Machine learning-based method and device for predicting pfas bioaccumulation factor

Abstract

This invention provides a machine learning-based method and apparatus for predicting PFAS bioaccumulation factors, belonging to the field of environmental risk assessment technology. The method uses a dataset of enrichment factor data, corresponding environmental, biological, and spatial features, and a bioaccumulation factor prediction model constructed using machine learning methods. By combining the interactions between bioaccumulation factors and environmental, biological, and spatial features, the method predicts the molecular structure information of the target PFAS, obtaining the bioaccumulation factors of the target PFAS in various candidate water environments and for various organisms, thus improving the accuracy of PFAS bioaccumulation factor prediction.

Description

Technical Field

[0001] This invention relates to the field of environmental risk assessment technology, and in particular to a method and apparatus for predicting PFAS bioaccumulation factors based on machine learning. Background Technology

[0002] Per- and polyfluoroalkyl substances (PFAS) are a class of synthetic organic pollutants with environmental persistence, mobility, and potential bioaccumulation. Current technologies often use bioaccumulation factors (BAFs) to characterize the bioaccumulation risk of PFAS in order to achieve environmental management and risk control.

[0003] However, existing PFAS bioaccumulation factors often rely on single physicochemical properties, empirical formulas, or finite variables for prediction, which has the drawback of inaccurate prediction and is not conducive to environmental management and risk control. Summary of the Invention

[0004] This invention proposes a machine learning-based method and apparatus for predicting PFAS bioaccumulation factors. It utilizes a bioaccumulation factor prediction model constructed using a dataset of enrichment factor data, corresponding environmental, biological, and spatial features, and machine learning methods. By combining the interactions between bioaccumulation factors and these environmental, biological, and spatial features, the model predicts the molecular structure information of the target PFAS, obtaining the bioaccumulation factors of the target PFAS in various candidate water environments and for multiple organisms. This improves the accuracy of PFAS bioaccumulation factor prediction and is beneficial for environmental management and risk control.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, this invention provides a machine learning-based method for predicting PFAS bioaccumulation factors, comprising: acquiring molecular structure information of a target PFAS; the molecular structure information includes chemical name, CAS number, and molecular descriptor. Based on the molecular structure information of the target PFAS and a bioaccumulation factor prediction model, the bioaccumulation factors of the target PFAS in multiple organisms under various candidate aquatic environments are predicted; wherein, the bioaccumulation factor prediction model is constructed using an enrichment feature dataset and machine learning methods; the enrichment feature dataset includes multiple sample data, the sample data including enrichment factor data of a PFAS and corresponding environmental, biological, and spatial features; the enrichment factor data includes bioaccumulation factors of PFAS in multiple organisms under various aquatic environments; the bioaccumulation factor is the ratio of PFAS concentration in the organism to the PFAS concentration in the aquatic environment.

[0006] In one implementation of the first aspect, the method involves predicting the bioaccumulation factors of the target PFAS in multiple candidate aquatic environments based on the molecular structure information of the target PFAS and a bioaccumulation factor prediction model. This includes: acquiring key features of each candidate aquatic environment corresponding to the target PFAS. For each candidate aquatic environment, the molecular structure information of the target PFAS and the key features of the candidate aquatic environment are input into the bioaccumulation factor prediction model, and the bioaccumulation factor prediction model outputs the bioaccumulation factor of each organism in the multiple aquatic environments for the target PFAS.

[0007] In one implementation of the first aspect, the process of determining key features includes: using a feature contribution analysis method to analyze the contribution of the input features of the biological enrichment factor prediction model to the prediction results, and determining the input feature with the largest contribution as the key feature.

[0008] In one implementation of the first aspect, the key features include key environmental features, key biological features, and key spatial features; the key environmental features include the features that contribute the most to the environmental features, the key biological features include the features that contribute the most to the biological features, and the key spatial features include the features that contribute the most to the spatial features.

[0009] In one implementation of the first aspect, the environmental characteristics include environmental parameters and PFAS concentration in the aquatic environment; the environmental parameters include water temperature, pH value, dissolved oxygen and salinity; the spatial characteristics include the longitude and latitude of the sampling location and the type of water body; the biological characteristics include species classification, biological group, trophic level, tissue type and PFAS concentration in the organism; and the organism is a freshwater aquatic organism.

