Application of fingerprint spectrum of e-waste pollution in human body exposure risk prediction

Abstract

This invention discloses the application of human exposure fingerprinting of electronic waste pollution in exposure risk prediction. The human exposure fingerprinting consists of three or more of the following exposure markers: 2,4,6-tribromophenol, 2,4,5-trichlorophenol, 4-bromophenol, 1-hydroxypyrene, and bisphenol A. This invention is the first to employ the systematic method of urinary contaminant exposure profiling to achieve comprehensive screening and characterization of contaminant exposure markers related to specific components and byproducts in workers exposed to electronic waste. This is currently the most systematic organic contaminant exposure study known in this field.

Description

Technical Field

[0001] This invention belongs to the field of environmental biological monitoring and risk prediction, specifically involving the application of human exposure fingerprint spectrum of electronic waste pollution in exposure risk prediction. Background Technology

[0002] Electronic waste pollution is characterized by its insidious nature, widespread impact, and serious harm. Developing rapid and accurate identification technologies to enable early identification and prevention of exposure hazards has become a critical technological bottleneck that urgently needs to be overcome. Human pollutant exposure fingerprinting is an emerging technology that can accurately distinguish between individuals' exposure risk by systematically analyzing the levels and compositional characteristics of pollutant exposure biomarkers. The core advantage of this technology lies in its ability to efficiently characterize the overall exposure risk faced by a population using a small number of carefully selected high-risk characteristic exposure biomarkers. This overcomes the drawbacks of traditional health risk assessment methods, such as the time-consuming, labor-intensive, and costly process of quantifying a vast array of pollutants and their effect biomarkers.

[0003] Besides VOCs and heavy metals, characteristic pollutants in e-waste include aromatic hydrocarbons such as polybrominated diphenyl ethers (PBDEs), polychlorinated biphenyls (PCBs), and polycyclic aromatic hydrocarbons (PAHs). Existing research has focused very little on exposure biomarkers for aromatic hydrocarbons in human urine, primarily due to the lack of highly sensitive analytical detection techniques, resulting in extremely low detection rates for these biomarkers. Furthermore, most studies focus only on a few pollutants, failing to systematically and comprehensively reveal the complete exposure characteristics of e-waste pollution.

[0004] In recent years, organic pollutant exocomics, as a core branch of exocomics, has received increasing attention from scholars. It aims to comprehensively and systematically identify and quantify the total amount and dynamic changes of all exogenous organic pollutants exposed to the human body through various pathways such as respiration, diet, and skin contact. Currently, although numerous studies have reviewed the environmental characteristics of pollutants in e-waste dismantling areas, research on monitoring exposure markers in the human body remains severely lacking, especially regarding the overall exposure omics of organic pollutants in e-waste dismantling areas, which is still a blank area. Summary of the Invention

[0005] The purpose of this application is to provide the application of electronic waste pollution human exposure fingerprint spectrum in exposure risk prediction.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] Application of human exposure fingerprinting of electronic waste pollution in exposure risk prediction;

[0008] The human body exposure fingerprint spectrum is composed of three or more of the following exposure markers: 2,4,6-TriBP (2,4,6-tribromophenol), 2,4,5-TriCP (2,4,5-trichlorophenol), 4-MonoBP (4-bromophenol), 1-OHPyr (1-hydroxypyrene), and BPA (bisphenol A).

[0009] The aforementioned electronic waste pollution refers to the pollution released during the recycling and dismantling of discarded electrical and electronic equipment;

[0010] The electrical and electronic devices mentioned include computers, mobile phones, televisions, etc.

[0011] Preferably, the human body exposure fingerprint spectrum is composed of four or more of 2,4,6-TriBP (2,4,6-tribromophenol), 2,4,5-TriCP (2,4,5-trichlorophenol), 4-MonoBP (4-bromophenol), 1-OHPyr (1-hydroxypyrene), and BPA (bisphenol A);

[0012] More preferably, the human body exposure fingerprint spectrum is one of the following combinations:

[0013] Combination 1: 2,4,5-TriCP (2,4,5-trichlorophenol), 4-MonoBP (4-bromophenol), and 2,4,6-TriBP (2,4,6-tribromophenol);

[0014] Combination 2: 2,4,5-TriCP (2,4,5-trichlorophenol), 4-MonoBP (4-bromophenol), and BPA (bisphenol A);

[0015] Combination 3: 2,4,5-TriCP (2,4,5-trichlorophenol), 2,4,6-TriBP (2,4,6-tribromophenol), and BPA (bisphenol A);

[0016] Combination 4: 4-MonoBP (4-bromophenol), 2,4,6-TriBP (2,4,6-tribromophenol), and BPA (bisphenol A);

[0017] Combination 5: 2,4,6-TriBP (2,4,6-tribromophenol), 2,4,5-TriCP (2,4,5-trichlorophenol), 4-MonoBP (4-bromophenol), and 1-OHPyr (1-hydroxypyrene);

[0018] Combination 6: 2,4,6-TriBP (2,4,6-tribromophenol), 2,4,5-TriCP (2,4,5-trichlorophenol), 4-MonoBP (4-bromophenol), and BPA (bisphenol A);

[0019] Combination 7: 2,4,5-TriCP (2,4,5-trichlorophenol), 4-MonoBP (4-bromophenol), 1-OHPyr (1-hydroxypyrene), and BPA (bisphenol A);

[0020] Combination 8: 2,4,6-TriBP (2,4,6-tribromophenol), 2,4,5-TriCP (2,4,5-trichlorophenol), 1-OHPyr (1-hydroxypyrene), and BPA (bisphenol A);

[0021] Combination 9: 2,4,6-TriBP (2,4,6-tribromophenol), 4-MonoBP (4-bromophenol), 1-OHPyr (1-hydroxypyrene) and BPA (bisphenol A);

[0022] Combination 10: 2,4,6-TriBP (2,4,6-tribromophenol), 2,4,5-TriCP (2,4,5-trichlorophenol), 4-MonoBP (4-bromophenol), 1-OHPyr (1-hydroxypyrene), and BPA (bisphenol A).

