Discrete metal coordination supramolecular cage structure prediction method based on artificial intelligence

Abstract

The invention provides a discrete metal coordination supramolecular cage structure prediction method based on artificial intelligence, and can be used in the technical field of material structure feature evaluation. The method comprises the following steps: acquiring a discrete metal coordination assembly topological structure data set; performing standardized preprocessing on the data set; performing multiple machine learning model training and evaluation on a training set in the data set, and selecting a model with optimal evaluation as a discrete metal coordination supramolecular cage structure prediction model; and inputting test set descriptor parameters in the data set into the discrete metal coordination supramolecular cage structure prediction model to obtain a discrete metal coordination supramolecular cage structure prediction result. According to the method provided by the invention, the assembly cage structure can be quickly predicted, and the prediction accuracy is relatively high; compared with a traditional chemical experiment, the method can save a large amount of experiment and time cost, and is beneficial to screening and application transformation processes of the supramolecular cage.

Description

Technical Field

[0001] This application relates to the field of material structure characteristic evaluation technology, and in particular to an artificial intelligence-based method for predicting discrete metal coordination supramolecular cage structures. Background Technology

[0002] Assembly structures have given rise to almost all living organisms and many substances. Inspired by nature, scientists have imitated and designed self-assembly systems, synthesizing more and more intricate supramolecular assemblies and endowing them with diverse functions. These carefully designed geometric structures have demonstrated excellent performance in functional applications such as molecular machines, sensors, gas adsorption, microreactors, chemical catalysis, and drug delivery. The key issue in supramolecular assembly design is to rationally grasp the assembly rules of the structure.

[0003] Assemblies typically consist of highly symmetrical and stable structures formed by repeating units. For simple structures, humans can predict them through experience and theory. However, as functional requirements increase, the factors involved become increasingly difficult to grasp accurately. Reverse design based on functional requirements to find the optimal structure is challenging. Many early studies relied on prior knowledge and experience and system fine-tuning to solve reverse design problems, involving numerous trial-and-error experiments and significant time and cost. For problems influenced by multiple factors, the relationship between each factor and the predicted structure is nonlinear, with varying weights that may change systematically. In supramolecular assemblies, subtle changes in nodes and ligands can cause significant structural changes. Currently, there are approximately one hundred topological structures in assemblies, making it a multi-objective optimization problem under nonlinear constraints. Artificial intelligence, as an effective tool, extends and amplifies human senses and abilities. Currently, AI technology has empowered various disciplines and achieved remarkable results. In recent years, machine learning and automation, as technologies in artificial intelligence, have attracted increasing attention from researchers. Machine learning is good at solving multi-factor problems, while automation can conduct high-throughput experiments, accelerating the extraction of effective information and replacing manual labor in a large amount of mechanical work. The combination of the two provides an effective method for the rational design of supramolecular cages.

[0004] How to leverage artificial intelligence technology to empower supramolecular assembly scientific research, solve multi-objective optimization problems under nonlinear constraints, and obtain robust validation has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides an artificial intelligence-based method for predicting the structure of discrete metal-coordinated supramolecular cages, in order to solve the problem that the orientation of existing discrete metal-coordinated supramolecular assemblies cannot be rationally designed.

[0006] According to the first aspect disclosed in this application, this application provides a method for predicting discrete metal coordination supramolecular cage structures based on artificial intelligence, including: A dataset of discrete metal coordination assembly topologies is obtained, and the dataset is divided into a training set and a test set; wherein, the dataset includes 26 constructed discrete metal coordination assembly topologies and descriptor parameters of the geometric and chemical features of the collected discrete metal coordination assembly topologies; The dataset is preprocessed using standardization. The training set in the dataset is used to train and evaluate various machine learning models, and the model with the best evaluation is selected as the discrete metal coordination supramolecular cage structure prediction model. Input the test set descriptor parameters from the dataset into the discrete metal coordination supramolecular cage structure prediction model to obtain the discrete metal coordination supramolecular cage structure prediction results.

[0007] In one feasible implementation, the AI-based discrete metal coordination supramolecular cage structure prediction method further includes: The prediction results are compared with the results of high-throughput experiments conducted using automated equipment to verify the accuracy of the model.

[0008] In one feasible implementation, the step of acquiring a discrete metal coordination assembly topology dataset and dividing the dataset into a training set and a test set includes: Twenty-six discrete metal coordination assembly topologies were constructed. The Python-based STK toolkit digitizes the locations and relationships of nodes and ligands in the topology of the assembly. Descriptor parameters for geometric and chemical features of 26 discrete metal coordination assemblies were collected to establish a dataset of discrete metal coordination assembly topologies. The descriptor parameters for geometric features include the number of homology groups, average duration, and maximum duration. The descriptor parameters for chemical features include atomic connection number, cell volume, charge density, molecular volume, temperature, ligand coordination number, and node coordination number.

