Prediction method of isowurtzitane cage compound reaction substrate based on machine learning

Abstract

The invention discloses a machine learning-based isowurtzitane cage compound reaction substrate prediction method, which comprises the following steps: establishing a database containing condensation reactions of glyoxal and primary amine reaction substrates, marking the condensation reaction of which the product is an isowurtzitane cage compound as 1, and marking the condensation reaction of which the product is other types of compounds as 0; the method comprises the following steps: performing quantum chemistry calculation on primary amine reaction substrates, collecting descriptors describing chemical information, screening feature descriptors, and performing normalization processing; taking the feature descriptor as an input variable, taking a condensation reaction product marking result as an output variable, and selecting a random forest and a factorization machine to carry out weight distribution to construct an optimal fusion model; and performing prediction by using the optimal fusion model. According to the method, the reactivity of the primary amine reaction substrate can be quickly and accurately predicted, the research efficiency is improved, the experiment cost is reduced, and effective guidance is provided for synthesis of the isowurtzitane cage compound.

Description

Technical Field

[0001] The present invention relates to the technical field of chemical reaction prediction, and more specifically, to a method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning. Background Art

[0002] As the most basic energy unit for constructing energetic materials, the performance of energetic compounds directly affects the combat effectiveness of weapons and equipment. The pursuit of high-energy density materials (HEDMs) with excellent performance has always been the main direction of the development of energetic materials. HEDMs refer to energetic materials with high detonation performance and energy density. As the main component of the formulation of explosives, propellants, etc., their higher energy release efficiency and stronger detonation power can significantly improve the damage effect and combat effectiveness of weapons and equipment, and meet the needs of modern weapon systems for more powerful, precise and efficient strike capabilities. Therefore, HEDMs are one of the key materials to promote the performance improvement of modern weapons and equipment and realize the "long-range strike" and "efficient damage" of weapons and equipment. It has far-reaching significance for the progress of modern military science and technology and the improvement of national strategic defense capabilities.

[0003] Hexanitrohexaazaisowurtzitane (CL-20) is a cage-structured energetic compound with the molecular formula C6H6N 12 O 12 , density 2.04 g·cm -3 , explosion velocity 9580 m·s -1 , explosion pressure 45.2 GPa. As a typical representative of high energy density materials and third-generation energetic materials, CL-20 outperforms 1,3,5,7-tetranitro-1,3,5,7-tetraazacyclooctane (HMX), an energetic material that has been widely used in weapons and equipment. Actual measured data show that the density of CL-20 is 8% higher than that of HMX, the detonation velocity is 6%, the detonation pressure is 8%, and the energy density is more than 10%. Due to its excellent energy characteristics, CL-20 is considered to be the single-element energetic material with the best comprehensive performance that has been successfully developed and industrialized so far, and has broad application prospects in new weapon and equipment charges.

[0004] The optimal synthesis and production of CL-20 is an important direction for the development of energetic materials. Since Nielsen of the United States first synthesized CL-20 in 1987, various countries have conducted extensive research on its process methods in order to achieve industrial production. After years of development, there are multiple routes for the synthesis of CL-20, but the CL-20 manufacturing routes that can be engineered currently all use hexabenzylhexaazaisowurtzitane (HBIW) as a precursor. However, HBIW is unstable in an acidic environment, so it cannot be directly nitrated to CL-20. Instead, it must be hydrogenated using a noble metal catalyst to convert the benzyl group into an acid-stable group (formyl, acetyl or amine), and then nitrated to obtain CL-20. This makes the preparation process of CL-20 complicated and the cost remains high. Therefore, if hexaazaisowurtzitane without a benzyl group and other leaving groups that are easy to nitrate can be synthesized as a nitration precursor, eliminating the hydrogenation step, it will be a huge breakthrough in the synthesis research of CL-20.