[0023] The aforementioned human exposure fingerprint spectrum (i.e., exposure markers) can be used to construct a predictive model for electronic waste pollution exposure risks;

[0024] The method for constructing the prediction model includes the following steps:

[0025] S1: Obtain multiple exposure biomarkers related to e-waste pollution in urine samples from populations in potentially and unexposed areas;

[0026] S2: Establish training and validation sample sets based on various exposure biomarkers; (the training samples in each training sample set belong to all exposure biomarkers measured in urine)

[0027] S3: Train machine learning models using training sample sets corresponding to various exposure biomarkers to obtain various primary diagnostic models; wherein, during the training process, the concentration of the exposure biomarker is used as the input feature of the machine learning model, and the exposure status of the urine sample collected for the exposure biomarker is used as the label; the exposure status includes exposed and unexposed.

[0028] The machine learning model mentioned can be a random forest model, a support vector machine, a neural network model, or others;

[0029] S4: For various primary diagnostic models, the accuracy is verified using validation sample sets corresponding to various exposure biomarkers;

[0030] S5: Select the primary diagnostic model with the highest accuracy as the final exposure risk prediction model.

[0031] The apparatus for the exposure risk prediction model includes:

[0032] The acquisition module is used to acquire multiple exposure markers in urine;

[0033] The sample construction module is used to build training and validation sample sets based on various exposure markers.

[0034] The training module trains machine learning models using training sample sets corresponding to various exposure biomarkers to obtain various primary diagnostic models. During training, the concentration of the exposure biomarker is used as the input feature of the machine learning model, and the exposure status of the urine sample collected for the exposure biomarker is used as the label. The exposure status includes exposed and unexposed.

[0035] The machine learning model mentioned can be a random forest model, a support vector machine, a neural network model, or others;

[0036] Validation module: For various primary diagnostic models, the accuracy is verified using validation sample sets corresponding to various exposure biomarkers;

[0037] Select Module: Choose the primary diagnostic model with the highest accuracy as the final exposure risk prediction model.

[0038] The beneficial effects of this invention are reflected in the following aspects:

[0039] 1. This invention is the first to employ a systematic approach using urinary contaminant exposure profiling to comprehensively screen and characterize contaminant exposure biomarkers related to specific components and byproducts in a group of workers exposed to e-waste. This is currently the most systematic organic contaminant exposure study known in this field. This invention makes a breakthrough discovery by finding that urinary BRFs and CRFs exposure biomarkers have a strong ability to indicate e-waste pollution exposure. Through rigorous model training and validation, multiple highly predictive human contaminant exposure fingerprint profiles (i.e., combinations of exposure biomarkers) have been identified, providing precise tools and scientific basis for risk assessment of e-waste pollution.

[0040] 2. This invention constructs a high-precision e-waste exposure prediction model, providing key technical support for the rapid identification and screening of informal e-waste contaminated sites, significantly improving regulatory efficiency. This technical framework possesses strong scalability and, by integrating multi-source pollution fingerprint features, can be developed into a universal screening system covering multiple scenarios. Ultimately, this invention can effectively promote the value mining of existing large-scale global human biomonitoring databases, transforming them into efficient regulatory tools, possessing significant socio-economic value in reducing public governance costs and enhancing health risk intervention capabilities. Attached Figure Description

[0041] Figure 1 This is a typical chromatographic separation diagram of the simultaneous determination of six urinary oxidative damage markers and five volatile organic compound exposure markers in the standard solution of Method 1.

[0042] Figure 2 This is a typical chromatographic separation diagram for the simultaneous determination of two volatile organic compounds, seven benzene compounds, 15 chlorinated flame retardants, nine brominated flame retardants, 13 polycyclic aromatic hydrocarbons, and three bisphenol exposure markers in the standard solution of Method 2.

[0043] Figure 3 This is a typical chromatographic separation diagram of the simultaneous determination of four organophosphates and ten phthalate exposure markers in the standard solution of Method 3.

[0044] Figure 4 The exposure feature spectrum score plot (A) and model permutation test cross-validation plot (B) are based on orthogonal partial least squares discriminant analysis (OPLS-DA). 2 For goodness of fit, Q 2 (For predicting goodness).

[0045] Figure 5 Principal component analysis was used to reveal the differences in urine contaminant exposure profiles between workers exposed to e-waste and control adults.

[0046] Figure 6 The receiver operating characteristic (ROC) curve is used to assess the predictive ability of electronic waste pollution exposure fingerprints for different categories of pollutants.

[0047] Figure 7 This study uses the random forest algorithm to assess the importance of different pollutants in predicting the risk of exposure to electronic waste pollution.

[0048] Figure 8 It is the process of selecting high-ranking features from high-dimensional data based on the support vector machine recursive feature elimination algorithm (the relationship between error rate and the number of features).

[0049] Figure 9 This represents the inter-group concentration level distribution of feature variables selected jointly by the random forest and support vector machine recursive feature elimination algorithms.

[0050] Figure 10 This is the accuracy of predicting and identifying e-waste pollution exposure using threshold concentrations of 2,4,6-TriBP (0.169 μg / g creatinine) and 2,4,5-TriCP (0.0044 μg / g creatinine).

[0051] Figure 11 The model for predicting exposure to electronic waste pollution is constructed based on five optimized exposure biomarkers.

[0052] Figure 12 The electronic waste pollution exposure prediction model is constructed based on a combination of four exposure markers (ACC1 and ACC2 are the prediction accuracies of the training set and validation set, respectively).

[0053] Figure 13 It is based on a combination of three exposure markers to construct a predictive model for electronic waste pollution exposure.

