A method and system for intelligent recommendation of C-N coupling reaction conditions

Abstract

The application relates to a C-N coupling reaction condition intelligent recommendation method, system, device and medium. The method comprises the following steps: acquiring target C-N coupling reactant characteristics and C-N coupling reaction condition characteristic sets, and splicing the two to obtain respective target characteristic vectors; inputting the target characteristic vectors into a yield prediction neural network model to obtain corresponding predicted yield data; based on Monte Carlo estimation of the model, calculating an uncertainty index on each target characteristic vector; based on the predicted yield data and the uncertainty index, calculating a recommendation index parameter of each reaction condition characteristic; and outputting an optimal recommended reaction condition according to the parameter. The method can provide yield prediction with uncertainty quantification, guide the model to continuously optimize itself in application, and effectively improve the credibility and adaptive ability of the recommendation system.

Description

Technical Field

[0001] This invention belongs to the field of CN coupling reaction, and in particular relates to a method and system for intelligent recommendation of CN coupling reaction conditions. Background Technology

[0002] Palladium-catalyzed CN coupling (Buchwald-Hartwig amination) is a crucial reaction in drug development. However, its outcome is highly dependent on reactant structure and reaction conditions (ligands, bases, solvents), making condition optimization a time-consuming and labor-intensive bottleneck. With the deep integration of cheminformatics and artificial intelligence, data-driven machine learning models have been widely applied in the field of chemical reaction yield prediction. This technology can automatically uncover the complex nonlinear relationship between reaction conditions and experimental results. Its core feature lies in achieving rapid virtual screening through pattern recognition, thus forming the current mainstream intelligent recommendation paradigm based on static models.

[0003] In traditional techniques, the standard processing flow is to first collect historical experimental data and complete feature encoding, and then train a fixed yield prediction model. When faced with a new synthesis target, the system inputs candidate reaction conditions into the model to obtain predicted yields, and then ranks and recommends conditions based on the yield values.

[0004] However, this traditional paradigm suffers from two significant, interconnected drawbacks. First, the predictions provided by the system lack necessary credibility assessment. The model can only offer a single yield estimate, completely failing to measure the inherent uncertainty of this prediction when faced with unfamiliar reactants or novel ligands, thus exposing researchers to a high risk of failure when exploring unknown chemical spaces. Second, another limitation is the system's lack of self-learning and continuous evolution capabilities. Once trained, the model's knowledge base becomes fixed, unable to automatically identify its own knowledge gaps in subsequent recommendation and experimental cycles, nor can it utilize newly generated experimental evidence to optimize model parameters and expand its cognitive boundaries. This static characteristic severely restricts the system's practical value in long-term and dynamically evolving scientific research activities.

[0005] Predictive models based on historical literature data suffer from severe data bias (uneven distribution of reagents and yields), resulting in extremely poor generalization ability when faced with new and unseen reactants, leading to unreliable predictions. They cannot learn from new experimental feedback and cannot adapt to new chemical spaces. Existing models typically only provide yield predictions but fail to inform users of the reliability of these predictions, making it difficult for users to identify erroneous predictions and resulting in low trust levels. Summary of the Invention

[0006] Based on this, it is necessary to provide a method, system, device, and medium for intelligent recommendation of CN coupling reaction conditions that can quantify and predict uncertainties and achieve self-iterative optimization, in order to address the above-mentioned technical problems.

[0007] Firstly, this application provides a method for intelligent recommendation of CN coupling reaction conditions, including:

[0008] S1. Obtain the target CN coupling reactant features and CN coupling reaction condition feature set, and concatenate the target CN coupling reactant features with each CN coupling reaction condition feature in the CN coupling reaction condition feature set to obtain the target CN coupling reactant feature vector corresponding to each target CN coupling reactant feature.

[0009] S2. Input the feature vectors of each target CN coupling reaction into the yield prediction neural network model to obtain the predicted CN coupling reaction yield data corresponding to the feature vectors of each target CN coupling reaction.

[0010] S3. Monte Carlo estimation based on the yield prediction neural network model, setting the uncertainty index of the yield prediction neural network model on the eigenvectors of each target CN coupled reaction;

[0011] S4. Based on the predicted yield data and uncertainty index of CN coupling reaction, the recommended index parameters of each CN coupling reaction condition feature in the CN coupling reaction condition feature set are calculated.

[0012] S5. Based on the recommended index parameters, the optimal CN coupling reaction condition features are selected from the CN coupling reaction condition feature set, and the CN coupling reaction conditions corresponding to the optimal CN coupling reaction conditions are set as the optimal recommended CN coupling reaction conditions for the target CN coupling reactant corresponding to the target CN coupling reactant features.