[0009] In one feasible implementation, the construction of 26 discrete metal coordination assembly topologies specifically involves: The STK toolkit was used to program the topology graph, constructing 26 discrete metal coordination assembly topologies. These 26 discrete metal coordination assembly topologies include 11 planar structures, 4 spherical structures, 2 tetrahedral structures, 3 hexahedral structures, and 6 cylindrical structures. In one feasible implementation, the Python-based STK toolkit digitizes the node and ligand positions and relationships in the assembly topology; specifically including: Each topology is configured to include a class, and each class has vertex variables and edge variables; the vertex variables include node data and ligand data; the edge variables are the connection data between nodes and ligands. In one feasible implementation, the standardization preprocessing of the dataset specifically includes: performing dimensionality reduction, normalization, and visualization processing on the dataset. Obtain coordinate data of atoms in ligands: A three-dimensional conformation is generated using ligand smiles, and the coordinate data of ligand atoms is determined using the MMFF force field optimization method: X-axis coordinate vector data 1*n, Y-axis coordinate vector data n*1, and Z-axis coordinate vector data 1*n; Obtain the adjacency matrix of the ligand molecule: 1 indicates that there is a chemical bond between atoms; 0 indicates that there is no chemical bond between atoms, thus obtaining the adjacency matrix vector data n*n of the molecule; Obtain the adjacency-Euclidean distance matrix of the ligand molecule: Calculate the chemical bond length between atoms using Euclidean distance and insert it into the adjacency matrix to obtain an n*n adjacency-Euclidean distance matrix; the formula for calculating the chemical bond length is as follows:

[0010] in, Represents the first point Axis coordinates; Indicates the second point Axis coordinates; Represents the first point Axis coordinates; Indicates the second point Axis coordinates; Represents the first point Axis coordinates; Indicates the second point Axis coordinates; The atomic number vector data n*1 is obtained by using RDKit to obtain the ordinal numbers of all atoms in the ligand molecule. Information is integrated from all atoms in the ligand molecule to obtain ligand information; the calculation formula is as follows:

[0011] in, A represents atomic number vector data. express One atom Axis coordinate vector data; express One atom Axis coordinate vector data; express One atom Axis coordinate vector data; Represents the adjacency-Euclidean distance matrix; Represents the first in the matrix OK, Represents the first in the matrix List; The results are input into the PCA dimensionality reduction algorithm to obtain n*1 vector data, denoted as... ; Normalize the data: Substitute the above n*1 vector data into the min-max normalization formula to obtain n*1 vector data. The calculation formula is as follows:

[0012] Among them, X max X represents the maximum value of the sample. min This represents the minimum value of the sample. The values ​​are the normalized values. Visualizing ligand curves: Ligand curves are visualized on a two-dimensional coordinate system; where, The axis represents the number of atoms in the molecule. The axis represents the normalized curve values. In one feasible implementation, the training set in the dataset is used to train and evaluate multiple machine learning models, and the model with the best evaluation is selected as the discrete metal coordination supramolecular cage structure prediction model; the method also includes: visualizing the model.

[0013] In one feasible implementation, the step involves training and evaluating multiple machine learning models on the training set of the dataset, and selecting the model with the best evaluation as the discrete metal coordination supramolecular cage structure prediction model; specifically: Six supervised learning algorithms and one unsupervised learning algorithm are employed. The six supervised learning algorithms include Catboost (classification boosting algorithm), RandomForest (random forest algorithm), DecisionTree (decision tree algorithm), xgboost (extreme gradient boosting algorithm), Support Vector Machine (SVM), and NaiveBayes (Bayesian algorithm). The unsupervised algorithm is K-Means (K-means algorithm). The grid search method was used to process six supervised learning algorithms to determine the optimal parameters for each algorithm. Specifically, the parameters for the classification boosting algorithm were selected as the number of decision trees to be built iteratively (iterations); the parameters for the random forest algorithm were selected as the number of decision trees (n_estimator); the parameters for the decision tree algorithm were selected as the maximum depth (max_depths); the parameters for the extreme gradient boosting algorithm were selected as the maximum depth (max_depths); the parameters for the support vector machine algorithm were selected as the penalty parameter C; and the parameters for the Bayesian algorithm were selected as the smoothing parameter alphas. Each algorithm was adjusted to its optimal parameters, and the training set in the dataset was then input into each of the six algorithms. The algorithms were evaluated using multiple metrics to obtain the algorithm with the best classification performance, thus obtaining a discrete metal coordination supramolecular cage structure prediction model.

[0014] In one feasible implementation, visualizing the model includes: Flask was used as the backend framework, CSS, HTML and JS as the frontend technology stack, and Python's standard library Werkzeug was used for frontend and backend information processing to obtain a visualized discrete metal coordination supramolecular cage structure prediction model.

[0015] In one feasible implementation, the process involves adjusting each algorithm to its optimal parameters, then inputting the training set from the dataset into the six algorithms respectively, evaluating the algorithms using multiple metrics, and obtaining the algorithm with the best classification performance to obtain a discrete metal coordination supramolecular cage structure prediction model; including: Each algorithm was adjusted to its optimal parameters, and the training set in the dataset was input into the six algorithms respectively. Four evaluation parameters were selected for the optimal algorithm selection process, resulting in the optimal algorithm classification boosting algorithm Catboost, which is a discrete metal coordination supramolecular cage structure prediction model. The four evaluation parameters include precision, recall, F1 score, and Kappa coefficient; their calculation formulas are as follows: Precision

[0016] Where TP represents the number of samples that are actually positive and are predicted as positive, and FP represents the number of samples that are actually negative but are predicted as positive. Recall

[0017] Where FN represents the number of samples that are actually positive but are predicted as negative; F1 Score

[0018]

[0019] Among them, P o P represents observational consistency, i.e., the proportion of classifier predictions that match the true labels; e It represents the expected consistency, that is, the proportion of predictions that match the true labels when guessing randomly;

[0020]

[0021] In one feasible implementation, comparing the prediction results with the results of high-throughput experiments conducted using automated equipment to verify the model accuracy includes: Develop automated equipment for high-throughput experimental verification; the automated equipment includes a pipetting device for liquid transfer, a syringe pump device for liquid dispensing, and a shaker device for reagent mixing; Based on the test set descriptor parameters of the prediction results, a high-throughput experiment was conducted using the automated equipment to perform a self-assembly experiment. The prediction results were then compared with the experimental results to verify the accuracy of the model.