[0054] Figure 14 It is a three-dimensional scatter plot that distinguishes between people exposed to e-waste (children and workers) and those not exposed (adult control group) based on three key predictive features.

[0055] Figure 15 An SVM classifier is constructed based on the optimal combination of three exposure markers (2,4,5-TriCP, 4-MonoBP, 2,4,6-TriBP), and a simplified function is derived (ACC1 and ACC2 are the prediction accuracies of the training set and validation set, respectively).

[0056] Figure 16 The accuracy of the simplified function prediction was verified based on children exposed to e-waste (independent validation set 1) (red indicates correctly identified as the exposure group, and blue indicates incorrectly identified as the control group).

[0057] Figure 17 The accuracy of the prediction of the simplified function 2 was verified by using eight different groups of people exposed to pollution in different industries in different regions (independent validation set 2). Detailed Implementation

[0058] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0059] Example

[0060] The screening of electronic waste pollution exposure biomarkers and their application in monitoring human body load includes the following steps:

[0061] (1) Recruitment of people

[0062] Training group recruitment: After the research protocol of this invention was approved by the ethics review committee, the recruitment of research subjects was completed between 2016 and 2019.

[0063] The training population included both the exposed group and the control group.

[0064] The exposure team consisted of 167 workers from the electronic waste dismantling industry in a certain region of province 1.

[0065] The control group consisted of 174 healthy adults from a non-electronic waste dismantling area in a region of province 1.

[0066] Validation set audience recruitment:

[0067] Validation set 1: Composed of 226 children who live within 3 kilometers of the electronic waste dismantling park, recruited during the same period as the training set.

[0068] Validation set 2 consisted of 1,453 ordinary adults recruited nationwide from 2019 to 2020. Participants came from five major geographical regions in China, specifically: Province 1 (N=158), Province 2 (N=176), Province 3 (N=156), Province 4 (N=145), Province 5 (N=142), Province 6 (N=198), Province 7 (N=266), and Province 8 (N=212). The recruitment covered employees from eight typical industrial parks, including: two e-waste dismantling parks (located in Provinces 1 and 2), three non-ferrous metal smelting / processing parks (located in Provinces 3, 4, and 5), two non-ferrous metal mining parks (located in Provinces 6 and 7), and one oil / coal processing park (located in Province 8).

[0069] The inclusion criteria for the study participants were as follows: participants in the training set were under 60 years of age, had no history of serious organic diseases, had not taken any medications within the two weeks prior to enrollment, and voluntarily provided biological samples. Workers were required to have at least one year of experience in the electronic waste dismantling industry.

[0070] The validation set population 1 consists of individuals aged 5 to 13 who have resided in the local area for at least two years and were recruited using an annual random sampling method.

[0071] All participants signed written informed consent forms and completed a comprehensive questionnaire covering demographic characteristics, dietary habits, and specific exposure information. Biosample collection strictly followed standardized operating procedures. After collecting a morning urine sample from each participant, it was immediately stored at –80°C for subsequent chemical analysis.

[0072] The specific characteristics of group 1 in the training and validation sets are shown in Table 1:

[0073] Table 1. Demographic characteristics of population 1 in the training and validation sets.

[0074] The mean age of e-waste dismantling workers was 40.9 ± 11.4 years, compared to 38.0 ± 6.06 years for adults in the control group. The mean age of children in the e-waste dismantling area was 8.34 ± 3.50 years. There were no significant differences between e-waste workers and adults in terms of sex ratio, body mass index (BMI), and smoking status.

[0075] (2) Characterization of organic pollutant exposure groups

[0076] Targeted screening: After targeted screening of more than 200 organic pollutants in the mixed urine of workers in the training set exposure group, 68 exposure biomarkers with high detection rates were identified, and an organic pollutant exposure group reflecting the characteristics of e-waste pollution was constructed based on this.

[0077] The biomarkers covered by this exposure group included the following categories: 7 volatile organic compounds (mVOCs), 7 benzene series compounds (mBTEX), 15 chlorinated flame retardants (mCRFs), 9 brominated flame retardants (mBRFs), 13 polycyclic aromatic hydrocarbons (mPAHs), 3 bisphenols (BPs), 4 organophosphates (OPEs), and 10 phthalates (PAEs). To further assess the potential impact of pollutants on human health, this study also measured 6 oxidative damage biomarkers (ODBs).

[0078] To address the differences in chemical and physical properties among different target compounds, the following three analytical methods were employed. The mass spectrometry detection and quantitative parameters for each target compound are shown in Table 2.

[0079] Table 2. Mass spectrometry detection parameters of pollutant exposure and oxidative damage biomarkers in targeted ion monitoring mode