[0005] To achieve this goal, many scientists have conducted a lot of exploration and experiments, but no progress has been made until 2004, when Cagnon et al. in France used glyoxal to react with primary amines with different substituents (2-thienylmethylamine, 2-furanmethylamine and allylamine, etc.) to obtain the corresponding hexaazaisowurtzitane cage compounds, and used nitric acid and sulfuric acid mixed acid for direct nitration to obtain a small amount of CL-20, but the yield was too low to be used in practice. Although there have been a few reports on the synthesis of new hexaazaisowurtzitane cage compounds, no substantial breakthroughs have been made. At present, HBIW is still considered to be the best precursor for the preparation of CL-20. The main reason is that the traditional experimental method mainly relies on empirical screening and trial and error to determine the appropriate primary amine substrates, which has the problems of long cycle, low efficiency and high cost, and it is difficult to systematically summarize the rules. In addition, although the existing computational chemistry methods can assist in understanding the relationship between substrates and products, they have large computational complexity and limited accuracy, making them difficult to be widely promoted in practical applications. To date, there is still no method that can accurately predict which primary amine substrates can react with glyoxal to form isowurtzitane cage structures. Summary of the invention

[0006] The present invention provides a method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning, which can achieve rapid and accurate prediction of the reactivity of primary amine reaction substrates, improve research efficiency, reduce experimental costs, and provide effective guidance for the synthesis of isowurtzitane cage compounds.

[0007] In order to achieve these purposes and other advantages according to the present invention, a method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning is provided, wherein the isowurtzitane cage compound is a nitration precursor compound of hexanitrohexaazaisowurtzitane, comprising: A database of condensation reactions of glyoxal and primary amine reaction substrates is established, and the condensation reactions whose products are isowurtzitane cage compounds are marked as 1, and the condensation reactions whose products are other types of compounds are marked as 0; Perform quantum chemical calculations on primary amine reaction substrates, collect descriptors that describe chemical information, screen the feature descriptors so that there is no obvious correlation between the feature descriptors, and normalize the feature descriptors; The feature descriptor is used as the input variable of machine learning, and the condensation reaction product labeling result is used as the output variable of machine learning. Random forest and factor decomposition machine are selected for weight allocation to build a fusion model, and the model is trained and evaluated to obtain the optimal fusion model. The optimal fusion model was used to predict whether glyoxal and new primary amine reaction substrates could generate isowurtzitane cage compounds through condensation reaction.

[0008] Preferably, the quantum chemical calculations include the steps of geometry optimization, vibration analysis and energy calculation.

[0009] Preferably, when correlation analysis is performed on the above descriptors, the Pearson correlation coefficients between the descriptors are all lower than 0.8, indicating that there is no obvious correlation between the descriptors, and they are used as feature descriptors.

[0010] Preferably, the characteristic descriptors include highest occupied molecular orbital energy, lowest unoccupied molecular orbital energy, bond length, bond angle, dipole moment, Hirshfeld charge, Fukui function, dual descriptor, electronegativity, nucleophilicity, Rydberg electron density, total electron density, van der Waals surface area, molecular volume, total energy, NMR chemical shift.

[0011] The method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning as claimed in claim 1, characterized in that the optimal parameter combination of random forest is: the maximum depth is 4, the maximum number of leaf nodes is 100, and the maximum number of bins is 128.

[0012] Preferably, the best parameter combination of the factorization machine is: learning rate 0.005, number of factors 5, L2 regularization coefficient 0.01, and number of training rounds 10.

[0013] Preferably, the random forest and the factor decomposition machine are weighted to construct a fusion model, the weight of the random forest is 0.8, and the weight of the factor decomposition machine is 0.2.

[0014] An electronic device comprises: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the described method.

[0015] A storage medium stores a computer program, which, when executed by a processor, implements the method described.

[0016] A computer program product comprises a computer program, which, when executed by a processor, implements the method described.

[0017] The present invention has at least the following beneficial effects: The present invention analyzes historical reaction data of condensation reactions between glyoxal and primary amine reaction substrates, combines high-throughput quantum chemical calculations with machine learning models, and automatically learns the complex relationship between substrate characteristics and reaction results. The constructed fusion model can not only process high-dimensional and nonlinear data, but also be applied to the prediction of new primary amine substrates under various conditions, and can accurately screen out primary amine substrates with higher possibilities in a short time, thereby guiding the implementation of actual experiments and avoiding blind synthesis. This "calculation first, experimental verification" strategy greatly improves research efficiency and saves experimental resources. The obtained isowurtzitane cage compound is used as a nitration precursor of CL-20, directly eliminating the hydrogenolysis step, and a huge breakthrough has been made in the synthesis research of CL-20. It has the practicality of being invested in the application research of new weapon equipment charges.