[0022] According to a second aspect disclosed in this application, this application provides an apparatus for predicting the structure of discrete metal coordination supramolecular cages based on artificial intelligence, comprising: The data acquisition module is used to acquire a dataset of discrete metal coordination assembly topologies and divide the dataset into a training set and a test set; wherein, the dataset includes 26 constructed discrete metal coordination assembly topologies and descriptor parameters of the geometric and chemical features of the collected discrete metal coordination assembly topologies; The dataset standardization preprocessing module is used to perform standardization preprocessing on the dataset; The model building module trains and evaluates various machine learning models on the training set in the dataset, and selects the model with the best evaluation as the discrete metal coordination supramolecular cage structure prediction model. The result prediction module is used to input the test set descriptor parameters from the dataset into the discrete metal coordination supramolecular cage structure prediction model to obtain the discrete metal coordination supramolecular cage structure prediction results.

[0023] According to a third aspect disclosed in this application, this application provides an electronic device, including a processor and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method described in any one of the first aspects.

[0024] According to the fourth aspect disclosed in this application, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the method described in any one of the first aspects.

[0025] Compared with the prior art, this application has the following advantages: This application provides an artificial intelligence-based method for predicting the structure of discrete metal-coordinated supramolecular cages. This method, based on chemical knowledge, machine learning, and automation skills, can quickly predict the cage structure of the assembly and has high prediction accuracy. Compared with traditional chemical experiments, the method of this invention is rational and efficient, saving a lot of experimental and time costs, and contributing to the screening and application transformation of supramolecular cages. Attached Figure Description

[0026] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0027] Figure 1 A flowchart illustrating an artificial intelligence-based discrete metal coordination supramolecular cage structure prediction method provided in this application embodiment; Figure 2 This is a schematic diagram of the topology of the discrete metal coordination supramolecular cage structure provided in the embodiments of this application; Figure 3 Volume and electrostatic potential diagrams of eight ligand molecules, as well as assembly predictions and single-crystal structure diagrams, are provided for embodiments of this application. Figure 4 A schematic diagram illustrating the conformation determination for MFF force field optimization provided in an embodiment of this application; Figure 5 Visualization curves of eight ligands for discrete metal-coordinated supramolecular cage structures and SMILES strings for the ligands are provided for embodiments of this application. Figure 6 Six algorithm iteration parameter-accuracy curves provided for embodiments of this application; Figure 7 Heatmaps for evaluating machine learning algorithms provided in embodiments of this application; Figure 8 Schematic diagrams of eight ligand structures provided in the embodiments of this application; Figure 9 The following are verification diagrams of Fujita and Clever crystals provided in this embodiment: (A) is a scatter plot comparing the Fujita single crystal structure and the predicted structure; (B) is a scatter plot comparing the Clever single crystal structure and the predicted structure. Figure 10A schematic diagram of the structure of the discrete metal coordination supramolecular cage structure prediction device based on artificial intelligence provided in the embodiments of this application; Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0028] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0029] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0030] This application proposes an artificial intelligence-based discrete metal coordination supramolecular cage structure prediction method. This method can quickly predict the cage structure of the assembly and has high prediction accuracy. Compared with traditional chemical experiments, the method of this invention is rational and efficient, saving a lot of experimental and time costs, and helping the screening and application transformation of supramolecular cages.

[0031] The technical solution of the discrete metal coordination supramolecular cage structure prediction method based on artificial intelligence provided in this application will be described in detail below through specific embodiments. It should be noted that the following embodiments may exist alone or in combination with each other, and the same or similar content may not be described again in different embodiments.

[0032] Figure 1 A flowchart illustrating an artificial intelligence-based discrete metal coordination supramolecular cage structure prediction method is provided in this application embodiment. (See attached diagram.) Figure 1 In some embodiments, the process of this AI-based discrete metal coordination supramolecular cage structure prediction method includes the following steps: S1. Obtain a dataset of discrete metal coordination assembly topologies and divide the dataset into a training set and a test set with a ratio of 8:2, containing more than 1200 usable data points. The dataset includes 26 constructed discrete metal coordination assembly topologies and descriptor parameters of the geometric and chemical features of the collected discrete metal coordination assembly topologies.

[0033] In one embodiment of this application, the above steps specifically include: S11. Create 26 discrete metal coordination assembly topology diagrams: Use the STK toolkit to program the topology diagrams and construct 26 discrete metal coordination assembly topology structures; such as Figure 2 As shown, according to the cavity type of the assembly's topology, it can be divided into five types: 11 planar types, 4 spherical types, 2 tetrahedral types, 3 hexahedral types, and 6 cylindrical types. The naming rules for topologies are illustrated using topology sequence number 16 as an example; the topology is named 3. 2 2 3 This indicates that the topology of the assembly consists of two 3-coordination nodes and three 2-tooth ligands.

[0034] S12. Use the Python-based STK toolkit to digitize the positions and relationships of nodes and ligands in the assembly topology. Specifically, each topology is assigned a class, and each class has variables for vertices (_vertex_prototypes) and edges (_edge_prototypes). Vertex variables are classified into node (NonLinearVertex) and ligand (LinearVertex) vertex data, and edge variables represent the connection data between nodes and ligands (Edge).