[0080] Targeted High-Throughput Quantitative Method 1: This method is used to simultaneously determine six oxidative damage biomarkers (ODBs) (8-OHG, 8-OHdG, CA, CDCA, GCA, GCDCA) and five mVOCs (TGA, AAMA, DHBMA, CYMA, 3HPMA). Urine samples were thawed at -20°C, centrifuged at 14,000 rpm for 5 minutes, and 1.5 mL of the supernatant was transferred to a plastic tube. 50 μL of internal standard solution, 50 μL of 1% formic acid, and 7 μL of a β-glucuronidase / arylsulfatase mixed enzyme solution were added sequentially, and the mixture was incubated at 37°C for 12 hours. Subsequently, the target analytes were enriched and purified using a polystyrene / divinylbenzene (PEP) solid-phase extraction column, and eluted with 2 mL of a formic acid-acetonitrile mixture (2:98, v / v). The eluent was collected, dried under nitrogen, and then reconstituted with 300 μL of methanol-water solution (1:9, v / v). 150 μL of the supernatant was used for instrumental analysis. Target analyte detection was performed using an ultra-high performance liquid chromatography-quadrupole / orbitrap-HRMS system (Exploris 240, UPLC-Orbitrap-HRMS). Chromatographic separation was performed on a Hypersil GOLD AQ C18 column (150 × 2.1 mm, 1.9 μm) at a flow rate of 0.25 mL / min, a column temperature of 35 °C, and an injection volume of 2 µL. The mobile phase consisted of an aqueous solution containing 0.1% formic acid (phase A) and methanol (phase B). The gradient program was set as follows: 0–1 min to maintain 2% B, 1–3 min to increase from 2% to 30% B, 3–11 min to increase from 30% to 100% B, 11–15 min to maintain 100% B, and 15.01–18 min to restore to 2% B and equilibrate the system. Mass spectrometry acquisition was performed using an electrospray ionization source in negative ion mode with a spray voltage of -2.5 kV, sheath gas, auxiliary gas, and purge gas flow rates of 45, 8, and 1 arbitrary unit, respectively. The ion transfer tube temperature was 320 °C, and the vaporizer temperature was 350 °C. The global mass spectrometry parameters were set as follows: HPLC peak width of 10 s was expected, and real-time mass calibration was performed using the EASY-IC™ system. To improve the capture efficiency of trace contaminants, data acquisition employed targeted ion monitoring mode, with an isolation window width of 1.5 m / z, RF lens voltage of 70%, orbital trap resolution of 45,000, automatic gain control target value in standard mode, and maximum injection time in automatic mode. Chromatograms of each target compound are shown below. Figure 1 .

[0081] Targeted High-Throughput Quantitative Method 2: This method is used to simultaneously detect 2 mVOCs (TTCA, 3MHBMA), 7 mBTEXs (1,2-DB, BMA, PGA, MA, 2-MHA, 3-MHA, 4-MHA), and 15 mCRFs (4CCT, 4-MonoCP, 2,5-DiCP, 3,4-DiCP, 3,5-DiCP, 2,4,5-TriCP, 2,3,4,6-TetraCP, 3,3'-DiCBPA, TCS, TCBPA, TCBPS, 3-MonoCBPA, 4-OHPCB18, 4-OHPCB76, 4-OHPCB18). 9) Nine types of mBRFs (4-MonoBP, 2,4,6-TriBP, 2,3,4,6-TetraBP, 3,3',5-TriBBPA, DiBBPA, TBBPA, TBBPS, 4'-OH-BDE_17, 6-OH-BDE_47), 13 types of mPAHs (2-OHNap, 1-OHNap, 2-NapCA, 2-OHFlu, 3-OHFlu, 1-OHPhe, 2-OHPhe, 3-OHPhe, 4-OHPhe, 9-OHPhe, 1-OHPyr, 6-OHChr, 1-PyrCA) and three types of BPs (BPA, BPS, BPAF). After centrifugation at 14,000 rpm, 1.5 mL of the supernatant was placed in a polypropylene tube, and 50 μL of 1% formic acid, 50 μL of internal standard solution, and 7 μL of β-glucuronidase / arylsulfatase mixture were added sequentially. Enzymatic digestion was performed overnight at 37°C. Subsequently, PEP solid-phase extraction was used for enrichment: the column was activated sequentially with 3 mL of methanol, 3 mL of ultrapure water, and 3 mL of 0.1% formic acid aqueous solution. After loading the sample, it was washed with 2 mL of 5% methanol aqueous solution, and finally eluted with 2 mL of formic acid-acetonitrile solution (2:98, v / v). The eluent was collected, concentrated to dryness by nitrogen blowing, and reconstituted with 300 μL of methanol. 200 μL of the supernatant was used for instrumental analysis. Chromatographic separation was performed on a Hypersil GOLD C18 column (100 × 2.1 mm, 1.9 μm) at 35 °C. The mobile phase consisted of an aqueous solution containing 0.05% acetic acid (phase A) and methanol (phase B) at a flow rate of 0.25 mL / min. The gradient elution program was set as follows: 0–1 min, from 5% to 10% B; 1–3 min, from 10% to 45% B; 3–15 min, from 45% to 78% B; 15–17 min, from 78% to 100% B; 17–20.5 min, maintaining 100% B; 20.5–23.5 min, returning to 5% B and equilibrating the system.Mass spectrometry detection employed negative ion electrospray ionization mode with a spray voltage of –2.5 kV, ion transfer tube temperature of 320°C, and vaporizer temperature of 350°C. Sheath gas, auxiliary gas, and purge gas flow rates were 45, 8, and 1 arbitrary unit, respectively. Data acquisition utilized targeted ion monitoring mode with an orbital trap resolution of 60,000 and RF lens voltage of 70% to enhance the sensitivity for trace contaminant detection. Real-time mass calibration was performed using EASY-IC™, and the expected peak width for liquid chromatography was 8 seconds. The chromatogram of the target compound is shown below. Figure 2 .

[0082] Validation of Method 2: We strictly followed the relevant guidelines of the U.S. Food and Drug Administration and the European Medicines Agency, and examined parameters including limit of detection, limit of quantitation, matrix effect, extraction recovery, and intra-day and inter-day precision.

[0083] The specific evaluation methods are as follows: the limit of detection is determined by the lowest standard concentration corresponding to a signal-to-noise ratio ≥3 (N=3); the limit of quantitation is defined as the lowest analyte concentration in the sample that can be reliably quantified, with a measurement deviation ≤20% (N=3), and the analyte signal intensity at this concentration is not less than 10 times that of the blank sample signal. The evaluation of matrix effect (N=6) and extraction recovery (N=6) refers to our team's previous research (…). Anal Bioanal Chem. 2019, 411(29): 7841–7855 ), and calculate according to Formula 1 and Formula 2 respectively.

[0084] The experiment was conducted with four sets of samples: Set 1 was a pure standard sample (the target analyte and internal standard were added to a methanol-water solution); Set 2 was a spiked sample before extraction (the target analyte and internal standard were added to the diluted mixed urine before solid-phase extraction); Set 3 was a mixed urine extraction sample (only the internal standard was added to the diluted mixed urine before solid-phase extraction); and Set 4 was a spiked sample after extraction (the target analyte and internal standard were added to the diluted mixed urine after solid-phase extraction).

