A method for predicting the adsorption properties of a metal organic framework

Abstract

The application discloses a kind of metal organic framework adsorption performance prediction method, belong to water treatment technical field;Including the following steps: obtaining the physical property parameters of metal organic framework, organic target pollutant structure data, adsorption condition parameter and adsorption capacity label value;Three kinds of molecular fingerprint encoding strategies are used to encode pollutant structure data, obtain three groups of fingerprint characteristic vectors;Physical property parameters and adsorption condition parameters are combined with each group of fingerprint vector respectively, to construct three overall inputs;Multiple machine learning models are trained respectively using three overall inputs, to determine the optimal fingerprint encoding strategy and the optimal prediction model;After the optimal fingerprint encoding of the data to be measured, it is combined with physical property parameters and adsorption condition parameters, input into the optimal prediction model, to obtain the adsorption capacity prediction value;The present application has high prediction accuracy, and is suitable for adsorption performance prediction, adsorbent screening and process parameter optimization in water treatment.

Description

Technical Field

[0001] This invention belongs to the field of water treatment technology, and specifically relates to a method for predicting the adsorption performance of metal-organic frameworks. Background Technology

[0002] Metal-organic frameworks (MOFs) have broad application prospects in gas adsorption and separation due to their high specific surface area, tunable pore size, and rich chemical functionalities. With the rapid increase in the number of MOFs, traditional experimental and computational methods are inefficient and costly. Machine learning methods, due to their efficient predictive capabilities, have been introduced into MOF adsorption performance prediction research.

[0003] In existing technologies, machine learning-based methods for predicting the adsorption performance of MOFs typically transform the molecular structure information of MOFs into numerical feature descriptors through specific molecular characterization methods. Machine learning models are then built based on these features to establish a mapping relationship between structure and adsorption performance, enabling rapid prediction of the adsorption performance of new MOF materials. However, different molecular characterization methods exhibit significant differences in their ability to express MOF structural information, and their applicability varies with the target adsorption system. Existing methods often build models based on only one or a few characterization methods, lacking systematic validation of the applicability of different characterization methods. This results in insufficient extraction and expression of molecular structural information, limiting the accuracy and generalization ability of the models in predicting adsorption performance. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting the adsorption performance of metal-organic frameworks, which has high prediction accuracy and is applicable to adsorption performance prediction, adsorbent screening and process parameter optimization in water treatment.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for predicting the adsorption performance of metal-organic frameworks, comprising the following steps: To obtain the physical properties of metal-organic frameworks, the chemical structure data of target organic pollutants, the experimental conditions of the adsorption process, and the known adsorption capacity label values; The chemical structure data of organic target pollutants are encoded using multiple molecular fingerprinting strategies to generate molecular fingerprint feature vectors corresponding to each molecular fingerprinting strategy. The physical properties of metal-organic frameworks and the experimental conditions of the adsorption process are used as common features and combined with the feature vectors of each molecular fingerprint to construct the overall input under each molecular fingerprint encoding strategy. Multiple machine learning models were trained using the overall input under each molecular fingerprint encoding strategy and combined with the adsorption capacity label value to obtain multiple candidate prediction models. The prediction performance of each candidate prediction model was compared and evaluated to determine the optimal molecular fingerprint encoding strategy and the optimal prediction model. The chemical structure data of the target organic pollutant to be predicted are encoded using the optimal molecular fingerprint encoding strategy to obtain the target molecular fingerprint feature vector. The target molecular fingerprint feature vector is combined with the physical property parameters of the metal-organic framework to be predicted and the experimental condition parameters of the adsorption process to be predicted, and then input into the optimal prediction model to obtain the predicted adsorption capacity value.

[0006] In some optional embodiments, the multiple molecular fingerprinting strategies include at least three, namely MACCS fingerprinting, binary Morgan fingerprinting, and counting Morgan fingerprinting; the optimal molecular fingerprinting strategy is binary Morgan fingerprinting.

[0007] In some optional embodiments, the plurality of machine learning models include CatBoost, LightGBM, XGBoost, random forest, support vector regression, backpropagation neural network, K-nearest neighbors, and linear regression; the optimal prediction model is the CatBoost model.

[0008] In some optional embodiments, the specific steps for comparing and evaluating the predictive performance of each candidate prediction model are as follows: The coefficient of determination, root mean square error, mean absolute error, and mean absolute percentage error were used as evaluation indicators to evaluate each candidate prediction model. The higher the coefficient of determination and the lower the root mean square error, mean absolute error, and mean absolute percentage error, the better the evaluation criteria were to determine the optimal molecular fingerprint coding strategy and the optimal prediction model.

