Environmental micro-plastic ecological risk evaluation method based on data driving and normalization

Abstract

The invention discloses an environmental micro-plastic ecological risk evaluation method based on data driving and normalization, and belongs to the technical field of micro-plastic environmental risk evaluation, and the method comprises the steps: collecting and integrating micro-plastic exposure experiment data, carrying out response ratio normalization processing and Meta analysis, carrying out p value inspection and effect quantity threshold inspection, and identifying significant response species. Screening sensitive species and sensitive effect indexes, traversing a machine learning classification algorithm, taking parameters of a micro-plastic exposure experiment as input features and taking whether the effect is significant as output, forming a supervised learning data set, training a machine learning classifier, carrying out deep analysis by utilizing a model interpretable method, and quantifying and normalizing a required effect threshold value. The method comprises the following steps: establishing a toxicity equivalent model, performing equivalent concentration conversion of microplastic combinations with different characteristics according to different effect threshold ratios, comparing the equivalent concentration with an effect threshold of standard microplastic, determining ecological risks, evaluating microplastic environmental risks, and providing a scientific basis for pollution management and control.

Description

Technical Field

[0001] This application belongs to the field of environmental risk assessment technology for microplastics, specifically, it relates to a data-driven and normalized method for assessing the ecological risk of microplastics in the environment. Background Technology

[0002] Microplastics (plastic particles with a diameter of <5 mm) have become a global emerging pollutant, widely distributed in water, soil, atmosphere, and organisms. They pose a potential threat to ecosystems and human health through the food chain. Existing research indicates that microplastics can exert multidimensional toxic effects on aquatic organisms through physical damage, chemical toxicity, and combined pollution pathways, including oxidative stress, immunosuppression, metabolic disorders, and reproductive impairment. Furthermore, their nanoscale particles can penetrate the Casparian strip barrier of plant roots, accumulating in the edible parts of vegetables, further expanding human exposure pathways.

[0003] The core bottleneck in current microplastic environmental risk assessment lies in the lack of a unified toxicity threshold standard. Traditional ecological risk assessment methods mainly rely on dose-response experiments of single species, making it difficult to integrate the complex toxicity effects of microplastics across multiple dimensions. Although meta-analyses have confirmed the sensitivity of fish to microplastic exposure, different experimental conditions lead to significant heterogeneity in effect data, hindering cross-study comparability. Furthermore, the diversity of microplastic forms in the environment and their synergistic effects with coexisting pollutants further increase the complexity of threshold setting.

[0004] Machine learning technology offers a new approach to solving these problems. In the field of environmental science, it has been successfully applied to pollutant concentration prediction, ecosystem health assessment, and risk source identification. Through multi-source data fusion and pattern mining, it can overcome the limitations of traditional statistical methods. Models based on algorithms such as random forests and support vector machines can effectively analyze the nonlinear relationship between high-dimensional features and toxic responses, and achieve accurate classification of risk levels. However, existing research has not yet established a threshold solution framework for multi-parameter coupling of microplastics, and in particular, it lacks a standardized method to link the interpretability of machine learning models with ecotoxicological endpoints. The lack of thresholds in microplastic environmental risk assessment leads to the inability to effectively assess the environmental risks of microplastics, which restricts the scientific assessment and effective management of microplastic pollution. Summary of the Invention

[0005] To address the aforementioned problems and technical deficiencies, this application adopts the following technical solution: a data-driven and normalized environmental microplastics ecological risk assessment method, comprising the following steps: Step 1: Based on a systematic literature retrieval strategy, conduct data mining on academic databases, collect and integrate experimental data on microplastic exposure of representative plants and animals at key trophic levels or with indicative functions in different ecosystems; Step 2: Normalize the experimental data using the natural logarithm response ratio method; Step 3: Perform meta-analysis on the normalized data, draw effect forest plots for each species, show the effect size distribution and confidence intervals, perform p-value tests and effect size threshold tests, and identify significant response species and indicators as sensitive species and indicators related to microplastics. Step 4: Iterate through the machine learning classification algorithms, taking the features of the microplastic exposure experiment as input and whether the effect is significant as output, to form a supervised learning dataset, train the selected machine learning classifier, and optimize the optimal model; Step 5: Using model interpretability methods, perform in-depth analysis on the trained and optimized machine learning classifier to quantify the effect threshold required for normalization. Step 6: Based on the effect threshold and different microplastic exposure characteristics, establish a toxicity equivalence model. According to the ratio of the effect threshold corresponding to different microplastic characteristics to the effect threshold of standard microplastics, convert the equivalent concentration of different microplastic combinations under the effect equivalence background. Compare the equivalent concentration with the effect threshold of standard microplastics to determine whether there is an ecological risk.