[0018] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a correlation matrix diagram of the feature descriptors of the present invention; Figure 2 Compare the prediction performance of various traditional machine learning models for the present invention; Figure 3 It is a ranking diagram of the importance of characteristic variables of the CSP model of the present invention; Figure 4 The figure is the distribution diagram of the chemical shift data of the amino nitrogen atom of all the primary amine substrates of the present invention. DETAILED DESCRIPTION

[0020] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0021] It should be understood that the terms such as “having”, “including” and “comprising” used herein do not exclude the existence or addition of one or more other elements or combinations thereof.

[0022] It should be noted that the experimental methods described in the following embodiments are all conventional methods unless otherwise specified, and the reagents and materials are all commercially available unless otherwise specified. They do not indicate or imply that the methods, reagents, materials or elements referred to must be specific, and therefore cannot be understood as limitations on the present invention.

[0023] The present invention provides a method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning. The isowurtzitane cage compound is a nitration precursor compound of hexanitrohexaazaisourtzitane CL-20, comprising: A database of condensation reactions of glyoxal with primary amine reaction substrates was established, and currently reported experimental literature was obtained from the SciFinder database. Condensation reactions whose products were isowurtzitane cage compounds were marked as 1, and condensation reactions whose products were other types of compounds were marked as 0.

[0024] The 3D initial molecular structure of the primary amine reaction substrate was obtained from the CCDC crystal database and the ZINC-20 database, and quantum chemical calculations were performed on the primary amine reaction substrate using Gaussian 16 software. The quantum chemical calculations included the following steps: geometric structure optimization, vibration analysis, and energy calculations. Descriptors describing chemical information were collected from the result files output by the quantum chemical calculations to form initial sample data. The correlation between the descriptors was analyzed, and then the feature descriptors were screened so that there was no obvious correlation between the feature descriptors. If the correlation coefficient was lower than a given threshold, it was considered that there was no obvious correlation between the two descriptors and they could be used as feature descriptors. The feature descriptors were normalized.

[0025] When collecting descriptors that describe chemical information, we combined relevant theoretical knowledge. For example, the frontier orbital theory believes that the reaction activity is mainly determined by the interaction between the highest occupied orbital (HOMO) and the lowest unoccupied orbital (LUMO) of the substrate. Previous studies have also shown that the electronic effect and steric hindrance effect of primary amine substrates are also major factors affecting the reaction results. Therefore, all possible influencing factors are fully considered in the selection of chemical descriptors to ensure that the collected descriptors can fully reflect the reactive characteristics of the substrate.

[0026] This application measures the degree of correlation between two variables by calculating the Pearson correlation coefficient between the descriptors. If the absolute value of the Pearson correlation coefficient between the two variables is greater than a given threshold, it is considered that the two features have a strong correlation. In order to reduce the redundancy and multicollinearity problems between features, it is necessary to filter out one of the features and keep only one. The Pearson correlation coefficient calculation formula is as follows: Characteristic descriptors include HOMO, LUMO, bond length and angle, atomic charge, NMR chemical shift, etc.

[0027] Central standardization is used to normalize all sample feature values, expressed as: in, It is In the sample The standardized value of the feature, It is In the sample The original value of the feature, It is The mean of the features, It is The standard deviation of each feature. The feature descriptor is used as the input variable of machine learning, and the condensation reaction product labeling result (1 or 0) is used as the output variable of machine learning. Random forest and factor decomposition machine are selected for weight allocation to build a fusion model CSP (Cage Structure Product), and the model is trained and evaluated. 5-fold cross validation is used for training and evaluation. The data set is divided into 5 parts, 4 of which are used as training sets and 1 as test sets in turn. The model is trained on the training set and evaluated on the test set. Finally, the results of the 5 times are averaged to obtain the final result, and compared according to its prediction performance on the given data set (such as accuracy, recall rate, F1 value, etc.) to obtain the optimal fusion model. The fusion model can combine the advantages of each other to improve the overall prediction effect.