[0035] S13. Collect descriptor parameters of geometric and chemical features that conform to the topological structures of 26 discrete metal coordination assemblies, and establish a dataset of discrete metal coordination assembly topological structures. The geometric features include three descriptors after topological data analysis (TDA): homology group number (NHG), average duration (AD), and maximum duration (MD). The chemical features include seven descriptors: atomic connection number (ANNC), cell volume (CV), charge density (ED), molecular volume (MV), temperature (T), ligand coordination number (LN), and node coordination number (NCN).

[0036] The descriptor parameters mentioned above were obtained through the following method: Installing the Linux system Ubuntu on Windows, installing the dionysus toolkit, and inputting the ligand SMILES to obtain topology analysis descriptor data; such as... Figure 3 As shown, matching atoms in a ligand molecule to discrete points in a topological structure results in a 0-dimensional simplex. As the radius increases with each discrete point as the center, when circles become tangent, edges connecting the centers are formed, corresponding to the establishment of chemical bonds—this is a 1-dimensional simplex. Edges may form connected graphs, considered as holes—this is a 2-dimensional simplex. In three-dimensional space, holes may also form—this is a 2-dimensional simplex. Finally, a fully connected graph is formed, the radius increase ends, and the topological structure is combined with the chemical molecular structure. S14. In the chemical characterization descriptor, the molecular volume is obtained from theoretical calculations. First, the ligands are modeled in Gaussian software, and the structure is optimized using the DFT method. Then, the optimized ligand volume is calculated using Multiwfn software. Finally, VMD software is used for visualization to obtain images. Then, the electrostatic potential diagram of each ligand is plotted, such as... Figure 3 As shown, (A) displays the molecular volume and electrostatic potential diagrams of ligands L1-8 from left to right and top to bottom; other descriptors can be obtained based on chemical experience.

[0037] S2. Perform standardization preprocessing on the dataset; Specifically, this includes: performing dimensionality reduction, normalization, and visualization processing on the dataset; S21. Obtaining the coordinate data of atoms in the ligand: Using ligand SMILEs to generate a three-dimensional conformation, optimizing a specific conformation using MMFF force field, obtaining the three-dimensional coordinates of atoms in the molecule, calculating the molecule's center of gravity, and then translating the conformation back to the origin to obtain the new atomic coordinates. Output the X-axis coordinate vector data (1*n), Y-axis coordinate vector data (n*1), and Z-axis coordinate vector data (1*n) respectively. Taking the benzene ring as an example... Figure 4 As shown; S22. Obtain the adjacency matrix of the ligand molecule: 1 indicates that there is a chemical bond between atoms; 0 indicates that there is no chemical bond between atoms. This gives the adjacency matrix vector data of the molecule (n*n). S23. Obtain the adjacency-Euclidean distance matrix of the ligand molecule: Calculate the chemical bond length between atoms using Euclidean distance and place it into the adjacency matrix to obtain the adjacency-Euclidean distance matrix (n*n); the formula for calculating the chemical bond length is as follows:

[0038] in, Represents the first point Axis coordinates; Indicates the second point Axis coordinates; Represents the first point Axis coordinates; Indicates the second point Axis coordinates; Represents the first point Axis coordinates; Indicates the second point Axis coordinates; S24. Obtain the ordinal numbers of all atoms in the ligand molecule using RDKit to obtain the atomic number vector data (n*1). S25. Integrate the information of all atoms in the ligand molecule to obtain the ligand information; the calculation formula is as follows:

[0039] in, express One atom Axis coordinate vector data; express One atom Axis coordinate vector data; express One atom Axis coordinate vector data; Represents the adjacency-Euclidean distance matrix; Represents the first in the matrix OK, Represents the first in the matrix List; S26. Input the resulting (n*n) matrix into the PCA dimensionality reduction algorithm to obtain (n*1) vector data. Integrate and normalize the data, then substitute the above (n*1) vector data into the min-max normalization formula to still obtain (n*1) vector data. The calculation formula is as follows:

[0040] in, This represents the maximum value of the sample. This represents the minimum value of the sample. The values ​​are the normalized values. S27. Visualize the ligand curves in a two-dimensional coordinate system; The axis represents the number of atoms in the ligand molecule. The axis represents the normalized curve values, such as... Figure 5 As shown.

[0041] S3. Train and evaluate various machine learning models on the training set in the dataset, and select the model with the best evaluation as the discrete metal coordination supramolecular cage structure prediction model. S31. Establish six supervised learning algorithms and one unsupervised learning algorithm. For the supervised algorithms, use a network search method to select the optimal parameters. Input the processed data into the algorithm calculations, evaluate the algorithms using multiple metrics, and finally select the algorithm with the best classification performance to obtain a discrete metal coordination supramolecular cage structure prediction model, such as... Figure 6As shown; the six supervised learning algorithms used include Catboost (classification boosting algorithm), RandomForest (random forest algorithm), DecisionTree (decision tree algorithm), XGBoost (extreme gradient boosting algorithm), Support Vector Machine (SVM), and NaiveBayes (Bayesian algorithm); the unsupervised algorithm is K-Means (K-means algorithm). S32. A grid search method was used to process six supervised learning algorithms to determine the optimal parameters for each algorithm. Accuracy curves were plotted as the parameter values ​​changed. Specifically, the parameters for the classification boosting algorithm were selected based on the number of decision trees (iterations); for the random forest algorithm, the number of decision trees (n_estimator); for the decision tree algorithm, the maximum depth (max_depths); for the extreme gradient boosting algorithm, the maximum depth (max_depths); for the support vector machine algorithm, the penalty parameter C; and for the Bayesian algorithm, the smoothing parameter alphas. The parameter increment curves for each algorithm are shown below. Figure 6 As shown; S33. After adjusting each algorithm to its optimal parameter values, input the processed data into the six algorithms respectively. Select four evaluation parameters as the basis for selecting the optimal algorithm: precision, recall, F1 score, and Kappa coefficient. Plot a heatmap as follows. Figure 11 As shown, the calculation formula is as follows: Precision Precision