[0006] Preferably, a multidimensional data tuple consisting of microplastic multidimensional exposure parameters is constructed based on the experimental data. The multidimensional data tuple includes: the species of the test organism, the concentration c of the microplastic, the size l, the material m, the shape s, the exposure time t, and the corresponding biological effect endpoint X. The effect endpoint X is derived from the measured effect data of the exposed group. Background data compared to the control group Composition, effect data Background data These are the raw values ​​used to calculate the effect size.

[0007] Further analysis of the normalized data includes: Collected effect data Background data of the control group Normalized to the effect change factor To eliminate incomparability caused by differences in experimental conditions on a uniform scale, the formula for the natural logarithmic response ratio is as follows:

[0008] Then, data volume and sensitivity analysis were performed on the normalized data to screen sensitive species and sensitive effect indicators.

[0009] Furthermore, among the species with significant responses, the effect indicator with the broadest data coverage is selected as the key effect. Subsequently, all sub-indicators within the same species are tested, requiring that all sub-indicators meet the following requirements. pValue tests and effect size thresholds were used to verify the consistency of the sub-indicators.

[0010] Furthermore, the training of the machine learning classifier uses the sensitivity effect index of sensitive species and the multidimensional exposure parameters of microplastics as input features, and uses whether the effect is significant as label data. The criterion for judging the significance of the effect is the significance marker in the original data.

[0011] Furthermore, during the training process of the machine learning classifier, a multi-dimensional evaluation system is constructed by fitting the relationship between features and labels in the training data. The model performance is ranked by accuracy, AUC, recall, precision, F1 index, Kappa index, and MCC index. The optimal model is determined by ranking, and the hyperparameters of the model are tuned to obtain the best classification performance.

[0012] Furthermore, the deep parsing process includes: Select a representative multidimensional feature combination from the input features, and use the principle of the controlled variable method to fix all features except the target feature as constant values. Then, sequentially traverse the target feature within a preset numerical range to construct a one-dimensional discriminant domain about the target feature. For each feature value within the one-dimensional discriminant domain, a pre-trained machine learning classifier is invoked to make a prediction, outputting the probability P of the corresponding significant effect class, and generating a series of values ​​corresponding to the predicted probabilities. The dense grid of points formed; The exploration range of target features and the grid density are defined, so that all grid points are fitted into a probability decision curve in a two-dimensional coordinate system. By analyzing the gradient change of this curve, the probability is found. Identify the mutation points where significant transitions occur and obtain the corresponding eigenvalues.

[0013] Furthermore, the characteristic value of the mutation point is a key parameter for comparing and converting complex and diverse environmental microplastic exposure scenarios with a unified benchmark. At the same time, the physical meaning of the mutation point is clarified. Under the condition that other multidimensional characteristics of microplastics are set to specific constant values, the target characteristic constitutes a key discrimination boundary value. The boundary value corresponds to the lowest visible harmful concentration or the maximum concentration with no visible harmful effect under specific conditions in toxicology.

[0014] Furthermore, the process of obtaining the effect threshold includes: Define a standard microplastic reference object. Based on the effective application domain of the machine learning classifier, select the center vector of all sample feature vectors within the application domain as the feature parameters of the standard microplastic. Based on the application domain of the model, select the center vector within the application domain as the standard microplastic. Using the multidimensional feature parameters of the standard microplastic, excluding the target feature, as fixed inputs, the aforementioned gridded prediction and decision boundary analysis steps are repeated. While keeping other features fixed, the target features are traversed to find the probability... The boundary value corresponding to the transition is the effect threshold of standard microplastics.