[0010] In one implementation of the first aspect, the machine learning method is a gradient boosting tree model, a random forest model, or a support vector machine model.

[0011] Secondly, this invention provides a machine learning-based PFAS bioaccumulation factor prediction device, comprising an acquisition module and a prediction module. The acquisition module is used to acquire the molecular structure information of the target PFAS; the molecular structure information includes the chemical name, CAS number, and molecular descriptor. The prediction module is used to predict the bioaccumulation factors of the target PFAS in various organisms under multiple candidate aquatic environments based on the molecular structure information of the target PFAS and a bioaccumulation factor prediction model; wherein, the bioaccumulation factor prediction model is constructed using an enrichment feature dataset and machine learning methods; the enrichment feature dataset includes multiple sample data, the sample data including enrichment factor data of a PFAS and the corresponding environmental, biological, and spatial characteristics of the enrichment factor data; the enrichment factor data includes the bioaccumulation factors of PFAS in various organisms under multiple aquatic environments; the bioaccumulation factor is the ratio of the PFAS concentration in the organism to the PFAS concentration in the aquatic environment.

[0012] Thirdly, the present invention provides an electronic device including a processor and a memory coupled to the processor; the memory is used to store computer instructions, and when the electronic device is running, the processor executes the computer instructions stored in the memory to cause the electronic device to perform the method described in the first aspect above or any implementation thereof.

[0013] Fourthly, the present invention provides a computer-readable storage medium including computer program instructions that, when executed by a computer, cause the computer to perform the method described in the first aspect above or any implementation thereof.

[0014] Fifthly, the present invention provides a computer program product, including computer program instructions, which, when executed on a computer, cause the computer to perform the method described in the first aspect above or any implementation thereof.

[0015] The technical effects corresponding to the second to fifth aspects and their possible implementations can be referred to the above description of the technical effects of the first aspect and its possible implementations, and will not be repeated here.

[0016] Compared with the prior art, the present invention has the following beneficial effects.

[0017] This invention provides a machine learning-based method for predicting PFAS bioaccumulation factors. First, the molecular structure information of the target PFAS is obtained. Then, a bioaccumulation factor prediction model is used to analyze and predict the molecular structure information of the target PFAS, obtaining the bioaccumulation factors of the target PFAS in various candidate water environments and for various organisms. In this process, the bioaccumulation factor prediction model is constructed based on an enrichment feature dataset and machine learning methods. The enrichment feature dataset contains enrichment factor data for various PFAS, as well as the corresponding environmental, biological, and spatial characteristics of the enrichment factor data. This allows the bioaccumulation factor prediction model to combine the influence relationships between enrichment factor data and environmental, biological, and spatial characteristics to predict bioaccumulation factors, thereby improving the accuracy of bioaccumulation factor prediction and benefiting environmental management and risk control. Attached Figure Description

[0018] Figure 1 This is one of the schematic diagrams of a machine learning-based PFAS biological enrichment factor prediction method provided in the embodiments of this application; Figure 2 This is a second schematic diagram of a PFAS biological enrichment factor prediction method based on machine learning provided in the embodiments of this application; Figure 3 These are (a) a feature importance ranking diagram and (b) a SHAP bee colony diagram from the SHAP analysis results of the enrichment factor prediction model provided in this application embodiment; Figure 4 This is a comparison chart of the prediction performance of (a) RF model, (b) XGBoost model and (c) SVM model constructed by different machine learning methods provided in the embodiments of this application; Figure 5 These are (a) a comparison chart of predicted and measured values ​​and (b) a chart of prediction residual analysis of the bioaccumulation factor prediction model provided in the embodiments of this application; Figure 6 This is a schematic diagram of the structure of a PFAS bioaccumulation factor prediction device based on machine learning provided in an embodiment of this application. Detailed Implementation

[0019] In the specification and claims of this invention, the terms "first" and "second," etc., are used to distinguish different objects, rather than to describe a specific order of objects.

[0020] In the embodiments of this application, "and / or" indicates a relationship between objects. For example, A and / or B can represent the following three situations: A exists alone, B exists alone, and A and B exist simultaneously.

