A function parameter optimization method based on molecular structure multibody effect descriptor

Abstract

The application relates to a function parameter optimization method of a molecular structure-based organic compound multibody descriptor, which automatically optimizes main function parameters for constructing a molecular structure multibody effect descriptor by using a Bayesian optimization method, and realizes rapid construction of a precise prediction model of typical properties of a novel single-molecule organic energetic compound. It is verified that the application has the characteristics of convenience, high efficiency and high reliability, is not limited to the molecular structure and constituent chemical elements of an organic compound, and has high universality. The method has important significance for design and construction research of an energetic molecule descriptor, and can be applied to training and optimization of a prediction model of target properties of various types of single-molecule organic compounds.

Description

Technical Field

[0001] This invention presents a method for optimizing function parameters of multibody descriptors for organic compounds based on molecular structure, applicable to reported or newly synthesized single-molecule organic compounds. Background Technology

[0002] In the prediction of properties and structural design of molecular energetic compounds, the descriptor system often determines whether the model can capture the key physicochemical relationship between density, energy, and stability. Existing descriptors generally suffer from two types of problems: First, they are too simplistic. For example, elemental composition, functional group counts, partial integral properties (tPSA, LogP), or traditional fingerprints, while computationally inexpensive and easy to generate in batches, essentially encode "what is there," failing to adequately express information such as the molecule's three-dimensional compactness, polar spatial distribution, local rigidity, and conformational freedom. Furthermore, they struggle to reflect intermolecular interactions and crystal packing differences, resulting in limited model differentiation capabilities in structurally similar systems, and interpretations tend to remain at a coarse-grained level of "more groups mean better / worse." Second, they are computationally too expensive. When higher-order features such as volume, polarizability, electrostatic potential surface characteristics, lattice energy, and condensation energy density obtained from quantitative chemistry or high-precision simulations are introduced, the physical meaning becomes clearer and the accuracy is usually better. However, data acquisition is highly dependent on computing power and processes, resulting in long computation times and limited repeatability and portability, which is not conducive to rapid screening and iterative optimization of massive candidate molecules. Faced with the contradiction between "simple but insufficient" and "accurate but too expensive," many-body descriptors provide a more reasonable compromise: taking the molecule as the background, the structural information is expanded hierarchically into first-order feature terms (inherent features of fragments / building blocks), second-order feature terms (coupled between fragments or atomic pairs, reflecting local polarity and spatial proximity), and third-order feature terms (synergistic effects of local environment, conformation, and interaction networks). This decomposition idea is consistent with the intuition of force fields and statistical physics, and can encode core factors affecting density and energy, such as "compactness, enhanced polarity, and conformational constraints," in a structured manner without relying on heavy computation, thereby improving prediction accuracy, generalization ability, and interpretability. Summary of the Invention

[0003] The purpose of this invention is to provide a method for optimizing function parameters of multi-body descriptors for organic compounds based on molecular structure. This method utilizes Bayesian hyperparameter optimization to optimize the monomer function parameters of the first-order characteristic term and the cutoff radius function parameters of the second-order characteristic term of the multi-body descriptor. Compared with traditional quantitative calculation optimization methods such as DFT optimization, semi-empirical quantization, and XTB optimization, it can significantly shorten the calculation time of descriptor parameters. For target properties with complex mechanisms, traditional quantitative calculation methods struggle to find corresponding physicochemical properties at the microscopic level and cannot optimize parameters through quantitative calculations. This method can optimize function parameters based on random initial guesses and has universal applicability.

[0004] To achieve the above objectives, the technical solution adopted by the present invention includes:

[0005] A method for optimizing function parameters of multibody descriptors for organic compounds based on molecular structure includes the following steps: Step 1: Construct a dataset containing the molecular structures of several organic compounds and the three-dimensional Cartesian coordinates of each atom within the molecule. Step 2: Determine the computational function for the multibody effect descriptors in the dataset; the multibody effect descriptors are mathematical vectors calculated based on the molecular structure, consisting of N first-order eigenvalues, M second-order eigenvalues, and L third-order eigenvalues; N is the number of atoms in the corresponding compound, and M is the sum of the number of different diatomic combinations among the N atoms, M=C N 2 L is the sum of the number of different triatomic combinations among N atoms, L=C N 3 ; in: Any element among the N first-order eigenvalues ​​is the sum of the contributions of all atoms belonging to class n atoms in the sequence of atoms constituting the corresponding compound to the target property. Class n atoms represent any one of the N types of atoms. The formula for calculating the first-order eigenvalue is: (1); in, is the first-order characteristic monomer parameter of the element, and n is the number of atoms of the element. There are a total of 4 monomer parameters in the CHON element system. For atom i, this is the first-order characteristic unit term under the many-body effect description rule; Any one of the M second-order eigenvalues ​​is: the sum of the van der Waals volumes of all diatomic combinations belonging to class m in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule of the organic compound. Class m diatomic combinations represent any one of the M types of diatomic combinations. (2); i and j are the atomic numbers of two atoms in the sequence of atoms that make up the corresponding compound, i≠j; The term is a second-order characteristic unit term of the diatomic combination composed of atom i and atom j under the many-body effect description rule.

[0006] A i and A j These are the first pre-exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. B i and B j These are the second pre-exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. a i and a j These are the first exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. b i and b j These are the second exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. α is the exponential factor for the function that calculates the cutoff radius of a diatomic combination; R ij It is the geometric spatial distance between atom number i and atom number j; and These are the cutoff radii for the diatomic combination of atoms i and j, respectively. Any one of the L third-order eigenvalues ​​is: the sum of the van der Waals volumes of all triatomic combinations belonging to class l in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule of the organic compound. Class l triatomic combination is any one of the L types of triatomic combinations. (3); i, j, and k are the atomic numbers of three atoms in the sequence of atoms that make up the corresponding compound, i≠j≠k; The term is the third-order characteristic unit term of the triatomic combination composed of atoms with serial numbers i, j, and k under the many-body effect description rule; C i C j and C k Calculate the first pre-exponential factor for the combined volume of the three atoms of sequence number i, sequence number j, and sequence number k, respectively; c i c j and c kCalculate the second pre-exponential factor for the combined volume of the three atoms of sequence number i, sequence number j, and sequence number k, respectively; D i D j and D k The intercept term is calculated for the combined volume of the three atoms i, j, and k, respectively. β is the exponential factor of the function for calculating the cutoff radius of a three-atom combination; R ij It is the geometric spatial distance between atom i and atom j; R ik R is the geometric spatial distance between atom i and atom k; kj It is the geometric spatial distance between atom k and atom j; , and These are the cutoff radii of the three-atom combination for atom i, atom j, and atom k, respectively. The parameters to be optimized are the first-order unit terms. and second-order unit terms Two types of parameters: cutoff radius; Step 3: Optimize the hyperparameters of the multibody effect descriptor to improve the fitting degree of the descriptor to the correlation between molecular structure and target properties, thereby improving the prediction accuracy of the structure-activity model. Step 4: After completing the hyperparameter optimization of the multibody effect descriptor, the descriptors of several organic compounds are calculated, and their target properties and descriptor data are fitted to obtain the target property prediction model of energetic compounds, where the target property is the dependent variable Y and the descriptor data is the independent variable X.

[0007] Optionally, step 3 specifically includes: A 3D structure descriptor generator is defined to characterize molecular structure by calculating the feature unit terms of various atomic combinations within the molecule and their contribution to the target value. The generator calculates a distance matrix based on the input atomic numbers and coordinates. Then, using predefined many-body effect descriptor function parameters, it calculates first-order, second-order, and third-order features respectively. Finally, it performs a merge summation based on the symmetry of element combinations, representing the contribution of a certain type of atomic combination to the target value. The final output is a dictionary with element combinations as keys and descriptor values ​​as values. The calculation process is shown below: First-order eigenvalues ​​are the sum of first-order eigenvalues: (4); The second-order eigenvalue is the sum of the distinct elements of the second-order eigenvalue: (5); Summation of identical elements with second-order characteristics: (6); In the formula, X and Y represent the element types of atoms within the molecule; Third-order features are calculated using the second-order feature method.

[0008] Optionally, the organic compound is an energetic organic compound.