[0015] Furthermore, the formula for calculating the equivalent concentration is as follows: Equivalent concentration = Actual detected concentration × (Characteristic value of mutation point / Effect threshold) The toxicity equivalence model formula is as follows:

[0016] Parameters were fitted using experimental data, based on the principle of toxicity equivalence. T 1= T 2. Enable concentration conversion; The effect of microplastics with known characteristic combinations; The effect of the combination of microplastic features to be converted; These are the concentration, size, material, shape, and exposure time of the microplastics to be converted.

[0017] Compared to existing technologies, the beneficial effects of this application are as follows: (1) This application uses a data-driven machine learning method to systematically collect data and process it scientifically to accurately determine the LOEC or MNOEC of microplastics, build a scientific quantitative threshold for the ecological risk assessment of microplastics, and use the accurate construction of risk thresholds to help make environmental protection policies more scientific and targeted. (2) This application eliminates the interference of data differences and explores complex potential relationships by using data normalization and machine learning algorithms, comprehensively assesses risks from multiple dimensions, improves the accuracy of risk assessment, breaks through the limitations of traditional assessment, makes the environmental risk assessment of microplastics more in line with reality, and provides a scientific basis for environmental management decisions. (3) Based on the established effect threshold and toxicity equivalence model, this application converts the concentration of microplastics with different characteristics into equivalent concentrations, realizes the equivalent conversion of concentration in complex environmental samples, intuitively presents the overall toxicity level of microplastics, and provides convenience for environmental monitoring and risk assessment. (4) This application integrates multi-dimensional exposure parameters and biological effect data, combined with ecological risk thresholds and risk assessment results, to scientifically assess and effectively manage microplastic pollution, provide a scientific basis for pollution control, point out the direction for microplastic pollution control, facilitate the rational allocation of resources, develop targeted control technologies, assist in monitoring work, and promote scientific and efficient control work. Attached Figure Description

[0018] In the attached diagram: Figure 1 This is a schematic diagram of the method flow of an embodiment of this application; Figure 2 This is a diagram showing the dimensions and scale of the aquatic organism dataset in this application embodiment; Figure 3 This is a graph showing the meta-analysis results of fish enzyme activity according to an embodiment of this application; Figure 4 The table below shows the performance of the effect discrimination model in the embodiments of this application. Figure 5 This is a table showing the characteristics and equivalents of the microplastic components in the leachate of this application embodiment. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this application, but not all embodiments. Generally, the components of the embodiments of this application described and shown in the accompanying drawings can be arranged and designed in various different configurations.

[0020] Example 1 like Figure 1 As shown, the data-driven and normalized method for assessing the ecological risk of microplastics in the environment includes the following steps: Data mining was conducted on academic databases based on a systematic literature retrieval strategy to collect and integrate experimental data on microplastic exposure of representative plants and animals at key trophic levels or with indicative functions in different ecosystems. Based on experimental data, a multidimensional data tuple consisting of microplastic multidimensional exposure parameters was constructed. The multidimensional data tuple includes: the species of the test organism, the concentration c of the microplastic, the size l, the material m, the shape s, the exposure time t, and the corresponding biological effect endpoint X. The effect endpoint X is derived from the measured effect data of the exposed group. Background data compared to the control group Composition, effect data Background data These are the raw values ​​used to calculate the effect size.

[0021] The experimental data were normalized using the natural logarithm response ratio method. Analysis of normalized data includes: Collected effect data Background data of the control group Normalized to the effect change factor To eliminate incomparability caused by differences in experimental conditions on a uniform scale, the formula for the natural logarithmic response ratio is as follows:

[0022] Then, data volume and sensitivity analysis were performed on the normalized data to screen sensitive species and sensitive effect indicators.

[0023] Meta-analysis was performed on the normalized data to draw effect forest plots for each species, showing the distribution of effect sizes and confidence intervals. P-value tests and effect size threshold tests were conducted to identify significant response species and indicators, which can be used as sensitive species and indicators related to microplastics. The effect measure with the broadest data coverage among the significantly responding species was selected as the key effect. Subsequently, all sub-measures within the same species were tested, requiring that all sub-measures meet the following requirements. p Value tests and effect size thresholds were used to verify the consistency of the sub-indicators.