[0021] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0022] In the description of this invention, unless otherwise stated, "multiple" means two or more. For example, multiple candidate aquatic environments refers to two or more candidate aquatic environments.

[0023] The methods and apparatus provided in this application relate to environmental risk assessment and can be used to predict the bioenrichment factors of target PFAS using a bioenrichment factor prediction model constructed through machine learning methods.

[0024] Understandably, perfluorinated / polyfluorinated alkyl substances (PFAS) are a class of man-made organic pollutants with environmental persistence, mobility, and potential bioaccumulation. Due to the presence of stable carbon-fluorine bonds in their molecular structure, PFAS are difficult to degrade in the natural environment and can persist in water bodies, sediments, and aquatic organisms for extended periods. They may also migrate and accumulate through the food chain, posing potential risks to aquatic ecosystems and human health.

[0025] Refining wastewater and related industrial wastewater are significant sources of emerging pollutants in the aquatic environment. Certain refining processes, the use of fluorinated additives, industrial cleaning, and wastewater treatment may lead to the entry of PFAS into wastewater treatment systems and receiving water bodies. Due to the complex composition of refining wastewater, environmental conditions such as organic matter, salinity, dissolved oxygen, and pH in receiving water bodies may change, thus affecting the bioaccumulation behavior of PFAS in aquatic organisms. Therefore, predictive analysis of the bioaccumulation risk of PFAS in receiving water bodies under the influence of refining wastewater discharge is of significant importance for environmental management and risk prevention.

[0026] Bioaccumulation factor (BAF) is an important indicator characterizing the accumulation of pollutants in organisms through the combined effects of water, food intake, and other exposure pathways in the real environment. Compared with bioconcentration factor (BCF), which mainly reflects the enrichment process under single-phase aqueous exposure conditions in the laboratory, BAF better reflects the bioaccumulation level of pollutants under multi-pathway exposure in real aquatic environments. Therefore, accurate prediction of BAF for PFAS is of great significance for ecological risk assessment, water quality benchmarking, priority pollutant screening, and environmental management.

[0027] However, existing bioaccumulation (BAF) data for PFAS mainly come from literature reports and field monitoring. Significant differences exist between different studies in terms of target compounds, test species, sampling areas, sampling times, tissue types, and environmental conditions, resulting in strong data heterogeneity. Furthermore, the bioaccumulation behavior of PFAS is not only influenced by molecular structural characteristics but is also closely related to environmental factors such as water temperature, pH, and dissolved oxygen, as well as species factors such as species classification and trophic level.

[0028] Furthermore, in typical pollution discharge scenarios such as fluorinated chemical and refining wastewater, the environmental occurrence characteristics and bioaccumulation risks of PFAS exhibit strong regional differences and environmental dependence, further increasing the complexity of BAF prediction and risk identification.

[0029] Existing BAF prediction methods mostly rely on single physicochemical properties, empirical formulas, or finite variables for estimation, making it difficult to fully integrate multi-source information such as pollutant molecular structure, aquatic environmental conditions, and biological characteristics. In recent years, machine learning methods have demonstrated strong nonlinear fitting capabilities in the simulation of complex environmental processes and can be used to characterize the multi-factor coupling relationships in the bioaccumulation process of PFAS.

[0030] To address the shortcomings of existing PFAS bioaccumulation factor prediction methods, which often rely on single physicochemical properties, empirical formulas, or finite variables, resulting in inaccurate predictions and hindering environmental management and risk control, this application provides a machine learning-based method and apparatus for predicting PFAS bioaccumulation factors. First, the molecular structure information of the target PFAS is acquired. Then, a bioaccumulation factor prediction model is used to analyze and predict the molecular structure information of the target PFAS, obtaining the bioaccumulation factors of the target PFAS in various candidate water environments and for various organisms. In this process, the bioaccumulation factor prediction model is constructed based on an enrichment feature dataset and machine learning methods. The enrichment feature dataset contains enrichment factor data for various PFAS, as well as corresponding environmental, biological, and spatial characteristics. This allows the bioaccumulation factor prediction model to combine the influence relationships between enrichment factor data and environmental, biological, and spatial characteristics to predict bioaccumulation factors, thereby improving the prediction accuracy and benefiting environmental management and risk control.