[0009] Optionally, the organic compound may contain carbon, hydrogen, nitrogen and oxygen, or other common chemical elements.

[0010] Optionally, in step 1, the three-dimensional Cartesian coordinates of the organic compound are obtained by processing the SMILES code of the energetic compound using the Rdkit program to obtain the initial three-dimensional structure, and then optimizing the initial three-dimensional structure using the GFN2-xTB method.

[0011] The advantages of this invention are: This invention obtains the first-order feature term single-unit contribution and the second-order feature term cutoff radius function parameters through the hyperparameter optimization method of the Bayesian optimization framework, thereby obtaining an optimized multi-body effect descriptor and establishing the relationship between molecular structure and target properties. This enables the acquisition of a sufficient amount of data that is highly correlated with the target properties in terms of numerical values ​​during the feature descriptor construction stage, providing a general method for constructing property prediction models, effectively improving model construction efficiency, reducing the amount of training sample data required to build accurate prediction models, and significantly expanding the applicability of multi-body effect descriptors. Attached Figure Description

[0012] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a graph showing the optimization curves of the single-function parameters in Example 1; Figure 2 This is a graph showing the optimized parameters of the cutoff radius function in Example 1; Figure 3 This is a graph showing the prediction model results from Example 1; Figure 4 The images show the results before and after optimization in Example 1. Detailed Implementation

[0013] Unless otherwise specified, the scientific and technical terms used in this article are intended for understanding by those skilled in the art.

[0014] This invention expresses the spatial occupancy and arrangement of molecules within a unit cell of an energetic compound in its crystalline state based on the concept of molecular volume. By optimizing first-order eigenvalue monomer function parameters and second-order eigenvalue cutoff radius, it calculates first-, second-, and third-order many-body effect characteristic unit terms and forms a structure descriptor. This descriptor is then used to construct a novel machine learning prediction model, reducing the crystal density prediction error of energetic compounds by 20%.

[0015] The method for optimizing function parameters of organic compound multibody descriptors based on molecular structure of the present invention includes the following steps: Step 1: Construct a dataset containing the molecular structures of several organic compounds and the three-dimensional Cartesian coordinates of each atom within the molecule. Step 2: Determine the computational function for the multibody effect descriptors in the dataset; the multibody effect descriptors are mathematical vectors calculated based on the molecular structure, consisting of N first-order eigenvalues, M second-order eigenvalues, and L third-order eigenvalues; N is the number of atoms in the corresponding compound, and M is the sum of the number of different diatomic combinations among the N atoms, M=C N 2 L is the sum of the number of different triatomic combinations among N atoms, L=C N 3 ; in: Any element among the N first-order eigenvalues ​​is the sum of the contributions of all atoms belonging to class n atoms in the sequence of atoms constituting the corresponding compound to the target property. Class n atoms represent any one of the N types of atoms. The formula for calculating the first-order eigenvalue is: (1); in, is the first-order characteristic monomer parameter of the element, and n is the number of atoms of the element. There are a total of 4 monomer parameters in the CHON element system. For atom i, this is the first-order characteristic unit term under the many-body effect description rule; Any one of the M second-order eigenvalues ​​is: the sum of the van der Waals volumes of all diatomic combinations belonging to class m in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule of the organic compound. Class m diatomic combinations represent any one of the M types of diatomic combinations. (2); i and j are the atomic numbers of two atoms in the sequence of atoms that make up the corresponding compound, i≠j; The term is a second-order characteristic unit term of the diatomic combination composed of atom i and atom j under the many-body effect description rule.