[0024] The machine learning classification algorithm is iterated through, taking the features of the microplastic exposure experiment as input and the significance of the effect as output, to form a supervised learning dataset. The selected machine learning classifier is trained and optimized for the best model. Training a machine learning classifier involves using sensitivity effect indicators of sensitive species and microplastic multidimensional exposure parameters as input features, and using the significance of the effect as label data. The criterion for judging the significance of the effect is the significance marker in the original data.

[0025] During the training process of a machine learning classifier, a multi-dimensional evaluation system is constructed by fitting the relationship between features and labels in the training data. The model performance is ranked by accuracy, AUC, recall, precision, F1 index, Kappa index, and MCC index. The optimal model is determined by ranking, and the hyperparameters of the model are tuned to obtain the best classification performance.

[0026] Using model interpretability methods, we perform deep analysis on machine learning classifiers that have been trained and optimized, and quantify the effect threshold required for normalization. The deep parsing process includes: Select a representative multidimensional feature combination from the input features, and use the principle of the controlled variable method to fix all features except the target feature as constant values. Then, sequentially traverse the target feature within a preset numerical range to construct a one-dimensional discriminant domain about the target feature. For each feature value within the one-dimensional discriminant domain, a pre-trained machine learning classifier is invoked to make a prediction, outputting the probability P of the corresponding significant effect class, and generating a series of values ​​corresponding to the predicted probabilities. The dense grid of points formed; The exploration range of target features and the grid density are defined, so that all grid points are fitted into a probability decision curve in a two-dimensional coordinate system. By analyzing the gradient change of this curve, the probability is found. Identify the mutation points where significant transitions occur and obtain the corresponding eigenvalues.

[0027] The eigenvalue of the mutation point is a key parameter for comparing and converting complex and diverse environmental microplastic exposure scenarios with a unified benchmark. At the same time, it clarifies the physical meaning of the mutation point. Under the condition that other multidimensional characteristics of microplastics are set to specific constant values, the target characteristic constitutes a key discrimination boundary value. In toxicology, the boundary value corresponds to the lowest concentration of visible harmful effects or the maximum concentration of no visible harmful effects under specific conditions.

[0028] The process of obtaining the effect threshold includes: Define a standard microplastic reference object. Based on the effective application domain of the machine learning classifier, select the center vector of all sample feature vectors within the application domain as the feature parameters of the standard microplastic. Based on the application domain of the model, select the center vector within the application domain as the standard microplastic. Using the multidimensional feature parameters of the standard microplastic, excluding the target feature, as fixed inputs, the aforementioned gridded prediction and decision boundary analysis steps are repeated. While keeping other features fixed, the target features are traversed to find the probability... The boundary value corresponding to the transition is the effect threshold of standard microplastics.

[0029] Based on the effect threshold and different microplastic exposure characteristics, a toxicity equivalence model is established. The equivalent concentration of different microplastic combinations under the effect equivalence background is converted by the ratio of the effect threshold corresponding to different microplastic characteristics to the effect threshold of standard microplastics. The equivalent concentration is compared with the effect threshold of standard microplastics to determine whether there is an ecological risk.

[0030] The formula for calculating equivalent concentration is as follows: Equivalent concentration = Actual detected concentration × (Characteristic value of mutation point / Effect threshold) The toxicity equivalence model formula is as follows:

[0031] Parameters were fitted using experimental data, based on the principle of toxicity equivalence. T 1= T 2. Enable concentration conversion; The effect of microplastics with known characteristic combinations; The effect of the combination of microplastic features to be converted; These are the concentration, size, material, shape, and exposure time of the microplastics to be converted.

[0032] Example 2 A data-driven machine learning method for solving the threshold of microplastic effects is proposed, and its specific implementation steps are as follows: Based on a systematic literature retrieval strategy, data mining was conducted on academic databases worldwide (PubMed, Web of Science, Google Scholar, IEEE Xplore, ScienceDirect) to collect and integrate experimental data on microplastic exposure of representative plants and animals at key trophic levels or with indicative functions in different ecosystems (including but not limited to aquatic and terrestrial ecosystems).

[0033] The data was constructed as multidimensional data tuples, including: the type of test organism and the concentration of microplastics. ,size Material ,shape Exposure time and the corresponding biological effect endpoint .