[0031] For example, the PFAS biological enrichment factor prediction method based on machine learning provided in this embodiment of the invention can be executed by an electronic device with processing capabilities, such as a computer or server. Taking a computer as an example, the hardware components of the computer may include: a processor, memory, a network interface, a user interface, a communication bus, etc.

[0032] The processor controls the electronic equipment to perform related processing and computation tasks, such as acquiring molecular structure information of the target PFAS and predicting the bioaccumulation factors of the target PFAS in various candidate aquatic environments. The processor may include a central processing unit (CPU), an AI processing unit (such as a GPU, TPU, NPU, LPU, etc.), or other processors. The processor can be single-core or multi-core; for example, the processor may include multiple CPUs or multiple GPUs.

[0033] Memory is used to store computer instructions and related data, such as molecular structure information of target PFAS, bioaccumulation factor prediction models, enrichment feature datasets, and bioaccumulation factors. Memory can be random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical storage, disk storage media, or other magnetic storage devices, or any other medium capable of storing program code or data accessible by a computer. Optionally, memory can be integrated into the processor, or it can be independent of the processor.

[0034] A network interface is used for communication between a computer and other devices or communication networks. A network interface can be a transceiver with transmit and receive capabilities. Optionally, a network interface may include standard wired interfaces or wireless interfaces (such as Wi-Fi interfaces, Bluetooth interfaces, and 5G interfaces).

[0035] The communication bus is used to enable communication between different components. For example, the processor, memory, network interface and user interface mentioned above can be interconnected through the communication bus.

[0036] The user interface may include a display screen and an input unit (such as a keyboard). Optionally, the user interface may also include a standard wired interface or a wireless interface.

[0037] Those skilled in the art will understand that the computer described above may include more or fewer components, or combine certain components, or have different component arrangements; the embodiments of this application do not limit this.

[0038] like Figure 1 As shown in the embodiments of this application, a PFAS biological enrichment factor prediction method based on machine learning includes S101-S102.

[0039] S101. Obtain the molecular structure information of the target PFAS.

[0040] In this embodiment, the target PFAS can be PFOS or PFOA, etc. The molecular structure information can include chemical name, CAS number, and molecular descriptor. The CAS number (CAS Registry Number) is a unique numerical identifier assigned to a chemical substance and is existing technology. The molecular descriptor can include molecular weight, lipid-water partition descriptor, topological polar surface area, number of hydrogen bond donors, number of hydrogen bond acceptors, number of rotatable bonds, number of rings, number of heavy atoms, and charge-related descriptor.

[0041] S102. Based on the molecular structure information of the target PFAS and the bioaccumulation factor prediction model, predict the bioaccumulation factors of the target PFAS in various organisms under various aquatic environments.

[0042] In this embodiment of the application, the above-mentioned biological enrichment factor prediction model is constructed using enrichment feature datasets and machine learning methods.

[0043] The aforementioned enrichment feature dataset includes multiple sample data sets. Each sample data set includes enrichment factor data for a specific PFAS, along with corresponding environmental, biological, and spatial characteristics. The enrichment factor data includes the bioaccumulation factor of the aforementioned PFAS for each organism in various aquatic environments. The bioaccumulation factor is the ratio of the PFAS concentration in the organism to the PFAS concentration in the aquatic environment. The organisms mentioned can be freshwater aquatic organisms. Freshwater aquatic organisms can include aquatic plants (such as emergent plants, floating-leaved plants, floating vegetation, and submerged plants), aquatic invertebrates (such as zooplankton and benthic animals), fish (such as cyprinid fish), amphibians (such as giant salamanders, oriental newts, and frogs), reptiles (such as Chinese pond turtles), and mammals (such as Yangtze finless porpoises), etc., but this embodiment does not limit the species.

[0044] For example, the aforementioned environmental characteristics may include environmental parameters and PFAS concentrations in the aquatic environment. The aforementioned environmental parameters may include water temperature, pH value, dissolved oxygen, and salinity. The aforementioned spatial characteristics may include the longitude and latitude of the sampling location and the type of water body. The aforementioned biological characteristics may include species classification, biological group (such as prokaryotes, protists, fungi, plants, invertebrates, vertebrates, etc.), trophic level (such as level 1 - producers, level 2 - primary consumers, level 3 - secondary consumers, level 4 - tertiary consumers, level 5 - apex consumers, etc.), tissue type, and PFAS concentration in the organism. Regarding the aforementioned tissue types, taking animals as an example, animal tissue types include epithelial tissue, connective tissue, muscle tissue, and nervous tissue; taking plants as an example, plant tissue types include meristematic tissue, parenchyma tissue, protective tissue, mechanical tissue, and vascular tissue. The aforementioned PFAS concentration in the organism may be PFOS concentration or PFOA concentration, etc.