[0009] In some optional embodiments, the physical properties of the metal-organic framework specifically include: Metal node type, organic ligand type, molar ratio of metal node to organic ligand, synthesis temperature, synthesis time, specific surface area, pore volume and pore diameter.

[0010] In some optional embodiments, the experimental conditions parameters for the adsorption process specifically include: Solution pH, water matrix type, adsorption time, adsorption temperature, stirring rate, amount of metal-organic framework, initial concentration of organic target pollutant, and solution volume.

[0011] In some optional embodiments, the organic target pollutant is a pharmaceutical product or personal care product, and the chemical structure data is a simplified molecular linear input canonical string corresponding to the organic target pollutant; the molecular fingerprint encoding is generated by processing the simplified molecular linear input canonical string based on an open-source cheminformatics toolkit.

[0012] In some optional embodiments, before combining the physical properties of the metal-organic framework and the experimental conditions of the adsorption process as common features with the feature vectors of each molecular fingerprint, a data preprocessing step is included for the physical properties of the metal-organic framework and the experimental conditions of the adsorption process. The specific steps are as follows: Missing values ​​were imputed and standardized for numerical variables in physical property parameters and experimental condition parameters, and categorical variables in physical property parameters and experimental condition parameters were encoded and converted.

[0013] In some optional embodiments, after determining the optimal molecular fingerprint encoding strategy and the optimal prediction model, the method further includes: The SHAP method is used to obtain the SHAP value of each input feature on the prediction result of the optimal prediction model. Based on the SHAP value, the magnitude and direction of the contribution of each input feature to the adsorption capacity are determined.

[0014] In some optional embodiments, after employing the SHAP method, the method further includes: The cumulative local effect diagram method was used to analyze the marginal effect of each input feature on the adsorption capacity, and to determine the recommended range of values ​​for each experimental condition parameter and physical property parameter that are conducive to improving the adsorption capacity.

[0015] The method for predicting the adsorption performance of metal-organic frameworks provided by this invention has at least the following beneficial effects: This invention utilizes the physical properties of metal-organic frameworks and experimental conditions of the adsorption process as common features within the same feature system. These features are combined with molecular fingerprint feature vectors generated by various molecular fingerprinting strategies to construct multiple overall inputs. The combinations of "fingerprinting strategy-machine learning model" are then systematically trained and compared to identify the optimal fingerprinting strategy and prediction model. This allows for the informed determination of the most suitable molecular characterization method and model configuration for the target adsorption system. This addresses the limitations in prediction accuracy and general predictive capability in existing technologies due to a lack of adaptability verification for different molecular characterization methods and insufficient synergistic utilization of multi-source adsorption information. It improves the accuracy and generalization ability of adsorption capacity prediction, providing a reliable data-driven foundation for screening high-efficiency adsorbents and optimizing adsorption processes in water treatment.

[0016] This invention further combines the SHAP method and the cumulative local effect diagram method to perform interpretability analysis on the optimal prediction model. It can quantitatively reveal the magnitude and direction of the contribution of each input feature to the adsorption capacity, and provide recommended value ranges for each experimental condition parameter that are conducive to improving the adsorption capacity. It realizes the technical leap from "black box prediction" to "mechanism elucidation" and "condition optimization", and can be directly used to guide the screening of high-efficiency adsorbents and the optimization of adsorption process parameters in water treatment. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0018] Figure 1 This is a flowchart of a method for predicting the adsorption performance of metal-organic frameworks according to an embodiment of the present invention; Figure 2 This is a Pearson correlation heatmap between quantitative variables provided according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the prediction performance of various models under the MACCS encoding method provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the prediction performance of various models under the counting-type Morgan fingerprint coding method according to an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the prediction performance of various models under the binary Morgan fingerprint encoding method according to an embodiment of the present invention; Figure 6 The results are SHAP analysis of experimental conditions (A), metal-organic framework characteristics (B), and molecular structural characteristics (C) of organic target pollutants provided in an embodiment of the present invention. Figure 7 The results are the cumulative local effect (ALE) analysis of experimental conditions on QE during MOF adsorption of PPCPs according to an embodiment of the present invention.