[0028] The optimal fusion model was used to predict whether glyoxal and new primary amine reaction substrates could generate isowurtzitane cage compounds through condensation reaction.

[0029] In order to further improve the prediction performance of the CSP model, an adaptive weight optimization strategy is introduced. By dynamically adjusting the weight distribution of the model, the optimal weight combination is found and applied to the fusion model. The data set is trained to obtain the corresponding model parameters and output the CSP model trained based on the existing data set.

[0030] In the above technical scheme, multiple descriptors that can describe the molecular structure, electronic and energy characteristics are obtained through quantum chemical calculations. Machine learning of the screened characteristic descriptors can achieve fast and accurate predictions, and accurately screen out primary amine substrates with a higher possibility of cage formation, thereby avoiding blind synthesis experiments, providing an effective guiding basis for the synthesis of isowurtzitane cage compounds, improving experimental efficiency, saving experimental resources, and ensuring accurate and reliable results.

[0031] <Example 1> A method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning, comprising: S1: Collect the experimental data of condensation reactions of glyoxal and primary amines published in the literature, a total of 91 chemical reactions, of which 37 reactions produced isowurtzitane cage compounds and 54 produced other types of compounds; S2: Model the molecular structure of the primary amine substrate in the above reaction. First, obtain the initial 3D molecular structure from the CCDC and ZINC-20 databases. Then, use the B3LYP / 6-31G* theoretical method in Gaussian16 software to perform geometry optimization and vibration analysis. Use the M062X / def2-TZVP theoretical method to calculate the energy of the optimized structure. Then, use Multiwfn software and Python code to analyze and collect the output result files.

[0032] Relevant chemical information was collected from the result files output by quantum chemical calculations to form initial sample data, and 28 relevant chemical descriptors were obtained, including HOMO, LUMO, bond length and angle, atomic charge, NMR chemical shift, etc. Feature selection was performed by correlation filtering, and the threshold was set to 0.8. Only one feature was retained from the feature pairs with correlation greater than 0.8, and 16 chemical descriptors were finally obtained, including the highest occupied molecular orbital energy, the lowest unoccupied molecular orbital energy, bond length, bond angle, dipole moment, Hirshfeld charge, Fukui function, dual descriptor, electronegativity, nucleophilicity, Rydberg electron density, total electron density, van der Waals surface area, molecular volume, total energy and NMR chemical shift.

[0033] S3: Perform correlation analysis on the above features. The correlation matrix of this embodiment is as follows: Figure 1 As shown in the figure, the correlation coefficients between the features are all lower than 0.8, indicating that there is no obvious correlation between the features.

[0034] S4: The above characteristic variables are used as the input variables of machine learning, and the labels of the reaction results (cage-like products are 1 and other products are 0) are used as the output variables of machine learning. Central standardization is used to normalize the characteristic values ​​of all samples.

[0035] S5: Model training and evaluation: First, the random forest and factor decomposition machine models were selected for testing. The input of the model was the characteristic data of all samples (16 descriptors), and the output was the label of the sample (cage compound was 1, and other products were 0).

[0036] Through grid search for hyperparameter tuning, the optimal parameter combination of random forest was determined to be: maximum depth of 4, maximum number of leaf nodes of 100, and maximum number of bins of 128; the optimal parameters of the factorization machine were: learning rate 0.005, number of factors 5, L2 regularization coefficient 0.01, and number of training rounds 10.

[0037] The Cage Structure Product (CSP) fusion model based on random forest and factor decomposition machine is selected. Through the adaptive weight optimization strategy, the weight distribution of the model is dynamically adjusted to find the optimal weight combination (the weight of random forest is 0.8, and the weight of factor decomposition machine is 0.2).