[0045] In one implementation, combined with Figure 1 ,like Figure 2 As shown, S102 includes S1021-S1022.

[0046] S1021. Obtain the key features of each candidate water environment in multiple candidate water environments corresponding to the target PFAS.

[0047] Optionally, the aforementioned candidate aquatic environments may include refining wastewater and reservoirs / lakes, etc. The aforementioned key features may include key environmental features, key biological features, and key spatial features. Specifically, the key environmental features may include the feature with the greatest contribution among environmental features, the key biological features may include the feature with the greatest contribution among biological features, and the key spatial features may include the feature with the greatest contribution among spatial features.

[0048] The process of determining the aforementioned key features may include: using a feature contribution analysis method to analyze the contribution of the input features of the bioaccumulation factor prediction model to the prediction results, and identifying the input features with the largest contribution as key features.

[0049] In some embodiments, the specific determination process of the above-mentioned key features includes the following steps 1-3.

[0050] Step 1: Use the feature contribution analysis method to calculate the contribution value of each input feature to the BAF prediction result of the bioaccumulation factor prediction model, and use it as the marginal contribution of each input feature to the model output.

[0051] Preferably, the above-mentioned feature contribution analysis method can be the SHAP method based on Shapley values, and the analysis results obtained are as follows: Figure 3 As shown.

[0052] It should be noted that, since the above-mentioned SHAP method based on Shapley values ​​is existing technology, the process of determining the above contribution values ​​will not be described in detail in the embodiments of this application.

[0053] Step 2: Summarize the feature contribution values ​​of each input feature in each sample data of the above enrichment feature dataset, calculate the average absolute contribution of each input feature, and obtain the feature importance ranking of the PFAS biological enrichment factor prediction model.

[0054] Step 3: Based on the ranking of feature importance, identify key environmental features, key biological features, and key spatial features.

[0055] The aforementioned key environmental features are the features that contribute the most to the environmental features, the aforementioned key biological features are the features that contribute the most to the biological features, and the aforementioned key spatial features are the features that contribute the most to the spatial features.

[0056] Understandably, the above model interpretation results may also include feature importance ranking plots, feature contribution direction plots, and lists of key driving factors, in order to support PFAS bioaccumulation mechanism analysis, ecological risk assessment, and subsequent environmental management.

[0057] S1022. For each candidate water environment among multiple candidate water environments, the molecular structure information of the target PFAS and the key features of the candidate water environment are input into the bioaccumulation factor prediction model. The bioaccumulation factor prediction model outputs the bioaccumulation factor of the target PFAS for each of the multiple organisms in the candidate water environment.

[0058] Therefore, the input features of the above bioaccumulation factor prediction model include the molecular structure information of the target PFAS and the key features of the candidate aquatic environment. The prediction results include the bioaccumulation factor of each organism in the candidate aquatic environment for the target PFAS (such as logBAF, which can be obtained by antilogarithmic transformation to obtain the BAF prediction value at the original scale).

[0059] The following describes the construction process of the above-mentioned biological enrichment factor prediction model in an application scenario, including the following steps 1-2.

[0060] Step 1: Construct an enriched feature dataset.

[0061] Specifically, the process of constructing the enriched feature dataset includes the following steps 1.1-1.10.

[0062] Step 1.1: Search for bioaccumulation factor data of PFAS in freshwater aquatic organisms through literature databases such as Web of Science, PubMed, ScienceDirect, and Google Scholar, and screen for research literature or monitoring data that contain effective BAF, water concentration, organism concentration, species name, sampling location or sampling time information.

[0063] Step 1.2: Extract and record the chemical substance name, CAS number, SMILES expression, species name, taxonomic unit, biological group, tissue type, sampling time, longitude and latitude of sampling location, water concentration, organism concentration and BAF value in each data obtained in Step 1.1 above, and construct the original database of PFAS bioaccumulation factors.

