it more suitable for the detection of pollutants.

[0063] Table 3 Internal Standard Information

[0064]

[0065] 4. Data Preprocessing

[0066] The raw data was converted to mzML format using ProteoWizard, and then non-targeted LC-QTOF-MS data analysis was performed on the R platform using the XCMS R package. Preliminary screening was performed based on retention time and mass-to-charge ratio (m / z) to remove abnormal signals that significantly deviated from the reasonable detection range, thus reducing interference in subsequent analyses. Peak alignment tolerances were set as follows: retention time deviation 0.2 min, mass deviation 10 ppm. Differences in retention time and mass-to-charge ratio between samples were corrected to improve peak matching accuracy. Peak extraction parameters were set as follows: mass deviation ≤ 10 ppm, relative standard deviation of signal intensity ≤ 30%, signal-to-noise ratio (S / N) ≥ 3, peak width range (s) = c(5,30), and peaks meeting these conditions were defined as stable characteristic peaks. Combined with total ion identification rules, the "centWave" algorithm was used to complete characteristic peak extraction and peak area quantification. Target ion integration was performed on the extracted characteristic peaks, and peak shapes corresponding to different ion states of the same compound (such as adduct ions and isotope peaks) were merged.

[0067] Based on the precise mass values ​​of molecular ion peaks and ion information of secondary fragments, chemical features were used to predict compound molecular formulas. Subsequently, the compound structure data were compared with those in the BiotreeDB 3.0 database to obtain the physicochemical information of each chemical feature and its corresponding characteristic peak information. Then, the confidence levels of the compounds were classified according to the Metabolomics Standards Initiative (2017), and the definitions of confidence levels are shown in Table 4. A total of 3833 chemical features were classified (Level 2-4). Only Level 2 chemical features were used for analysis. After data cleaning to remove interfering features and false positive signals introduced during the experiment, 1304 Level 2 chemical features were retained as high-confidence chemical features for subsequent analysis.

[0068] Table 4 Confidence Levels

[0069]

[0070] The data cleaning method is as follows: First, stable characteristic peaks with a relative standard deviation (RSD) of <30% were retained based on the QC samples. Second, to remove background interference, chemical characteristics meeting all of the following conditions were further screened: ① missing in blank samples but present in target samples, with a within-group missing rate of <50%; ② abundance in target samples 5 times or more higher than in blank samples. Subsequently, KNN imputation was used to fill missing values ​​to ensure data integrity. Finally, the data was processed using probability quotient normalization to reduce fluctuations in sample preparation and instrument response; logarithmic transformation and scaling were also performed on the data before analysis to reduce the impact of noise and high variance, thus laying the foundation for subsequent multivariate statistical analysis.

[0071] 5. Identification of toxicity drivers

[0072] 5.1 Multivariate statistical screening (Level 1)

[0073] Then, the peak area data of the high-confidence chemical characteristics obtained in Section 4 were analyzed using the ropls package in R (version 4.5.1).

[0074] First, unsupervised PCA (principal component analysis) was performed, and the results are as follows: Figure 3 As shown, the exudate data points of different materials (polymers) form independent clusters, revealing significant inter-group differences in their chemical fingerprints (except for PP and PS, which are relatively close together), while the QC samples are tightly clustered, indicating that the data quality is reliable.

[0075] Subsequently, supervised OPLS-DA (orthogonal partial least squares discriminant analysis) was performed to maximize between-group differences. The model was validated for effectiveness after 499 permutation tests (R²Y = 0.856, Q² = 0.776). We set comprehensive thresholds, including compounds with variable importance projection (VIP > 1.5), effect size (|Cohen's d| > 1.5), statistical significance (FDR < 0.01), and abundance difference fold (|Median Diff| > 2.0) as "difference factors". Results are as follows... Figure 4 As shown, a total of 507 differential factors were selected. The variable importance projection (VIP) distribution of these differential factors is as follows: Figure 5 .