[0085] A matrix effect greater than zero indicates enhanced ionization, while a matrix effect less than zero indicates inhibition. Quality control samples were prepared by adding the target analyte and internal standard to the diluted mixed urine before solid-phase extraction. To assess precision, six replicate quality control samples were prepared daily for six consecutive days, and intra-day and inter-day precision were expressed as relative standard deviation. The mixed urine samples used were prepared by mixing equal volumes of 20 randomly selected urine samples.

[0086] The method validation results are shown in Table 3: all target analytes were effectively separated in the chromatograms, with limits of quantitation ranging from 0.001 to 4 μg / L. Except for a few substances, the extraction recoveries of most analytes ranged from 71.4% to 114%, with relative standard deviations (RSDs) all below 23.1%. The matrix effect ranged from -26.1% to 26.4%, and the relative standard deviations of intra-day and inter-day precision were both controlled within 12.8%. Based on all validation indicators, Method 2 demonstrates good reproducibility and analytical stability.

[0087] Formula 1 Matrix effect (%) =

[0088] Formula 2 Extraction recovery rate (%) =

[0089] Table 3. Limit of quantitation, extraction recovery, matrix effect, and intra-day / inter-day precision validation results for Method 2 (N=6)

[0090] Targeted High-Throughput Quantitative Method 3: This method focuses on the qualitative and quantitative analysis of OPEs and PAEs. We first established a targeted screening list containing 41 OPEs and 42 PAEs (Table 4), and based on their precise mass-to-charge ratios, performed preliminary screening on mixed urine samples in negative ion full-scan mode. The resolution was set to 120,000, and the scan range was set to 60–800 m / z. The targeted screening analysis identified 4 OPEs (DiMPho, DiEPho, 4HPDiPPho, DiPPho) and 10 PAEs (MMPht, DiMPht, MECPPht, MnBPht, MiBPht, DiEPht, MEHHPht, MBzPht, MECPTPht, MiNPht), which were then quantitatively analyzed using Method 3. This method follows Method 2 in sample pretreatment and mass spectrometry parameters, with only appropriate optimization of the chromatographic conditions. The specific chromatographic conditions were as follows: A Hypersil GOLD C18 column (100 × 2.1 mm, 1.9 μm) was used for chromatographic separation, and the column temperature was maintained at 40℃. The mobile phase was an aqueous solution containing 0.1% acetic acid (phase A) and methanol (phase B), with a flow rate of 0.3 mL / min. The gradient elution program was optimized as follows: 0–1 min, 1% B; 1–3 min, from 1% to 70% B; 3–10 min, from 70% to 99% B; 10–12 min, maintain 99% B; 12.1–15 min, return to 1% B. The chromatogram of the target compound is shown below. Figure 3 .

[0091] Table 4. Targeted Screening List for Organophosphates (OPEs) and Phthalate Esters (PAEs) in Urine

[0092] (3) Quality Assurance and Control (QA / QC) for Characterization of Organic Pollutant Exposure Groups

[0093] To ensure the reliability of the analytical results, this invention implemented a systematic quality assurance and control procedure. First, isotope-labeled internal standards were added to the water blank and process blank to monitor and correct for background contamination introduced by reagents; based on this, MBzPht and MiNPht, which were detected as contaminants in the process blank, were excluded. Second, quality control samples with spiked concentrations of 4.0–4000 μg / L were used for evaluation. The results showed that, except for GCA (20.3%) and 4'-OH-BDE_17 (27.9%), the average relative recoveries of all other target analytes were between 85.9% and 121%, with relative standard deviations not exceeding 15% (see Table 5), indicating that the method possesses good accuracy and precision.

[0094] Table 5. Relative recovery rates of spiked quality control samples in urine analysis (N=4)

[0095] Further repeat analysis was performed on 5% of random urine samples to assess reproducibility. Except for BPS, DiBBPA, DiEPho, MMPht, and 3,3'-DiCBPA, the coefficients of variation for the remaining targets were all no higher than 18.8% (see Table 6). In summary, all quality control results met the requirements of routine analysis, confirming that this method has excellent reproducibility and robustness.

[0096] Table 6. Reproducibility of methods assessed based on repeated urine sample assays

[0097] Of the 68 urinary exposure and effect biomarkers detected, 44 biomarkers had a detection frequency higher than 80% in e-waste plant workers (Table 7). Compared with the control group, except for 1,2-DB, PGA, BMA, 4-MonoCP, 2,5-DiCP, BPAF, DiMPht, and MECPTPht, the geometric mean concentrations of other pollutants in the urine of e-waste plant workers were significantly increased, indicating that VOCs, BTEX, CRFs, BRFs, BPs, OPEs, and PAEs can serve as potential characteristic pollutants of e-waste exposure. Furthermore, the geometric mean levels of 8-OHG, 8-OHdG, GCA, CA, and GCDCA in the urine of e-waste plant workers were 1.71–4.31 times higher than those in the control group. p The value was <0.01, indicating that this population had higher levels of nucleic acid and cholesterol oxidative damage.

[0098] More importantly, this invention is the first to systematically characterize the metabolic profiles of basal bioflavonoids (BRFs) and cerebral bioflavonoids (CRFs) in the urine of e-waste plant workers. It also identified several to ten times higher concentrations of various chlorophenols (4CCT, 2,4,5-TriCP, 2,3,4,6-TetraCP, and 3,3'-DiCBPA) and the metabolic transformation products of TBBPA (DiBBPA and 3,3',5-TriBBPA) compared to the control group, suggesting that these exposure markers are potential specific exposure markers for e-waste pollution. In summary, e-waste plant workers face a high risk of combined exposure to multiple pollutants, particularly BRFs, CRFs, BTEX, and PAHs.