[0019] Figure 8 This is the result of the cumulative local effect (ALE) analysis of physical property parameters on QE during the MOF adsorption of PPCPs according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0021] One embodiment of the present invention relates to a method for predicting the adsorption performance of metal-organic frameworks. The implementation details of the method for predicting the adsorption performance of metal-organic frameworks in this embodiment are described in detail below. The following implementation details are provided for ease of understanding and are not necessary for implementing this solution.

[0022] The specific process of the metal-organic framework adsorption performance prediction method in this embodiment can be as follows: Figure 1 As shown, it includes: Step 101: Obtain the physical property parameters of the metal-organic framework, the chemical structure data of the target organic pollutant, the experimental condition parameters of the adsorption process, and the known adsorption capacity label value. In this step, the physical properties of the metal-organic framework include at least one of the following: metal node type, organic ligand type, molar ratio of metal nodes to organic ligands, synthesis temperature, synthesis time, specific surface area, pore volume, and pore diameter. The experimental conditions for the adsorption process include at least one of the following: solution pH, aqueous matrix type, adsorption time, adsorption temperature, stirring rate, amount of metal-organic framework used, initial concentration of the target organic pollutant, and solution volume. The target organic pollutant is a pharmaceutical or personal care product, and its chemical structure data includes at least the corresponding simplified molecular linear input canonical string.

[0023] The data obtained in this step was specifically based on the Web of Science database. The system searched for keywords such as "PPCPs," "Antibiotics," "Adsorption," and "MOF," extracted data, and removed outliers, resulting in 7075 data points containing 113 metal-organic frameworks and 39 pollutants. Some data were obtained directly from literature, while others were extracted from images using Origin 2021. Input features covered three categories of parameters: metal-organic framework properties, operating parameters, and molecular characteristics of the target organic pollutant. The output variable was adsorption capacity. Simplified linear input canonical strings corresponding to the target organic pollutants were obtained from the PubChem website. Specific information for each feature is shown in Table 1.

[0024] Table 1 Model Feature Information Step 102: The chemical structure data of the organic target pollutant are encoded using multiple molecular fingerprint encoding strategies to generate molecular fingerprint feature vectors corresponding to each molecular fingerprint encoding strategy. In this step, at least three molecular fingerprinting strategies are employed: MACCS fingerprinting, binary Morgan fingerprinting, and counting Morgan fingerprinting. The molecular fingerprinting is generated by processing the simplified linear input canonical string of the molecule using an open-source cheminformatics toolkit. Specifically, the MACCS fingerprint is obtained using the MACCSkeys.GenMACCSKeys method; the binary Morgan fingerprint is obtained using the AllChem.GetMorganFingerprintAsBitVect method; and the counting Morgan fingerprint is obtained using the AllChem.GetHashedMorganFingerprint method. The fingerprint radii for the binary Morgan fingerprint and the counting Morgan fingerprint are 1 and 2, respectively, and the fingerprint lengths are 128, 256, 512, and 1024.

[0025] Before this step, the acquired raw data needs to be preprocessed. Specifically, the physical properties of the metal-organic framework and the experimental conditions of the adsorption process are preprocessed: missing value imputation and standardization are performed on the numerical variables in the physical properties and experimental conditions, and categorical variables in the physical properties and experimental conditions are encoded and converted.

[0026] The missing value imputation method was as follows: for solution pH and adsorption temperature, fixed values ​​of 7 and 25℃ were used for data interpolation; for features with missing values, such as stirring rate, pore volume, and pore diameter, the mean of the feature across all samples was used for imputation. Standardization was performed using Z-score standardization, independently applied to each numerical variable according to the formula Z=(X-μ) / σ, where X represents the original value of the variable, μ represents the mean of the variable in the dataset, and σ represents the standard deviation of the variable in the dataset. The encoding transformation of categorical variables (metal node M, organic ligand OL, and aqueous matrix type WT) used a one-heat encoding method.

[0027] After imputation of missing values, standardization, and encoding transformation, the Pearson correlation coefficient matrix was used to analyze the quantitative variables in the dataset. The analysis results showed a highly significant correlation between the specific surface area (SSA) and pore volume (Vp) of the metal-organic framework (MOF) (p < 0.01). Since SSA and Vp are important structural parameters of MOFs, no features were removed during model training. The Pearson correlation heatmap between the quantitative variables is shown below. Figure 2As shown in the figure, SSA and Vp exhibit a strong positive correlation, and no significant collinearity exceeding this order of magnitude was detected among the other variables.