[0034] The endpoint of the effect Specifically, based on the measured effect data of the exposure group. Background data compared to the control group Composition, including effect data Background data These are the raw values ​​used to calculate the effect size.

[0035] use The (natural logarithmic response ratio) method collects effect data. Background data of the control group Normalized to the factor of change of effect This unified scale eliminates the incomparability caused by differences in experimental conditions; Based on normalized data, the sensitivity species and sensitivity effect indicators were jointly determined through data volume and sensitivity effect significance analysis.

[0036] Based on normalized data, sensitive species and sensitive effect indicators are identified.

[0037] Meta-analysis of normalized data was performed to create effect forest plots for each species, visually displaying the distribution of effect sizes and confidence intervals. p Value test ( p<0.05 The dual criteria of effect size threshold (|lnRR|>0.2) were used to identify species with significant responses.

[0038] Among species with significant responses, the effect indicator with the broadest data coverage was selected as the key effect. Subsequently, all sub-indicators within the same species were tested, requiring all to be significant. p<0.05 The stable effect size (|lnRR|>0.2) verifies the consistency of the sub-indicators.

[0039] Next, the sensitivity effect indicators of sensitive species and the multidimensional exposure parameters of microplastics (concentration, size, material, shape, time) are used as input features of the machine learning model, and whether the effect is significant is used as the label. Here, the criterion for judging "significant effect" is the significance marker in the original data, and the machine learning classifier is trained.

[0040] The machine learning classification algorithms are traversed, including Logistic Regression, RidgeClassifier, Linear Discriminant Analysis, Random Forest Classifier, LightGradient Boosting Machine, K Neighbors Classifier, Extra Trees Classifier, Gradient Boosting Classifier, Ada Boost Classifier, Decision Tree Classifier, Naive Bayes, Dummy Classifier, Quadratic Discriminant Analysis, and SVM-LinearKernel. The selected machine learning classifier is trained using the determined input features and label data as a supervised learning dataset.

[0041] During model training, a multi-dimensional evaluation system is constructed by fitting the relationship between features and labels in the training data. Model performance is ranked using accuracy, AUC, recall, precision, F1 score, Kappa score, and MCC score. The optimal model is determined by ranking, and its hyperparameters are then tuned to achieve the best classification performance.

[0042] By utilizing model interpretability methods, such as feature space gridding prediction and decision boundary visualization methods, we can perform in-depth analysis on machine learning classifiers that have been trained and validated to quantify the contribution of each input feature to the model's prediction results. The interpretable model method selects a representative multidimensional feature combination from the input features (concentration c, size l, material m, shape s, exposure time t) and adopts the principle of the control variable method. For example, all features other than concentration feature c (size l, material m, shape s, exposure time t) are fixed as constant values, and then concentration feature c is sequentially traversed within a preset numerical range to construct a one-dimensional discriminant domain for concentration. For each concentration value within the discrimination domain, a pre-trained machine learning classifier is invoked to make a prediction, outputting the corresponding "significant effect" category probability P, thereby generating a series of... With predicted probability The dense grid points are formed, and the range of values ​​for the concentration feature c is assumed to be... And set the mesh density to n (where n is a positive integer, with a default value of 100), then all mesh points A set can be fitted to form a probability decision curve in a two-dimensional coordinate system; By analyzing the gradient change of the curve, we can find the probability. The point of significant transition, the concentration value corresponding to this point is... ; This mutation point With a clear physical meaning, the concentration of microplastics, under the condition that other multidimensional characteristics (such as size, material, shape, and exposure time) are set to specific constant values, has a specific physical meaning. This constitutes a key discrimination threshold, which toxicologically corresponds to the lowest concentration with visible harmful effects or the highest concentration with no visible harmful effects under specific conditions. These parameters constitute the key parameters for comparing and converting complex and diverse environmental microplastic exposure scenarios with a unified benchmark.