[0042] Where TP represents the number of samples that are actually positive and are predicted as positive, and FP represents the number of samples that are actually negative but are predicted as positive. Recall Recall

[0043] Where FN represents the number of samples that are actually positive but are predicted as negative; F1 Score F1 Score

[0044] Kappa coefficient

[0045] Among them, P o P represents observational consistency (i.e., the proportion of classifier predictions that match the true labels). eThis indicates expected consistency (i.e., the proportion of predictions that match the true labels when guessing randomly).

[0046]

[0047] like Figure 7 As shown, the four evaluation parameters of the six algorithms are presented. Among them, the CatBoost algorithm has a value of 0.8 or higher. Therefore, the CatBoost algorithm was selected as the optimal algorithm for the next testing and prediction tasks.

[0048] According to one embodiment of this application, it further includes visualizing the prediction model of discrete metal-coordinated supramolecular cage structures; specifically including: Develop a visualization-based discrete metal coordination supramolecular cage structure prediction system: using Flask as the backend framework and CSS, HTML, and JS as the frontend technology stack, relying on Python's standard library Werkzeug to handle HTTP protocol and routing, and acting as a development server for frontend and backend information interaction processing, to obtain a visualization-based discrete metal coordination supramolecular cage structure prediction model.

[0049] S4. Input the test set descriptor parameters in the dataset into the discrete metal coordination supramolecular cage structure prediction model to obtain the discrete metal coordination supramolecular cage structure prediction results. The AI-based discrete metal coordination supramolecular cage structure prediction method of this application further includes the following steps: S5: Compare the predicted results with the results of high-throughput experimental verification using automated equipment to verify the model's accuracy; specifically: construct automated equipment for high-throughput experimental verification; the automated equipment includes a pipetting device for liquid transfer, a syringe pump device for liquid dispensing, and a shaker device for reagent mixing; the pipetting device includes three stepper motors, drivers, and pipettes connected in sequence; the syringe pump device includes a 1mL syringe, a stepper motor, a driver, and a solenoid valve connected in sequence; the shaker device includes a stepper motor, a driver, and a reaction bed connected in sequence; wherein, the driver is a Leadshine Intelligent DM2C series integrated stepper driver (RS485); S51. Use the assembled automated equipment to conduct assembly experiments to verify the accuracy of the prediction system: Automated equipment is built according to the assembly experiment requirements. There are three modules: a pipetting device, a syringe pump device, and a shaker device, which correspond to the three functions of pipetting, liquid picking, and shaking. Some parts of each device are obtained by 3D printer according to the design drawings. S52. Design organic ligands and single-atom structures based on dibromobenzene. Before starting the formal synthesis experiment, input the designed candidate ligands and related descriptor information into the prediction system for structure prediction. For ligands and nodes whose target and predicted structures match, begin the chemical synthesis operation; conversely, for those that do not match, do not proceed with the synthesis operation, and redesign the ligands and nodes. Figure 3 As shown in Figure (B), the predicted assembly structure is obtained by the prediction system after inputting ligand and node information. S53. Chemically synthesize the ligands and nodes screened by the prediction system; According to an embodiment of this application, the synthesis process is as follows: Starting from the bidentate phenyl bromide substrate, imidazole undergoes nucleophilic bromine substitution to generate imidazole onium salt; under alkaline conditions, sulfur powder / selenium powder undergoes nucleophilic substitution with it to generate bidentate organosulfur / selenium ligands.

[0050] S54. Set experimental strategies on the front-end page of the automated system to explore the crystallization boundary of the assembly structure formed by ligands and nodes.

[0051] S55. Perform single-crystal structure diffraction characterization on the obtained crystal; such as... Figure 3 As shown in Figure (C), the assembled structure after single-crystal testing is illustrated.

[0052] The overall process of the AI-based discrete metal coordination supramolecular cage structure prediction method in this application includes: first, building a database and collecting relevant data according to the target to prepare for machine learning algorithm learning; then designing multiple ligand and node candidate substances; next, inputting the candidate substances into the prediction model for prediction; for assemblies that conform to the target structure, the next step of automated experiment can be carried out, and for assemblies that do not conform to the target structure, the previous step is returned and redesigned; finally, using automated equipment, high-throughput experiments are carried out according to the designed strategy to verify the prediction results, and data with accurate experimental results are added to the database to realize a cycle.