[0028] Step 103: The physical properties of the metal-organic framework and the experimental conditions of the adsorption process are used as common features and combined with the feature vectors of each molecular fingerprint to construct the overall input under each molecular fingerprint encoding strategy. In this step, the physical properties of the metal-organic framework and the experimental conditions of the adsorption process are used as common features and combined with the feature vectors of each molecular fingerprint to construct the overall input under each molecular fingerprint encoding strategy.

[0029] Step 104: Using the overall input under each molecular fingerprint encoding strategy and combining it with the adsorption capacity label value, train multiple machine learning models to obtain multiple candidate prediction models; compare and evaluate the prediction performance of each candidate prediction model, and determine the optimal molecular fingerprint encoding strategy and the optimal prediction model. In this step, the multiple machine learning models include CatBoost, LightGBM, XGBoost, Random Forest, Support Vector Regression, Backpropagation Neural Network, K-Nearest Neighbors, and Linear Regression.

[0030] The dataset was randomly divided into training and test sets in an 8:2 ratio. During the training phase, five-fold cross-validation was used to evaluate model stability, and Bayesian optimization was combined to optimize the hyperparameters of each model. Finally, the generalization performance of the model was evaluated using an independent test set.

[0031] Model performance was evaluated using the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Higher R², lower RMSE, and lower MAE were used as evaluation criteria to determine the optimal molecular fingerprint encoding strategy and the optimal prediction model. The formulas for calculating the evaluation metrics are as follows:

[0032] ; ; ; ; Where n represents the total number of data points in the dataset, and ŷᵢ and yᵢ are the best predicted value and the true value of the i-th data point, respectively. This represents the average of all true values.

[0033] The prediction performance diagram of each model under the MACCS encoding method is shown below. Figure 3As shown, under the MACCS fingerprint encoding strategy, the CatBoost model performs best, with a test set R² of 0.9703, RMSE of 29.8563, MAE of 15.0551, and MAPE of 15.3676. The prediction performance of each model under the counting-based Morgan fingerprint encoding method is illustrated in the diagram below. Figure 4 As shown, the CatBoost model achieves better results when the fingerprint radius is 1 and the fingerprint length is 512, with a test set R² of 0.9711, RMSE of 29.4737, MAE of 14.2224, and MAPE of 14.9380. A schematic diagram illustrating the prediction performance of each model under the binary Morgan fingerprint encoding scheme is shown below. Figure 5 As shown, the CatBoost model achieves the global optimum when the fingerprint radius is 1 and the fingerprint length is 512, with R² of 0.9723, RMSE of 28.8352, MAE of 15.0873, and MAPE of 18.0469 on the test set.

[0034] Based on the prediction results of the three types of molecular fingerprints, the combination of binary Morgan fingerprint and CatBoost model shows the best performance. The combination of binary Morgan fingerprint with radius 1 and length 512 with CatBoost model achieves a good balance between prediction accuracy and generalization ability, and is therefore selected as the optimal molecular fingerprint encoding strategy and optimal prediction model.

[0035] Step 105: The chemical structure data of the target organic pollutant to be predicted is encoded using the optimal molecular fingerprint encoding strategy to obtain the target molecular fingerprint feature vector; the target molecular fingerprint feature vector is combined with the physical property parameters of the metal-organic framework to be predicted and the experimental condition parameters of the adsorption process to be predicted, and input into the optimal prediction model to obtain the predicted adsorption capacity value.

[0036] In this embodiment, after determining the optimal molecular fingerprint encoding strategy and the optimal prediction model, a model interpretability analysis step is also included. Specifically:

[0037] The SHAP method was used to obtain the SHAP values ​​of each input feature for the prediction results of the optimal prediction model. Based on these SHAP values, the magnitude and direction of each input feature's contribution to the adsorption capacity were determined. The SHAP analysis results for experimental conditions (A), metal-organic framework features (B), and molecular structure features of the target organic pollutant (C) are shown below. Figure 6 As shown, the top 20 most important features determined by SHAP values ​​include 7 experimental parameters, 8 parameters related to metal-organic frameworks, and 5 molecular fingerprint sites related to organic target pollutants.