[0064] Step 1.3: Perform quality control and standardization on the original database, including unifying BAF units, water concentration and organism concentration units, unifying species scientific names, removing data lacking necessary information, processing duplicate records, and removing obvious outliers, and performing logarithmic transformation on the BAF values ​​to obtain the target variable logBAF.

[0065] Step 1.4: For data that do not directly report BAF but report both water concentration and organism concentration, calculate BAF using formula (1) to supplement BAF data that can be used for model training.

[0066] (1) Among them, C bio C represents the concentration of pollutants in an organism. water This refers to the concentration of pollutants in the sampled water.

[0067] Step 1.5: Calculate the molecular structure descriptor based on the SMILES expression or molecular structure information of PFAS. The molecular descriptor includes, but is not limited to, MolWt, MolLogP, TPSA, NumHDonors, NumHAcceptors, NumRotatableBonds, RingCount, HeavyAtomCount, etc.

[0068] Step 1.6: Based on the sampling latitude and longitude and sampling time corresponding to each BAF data point, obtain multi-source environmental parameter data for the corresponding area. These environmental parameters include water temperature, pH, dissolved oxygen, and salinity. The environmental parameter data can be sourced from water environment monitoring databases, watershed water quality monitoring data, public environmental databases, or on-site monitoring data from typical industrial discharge-affected areas such as refining wastewater receiving bodies.

[0069] Step 1.7: For data with clear sampling latitude and longitude, set the spatial matching range with the sampling point as the center, and extract candidate environmental data within the preset distance threshold from the environmental parameter database. The spatial distance threshold can be set according to the environmental parameter type and sample coverage.

[0070] Step 1.8: For data with a clear sampling time, match environmental parameters according to the sampling date, sampling month, sampling season, or sampling year; for data without a specific sampling date but with a sampling year, use the annual average as the representative environmental parameter.

[0071] Step 1.9: When multiple candidate environmental parameter values ​​exist for the same sampling point within a preset spatial and temporal window, the nearest neighbor matching and spatiotemporal weighting method is used to determine the representative environmental parameter.

[0072] Step 1.10: Merge environmental parameters with BAF data, PFAS molecular descriptors, biological features, and spatial features to obtain an enriched feature dataset.

[0073] The above-mentioned enriched feature dataset includes: (1) target variable: logBAF; (2) molecular features: PFAS molecular structure descriptor; (3) environmental features: water temperature, pH, dissolved oxygen, environmental concentration, etc.; (4) biological features: species classification, biological group, trophic level, tissue type, etc.; (5) spatial features: longitude, latitude, water body type.

[0074] Step 2: Construct a biological enrichment factor prediction model using enriched feature datasets and machine learning algorithms.

[0075] Specifically, the construction process of the above-mentioned bioaccumulation factor prediction model may include the following steps 2.1-2.5.

[0076] Step 2.1: Perform data preprocessing on the above-mentioned enriched feature dataset.

[0077] The data preprocessing process described above can be as follows: For continuous variables, categorical variables, and missing values ​​in the enriched feature dataset, standardization is used; for categorical variables, one-hot encoding is used for transformation; and for missing values, deletion is performed. A consistency check is then performed on the target variable logBAF and the input features, removing data records lacking the target variable or key features, and constructing the feature matrix X and the target variable vector y for model training.

[0078] Step 2.2: Divide the preprocessed enriched feature dataset into a training set and a test set.

[0079] For example, the training and test sets are divided in an 8:2 ratio, and the training effect of the model is validated using five-fold cross-validation.

[0080] Step 2.3: Construct a base model using machine learning methods.

[0081] The first basis model is constructed using the gradient boosting tree model (XGBoost), the second basis model is constructed using the random forest model (RF), and the third basis model is constructed using the support vector machine model (SVM).

[0082] Step 2.4: Train the base model using the training set to obtain the trained base model.

[0083] The first basis model, the second basis model, and the third basis model were trained using the training set, respectively, to obtain the trained first basis model, the second basis model, and the third basis model.