[0016] A i and A j These are the first pre-exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. B i and B j These are the second pre-exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. a i and a j These are the first exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. b i and b j These are the second exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. α is the exponential factor for the function that calculates the cutoff radius of a diatomic combination; R ij It is the geometric spatial distance between atom number i and atom number j; and These are the cutoff radii for the diatomic combination of atoms i and j, respectively. Any one of the L third-order eigenvalues ​​is: the sum of the van der Waals volumes of all triatomic combinations belonging to class l in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule of the organic compound. Class l triatomic combination is any one of the L types of triatomic combinations. (3); i, j, and k are the atomic numbers of three atoms in the sequence of atoms that make up the corresponding compound, i≠j≠k; The term is the third-order characteristic unit term of the triatomic combination composed of atoms with serial numbers i, j, and k under the many-body effect description rule; C i C j and C k Calculate the first pre-exponential factor for the combined volume of the three atoms of sequence number i, sequence number j, and sequence number k, respectively; c i c j and c k Calculate the second pre-exponential factor for the combined volume of the three atoms of sequence number i, sequence number j, and sequence number k, respectively; D i D j and D k The intercept term is calculated for the combined volume of the three atoms i, j, and k, respectively. β is the exponential factor of the function for calculating the cutoff radius of a three-atom combination; R ij It is the geometric spatial distance between atom i and atom j; R ik R is the geometric spatial distance between atom i and atom k; kj It is the geometric spatial distance between atom k and atom j; , and These are the cutoff radii of the three-atom combination for atom i, atom j, and atom k, respectively. In this optimization method, the optimized parameter is the first-order element term. and second-order unit terms Cutoff radius is one of the two types of parameters.

[0017] Step 3 involves optimizing several hyperparameters to improve the fitting of the descriptor to the correlation between molecular structure and target properties, thereby enhancing the prediction accuracy of the structure-activity model. First, a three-dimensional structure descriptor generator is defined. This generator characterizes the molecular structure by calculating the feature unit terms of various atomic combinations within the molecule and their contribution to the target value. The distance matrix is ​​calculated based on the input atomic number and coordinates. Then, using predefined multi-body effect descriptor function parameters such as element-related monomer contribution parameters, cutoff radius parameters, two-body contribution parameters, and three-body contribution parameters, first-order, second-order, and third-order features are calculated respectively. Finally, the elements are merged and summed according to their symmetry (same / different elements), representing the contribution of a certain type of atomic combination to the target value. The final output is a dictionary with element combinations as keys and descriptor values ​​as values. The calculation process is shown below: First-order eigenvalues ​​are the sum of first-order eigenvalues: (4); The second-order eigenvalue is the sum of the second-order eigenvalues ​​(distinct elements): (5); Summation of second-order features (identical elements): (6); In the formula, X and Y represent the element types of atoms within the molecule; The third-order features are calculated using the second-order feature method.

[0018] The first-order eigenvalues ​​above reflect the contribution of a single atom to the target value. For second-order eigenvalues, which are determined by the element types of the two atoms and their interactions, it is necessary to consider the attenuation of their contribution as the distance increases. Therefore, different cutoff radius parameters are defined for each element atom, and the final output descriptor has translational and rotational symmetry.

[0019] After obtaining the initial guess value of the descriptor, it is necessary to establish a hyperparameter optimization method based on the Bayesian optimization framework to optimize the individual contribution and cutoff radius in the descriptor generation process: the function randomly tries several sets (≥5) of different hyperparameter combinations, i.e., the observation set; establish a probabilistic model of the current optimal solution, and use the model to predict the current best function parameter combination, and then add the current best parameter combination to the observation set to continue updating the probabilistic model of the optimal solution until the model effect reaches the best, and obtain the optimal parameter combination that can characterize the target property and predict the target property.

[0020] Step 4: After completing the hyperparameter optimization of the multibody effect descriptor, the descriptors of several organic compounds are calculated, and their target properties and descriptor data are fitted to obtain the target property prediction model of energetic compounds, where the target property is the dependent variable (Y) and the descriptor data is the independent variable (X).

[0021] In this invention, the organic compound is an energetic organic compound.

[0022] In this invention, the organic compounds contain four elements: carbon, hydrogen, nitrogen, and oxygen, and can be extended to other common chemical elements.

[0023] In this invention, the three-dimensional Cartesian coordinates of the organic compound are obtained by processing the SMILES code of the energetic compound using the Rdkit program to obtain the initial three-dimensional structure, and then optimizing the initial three-dimensional structure using the GFN2-xTB method.