[0038] 5-fold cross validation is used for data partitioning, model training and evaluation. The entire data set is randomly divided into 5 folds of equal size. Four of them are used as training sets and one as a test set in turn. The model is trained on the training set, and the model performance is evaluated on the test set. The relevant indicators (accuracy, recall rate and F1 value) are calculated.

[0039] S6: Repeat the above process several times, and take the average of the accuracy, recall and F1 values ​​obtained multiple times. In the end, the best performing CSP model had an accuracy of 91.4%, an F1 value of 89.1%, and a recall of 91.4%, indicating that the model had good prediction performance. The accuracy of the RF model was close to that of the CSP model, but its recall and F1 values ​​were poor. Although the recall of the FM model was higher than that of the CSP model, its accuracy and F1 values ​​were low. The CSP model showed stronger stability and comprehensiveness in various indicators. It not only maintained a high accuracy rate, but also took into account the good performance of recall and F1 values, reflecting its advantage in balancing prediction performance. Accuracy, recall and F1 values ​​are used to quantitatively evaluate the prediction performance of the model, and the optimal prediction model is saved according to the model performance. The calculation formula is as follows, and the explanation is as shown in Table 1.

[0040] Table 1 Actual positive example Actual negative examples Predict positive examples TP FP Predict negative examples FN TN This embodiment selects a variety of traditional machine learning models for training, including random forest, factorization machine, linear regression, support vector machine and gradient boosting tree, and uses grid search to tune hyperparameters to determine the best parameter combination for each model. Then, 5-fold cross validation is used for training and evaluation. The data set is divided into 5 parts, 4 of which are used as training sets and 1 as test sets in turn. The model is trained on the training set and evaluated on the test set. Finally, the results of the 5 times are averaged to obtain the final result, and compared based on their prediction performance on the given data set (such as accuracy, recall rate, F1 value, etc.). The prediction performance of all models is as follows: Figure 2 As shown in the figure, random forest RF and factorization machine FM performed relatively well. The fusion model CSP performed the best.

[0041] S7: According to the importance analysis results given by the fusion model CSP, it was found that the chemical shift of amino nitrogen atoms, amino surface area and total electron density of amino nitrogen atoms contribute greatly to the accuracy of the model output, and are important feature descriptors that affect reaction prediction. The MDA (Mean Decrease Accuracy) method is used to measure the contribution of each feature variable to the accuracy of the output variable and to rank the importance of the feature variables. Each feature descriptor is shuffled in turn, that is, the value of the feature in the test set is randomly rearranged while keeping other features unchanged, and then the accuracy of the model is re-evaluated. The contribution of the feature to the overall prediction performance of the model is evaluated by measuring the degree of decrease in model accuracy when the feature is randomly shuffled, such as Figure 3 shown.

[0042] The chemical shift of the most important characteristic amino nitrogen atom was further analyzed. Figure 4 The distribution difference of the chemical shift of amino nitrogen atoms between the two groups of samples (Group 0 and Group 1) is shown. As can be seen from the figure, the chemical shift distribution of amino nitrogen atoms in Group 0 (the product is not a cage compound) is wider and has greater discreteness, with values ​​ranging from approximately 160 to 240. This more dispersed distribution indicates that the amino nitrogen atoms in this group of samples may have diverse electronic environments and reactivity. The chemical shift of amino nitrogen atoms in Group 1 (the product is a cage compound) is concentrated in a narrower range, with values ​​concentrated between approximately 190-210, and the distribution is more concentrated, showing less discreteness. This more concentrated chemical shift distribution may indicate that the amino nitrogen atoms in this group have similar electronic environments and reaction site characteristics, which is conducive to the occurrence of cage-forming reactions.

[0043] The results of importance analysis are expected to be applied to further optimize the CSP of the fusion model. By removing unimportant features or assigning higher weights to important features, the decision-making process can be more dependent on important features, thereby improving prediction accuracy. At the same time, the interaction of these key features can be analyzed to explore whether there are potential nonlinear relationships or combined features between them, thereby enhancing model performance.