[0084] Specifically, logBAF is used as the output target of the model, and multi-source fusion features (i.e., molecular features, environmental features, biological features, and spatial features) are used as the input variables to train the model. The optimal parameters are determined by adjusting the parameters to optimize the model performance.

[0085] Step 2.5: Validate the trained first basis model, second basis model, and third basis model using the test set, and calculate the coefficient of determination R between the model predictions and the measured values. 2 Root Mean Square Error (RMSE) is an evaluation index.

[0086] The predictive performance of the three basis models mentioned above is as follows: Figure 4 As shown, by Figure 4 It can be seen that the first basis model constructed by XGBoost (corresponding to Figure 4 The XGBoost model (among others) showed the best prediction performance, therefore XGBoost is the preferred machine learning method for predicting bioaccumulation factors, and the first basis model is used as the bioaccumulation factor prediction model. The prediction results of the above bioaccumulation factor prediction models are as follows: Figure 5 As shown, by Figure 5 It can be seen that the above-mentioned bioaccumulation factor prediction model can accurately predict BAF.

[0087] In summary, the PFAS bioaccumulation factor prediction method based on machine learning provided in this application first obtains the molecular structure information of the target PFAS, and then uses a bioaccumulation factor prediction model to analyze and predict the molecular structure information of the target PFAS, thereby obtaining the bioaccumulation factors of the target PFAS in various candidate water environments and various organisms. In the above process, the bioaccumulation factor prediction model is constructed based on an enrichment feature dataset and machine learning methods. The enrichment feature dataset contains enrichment factor data of various PFAS, as well as the corresponding environmental, biological, and spatial characteristics of the enrichment factor data. This allows the bioaccumulation factor prediction model to combine the influence relationship between enrichment factor data and environmental, biological, and spatial characteristics to predict bioaccumulation factors, thereby improving the prediction accuracy of bioaccumulation factors and thus benefiting environmental management and risk prevention.

[0088] Accordingly, embodiments of this application provide a PFAS bioaccumulation factor prediction device based on machine learning, such as... Figure 6 As shown, it includes an acquisition module 501 and a prediction module 502.

[0089] The acquisition module 501 is used to acquire the molecular structure information of the target PFAS; the molecular structure information includes the chemical name, CAS number, and molecular descriptor. For example, the acquisition module 501 is used to implement S101 of the above method.

[0090] The prediction module 502 is used to predict the bioaccumulation factors of the target PFAS in various organisms under various aquatic environments, based on the molecular structure information of the target PFAS and the bioaccumulation factor prediction model. The bioaccumulation factor prediction model is constructed using an enrichment feature dataset and machine learning methods. The enrichment feature dataset includes enrichment factor data for various PFAS, as well as the corresponding environmental, biological, and spatial characteristics. The enrichment factor data includes the bioaccumulation factors of various PFAS in various organisms under various aquatic environments. The bioaccumulation factor is the ratio of the PFAS concentration in the organism to the PFAS concentration in the aquatic environment. For example, the prediction module 502 is used to implement step S102 of the above method.

[0091] Optionally, the prediction module 502 is specifically used to: acquire the key features of each candidate water environment in multiple candidate water environments corresponding to the target PFAS. For each candidate water environment in the multiple candidate water environments, the molecular structure information of the target PFAS and the key features of the candidate water environment are input into the bioaccumulation factor prediction model, and the bioaccumulation factor prediction model outputs the bioaccumulation factor of each organism in the multiple organisms in the candidate water environment for the target PFAS. For example, the prediction module 502 is specifically used to implement S1021-S1022 of the above method.

[0092] The modules of the aforementioned machine learning-based PFAS biological enrichment factor prediction device can also be used to perform other steps in the above method embodiments. All relevant content involved in the above method embodiments can be referred to in the functional description of the corresponding functional module, and will not be repeated here.

[0093] This application also provides an electronic device, including: a processor and a memory coupled to the processor; the memory is used to store computer instructions, and when the electronic device is running, the processor executes the computer instructions stored in the memory to cause the electronic device to perform the methods in the above embodiments. The processor can implement the acquisition module 501 and the prediction module 502 described above; the memory can also be used to store molecular structure information of the target PFAS, a biological enrichment factor prediction model, an enrichment feature dataset, and biological enrichment factors, etc.