[0024] The parameter optimization method of this invention is applicable to most common machine learning and deep learning algorithms, including Kernel Ridge Regression, Adaboost Regressor, Extra Trees Regressor, and Transformed Target Regressor. Steps 1, 2, and 3 are used to obtain the function parameters of the molecular structure multi-body effect descriptor, which is then used to calculate the multi-body effect feature descriptor data of the molecular structure. This feature descriptor data is then used as input to the model constructed by the method to predict the target properties of the organic compound.

[0025] It should be noted that the energetic compounds in step 1 of the present invention are selected according to the object to be predicted. Specifically, multiple compounds can be selected according to the constituent atoms and crystal characteristics of the object to be predicted. In order to ensure the accuracy of the prediction results, the series of organic compounds should be selected as much as possible as compounds with relevant atomic types and reasonable crystal density parameter values ​​to the object to be predicted.

[0026] In the present invention All of these are statistical parameters, meaning they were obtained through statistical analysis of a series of organic compounds.

[0027] In step 2 of this invention, the 3D structure descriptor generator is defined as a callable function. First, the function parameter ranges are defined: four first-order feature unit function parameters, C_1, H_1, N_1, and O_1, where H_1 ranges from (2.0 to 8.0), and C_1, N_1, and O_1 range from (5.0 to 20.0); and two second-order feature cutoff radius function parameters, Hrc, Nrc, Orc, and Crc, where Hrc ranges from (0.1 to 0.5), and Nrc, Orc, and Crc range from (0.4 to 1.0). Next, a 3D structure descriptor is created by calling the 3D structure descriptor generator function and generating descriptor features. The third step involves training the feature data using a machine learning model and employing five-fold cross-validation. Finally, Bayesian optimization is called to find the optimal parameters, and the parameter optimization process and the optimal error value are output.

[0028] In a specific embodiment, the organic compounds or energetic compounds used in the model construction of this invention can be selected from compounds that have been synthesized and publicly reported in the prior art. The specific implementation process can be achieved with the help of existing related software.

[0029] The following are specific embodiments of the present invention. It should be noted that these embodiments are preferred examples of the present invention. Those skilled in the art can optimize the features, formulas, parameters, and fitting methods of special data within the scope of the present invention to obtain other implementation schemes that fall within the protection scope of the present invention. The present invention is not limited to the following examples. Unless otherwise specified, the methods described are conventional methods.

[0030] Example 1: Step 1: The data for this embodiment is selected from approximately 1300 energetic compounds in the CCDC database, containing only four elements: carbon, hydrogen, nitrogen, and oxygen. The crystal density of these energetic compounds ranges from 1.20 to 2.00 g / cm³. 3 .

[0031] Step 2: (Optimization of individual parameters for first-order features) The method is as follows: The molecular structure was optimized using the GFN2 method in the XTB software, and the optimization results were saved in an XYZ format file, ensuring that the new structure data overwrote the original structure data. The monomer parameters C_1, H_1, N_1, and O_1 were optimized using Bayesian optimization, with initial guesses of 5, 10, 10, and 10 respectively. The number of iterations was set to 200, and the target value was -RMSE, aiming to find the lowest RMSE. The four parameter values ​​corresponding to this lowest RMSE were marked as the current optimal values ​​until the optimal monomer parameters, i.e., the lowest RMSE value, were obtained. After the iterations, the optimal monomer parameter combination was output. Table 1 shows the iterative process of finding the optimal parameters. Figure 1 The curve showing the change of the loss function after optimizing the hyperparameters of a single entity.

[0032] Table 1

[0033] Step 3: Optimization of the cutoff radius parameter for the second-order feature This embodiment describes a Bayesian optimization method for descriptor hyperparameters. The specific method is as follows: The three-dimensional Cartesian coordinates of each energetic compound were obtained by processing the SMILES codes of the energetic compounds using the Rdkit program to obtain the initial three-dimensional structure, and then optimizing the initial three-dimensional structure using the xtb optimization method. Afterwards, the descriptor data vector for each molecule was calculated. A descriptor generation function was called for each molecular structure to generate a many-body effect descriptor.

[0034] Perform Bayesian optimization settings. Define four cutoff radius hyperparameters H_rc, N_rc, O_rc, and C_rc, and assign them initial values: 'H_rc': 0.90, 'C_rc': 1.15, 'N_rc': 1.15, 'O_rc': 1.25. See Table 2 for details.