[0043] To achieve this conversion, a standard microplastic reference is defined as follows: based on the effective application domain of the machine learning model, the center vector of all sample feature vectors within that application domain (i.e., the combination of the mean values ​​of each feature, including standard concentration, standard size, standard material, standard shape, and standard exposure time) is selected as the feature parameter of the standard microplastic. Based on the model's application domain, the central vector within the application domain (combination of concentration, size, material, shape, and mean values ​​of each feature) is selected as the standard microplastic, and its effect threshold is similarly calculated. The effect threshold of this standard microplastics The solution logic is the same as the aforementioned The solution process is completely consistent. Using the multidimensional characteristic parameters of the standard microplastic (except for concentration characteristics) as fixed inputs, the aforementioned gridded prediction and decision boundary analysis steps are repeated. That is, under the premise of fixed standard size, standard material, standard shape, and standard exposure time, the concentration characteristics are traversed to find the probability... The concentration threshold value at which the transition occurs is the effect threshold of standard microplastics. .

[0044] Based on the established effect threshold and different microplastic exposure characteristics (concentration, size, material, time), a toxicity equivalence model is established. According to different effect threshold ratios ( This allows for the conversion of equivalent concentrations of different microplastic combinations under the condition of equivalent effect. Equivalent concentration = actual detected concentration × The effect threshold compared to standard microplastics. In comparison, it can be determined whether there are ecological risks.

[0045] Example 3 To assess the ecological risk of microplastics in river water, real aquatic environment microplastic samples were used. The equivalent of microplastics in the aquatic environment was converted to that of standard microplastics. Based on the equivalent of standard microplastics and the effect threshold, it was determined whether there was an ecological risk from the microplastics in the aquatic environment (48 h).

[0046] After testing, the microplastics in the aquatic environment were found to be spherical in shape, with a concentration of 4.1 n / L, a size range of about 120 μm, and all made of PS.

[0047] First, an aquatic biological effect dataset was constructed to identify sensitive organisms and receptors. Common species such as zebrafish and crucian carp were selected as fish samples, and water hyacinth and duckweed were selected as aquatic plant samples. For marine ecosystems, shellfish such as mussels and oysters, as well as planktonic organisms such as phytoplankton were selected.

[0048] For each biological sample, based on meta-analysis, detailed environmental characteristics of its exposure experiment are recorded, such as water temperature, salinity, and exposure time. Detailed information on microplastic samples was recorded, including material, size, shape, and concentration. Various materials were included, such as polyethylene (PE), polypropylene (PP), polystyrene (PS), and polyvinyl chloride (PVC); microplastic sizes ranged from micrometers (e.g., 1 micrometer) to millimeters (e.g., 5 millimeters); and concentration gradients ranged from low (0.1 n / L) to high (100 n / L). Furthermore, changes in physiological and biochemical indicators of model organisms were collected, and the categories, values, and significance of the control and experimental groups were identified.

[0049] After logarithmic response ratio processing, such as Figure 2As shown, the scale of the species dimension of the dataset is as follows: total number of samples 3835, including fish (1607), arthropods (774), algae (754), mollusks (450), and bacteria (250). Since machine learning requires a large amount of data support, fish were selected as the key species. Among all fish samples, the scale of the effect dimension of the dataset is as follows: oxidative stress (510), immune system (466), metabolism (216), motor nerves (143), oxidative stress genes (98), excretion (89), growth (78), and reproduction (7).

[0050] Based on the discrimination principle, the meta-analysis results of fish enzyme activity are as follows: Figure 3 As shown, since all sub-dimensions of oxidative stress have lnRR ranging from 0.25 to 0.49, they are all sensitive to microplastics and have the largest amount of effect data. Therefore, oxidative stress is determined to be a sensitive indicator for fish.

[0051] The input features and label data are divided into training and test sets in a ratio of 7:3.

[0052] In Python, the Pycaret library iterates through all classifier models, and the performance of the effect discrimination model is as follows: Figure 4 As shown, the optimal model is determined to be the limit decision tree ET.

[0053] By traversing the grid relationship of the microplastic feature combination in this aquatic environment as a function of concentration, the model gives a LOEC of 6.3 n / L. The standard microplastic feature represented by the domain center vector in this dataset is 200 μm, spherical, made of PP material, with a LOEC of 10 n / L.

[0054] The effect threshold concentration of environmental microplastics The value is 6.3 n / L, which is the standard microplastic effect threshold. The value is 10 n / L. Based on the equivalent effect conversion, the standard equivalent of environmental microplastics is 4.1 * (10 / 6.3) n / L = 6.508 n / L, which is close to the standard microplastic effect threshold. In comparison, no significant effect was observed.