[0076] 5.2 Calculation of initial toxicological screening (Level 2)

[0077] The SMILES structures of the differential factors were input into the ProTox-3.0 online platform, and the core predictive toxicity parameters for each predictor were obtained using the GHS toxicity classification as the grading standard: organ toxicity, toxic endpoint, TOX21 receptor pathway, TOX21 stress response pathway, molecular initiation event, metabolic mechanism, and acute oral toxicity (LD50). 50 The system outputs toxicity levels (1-6). Toxicity prediction levels are categorized as follows: highly toxic (1-2), moderately toxic (3-4), and low or non-toxic (5-6); the grading standard is the GHS toxicity classification. Acute oral toxicity prediction relies primarily on two core logics: first, analyzing the "2D structural similarity" of compounds (inferring toxicity associations through structural matching); and second, identifying "toxic fragments" (specific molecular fragments in the compound that may cause toxicity). Analysis of substances with toxicity levels 1-4 among the differential factors reveals differences in the distribution of these markers in toxicity prediction levels, as shown below. Figure 6 The Kruskal-Wallis test (p=0.18) indicates that LD values ​​for different material types... 50 The distributions showed no statistically significant difference. Figure 7 ).

[0078] Compounds with a predicted toxicity level of 1-2 (high toxicity) were designated as "high-risk factors," and a total of 49 high-risk factors were identified. Based on the aforementioned core toxicity parameters, a heatmap of the toxicity pathway activation patterns of these high-risk factors was generated, as shown below. Figure 8 As shown (where Figure 8 (The x-axis is shown in Tables 5-1 to 5-3, and the y-axis is shown in Table 6.) The toxic effects of these chemical features showed significant high-activation clusters in pathways such as respiratory toxicity, immunotoxicity, nutritional toxicity, and PXR, as well as toxic endpoints. This finding suggests that extracts from disposable food packaging may generally exhibit activation in specific pathways. Overall, most chemical features showed a low-toxicity background, and several high-activation hotspots shown in the spectra require further targeted verification to clarify their specific toxic risks under physiological conditions. Therefore, further molecular docking simulations are needed for the "core toxicity targets," based on... Figure 6 Based on the analysis, PXR was selected as the "core toxicity target" in this embodiment (see Section 5.3).

[0079] Table 5-1 Figure 8 Chemical characteristics / compounds involved 1 (in order from top to bottom on the vertical axis)

[0080]

[0081] Table 5-2 Figure 8 Chemical characteristics / compounds involved 2 (in order from top to bottom on the vertical axis)

[0082]

[0083] Table 5-3 Figure 8 Chemical characteristics / compounds involved 3 (in order from top to bottom on the vertical axis)

[0084]

[0085] Table 6 Figure 8 The toxic targets / toxic endpoints involved (in order from left to right on the x-axis).

[0086]

[0087] 5.3 Molecular docking fine screening (third stage)

[0088] 5.3.1 Receptor Preparation

[0089] Download the crystal structure of the human PXR protein from the RCSB PDB database (PDB ID: 1ILH). Use AutoDockTools to remove water molecules, add hydrogen atoms, and calculate the Gasteiger charge.

[0090] 5.3.2 Ligand Preparation

[0091] All SMILEs with high-confidence chemical features (1304 in total) screened in Section 4 were converted into 3D structures and their energy was minimized.

[0092] 5.3.3 Molecular docking

[0093] Semi-flexible molecular docking was performed using AutoDock Vina software. The center of the docking box was defined as the center of the PXR ligand binding pocket, and the box size was set to cover the entire active pocket. Ten docking calculations were performed for each compound, and the binding free energy of the optimal binding conformation was taken.

[0094] 5.3.4 Results Analysis

[0095] The binding energy distribution of all chemical characteristics with PXR is as follows: Figure 9 As shown in Table 7, a more negative binding energy indicates a more stable binding. The results indicate that 428 "potentially active compounds" with binding energies ≤ -7.0 kcal / mol successfully underwent molecular docking, among which 6 were "potentially high-quality active compounds" with binding energies ≤ -10.0 kcal / mol (see Table 7). We define compounds with binding energies ≤ -7.0 kcal / mol as "toxicity drivers" and output a list of binding energies between toxicity drivers and core toxicity targets.

[0096] Table 7 Potential high-quality active compounds

[0097]

[0098] Toxicity drivers have a strong potential interference effect on the identified core toxicity targets and pose a high toxicity risk under physiological conditions. They can serve as the core objects for subsequent toxicity mechanism research, replacing a large number of candidate compounds in traditional high-throughput drug screening and significantly reducing R&D costs.