[0035] Table 2

[0036] Bayesian optimization is used to find the optimal cutoff radius parameters to minimize the error between the predicted and actual volumes. The result outputs the optimal combination of parameters for the cutoff radius. Figure 2 The curve showing the change of the loss function after optimizing the truncation radius hyperparameter.

[0037] Predictive model building This embodiment describes a method for constructing a crystal density prediction model for energetic compounds. The energetic compounds used in constructing the model are selected from approximately 1300 known and publicly reported energetic compounds, limited to the four elements of carbon, hydrogen, nitrogen, and oxygen. These compounds include nitrate esters, ammonium nitrates, cubane compounds, azoles, and azides. The crystal density range of the energetic compounds involved is 1.20~2.00 g / cm³. 3 The specific method is as follows: The three-dimensional Cartesian coordinates of each energetic compound were obtained by processing the SMILES codes of the energetic compounds using the Rdkit program to obtain the initial three-dimensional structure, and then optimizing the initial three-dimensional structure using the xtb optimization method. Then, the descriptor data vectors of each molecule were calculated. Using the optimized descriptors, the kernel ridge regression algorithm was trained. The machine learning software used for construction was the open-source program scikit-learn. Specifically, according to the settings in scikit-learn, building the kernel ridge regression algorithm model requires setting six parameters: kernel, alpha, gamma, degree, coef0, and kernel_params. In this embodiment, kernel='rbf', the values ​​of alpha and gamma were obtained through grid search based on the content of the training set, degree=3.0, coef0=1.0, and kernel_params='None'. The model validation method was 5-fold cross-validation, and the training set consisted of crystals with a density greater than 1.20 g / cm³ from the Cambridge CCDC crystal database. 3 The set of 1018 energetic compounds; the validation set is the Cambridge CCDC crystal database with crystal densities greater than 1.20 g / cm³. 3 Molecules of 255 energetic compounds were analyzed; the mean absolute percentage error (MAE) for the training set was 1.99%, and for the test set it was 2.7%. The mean absolute prediction error for the crystal density of various energetic molecules reached 0.036 g / cm³. 3 level. Figure 3 For the optimized model training and testing parity graph, Figure 4 The image shows the odd / even plots for training and testing the model before optimization. Prediction accuracy was improved by 20%.

[0038] Although the illustrative specific embodiments of the present invention have been described above to enable those skilled in the art to understand the invention, it should be understood that the invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the invention as defined and determined by the appended claims, and all inventions utilizing the concept of the present invention are protected.