[0053] Example This application designs multiple organic ligands and metal nodes. Through the above model prediction and automated equipment experiments, a total of 8 single crystals were obtained. The following uses the results of these eight single crystals as an example to demonstrate the process of rationally designing discrete metal-coordinated supramolecular cage structures: This embodiment discloses a discrete metal-coordination supramolecular cage structure prediction method based on artificial intelligence. The L1-L8 ligand structures are as follows: Figure 8 The structure shown has two types of nodes: Pd(MeCN)4PF6 and Cu(MeCN)4BF4, and includes the following steps: Example 1: (1) Prediction phase using a structural prediction system The L1 ligand and Pd(MeCN)4PF6 node information are input into a machine learning-based discrete supramolecular structure prediction system for structure prediction. Except for the descriptor molecular volume, which requires theoretical calculation, and the structure descriptor, which is obtained through code execution, other descriptors can be derived from chemical experience. The input information on the front-end page is shown in the table below:

[0054] After the prediction is completed, the page displays topology number 19, which can be queried. Figure 2 .

[0055] Click the 'Download' button to download a file named example.mol. The structure displayed using relevant software is as follows:

[0056] (2) Synthesis stage of L1 ligand If the predicted assembly structure is the target structure, the ligand synthesis experiment can be formally started. (3) Assembly experiment stage using automated equipment Experimental variables were designed, and the synthesized ligands were assembled with metal nodes using automated equipment.

[0057] Preparation before experiment 1. Solution preparation: (1) Take 5 ml of acetonitrile and dichloromethane solution respectively, and prepare a 1:1 acetonitrile / dichloromethane solution, totaling 10 ml; (2) Take 29.158 mg of Pd(MeCN)4PF6 (0.1 mmol) and add it to the above solution to form a node solution; (3) Take (0.2 mmol) of bidentate ligand and dissolve it in 10 ml of acetonitrile to make a ligand solution; 2. Automated parameter setting Liquid volume per sample: 300,000 rpm (0.5 ml); Waste liquid tank location: 00; Single shaker duration: 60 s; Injection pump No. 1: node solution; Injection pumps 2-5: Ligand solution.

[0058] 3. Formal Experiment: Design of automated front-end page experimental steps, front-end page input information: (1) Automatic cleaning and emptying equipment; (2) Add the node solution to the first row: Each row has 6 liquid filling positions, which need to be set 6 times on the front-end page. First, select the "Liquid Filling" function with parameter 1, and then select the "Move" function with the position parameter, as shown in the example below:

[0059] (3) Add the ligand solution in the first row. The process is the same as the previous step, but the parameter is set to 2 for the liquid addition function, as shown in the example below:

[0060] (4) Shaking table vibration for 120 s Select the "Shake" function without parameters. (5) Recycling Select the parameters for the "Recovery" function based on the usage of the syringe pump. (6) Automatic equipment cleaning (7) Selecting a poor solvent, single-crystal cultivation was performed using the diffusion method to obtain metal coordination cage 1, as shown in the figure. Figure 3 As shown in (C).

[0061] The crystal structure of the actual assembly can be obtained using a single-crystal diffractometer, and it is found to be the same as the predicted structure, which demonstrates the accuracy of the prediction system.

[0062] Examples 2-8: An artificial intelligence-based method for predicting discrete metal-coordination supramolecular cage structures includes the following steps: The preparation method is the same as in Example 1, except that assembly experiments were performed sequentially on L2 and Pd(MeCN)4PF6 nodes; assembly experiments were performed on L2-8 and Cu(MeCN)4BF4; the ligand input information in the structure prediction system was changed; and different injection pumps were selected, with corresponding changes to the automation strategy information.

[0063] Experimental example: (1) Taking coordination cage 1 as an example, the predicted structure of the structure prediction system in Example 1 and the single crystal structure obtained by the automated equipment experiment were compared; it was found that the topology of the prediction system and the actual single crystal structure were similar, indicating that the accuracy of the prediction system and the verification of the automated equipment were correct.

[0064] In addition to verifying the accuracy of the structural prediction through experiments, this invention collected multiple experimental data from Fujita and Clever experts in the field of discrete metal coordination cages to verify the accuracy of the prediction; the experimental results are as follows: Figure 9 As shown: A total of 60 Fujita literature data were collected. When the predicted values ​​were compared with the actual values, two inconsistencies were found, and the accuracy rate was 96.67%. A total of 87 Clever literature data were collected. When the predicted values ​​were compared with the actual values, two inconsistencies were found, and the accuracy rate was 97.70%.

[0065] In summary, the AI-based discrete metal coordination supramolecular cage structure prediction method proposed in this application achieves higher accuracy than... Figure 10 This is a schematic diagram of the device for predicting the structure of discrete metal coordination supramolecular cages based on artificial intelligence, as provided in this application. The device 1000 for predicting the structure of discrete metal coordination supramolecular cages based on artificial intelligence provided in this application embodiment is used to execute the technical solution provided in the aforementioned embodiment of the method for predicting the structure of discrete metal coordination supramolecular cages based on artificial intelligence. Its implementation principle and technical effects are similar to those in the aforementioned method embodiment, and will not be repeated here.

[0066] It includes: a data acquisition module, used to acquire a dataset of discrete metal coordination assembly topologies, and to divide the dataset into a training set and a test set; wherein, the dataset includes 26 constructed discrete metal coordination assembly topologies and descriptor parameters of the geometric and chemical features of the collected discrete metal coordination assembly topologies; The dataset standardization preprocessing module is used to perform standardization preprocessing on the dataset; The model building module trains and evaluates various machine learning models on the training set in the dataset, and selects the model with the best evaluation as the discrete metal coordination supramolecular cage structure prediction model. The result prediction module is used to input the test set descriptor parameters from the dataset into the discrete metal coordination supramolecular cage structure prediction model to obtain the discrete metal coordination supramolecular cage structure prediction results.