[0038] Among the main adsorption conditions, the initial concentration (IC) of the organic target pollutant has the largest average absolute SHAP value, which contributes most significantly to the predicted adsorption capacity. The SHAP value distribution shows a highly consistent monotonically positive correlation, indicating that the adsorption capacity increases significantly with increasing initial concentration. Increasing the amount of metal-organic framework (AMB) (AD) leads to a decrease in adsorption capacity, and excessive addition leads to a decrease in adsorption capacity per unit mass. High-value samples of adsorption time (Adt) are mostly located in the positive SHAP value region, indicating that extending the adsorption time helps to improve the adsorption capacity. The molar ratio (MO) of metal nodes to organic ligands is one of the material structure variables with a high contribution, and higher MO values ​​are mainly distributed in the positive SHAP value region. Pore diameter (Pd) and pore volume (Vp) also have a high contribution to the adsorption capacity, with larger Pd and Vp generally producing a positive SHAP contribution. In terms of molecular structure characteristics, the diarylamine structure characterized by fp_491, the Na⁺ structure characterized by fp_189, and the anionic oxygen associated with fp_203 generally significantly improve the model-predicted adsorption capacity, while the ether bond associated with fp_183 generally has a negative contribution.

[0039] Following the SHAP analysis, the cumulative local effects plot method was used to analyze the marginal effect of each input feature on the adsorption capacity, and to determine the recommended range of values ​​for each experimental condition parameter and physical property parameter that is conducive to improving the adsorption capacity. The recommended range of values ​​for each experimental condition parameter and physical property parameter is shown in Table 2.

[0040] Table 2 Recommended intervals given by cumulative local effects analysis Specifically, the recommended range for the initial concentration (IC) of the target organic pollutant is 76–600 mg / L; the recommended range for the amount of metal-organic framework (AD) is 1–8 mg; the recommended range for the adsorption temperature (Adtem) is 15–26 °C; the recommended range for the stirring rate (SR) is 188–225 rpm; the recommended range for the adsorption time (Adt) is 105–2880 min; the recommended range for the solution pH is 2.5–7.9; and the recommended range for the solution volume (SV) is 37–600 mL. Regarding MOF physical properties, the recommended range for the molar ratio of metal nodes to organic ligands (MO) is 3.664–64.849; the recommended range for synthesis temperature (Sytem) is 42.5–67 °C or 168–220 °C; the recommended range for synthesis time (Syt) is 2.1–17.7 h or 32–70 h; the recommended range for specific surface area (SSA) is 830–3398 m² / g; the recommended range for pore volume (Vp) is 0.286–2.25 cm³ / g; and the recommended range for pore diameter (Pd) is 3.047–28.67 nm.

[0041] The results of the analysis of the cumulative local effect (ALE) of experimental conditions on QE during MOF adsorption of PPCPs are as follows: Figure 7As shown in the figure, the marginal effect of various experimental parameters on adsorption capacity within different value ranges is illustrated. These parameters include the initial concentration of the target organic pollutant, the amount of metal-organic framework used, adsorption temperature, stirring rate, adsorption time, solution pH, and solution volume. The ALE method can reflect the local response of each experimental condition within its characteristic actual distribution range, identify the positive or negative influence trend of each parameter on adsorption capacity, and provide a basis for optimizing adsorption experimental conditions.

[0042] The results of the analysis of the cumulative local effect (ALE) of physical property parameters on QE during MOF adsorption of PPCPs are as follows: Figure 8 As shown in the figure, the marginal effects of various metal-organic framework (MOF) properties on adsorption capacity within different value ranges are illustrated, including the molar ratio of metal nodes to organic ligands, synthesis temperature, synthesis time, specific surface area, pore volume, and pore diameter. The ALE method can be used to identify the influence and nonlinear characteristics of each property on adsorption capacity, providing a reference for the synthesis and screening of high-adsorption-performance MOFs.

[0043] In this embodiment, the potential impact of ionic factors on adsorption performance in real aquatic environments has not been considered. In future studies, relevant ions can be included in the dataset to simulate real aquatic environments and further improve the model's predictive performance.

[0044] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the protection scope of this invention. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, without changing the core design of the algorithm and process, are also within the protection scope of this invention.

[0045] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.