Claims

1. A method for optimizing function parameters based on multi-body descriptors of organic compounds with molecular structure, characterized in that, Includes the following steps: Step 1: Construct a dataset containing the molecular structures of several organic compounds and the three-dimensional Cartesian coordinates of each atom within the molecule. Step 2: Determine the computational function for the multibody effect descriptors in the dataset; the multibody effect descriptors are mathematical vectors calculated based on the molecular structure, consisting of N first-order eigenvalues, M second-order eigenvalues, and L third-order eigenvalues; N is the number of atoms in the corresponding compound, and M is the sum of the number of different diatomic combinations among the N atoms, M=C N 2 L is the sum of the number of different triatomic combinations among N atoms, L=C N 3 ; in: Any element among the N first-order eigenvalues ​​is the sum of the contributions of all atoms belonging to class n atoms in the sequence of atoms constituting the corresponding compound to the target property. Class n atoms represent any one of the N types of atoms. The formula for calculating the first-order eigenvalue is: (1)； in, is the first-order characteristic monomer parameter of the element, and n is the number of atoms of the element. There are a total of 4 monomer parameters in the CHON element system. For atom i, the first-order characteristic unit term under the many-body effect description rule; Any one of the M second-order eigenvalues ​​is: the sum of the van der Waals volumes of all diatomic combinations belonging to class m in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule of the organic compound. Class m diatomic combinations represent any one of the M types of diatomic combinations. (2)； i and j are the atomic numbers of two atoms in the sequence of atoms that make up the corresponding compound, i≠j; The term is a second-order characteristic unit term of the diatomic combination composed of atom i and atom j under the many-body effect description rule. A i and A j These are the first pre-exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. B i and B j These are the second pre-exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. a i and a j These are the first exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. b i and b j These are the second exponential factors of the function for calculating the volume of a diatomic combination of atoms i and j, respectively. α is the exponential factor for the function that calculates the cutoff radius of a diatomic combination; R ij It is the geometric spatial distance between atom number i and atom number j; and These are the cutoff radii for the diatomic combination of atoms i and j, respectively. Any one of the L third-order eigenvalues ​​is: the sum of the van der Waals volumes of all triatomic combinations belonging to class l in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule of the organic compound. Class l triatomic combination is any one of the L types of triatomic combinations. (3)； i, j, and k are the atomic numbers of three atoms in the sequence of atoms that make up the corresponding compound, i ≠ j ≠ k; The term is the third-order characteristic unit term of the triatomic combination composed of atoms with serial numbers i, j, and k under the many-body effect description rule; C i C j and C k Calculate the first pre-exponential factor for the combined volume of the three atoms of sequence number i, sequence number j, and sequence number k, respectively; c i c j and c k Calculate the second pre-exponential factor for the combined volume of the three atoms of sequence number i, sequence number j, and sequence number k, respectively; D i D j and D k The intercept term is calculated for the combined volume of the three atoms i, j, and k, respectively. β is the exponential factor of the function for calculating the cutoff radius of a three-atom combination; R ij It is the geometric spatial distance between atom i and atom j; R ik R is the geometric spatial distance between atom i and atom k; kj It is the geometric spatial distance between atom k and atom j; , and These are the cutoff radii of the three-atom combination for atom i, atom j, and atom k, respectively. The parameters to be optimized are the first-order unit terms. and second-order unit terms Two types of parameters: cutoff radius; Step 3: Optimize the hyperparameters of the multibody effect descriptor to improve the fitting degree of the descriptor to the correlation between molecular structure and target properties, thereby improving the prediction accuracy of the structure-activity model. Step 4: After completing the hyperparameter optimization of the multibody effect descriptor, the descriptors of several organic compounds are calculated, and their target properties and descriptor data are fitted to obtain the target property prediction model of energetic compounds, where the target property is the dependent variable Y and the descriptor data is the independent variable X.

2. The method for optimizing function parameters of organic compounds based on molecular structure multi-body descriptors according to claim 1, characterized in that, Step 3 specifically includes: A 3D structure descriptor generator is defined to characterize molecular structure by calculating the feature unit terms of various atomic combinations within the molecule and their contribution to the target value. The generator calculates a distance matrix based on the input atomic numbers and coordinates. Then, using predefined many-body effect descriptor function parameters, it calculates first-order, second-order, and third-order features respectively. Finally, it performs a merge summation based on the symmetry of element combinations, representing the contribution of a certain type of atomic combination to the target value. The final output is a dictionary with element combinations as keys and descriptor values ​​as values. The calculation process is shown below: First-order eigenvalues ​​are the sum of first-order eigenvalues: (4)； The second-order eigenvalue is the sum of the distinct elements of the second-order eigenvalue: (5)； Summation of identical elements with second-order characteristics: (6)； In the formula, X and Y represent the element types of atoms within the molecule; Third-order features are calculated using the second-order feature method.

3. The method for optimizing function parameters of organic compounds based on molecular structure multi-body descriptors according to claim 1 or 2, characterized in that, The organic compound is an energetic organic compound.

4. The method for optimizing function parameters of organic compounds based on molecular structure multibody descriptors according to claim 1 or 2, characterized in that, The organic compounds contain four elements: carbon, hydrogen, nitrogen, and oxygen, or other common chemical elements.

5. The method for optimizing function parameters of organic compounds based on molecular structure multibody descriptors according to claim 1 or 2, characterized in that, In step 1, the three-dimensional Cartesian coordinates of the organic compound are obtained by processing the SMILES code of the energetic compound using the Rdkit program to obtain the initial three-dimensional structure, and then optimizing the initial three-dimensional structure using the GFN2-xTB method.

Images