[0099] Furthermore, the list of toxicity drivers can guide: targeted procurement of standards for confirmatory quantification; design of in vitro biological experiments (such as reporter gene experiments) for mechanism verification; use as high-risk structural templates to be avoided in the development of new materials; and replace traditional large-scale cytotoxicity screening, significantly reducing verification costs.

[0100] The above description is merely a preferred embodiment of the present invention, and therefore should not be construed as limiting the scope of the present invention. All equivalent changes and modifications made in accordance with the scope of the patent and the contents of the specification should still fall within the scope of the present invention.

Claims

1. A method for non-targeted leachate screening and toxicity driver identification based on IDA-UPLC-QTOF-MS / MS, characterized in that, Includes the following steps: (1) The sample was made into a sample powder with a particle size of less than 500 μm, and then extracted with ultrapure water, filtered and lyophilized to obtain lyophilized powder. (2) The solution of the lyophilized powder was analyzed by ultra-high resolution liquid chromatography-tandem mass spectrometry based on information-dependent acquisition mode to obtain mass spectrometry data containing all detectable chemical features; (3) The stable characteristic peaks obtained from the mass spectrometry data are matched with the database for qualitative analysis and confidence level classification, and their physicochemical information is obtained; (4) Perform data cleaning on the chemical features to remove interfering features and false positive signals, and obtain high-confidence chemical features for subsequent screening; (5) The chemical characteristics are screened in three levels as follows: (5.1) Multivariate statistical screening: Through orthogonal partial least squares discriminant analysis, differential factors with significant differences between different samples were screened from the high-confidence chemical characteristics; (5.2) Calculate the initial toxicological screening: perform toxicity prediction on the differential factors and screen out high-risk factors with predicted toxicity level of high toxicity; (5.3) Molecular docking fine screening: Taking the toxic pathways in which the high-risk factors generally show an activated state as the core toxicity targets, all the high-confidence chemical features are used as ligands to perform molecular docking simulation with the protein structure of the core toxicity targets, and the chemical features with binding energy lower than the preset value are the toxicity driving factors.

2. The method for screening non-targeted leachate and identifying toxicity drivers as described in claim 1, characterized in that, The ultrapure water extraction temperature is room temperature, and the ultrapure water extraction time is 7 days.

3. The method for screening non-targeted leachate and identifying toxicity drivers as described in claim 1, characterized in that, The stable characteristic peak has a quality deviation of 10 ppm, a signal strength deviation of 30%, a signal-to-noise ratio (S / N) ≥ 3, and a peak width range (s) = c(5,30).

4. The method for screening non-targeted leachate and identifying toxicity drivers as described in claim 1, characterized in that, The screening criteria for the high-confidence chemical features are as follows: (1) The confidence level of the Metabolomics Standards Initiative is Level 2; (2) Stable feature peaks with a relative standard deviation (RSD) of <30% are retained based on quality control; (3) The feature peaks are missing in the blank sample but present in the target sample and the intragroup deletion rate is <50%, or the abundance of the target sample is 5 times or more higher than that of the blank sample.

5. The method for screening non-targeted leachate and identifying toxicity drivers as described in claim 1, characterized in that, The screening thresholds for the differential factors are as follows: variable importance projection > 1.5, effect size > 1.5, statistical significance < 0.01, and abundance difference fold > 2.

0.

6. The method for screening non-targeted leachate and identifying toxicity drivers as described in claim 1, characterized in that, Step (5.2) performs toxicological predictions using a computational toxicology platform, then evaluates the toxicity level of the differential factors according to the GHS toxicity classification standard, and defines the factors with a predicted toxicity level of 1-2 as high-risk factors.

7. The method for screening non-targeted leachate and identifying toxicity drivers as described in claim 5, characterized in that, The computational toxicology platform is ProTox-3.

0.

8. The method for screening non-targeted leachate and identifying toxicity drivers as described in claim 1, characterized in that, The molecular docking simulation was performed using AutoDock Vina software, and the preset value of the binding energy was -7.0 kcal / mol; the orthogonal partial least squares discriminant analysis was performed using the ropls package in R language.

9. The method for screening non-targeted leachate and identifying toxicity drivers as described in claim 1, characterized in that, The sample was a disposable food package, and the core toxicity target was the pregnane X receptor.

10. The method for screening non-targeted leachate and identifying toxicity drivers as described in claim 9, characterized in that, The disposable food packaging includes at least one of polyolefins, biodegradable plastics, and paper-plastic composites.

Images