[0099] Table 7. Concentrations of contaminants and biomarkers of oxidative damage in urine of e-waste plant workers (N=167) and control adults (N=174)

[0100] (4) Construction and validation of electronic waste pollution exposure prediction model

[0101] Data preprocessing and feature screening: All urinary biomarker concentrations were corrected for creatinine levels to eliminate the influence of individual differences in urine dilution. Given the non-normal distribution of biomarker data, the Mann-Whitney U test was used to analyze the concentration differences between e-waste exposed workers and the control group, with a statistical significance level set at [value missing]. p <0.05 (two-tailed test). Pollutants with significantly higher levels in the exposure group were selected as candidate features for modeling.

[0102] Discriminant analysis between exposed and unexposed groups: First, the study subjects were clearly divided into an exposed group (e-waste exposed workers) and an unexposed group (control adults) as the discriminant targets for the binary classification model. To assess the overall differences in urinary biomarker profiles between the two groups, orthogonal partial least squares-discriminant analysis (OPLS-DA) was used for modeling, and the model's effectiveness was validated through a permutation test with 200 replicates. Based on this, the combination of biomarkers with optimal discriminant ability was identified, and principal component analysis (PCA) was further used to examine its separation within the groups, thus initially screening out biomarkers with potential importance for identifying e-waste exposure.

[0103] The OPLS-DA score plot showed a significant divergence trend between the control group and the exposure group in terms of urine biomarker profiles. Figure 4 A). Cross-validation results of the OPLS-DA model generated through 200 repeated permutation tests show that both the goodness of fit and the predicted goodness of fit parameters meet the reliability requirements, and no overfitting is observed. Figure 4 B).

[0104] Figure 5 The diagram presents a three-dimensional principal component analysis (PCA) plot illustrating the differences in pollutant exposure and oxidative damage biomarker profiles between groups. Spheres represent the spatial distribution of samples based on the first three principal components; smaller confidence ellipsoid overlap indicates greater inter-group differences. The results showed significant differences between the two groups in both pollutant exposure characteristics and oxidative damage biomarker profiles, with mPAHs exhibiting the smallest ellipsoidal overlap region within the 95% confidence interval. Figure 5 ).

[0105] Predictive Model Construction and Performance Evaluation: Based on the above screening results, six exposure prediction models with different biomarker combinations were constructed using machine learning algorithms. By comparing algorithms such as random forest, support vector machine (SVM), and neural network models, the SVM algorithm was found to be generally superior. Therefore, subsequent models were all constructed using the SVM algorithm, specifically: Model 1 (ODBs), Model 2 (mVOCs and mBTEX), Model 3 (mPAHs), Model 4 (mBRFs), Model 5 (mCRFs), and Model 6 (BPs, OPEs, and PAEs). All input data underwent logarithmic transformation and z-score standardization to improve model comparability and convergence efficiency. Model performance was evaluated using receiver operating characteristic (ROC) curves, with accuracy (ACC) and area under the curve (AUC) as the core evaluation metrics.

[0106] ROC analysis was used to quantitatively evaluate the predictive performance of exposure fingerprint models for different types of pollutants. Figure 6 The metrics included in each model are shown in Table 8.

[0107] Table 8. Top three principal component loadings in principal component analysis of different combinations of exposure biomarkers

[0108] The model performance comparison results show that Model 3 (mPAHs) has the best predictive effect on e-waste pollution (average AUC: 0.97, ACC: 0.87), followed by Model 4 (mBRFs) (AUC: 0.97, ACC: 0.77), Model 5 (mCRFs) (AUC: 0.95, ACC: 0.85), Model 2 (mVOCs & mBTEX) (AUC: 0.94, ACC: 0.86), Model 6 (BPs & OPEs & PAEs) (AUC: 0.86, ACC: 0.79), and Model 1 (ODBs) (AUC: 0.85, ACC: 0.78).

[0109] The results above indicate that the prediction model based on pollutant exposure fingerprints (especially PAHs, BRFs, and CRFs) is significantly better than the discriminant model based on oxidative damage effects in distinguishing e-waste exposure.

[0110] Cross-validation and parameter optimization: To effectively avoid overfitting and underfitting, a stratified 5-fold cross-validation method was used to evaluate model performance. The dataset was evenly divided into 5 subsets, with each subset used as the test set and the rest as the training set. This process was repeated 5 times, and the average AUC and ACC were used as the final evaluation criteria for model performance. Key parameters were set during model construction as follows: class weights were set to "balanced" to handle sample imbalance; a radial basis function kernel was selected, with a convergence tolerance of 0.000001; the regularization parameter C was set to 10; and the kernel parameter gamma was set to 1.

[0111] To optimize the model's discriminative performance and eliminate the interference of redundant variables, this invention further employs two algorithms, Random Forest and SVM-RFE, to systematically evaluate the importance of each feature in predicting electronic waste exposure.

[0112] Random forest algorithm displays importance ranking based on Gini index ( Figure 7 The top ten key variables are as follows: 2,4,6-TriBP>2,4,5-TriCP>4-MonoBP>3-OHF>BPA>2-MHA>1-OHPyr>2,3,4,6-TetraCP>3,4-DiCP>MA.

[0113] Under 10-fold cross-validation, the SVM-RFE algorithm achieved the lowest model error rate (2.06%) when selecting 7 features. The average importance scores were ranked as follows: 2,4,5-TriCP > 1-OHPyr > 1-&9-OHPhe > 2,4,6-TriBP > DiEPho > BPA > 4-MonoBP. Figure 8 ).

[0114] Through cross-matching using dual algorithms, five core exposure biomarkers were ultimately selected: 2,4,6-TriBP (2,4,6-tribromophenol), 2,4,5-TriCP (2,4,5-trichlorophenol), 4-MonoBP (4-bromophenol), 1-OHPyr (1-hydroxypyrene), and BPA (bisphenol A). Intergroup concentration comparisons showed that all five exposure biomarkers were significantly elevated in the population exposed to e-waste pollution, with significant differences in overall concentration distribution. The difference between 2,4,6-TriBP and 2,4,5-TriCP was the most significant. Figure 9 ).