[0094] This application also provides a computer-readable storage medium including a computer program that, when run on a computer, performs the methods described in the above embodiments.

[0095] This application also provides a computer program product, which includes computer program instructions that, when run on a computer, execute the methods described in the above embodiments.

[0096] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A machine learning-based PFAS bioconcentration factor prediction method, characterized by, include: Obtain the molecular structure information of the target PFAS; the molecular structure information includes the chemical name, CAS number, and molecular descriptor; Based on the molecular structure information of the target PFAS and the bioaccumulation factor prediction model, the bioaccumulation factors of the target PFAS in various candidate water environments of various organisms are predicted; wherein, the bioaccumulation factor prediction model is constructed using an enrichment feature dataset and machine learning methods; the enrichment feature dataset includes multiple sample data, and the sample data includes the enrichment factor data of one PFAS and the corresponding environmental features, biological features and spatial features of the enrichment factor data; The enrichment factor data includes the bioaccumulation factors of PFAS in various organisms under various aquatic environments; the bioaccumulation factor is the ratio of the PFAS concentration in the organism to the PFAS concentration in the aquatic environment.

2. The method of claim 1, wherein, The method involves predicting the bioaccumulation factors of the target PFAS in various organisms under multiple candidate aquatic environments based on the molecular structure information and bioaccumulation factor prediction model, including: Obtain the key features of each candidate water environment in a variety of candidate water environments corresponding to the target PFAS; For each of the multiple candidate aquatic environments, the molecular structure information of the target PFAS and the key features of the candidate aquatic environment are input into the bioaccumulation factor prediction model. The bioaccumulation factor prediction model outputs the bioaccumulation factor of each organism in the multiple organisms in the candidate aquatic environment for the target PFAS.

3. The method as described in claim 1 or 2, characterized in that, The process of determining the key features includes: The contribution of the input features of the bioaccumulation factor prediction model to the prediction results is analyzed using the feature contribution analysis method, and the input feature with the largest contribution is identified as the key feature.

4. The method as described in claim 3, characterized in that, The key features include key environmental features, key biological features, and key spatial features; the key environmental features include the feature with the greatest contribution among the environmental features, the key biological features include the feature with the greatest contribution among the biological features, and the key spatial features include the feature with the greatest contribution among the spatial features.

5. The method of claim 1, wherein, The method further includes: The environmental characteristics include environmental parameters and PFAS concentration in the water body; the environmental parameters include water temperature, pH value, dissolved oxygen, and salinity; the spatial characteristics include the longitude and latitude of the sampling location and the water body type; the biological characteristics include species classification, biological group, trophic level, tissue type, and PFAS concentration in the organism; the organism is a freshwater aquatic organism.

6. The method as described in claim 1, characterized in that, The molecular descriptors include molecular weight, lipid-water partition-related descriptors, topological polar surface area, number of hydrogen bond donors, number of hydrogen bond acceptors, number of rotatable bonds, number of rings, number of heavy atoms, and charge-related descriptors.

7. The method as described in claim 1, characterized in that, The machine learning method mentioned is a gradient boosting tree model, a random forest model, or a support vector machine model.

8. A machine learning based PFAS bioconcentration factor prediction apparatus, characterized by, Includes an acquisition module and a prediction module; The acquisition module is used to acquire the molecular structure information of the target PFAS; the molecular structure information includes the chemical name, CAS number, and molecular descriptor; The prediction module is used to predict the bioaccumulation factors of the target PFAS in various organisms under multiple candidate aquatic environments, based on the molecular structure information of the target PFAS and a bioaccumulation factor prediction model. The bioaccumulation factor prediction model is constructed using an enrichment feature dataset and machine learning methods. The enrichment feature dataset includes multiple sample data, each including enrichment factor data for a PFAS and corresponding environmental, biological, and spatial characteristics. The enrichment factor data includes the bioaccumulation factors of the PFAS in various organisms under multiple aquatic environments. The bioaccumulation factor is the ratio of the PFAS concentration in the organism to the PFAS concentration in the aquatic environment.

9. An electronic device, comprising: The device includes a processor and a memory coupled to the processor; the memory is used to store computer instructions, which, when the electronic device is running, are executed by the processor to cause the electronic device to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It includes computer program instructions that, when executed by a computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images