[0046] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0047] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing the present invention, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A method for predicting the adsorption performance of metal-organic frameworks, characterized in that, The method includes: To obtain the physical properties of metal-organic frameworks, the chemical structure data of target organic pollutants, the experimental conditions of the adsorption process, and the known adsorption capacity label values; The chemical structure data of organic target pollutants are encoded using multiple molecular fingerprinting strategies to generate molecular fingerprint feature vectors corresponding to each molecular fingerprinting strategy. The physical properties of metal-organic frameworks and the experimental conditions of the adsorption process are used as common features and combined with the feature vectors of each molecular fingerprint to construct the overall input under each molecular fingerprint encoding strategy. Multiple machine learning models were trained using the overall input under each molecular fingerprint encoding strategy and combined with the adsorption capacity label value to obtain multiple candidate prediction models. The prediction performance of each candidate prediction model was compared and evaluated to determine the optimal molecular fingerprint encoding strategy and the optimal prediction model. The chemical structure data of the target organic pollutant to be predicted are encoded using the optimal molecular fingerprint encoding strategy to obtain the target molecular fingerprint feature vector. The target molecular fingerprint feature vector is combined with the physical property parameters of the metal-organic framework to be predicted and the experimental condition parameters of the adsorption process to be predicted, and then input into the optimal prediction model to obtain the predicted adsorption capacity value.

2. The method for predicting the adsorption performance of metal-organic frameworks as described in claim 1, characterized in that, The multiple molecular fingerprinting strategies include at least three types: MACCS fingerprinting, binary Morgan fingerprinting, and counting Morgan fingerprinting; the optimal molecular fingerprinting strategy is binary Morgan fingerprinting.

3. The method for predicting the adsorption performance of metal-organic frameworks as described in claim 1, characterized in that, The multiple machine learning models include CatBoost, LightGBM, XGBoost, Random Forest, Support Vector Regression, Backpropagation Neural Network, K-Nearest Neighbors, and Linear Regression; the optimal prediction model is CatBoost.

4. The method for predicting the adsorption performance of metal-organic frameworks as described in claim 1, characterized in that, The specific steps for comparing and evaluating the predictive performance of each candidate prediction model are as follows: The coefficient of determination, root mean square error, mean absolute error, and mean absolute percentage error were used as evaluation indicators to evaluate each candidate prediction model. The higher the coefficient of determination and the lower the root mean square error, mean absolute error, and mean absolute percentage error, the better the evaluation criteria were to determine the optimal molecular fingerprint coding strategy and the optimal prediction model.

5. The method for predicting the adsorption performance of metal-organic frameworks as described in claim 1, characterized in that, The physical properties of the metal-organic framework specifically include: Metal node type, organic ligand type, molar ratio of metal node to organic ligand, synthesis temperature, synthesis time, specific surface area, pore volume and pore diameter.

6. The method for predicting the adsorption performance of metal-organic frameworks as described in claim 1, characterized in that, The experimental conditions and parameters for the adsorption process specifically include: Solution pH, water matrix type, adsorption time, adsorption temperature, stirring rate, amount of metal-organic framework, initial concentration of organic target pollutant, and solution volume.

7. The method for predicting the adsorption performance of metal-organic frameworks as described in claim 1, characterized in that, The organic target pollutant is a pharmaceutical or personal care product, and the chemical structure data is a simplified molecular linear input canonical string corresponding to the organic target pollutant; the molecular fingerprint encoding is generated by processing the simplified molecular linear input canonical string based on an open-source cheminformatics toolkit.

8. The method for predicting the adsorption performance of metal-organic frameworks as described in claim 1, characterized in that, Before combining the physical properties of the metal-organic framework and the experimental conditions of the adsorption process as common features with the feature vectors of each molecular fingerprint, a data preprocessing step is included for the physical properties of the metal-organic framework and the experimental conditions of the adsorption process. The specific steps are as follows: Missing values ​​were imputed and standardized for numerical variables in physical property parameters and experimental condition parameters, and categorical variables in physical property parameters and experimental condition parameters were encoded and converted.

9. The method for predicting the adsorption performance of metal-organic frameworks as described in claim 1, characterized in that, After determining the optimal molecular fingerprint encoding strategy and the optimal prediction model, the following steps are also included: The SHAP method is used to obtain the SHAP value of each input feature on the prediction result of the optimal prediction model. Based on the SHAP value, the magnitude and direction of the contribution of each input feature to the adsorption capacity are determined.

10. The method for predicting the adsorption performance of metal-organic frameworks as described in claim 9, characterized in that, Following the adoption of the SHAP method, the following is also included: The cumulative local effect diagram method was used to analyze the marginal effect of each input feature on the adsorption capacity, and to determine the recommended range of values ​​for each experimental condition parameter and physical property parameter that are conducive to improving the adsorption capacity.

Images