[0115] Quantitative analysis showed that the geometric mean concentration of 2,4,6-TriBP in the exposure group was approximately 24 times that of the control group, and that of 2,4,5-TriCP was 12 times higher (Table 7). Notably, the 5th percentile values ​​of both biomarkers in the exposure group exceeded the 95th percentile values ​​in the control group, demonstrating extremely strong potential for exposure discrimination. Based on this, the present invention further established a discrimination method based on single biomarker thresholds: the threshold for 2,4,6-TriBP was set at 0.169 μg / g creatinine, and the threshold for 2,4,5-TriCP was set at 0.0044 μg / g creatinine, according to the aforementioned percentile values. Validation results showed that using the 2,4,6-TriBP threshold could accurately identify 95.2% of the exposed samples and 98.3% of the control samples, with an overall accuracy of 96.7%; 2,4,5-TriCP also showed a similar accuracy (…). Figure 10 Therefore, 2,4,6-TriBP and 2,4,5-TriCP can serve as specific diagnostic indicators for e-waste exposure, providing an effective technical means for achieving rapid and accurate exposure assessment.

[0116] External validation of model generalization ability: To test the model's generalization ability, an independent dataset containing 226 children from e-waste dismantling areas was introduced as a validation set (presumably the exposed group). If the model can accurately determine that this group is exposed to e-waste pollution, the prediction is considered correct; otherwise, it is considered incorrect. This is used to evaluate the model's applicability to unknown populations.

[0117] (5) Interpretability, dimensionality reduction and simplified function derivation of the electronic waste pollution exposure prediction model

[0118] A systematic, hierarchical dimensionality reduction strategy was adopted to simplify the interpretability of the high-dimensional e-waste pollution exposure risk prediction model. The specific steps include:

[0119] Key Feature Identification and Preliminary Dimensionality Reduction: Based on a fully trained optimal high-dimensional prediction model, two algorithms—Random Forest (RF) and Support Vector Machine Recursive Feature Elimination (SVM-RFE)—are used to analyze feature importance. Random Forest assesses a feature's contribution by calculating the average Gini index decrease it brings to the decision tree; SVM-RFE ranks features through iterative training and removal of the lowest-weighted features. Combining the results of multiple iterations of both methods, the average rank (AvgRank) of the features is used as the overall importance criterion. Based on this, key features are selected, completing the initial dimensionality reduction from the high-dimensional model to a five-dimensional model.

[0120] based on Figure 9 Based on the five identified biomarkers, this invention constructs a support vector machine e-waste exposure prediction model, which exhibits excellent discriminative performance (mean AUC = 1.00, ACC = 0.97). Figure 11 ), with ACC>0.95 in the independent validation set (children exposed to e-waste pollution).

[0121] Hierarchical optimal model construction: Based on the five-dimensional model, a single-feature elimination method is used to systematically evaluate all four-feature combinations, constructing five feature subsets containing four feature variables, and training four-dimensional support vector machine models based on radial basis function kernel functions for each subset. By comparing the AUC and ACC of each model, the optimal four-dimensional model is determined.

[0122] To improve the applicability of the model, a single feature exclusion method was used to systematically evaluate all feature combinations (Table 9), achieving a stepwise dimensionality reduction from five-dimensional to four-dimensional models.

[0123] Table 9. Combinations of different biomarkers

[0124] The results showed that all five combinations of e-waste pollution exposure prediction models exhibited excellent predictive performance. Their ACC1 and AUC on the test set were higher than 0.94 and 0.99, respectively; in the independent validation set (children exposed to e-waste pollution), ACC2 also reached 0.71–0.91. Figure 12 This indicates that all five models can be effectively used to predict e-waste pollution exposure. Among them, combination 2 (2,4,5-TriCP, 4-MonoBP, 2,4,6-TriBP and BPA) has the best performance (mean AUC = 1.00, test set ACC = 0.97, validation set ACC = 0.87).

[0125] Subsequently, using the same strategy, four more three-dimensional models were constructed and the optimal three-dimensional model was selected to form a three-dimensional discriminative hypersurface containing complex nonlinear decision boundaries.

[0126] When the dimensionality was further reduced to a combination of three biomarkers, all four combinations of e-waste pollution exposure prediction models showed excellent predictive performance, with ACC and AUC values ​​exceeding 0.95 and 0.99, respectively. This indicates that all four models can be effectively used for predicting e-waste pollution exposure. Figure 13 Among them, combination 1 (2,4,5-TriCP, 4-MonoBP, and 2,4,6-TriBP) was the optimal configuration (mean AUC = 1.00, ACC = 0.97). Three-dimensional scatter plot analysis showed that this combination could effectively distinguish between the e-waste exposed population and the non-exposed control group, achieving accurate separation between adult controls and exposed workers / children. Figure 14 ).

[0127] Visualization and simplified function representation of decision boundaries: Orthogonal projection is used to map the optimal 3D decision hypersurface to a 2D coordinate plane composed of pairwise features. On each projection plane, the intersection of the 3D hypersurface and the feature plane forms the 2D decision boundary. These boundary lines can be further expressed as linear discriminant functions, whose parameters are determined through mathematical transformations of the original 3D decision function.

[0128] To simplify the application of the model, the optimal 3D model is decomposed into three 2D support vector machine classifiers. Figure 15 Based on the decision boundary, the following simplified discriminant function is derived:

[0129] Function 1: When – 0.34 × Ln(2,4,5-TriCP) – Ln(2,4,6-TriBP) – 3.96 < 0, it is determined to be electronic waste exposure (test set ACC1 = 0.988, validation set ACC2 = 0.938).