[0044] The present invention also provides an electronic device, comprising: at least one processor, and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the method. The electronic device can be any terminal device including a mobile phone, a laptop computer, a desktop computer, a tablet computer, a PDA (Personal Digital Assistant), a POS (Point of Sales), a vehicle-mounted computer, etc.

[0045] The present invention also provides a storage medium on which a computer program is stored, and when the program is executed by a processor, the method described is implemented. Through the description of the above implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software plus necessary general hardware, and of course it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present invention, software program implementation is a better implementation method in more cases. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0046] The present invention also provides a computer program product, which includes a computer program, which, when executed by a processor, implements the method described. The computer program (which may also be referred to or described as a program, software, software application, module, software module, script or code) can be written in any form of programming language, including a compiled language or an interpreted language or a declarative language or a procedural language, and the computer program can be expanded in any form, including as an independent program or as a module, component, subroutine or other unit suitable for use in a computing environment. The computer program may but does not necessarily correspond to a file in a file system. The program may be stored in a part of a file that stores other programs or data, for example, one or more scripts stored in the following: in a markup language document; in a single file dedicated to a related program; or in multiple collaborative files, for example, files storing one or more modules, subroutines or code portions. The computer program can be expanded to execute on one or more computers, the computers being located in one place, or distributed to multiple locations and interconnected through a communication network. The processing and logic flow described in this specification can be performed by one or more programmable computers, which execute one or more computer programs by operating input data and generating output to run functions.

[0047] The number of devices and processing scales described here are used to simplify the description of the present invention. Applications, modifications and variations of the present invention will be obvious to those skilled in the art.

[0048] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning, wherein the isowurtzitane cage compound is a nitration precursor compound of hexanitrohexaazaisowurtzitane, characterized in that: include: A database of condensation reactions of glyoxal and primary amine reaction substrates is established, and the condensation reactions whose products are isowurtzitane cage compounds are marked as 1, and the condensation reactions whose products are other types of compounds are marked as 0; Perform quantum chemical calculations on primary amine reaction substrates, collect descriptors that describe chemical information, screen the feature descriptors so that there is no obvious correlation between the feature descriptors, and normalize the feature descriptors; The feature descriptor is used as the input variable of machine learning, and the condensation reaction product labeling result is used as the output variable of machine learning. Random forest and factor decomposition machine are selected for weight allocation to build a fusion model, and the model is trained and evaluated to obtain the optimal fusion model. The optimal fusion model was used to predict whether glyoxal and new primary amine reaction substrates could generate isowurtzitane cage compounds through condensation reaction.

2. The method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning according to claim 1, characterized in that: Quantum chemical calculations include the following steps: geometry optimization, vibration analysis, and energy calculations.

3. The method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning according to claim 1, characterized in that: The correlation analysis of the above descriptors shows that the Pearson correlation coefficients between the descriptors are all lower than 0.8, indicating that there is no obvious correlation between the descriptors as feature descriptors.

4. The method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning according to claim 1, characterized in that: Characteristic descriptors include highest occupied molecular orbital energy, lowest unoccupied molecular orbital energy, bond length, bond angle, dipole moment, Hirshfeld charge, Fukui function, dual descriptor, electronegativity, nucleophilicity, Rydberg electron density, total electron density, van der Waals surface area, molecular volume, total energy, and NMR chemical shift.

5. The method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning according to claim 1, characterized in that: The best parameter combination for random forest is: maximum depth of 4, maximum number of leaf nodes of 100, and maximum number of bins of 128.

6. The method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning according to claim 1, characterized in that: The best parameter combination for the factorization machine is: learning rate 0.005, number of factors 5, L2 regularization coefficient 0.01, and number of training rounds 10.

7. The method for predicting reaction substrates of isowurtzitane cage compounds based on machine learning according to claim 1, characterized in that: The random forest and factor decomposition machine are weighted to construct a fusion model, with the weight of the random forest being 0.8 and the weight of the factor decomposition machine being 0.

2.

8. An electronic device, characterized in that include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor performs the method described in any one of claims 1 to 7.

9. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method described in any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, characterized in that When the program is executed by a processor, the method described in any one of claims 1 to 7 is implemented.

Images