[0067] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls, entirely in hardware, or partially in software calls via processing elements and partially in hardware. For example, the model building module 1003 can be a separate processing element, or it can be integrated into a chip in the above device. Alternatively, it can be stored as program code in the memory of the above device, and its functions can be called and executed by a processing element of the device. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0068] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. (See attached diagram.) Figure 11 The electronic device 1100 includes a processor 1101 and a memory 1102 communicatively connected to the processor 1101; Memory 1102 stores computer-executed instructions; The processor 1101 executes computer execution instructions stored in the memory 1102 to implement the technical solution of the automated assembly strategy.

[0069] In the aforementioned electronic device 1100, the memory 1102 and the processor 1101 are electrically connected directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines, such as bus connections. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be classified as address buses, data buses, control buses, etc., but this does not mean that there is only one bus or one type of bus. The memory 1102 stores computer execution instructions for implementing automated assembly strategies, including at least one software functional module that can be stored in the memory 1102 in the form of software or firmware. The processor 1101 executes various functional applications and data processing by running the software programs and modules stored in the memory 1102.

[0070] The memory 1102 includes at least one type of readable storage medium, not limited to Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory 1102 stores the program 11, and the processor 1101 executes the program after receiving the execution instruction. Furthermore, the software programs and modules within the memory 1102 may also include an operating system, which may include various software components and / or drivers for managing system tasks (e.g., memory management, storage device control, power management, etc.), and can communicate with various hardware or software components to provide an operating environment for other software components.

[0071] Processor 1101 can be an integrated circuit chip with signal processing capabilities. The aforementioned processor 1101 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), etc. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor, or processor 1101 can be any conventional processor.

[0072] The electronic device 1100 is used to execute the technical solution provided in the aforementioned prediction method embodiment based on supramolecular cage structure. Its implementation principle and technical effect are similar to those in the aforementioned method embodiment, and will not be repeated here.

[0073] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the aforementioned technical solution of the method for predicting the discrete cage structure of an assembly based on a machine learning algorithm.

[0074] The aforementioned computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The computer-readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0075] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in an automated control device.

[0076] This application also provides a computer program product, including a computer program that, when executed, is used to implement the aforementioned technical solution for predicting and verifying assembly structures based on machine learning algorithms and automation.

[0077] In the above embodiments, those skilled in the art will understand that the above method embodiments can be implemented entirely or partially by software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented entirely or partially in the form of a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless network, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0078] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0079] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the appended claims.

[0080] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for predicting discrete metal-coordinated supramolecular cage structures based on artificial intelligence, characterized in that, include: A dataset of discrete metal coordination assembly topologies is obtained, and the dataset is divided into a training set and a test set; wherein, the dataset includes 26 constructed discrete metal coordination assembly topologies and descriptor parameters of the geometric and chemical features of the collected discrete metal coordination assembly topologies; The dataset is preprocessed using standardization. The training set in the dataset is used to train and evaluate various machine learning models, and the model with the best evaluation is selected as the discrete metal coordination supramolecular cage structure prediction model. Input the test set descriptor parameters from the dataset into the discrete metal coordination supramolecular cage structure prediction model to obtain the discrete metal coordination supramolecular cage structure prediction results.

2. The method according to claim 1, characterized in that, Also includes: The prediction results are compared with the results of high-throughput experiments conducted using automated equipment to verify the accuracy of the model.

3. The method according to claim 1, characterized in that, The process of acquiring a discrete metal coordination assembly topology dataset and dividing the dataset into a training set and a test set includes: Twenty-six discrete metal coordination assembly topologies were constructed. The Python-based STK toolkit digitizes the locations and relationships of nodes and ligands in the topology of the assembly. Descriptor parameters for geometric and chemical features of 26 discrete metal coordination assemblies were collected to establish a dataset of discrete metal coordination assembly topologies. The descriptor parameters for geometric features include the number of homology groups, average duration, and maximum duration. The descriptor parameters for chemical features include atomic connection number, cell volume, charge density, molecular volume, temperature, ligand coordination number, and node coordination number.

4. The method according to claim 3, characterized in that, The construction of 26 discrete metal coordination assembly topologies is as follows: The STK toolkit was used to program the topology graph, constructing 26 discrete metal coordination assembly topologies. These 26 discrete metal coordination assembly topologies include 11 planar structures, 4 spherical structures, 2 tetrahedral structures, 3 hexahedral structures, and 6 cylindrical structures.

5. The method according to claim 4, wherein the Python-based STK toolkit digitizes the node and ligand positions and relationships in the assembly topology; specifically including: Each topology is configured to include a class, and each class has vertex variables and edge variables; The vertex variables include node data and ligand data; The edge variables are the connection data between nodes and ligands.