[0130] Function 2: When – 0.32 × Ln(4-MonoBP) – Ln(2,4,6-TriBP) – 2.86 < 0, it is determined to be electronic waste exposure (test set ACC1 = 0.979, validation set ACC2 = 0.947).

[0131] Function 3: When – 1.91 × Ln(2,4,5-TriCP) – Ln(4-MonoBP) – 13.16 < 0, it is determined to be electronic waste exposure (test set ACC1 = 0.976, validation set ACC2 = 0.686).

[0132] The compounds in the function refer to their concentration in urine, expressed in μg / g creatinine;

[0133] Using a three-function joint discrimination method, only 4 out of 266 exposed children were consistently misclassified, resulting in an error rate as low as 1.77%. Figure 16 This demonstrates that the simplified function derived in this invention achieves a discrimination accuracy comparable to the aforementioned high-dimensional models. This indicates that while maintaining the predictive performance of high-dimensional SVM classifiers, it significantly improves the practicality and operability of the model.

[0134] Cross-regional validation and application performance evaluation: Based on the excellent predictive performance of function 2, this invention further employs a multi-regional validation set population system to evaluate its generalization ability in identifying electronic waste pollution exposure. Figure 17 Cross-regional validation results show that the accuracy rates of samples from e-waste dismantling areas in provinces 1 and 2 in correctly identifying them as "e-waste exposed" were 0.918 and 0.983, respectively; the accuracy rates of samples from non-ferrous metal smelting / processing areas in provinces 3, 4, and 5 in correctly identifying them as "non-e-waste exposed" were >0.880; and the accuracy rates of samples from non-ferrous metal mining and oil / coal processing areas in provinces 6 and 7 in correctly classifying them as "non-e-waste exposed" were >0.703. Overall prediction performance shows that Function 2 has an overall prediction accuracy rate of 0.842 for all eight regions, indicating its excellent ability to identify and generalize e-waste pollution exposure. In summary, Function 2 demonstrates stable discrimination performance in validations across different geographical regions and industrial backgrounds, confirming its effectiveness as a tool for identifying e-waste pollution sources and providing a new technical means for tracing regional environmental pollution sources.

[0135] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. The application of human exposure fingerprint spectrum of electronic waste pollution in exposure risk prediction, characterized by: The human exposure fingerprint spectrum is composed of three or more of the following exposure markers in urine: 2,4,6-tribromophenol, 2,4,5-trichlorophenol, 4-bromophenol, 1-hydroxypyrene, and bisphenol A.

2. Use according to claim 1, characterized in that: The human exposure fingerprint spectrum is composed of four or more of the following exposure markers in urine: 2,4,6-tribromophenol, 2,4,5-trichlorophenol, 4-bromophenol, 1-hydroxypyrene, and bisphenol A.

3. The application according to claim 1, characterized in that: The aforementioned human exposure fingerprint spectrum is one of the following combinations of exposure markers in urine: Combination 1: 2,4,5-trichlorophenol, 4-bromophenol, and 2,4,6-tribromophenol; Combination 2: 2,4,5-trichlorophenol, 4-bromophenol, and bisphenol A; Combination 3: 2,4,5-trichlorophenol, 2,4,6-tribromophenol, and bisphenol A; Combination 4: 4-bromophenol, 2,4,6-tribromophenol, and bisphenol A; Combination 5: 2,4,6-tribromophenol, 2,4,5-trichlorophenol, 4-bromophenol, and 1-hydroxypyrene; Combination 6: 2,4,6-tribromophenol, 2,4,5-trichlorophenol, 4-bromophenol, and bisphenol A; Combination 7: 2,4,5-trichlorophenol, 4-bromophenol, 1-hydroxypyrene, and bisphenol A; Combination 8: 2,4,6-tribromophenol, 2,4,5-trichlorophenol, 1-hydroxypyrene, and bisphenol A; Combination 9: 2,4,6-tribromophenol, 4-bromophenol, 1-hydroxypyrene, and bisphenol A; Combination 10: 2,4,6-tribromophenol, 2,4,5-trichlorophenol, 4-bromophenol, 1-hydroxypyrene, and bisphenol A.

4. The application according to claim 1, characterized in that: The aforementioned human exposure fingerprint spectrum is used to construct a predictive model for electronic waste pollution exposure risk.

5. The application according to claim 4, characterized in that: The method for constructing the exposure risk prediction model includes the following steps: S1: Obtain multiple exposure biomarkers related to e-waste pollution in urine samples from populations in potentially and unexposed areas; S2: Establish training and validation sample sets based on various exposure markers; S3: Train machine learning models using training sample sets corresponding to various exposure biomarkers to obtain various primary diagnostic models; wherein, during the training process, the concentration of the exposure biomarker is used as the input feature of the machine learning model, and the exposure status of the urine sample collected for the exposure biomarker is used as the label; the exposure status includes exposed and unexposed. S4: For various primary diagnostic models, the accuracy is verified using validation sample sets corresponding to various exposure biomarkers; S5: Select the primary diagnostic model with the highest accuracy as the final exposure risk prediction model.

6. The application according to claim 4, characterized in that: The apparatus for the exposure risk prediction model includes: The acquisition module is used to acquire multiple exposure markers in urine; The sample construction module is used to build training and validation sample sets based on various exposure markers. The training module trains machine learning models using training sample sets corresponding to various exposure biomarkers to obtain various primary diagnostic models. During training, the concentration of the exposure biomarker is used as the input feature of the machine learning model, and the exposure status of the urine sample collected for the exposure biomarker is used as the label. The exposure status includes exposed and unexposed. Validation module: For various primary diagnostic models, the accuracy is verified using validation sample sets corresponding to various exposure biomarkers; Select Module: Choose the primary diagnostic model with the highest accuracy as the final exposure risk prediction model.

7. The application according to claim 5 or 6, characterized in that: The machine learning model mentioned is a random forest model, support vector machine, or neural network model.

Images