[0055] Example 4 To assess the ecological risk of environmental microplastics in landfill leachate, we used real microplastic samples from leachate, where the composition of microplastics is more complex. We planned to convert the equivalent of microplastics in this environment to that of standard microplastics, and based on the equivalent of standard microplastics and the effect threshold, we determined whether there is an ecological risk from microplastics in this environment (48 h).

[0056] After testing, the microplastics in the leachate were characterized as follows: the shape was spherical (50%) and particulate (50%), with concentrations of 20 n / L and 30 n / L, respectively; the size was divided into three parts, around 200 (20%), 500 (50%) and 1000 (30%) μm; and the material was PS (50%) and PP (50%).

[0057] An aquatic organism dataset was established, and the key sensitive indicator was identified as oxidative stress in fish. An optimal RF random forest classifier was constructed. Since this process is based on meta-analysis data, the specific process and results are exactly the same as in Example 3.

[0058] Traverse the grid relationship of the microplastic feature combination in the leachate as a function of concentration, and record the effect threshold LOEC given by the model within each sub-feature combination, such as... Figure 5 As shown.

[0059] The standard microplastic characteristics of this dataset are 200 μm, spherical, PP material, and its effect threshold cs is 10 n / L.

[0060] Calculate the concentration of microplastic sub-components in each type of endorheic filtrate and record it as the detection concentration. According to the equivalent conversion formula, the standard microplastic equivalent of this component = detection concentration × (cs / effect threshold). The total equivalent concentration of each component in the leachate was calculated based on the toxicity equivalence model, and the total equivalent concentration was obtained by using the concentration summation model (without interaction).

[0061] based on Figure 5 The equivalent concentration of each component was calculated, and the equivalent concentration of microplastics in the leachate was found to be 42.19 n / L. Compared with the effect threshold cs of standard microplastics, this poses a significant ecological risk to fish.

[0062] Therefore, the conclusion of this method is that the equivalent of microplastics obtained from the leachate sampling is 42.19 n / L, which poses an ecological risk to fish.

[0063] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications, improvements, and substitutions without departing from the concept of this application, and these all fall within the protection scope of this application.

Claims

1. A data-driven and normalized method for assessing the ecological risk of microplastics in the environment, characterized in that, Includes the following steps: Step 1: Based on a systematic literature retrieval strategy, conduct data mining on academic databases, collect and integrate experimental data on microplastic exposure of representative plants and animals at key trophic levels or with indicative functions in different ecosystems; Step 2: Normalize the experimental data using the natural logarithm response ratio method; Step 3: Perform meta-analysis on the normalized data, draw effect forest plots for each species, showing the distribution of effect sizes and confidence intervals, and conduct... p Value tests and effect size threshold tests were used to identify species and indicators with significant responses, which can be used as sensitive species and sensitive indicators related to microplastics. Step 4: Iterate through the machine learning classification algorithms, taking the features of the microplastic exposure experiment as input and whether the effect is significant as output, to form a supervised learning dataset, train the selected machine learning classifier, and optimize the optimal model; Step 5: Using model interpretability methods, perform in-depth analysis on the trained and optimized machine learning classifier to quantify the effect threshold required for normalization. Step 6: Based on the effect threshold and different microplastic exposure characteristics, establish a toxicity equivalence model. According to the ratio of the effect threshold corresponding to different microplastic characteristics to the effect threshold of standard microplastics, convert the equivalent concentration of different microplastic combinations under the effect equivalence background. Compare the equivalent concentration with the effect threshold of standard microplastics to determine whether there is an ecological risk.

2. The environmental microplastics ecological risk assessment method based on data-driven and normalized methods according to claim 1, characterized in that, Based on the experimental data, a multidimensional data tuple consisting of microplastic multidimensional exposure parameters was constructed. The multidimensional data tuple includes: the species of the test organism, the concentration c of the microplastic, the size l, the material m, the shape s, the exposure time t, and the corresponding biological effect endpoint X. The effect endpoint X is derived from the measured effect data of the exposed group. Background data compared to the control group Composition, effect data Background data These are the raw values ​​used to calculate the effect size.