6. The method according to claim 5, characterized in that, The standardization preprocessing of the dataset specifically includes: dimensionality reduction, normalization, and visualization of the dataset. Obtain coordinate data of atoms in ligands: A three-dimensional conformation is generated using ligand smiles, and the coordinate data of ligand atoms is determined using the MMFF force field optimization method: X-axis coordinate vector data 1*n, Y-axis coordinate vector data n*1, and Z-axis coordinate vector data 1*n; Obtain the adjacency matrix of the ligand molecule: 1 indicates that there is a chemical bond between atoms; 0 indicates that there is no chemical bond between atoms, thus obtaining the adjacency matrix vector data n*n of the molecule; Obtain the adjacency-Euclidean distance matrix of the ligand molecule: Calculate the chemical bond length between atoms using Euclidean distance and insert it into the adjacency matrix to obtain an n*n adjacency-Euclidean distance matrix; the formula for calculating the chemical bond length is as follows: in, Represents the first point Axis coordinates; Indicates the second point Axis coordinates; Represents the first point Axis coordinates; Indicates the second point Axis coordinates; Represents the first point Axis coordinates; Indicates the second point Axis coordinates; The atomic number vector data n*1 is obtained by using RDKit to obtain the ordinal numbers of all atoms in the ligand molecule. Information is integrated from all atoms in the ligand molecule to obtain ligand information; the calculation formula is as follows: in, A represents atomic number vector data; express One atom Axis coordinate vector data; express One atom Axis coordinate vector data; express One atom Axis coordinate vector data; Represents the adjacency-Euclidean distance matrix; Represents the first in the matrix OK, Represents the first in the matrix List; The results are input into the PCA dimensionality reduction algorithm to obtain n*1 vector data, denoted as... ; Normalize the data: Substitute the above n*1 vector data into the min-max normalization formula to obtain n*1 vector data. The calculation formula is as follows: in, This represents the maximum value of the sample. This represents the minimum value of the sample. The values ​​are the normalized values. Visualizing ligand curves: Ligand curves are visualized on a two-dimensional coordinate system; where, The axis represents the number of atoms in the ligand molecule. The axis represents the normalized curve values.

7. The method according to claim 1, characterized in that, The process involves training and evaluating various machine learning models on the training set of the dataset, and selecting the optimal model as the discrete metal coordination supramolecular cage structure prediction model; specifically: Six supervised learning algorithms and one unsupervised learning algorithm are employed. The six supervised learning algorithms include Catboost (classification boosting algorithm), RandomForest (random forest algorithm), DecisionTree (decision tree algorithm), xgboost (extreme gradient boosting algorithm), Support Vector Machine (SVM), and NaiveBayes (Bayesian algorithm). The unsupervised algorithm is K-Means (K-means algorithm). The grid search method was used to process six supervised learning algorithms to determine the optimal parameters for each algorithm. Specifically, the parameters for the classification boosting algorithm were selected as the number of decision trees to be built iteratively (iterations); the parameters for the random forest algorithm were selected as the number of decision trees (n_estimator); the parameters for the decision tree algorithm were selected as the maximum depth (max_depths); the parameters for the extreme gradient boosting algorithm were selected as the maximum depth (max_depths); the parameters for the support vector machine algorithm were selected as the penalty parameter C; and the parameters for the Bayesian algorithm were selected as the smoothing parameter alphas. Each algorithm was adjusted to its optimal parameters, and the training set in the dataset was then input into each of the six algorithms. The algorithms were evaluated using multiple metrics to obtain the algorithm with the best classification performance, thus obtaining a discrete metal coordination supramolecular cage structure prediction model.

8. The method according to claim 7, characterized in that, The process involves adjusting each algorithm to its optimal parameters, then inputting the training set from the dataset into each of the six algorithms, evaluating the algorithms using multiple metrics, and finally obtaining the algorithm with the best classification performance to arrive at a discrete metal coordination supramolecular cage structure prediction model; including: Each algorithm was adjusted to its optimal parameters, and the training set in the dataset was input into the six algorithms respectively. Four evaluation parameters were selected for the optimal algorithm selection process, resulting in the optimal algorithm classification boosting algorithm Catboost, which is a discrete metal coordination supramolecular cage structure prediction model. The four evaluation parameters include precision, recall, F1 score, and Kappa coefficient; their calculation formulas are as follows: Precision Where TP represents the number of samples that are actually positive and are predicted as positive, and FP represents the number of samples that are actually negative but are predicted as positive. Recall Where FN represents the number of samples that are actually positive but are predicted as negative; F1 Score Among them, P o P represents observational consistency, i.e., the proportion of classifier predictions that match the true labels; e It represents the expected consistency, that is, the proportion of predictions that match the true labels when guessing randomly; 。 9. The method according to claim 1, characterized in that, The step of comparing the prediction results with the results of high-throughput experiments conducted using automated equipment to verify the model accuracy includes: Develop automated equipment for high-throughput experimental verification; the automated equipment includes a pipetting device for liquid transfer, a syringe pump device for liquid dispensing, and a shaker device for reagent mixing; Based on the test set descriptor parameters of the prediction results, a high-throughput experiment was conducted using the automated equipment to perform a self-assembly experiment. The prediction results were then compared with the experimental results to verify the accuracy of the model.

10. An apparatus for predicting discrete metal-coordinated supramolecular cage structures based on artificial intelligence, characterized in that, include: The data acquisition module is used to acquire a dataset of discrete metal coordination assembly topologies and divide the dataset into a training set and a test set; wherein, the dataset includes 26 constructed discrete metal coordination assembly topologies and descriptor parameters of the geometric and chemical features of the collected discrete metal coordination assembly topologies; The dataset standardization preprocessing module is used to perform standardization preprocessing on the dataset; The model building module trains and evaluates various machine learning models on the training set in the dataset, and selects the model with the best evaluation as the discrete metal coordination supramolecular cage structure prediction model. The result prediction module is used to input the test set descriptor parameters from the dataset into the discrete metal coordination supramolecular cage structure prediction model to obtain the discrete metal coordination supramolecular cage structure prediction results.

Images