3. The environmental microplastics ecological risk assessment method based on data-driven and normalized methods according to claim 2, characterized in that, The analysis of the normalized data includes: Collected effect data Background data of the control group Normalized to the effect change factor To eliminate incomparability caused by differences in experimental conditions on a uniform scale, the formula for the natural logarithmic response ratio is as follows: Then, data volume and sensitivity analysis were performed on the normalized data to screen sensitive species and sensitive effect indicators.

4. The environmental microplastics ecological risk assessment method based on data-driven and normalized methods according to claim 3, characterized in that, Among the species with significant responses, the effect indicator with the broadest data coverage was selected as the key effect. Subsequently, all sub-indicators within the same species were tested, requiring that all sub-indicators meet the following requirements. p Value tests and effect size thresholds were used to verify the consistency of sub-indicators.

5. The data-driven and normalized environmental microplastics ecological risk assessment method according to claim 3, characterized in that, The training of the machine learning classifier uses the sensitivity effect index of sensitive species and the multidimensional exposure parameters of microplastics as input features, and the significance of the effect as label data. The criterion for judging the significance of the effect is the significance marker in the original data.

6. The data-driven and normalized environmental microplastics ecological risk assessment method according to claim 5, characterized in that, During the training process of the machine learning classifier, a multi-dimensional evaluation system is constructed by fitting the relationship between features and labels in the training data. The model performance is ranked by accuracy, AUC, recall, precision, F1 index, Kappa index, and MCC index. The optimal model is determined by ranking, and the hyperparameters of the model are tuned to obtain the best classification performance.

7. The environmental microplastics ecological risk assessment method based on data-driven and normalized methods according to claim 6, characterized in that, The deep parsing process includes: Select a representative multidimensional feature combination from the input features, and use the principle of the controlled variable method to fix all features except the target feature as constant values. Then, sequentially traverse the target feature within a preset numerical range to construct a one-dimensional discriminant domain about the target feature. For each feature value within the one-dimensional discriminant domain, a pre-trained machine learning classifier is invoked to make a prediction, outputting the probability P of the corresponding significant effect class, and generating a series of values ​​corresponding to the predicted probabilities. The dense grid of points formed; The exploration range of target features and the grid density are defined, so that all grid points are fitted into a probability decision curve in a two-dimensional coordinate system. By analyzing the gradient change of this curve, the probability is found. Identify the mutation points where significant transitions occur and obtain the corresponding eigenvalues.

8. The data-driven and normalized environmental microplastics ecological risk assessment method according to claim 7, characterized in that, The characteristic value of the mutation point is a key parameter for comparing and converting complex and diverse environmental microplastic exposure scenarios with a unified benchmark. At the same time, it clarifies the physical meaning of the mutation point. Under the condition that other multidimensional characteristics of microplastics are set to specific constant values, the target characteristic constitutes a key discrimination boundary value. The boundary value corresponds to the lowest visible harmful concentration or the maximum concentration with no visible harmful effect under specific conditions in toxicology.

9. The environmental microplastics ecological risk assessment method based on data-driven and normalized methods according to claim 7, characterized in that, The process of obtaining the effect threshold includes: Define a standard microplastic reference object. Based on the effective application domain of the machine learning classifier, select the center vector of all sample feature vectors within the application domain as the feature parameters of the standard microplastic. Based on the application domain of the model, select the center vector within the application domain as the standard microplastic. Using the multidimensional feature parameters of the standard microplastic, excluding the target feature, as fixed inputs, the aforementioned gridded prediction and decision boundary analysis steps are repeated. While keeping other features fixed, the target features are traversed to find the probability... The boundary value corresponding to the transition is the effect threshold of standard microplastics.

10. The data-driven and normalized environmental microplastics ecological risk assessment method according to claim 9, characterized in that, The formula for calculating the equivalent concentration is as follows: Equivalent concentration = Actual detected concentration × (Characteristic value of mutation point / Effect threshold) The toxicity equivalence model formula is as follows: Parameters were fitted using experimental data, based on the principle of toxicity equivalence. T 1= T 2. Enable concentration conversion; The effect of microplastics with known characteristic combinations; The effect of the combination of microplastic features to be converted; These are the concentration, size, material, shape, and exposure time of the microplastics to be converted.

Images