[0013] In one embodiment, after inputting the feature vectors of each target CN-coupled reaction into a yield prediction neural network model to obtain the predicted CN-coupled reaction yield data corresponding to the feature vectors of each target CN-coupled reaction, the method further includes:

[0014] S11. Map the target CN coupling reaction feature vector to the target CN coupling reaction chemical space t-SNE diagram to obtain the target CN coupling reaction point;

[0015] S12. Calculate the minimum Euclidean distance between the target CN coupling reaction point and the nearest neighbor CN coupling reaction point in the t-SNE diagram of the chemical space of the target CN coupling reaction point;

[0016] S13. Input the minimum Euclidean distance of the target point into the uncertainty regression prediction model to generate the prediction error estimate of the yield prediction neural network model under the target CN coupled reaction feature vector.

[0017] S14. If the estimated prediction error is lower than the yield prediction reliability threshold, generate a reliability warning message for the target CN coupled reaction feature vector.

[0018] Based on the above embodiments, the method further includes:

[0019] S21. Map the characteristics of each CN coupling reactant to the t-SNE diagram of CN coupling reaction chemical space to obtain the CN coupling reaction sites;

[0020] S22. Calculate the minimum Euclidean distance between each CN-coupled reaction point and the nearest neighbor CN-coupled reaction point in the t-SNE diagram of the chemical space of CN-coupled reactions.

[0021] S23. Calculate the actual absolute error of the yield prediction neural network model on the CN-coupled reaction training samples in the CN-coupled reaction training set corresponding to the CN-coupled reaction point. The actual absolute error of the prediction is used to characterize the uncertainty of the yield prediction neural network model on the CN-coupled reaction training samples.

[0022] S24. An uncertainty regression prediction model is constructed based on the minimum Euclidean distance between training points and the actual absolute error of prediction.

[0023] In one embodiment, the method further includes:

[0024] S31. Identify the target CN coupling reaction feature vector with an uncertainty index higher than the uncertainty index threshold as a high-uncertainty target CN coupling reaction feature.

[0025] S32. Analyze the characteristics of the highly uncertain target CN coupling reaction and obtain the experimental verification parameters of the CN coupling reaction;

[0026] S33. Obtain the actual CN coupling reaction yield data corresponding to the experimental verification parameters of CN coupling reaction characteristics of each high-uncertainty target, and construct the CN coupling reaction enhancement samples based on the CN coupling reaction characteristics of each high-uncertainty target and the actual CN coupling reaction yield data corresponding to the CN coupling reaction characteristics of each high-uncertainty target.

[0027] S34. Using the high-uncertainty target CN coupling reaction characteristics of each CN coupling reaction enhancement sample as the model optimization input of the yield prediction neural network model, and using the real CN coupling reaction yield data of each CN coupling reaction enhancement sample as the model optimization label of the yield prediction neural network model, the yield prediction neural network model is updated and optimized.

[0028] In one embodiment of the present invention, obtaining the target CN coupling reactant characteristics and the CN coupling reaction condition characteristic set includes:

[0029] S41. Obtain structural information of the target reactant and candidate ligands;

[0030] S42. Based on the target reactant structure information and the candidate ligand structure information, construct the target reactant structure Morgan fingerprint vector and the candidate ligand structure Morgan fingerprint vector respectively, and splice the target reactant structure Morgan fingerprint vector and the candidate ligand structure Morgan fingerprint vector to obtain the target CN-coupled reactant features.

[0031] S43. Obtain the set of CN coupling reaction conditions, perform one-hot encoding on each CN coupling reaction condition in the set of CN coupling reaction conditions to obtain the characteristics of each CN coupling reaction condition, and summarize the characteristics of each CN coupling reaction condition to construct the CN coupling reaction condition feature set.

[0032] Secondly, this application also provides a CN-coupled reaction condition intelligent recommendation system for implementing the method described in the first aspect, comprising:

[0033] The feature acquisition and splicing module is used to acquire the target CN coupling reactant features and the CN coupling reaction condition feature set, and splice the target CN coupling reactant features with each CN coupling reaction condition feature in the CN coupling reaction condition feature set to obtain the target CN coupling reactant feature vectors corresponding to each target CN coupling reactant feature.

[0034] The yield prediction module is used to input the feature vectors of each target CN coupling reaction into the yield prediction neural network model to obtain the predicted CN coupling reaction yield data corresponding to the feature vectors of each target CN coupling reaction.

[0035] The uncertainty prediction module is used for Monte Carlo estimation based on the yield prediction neural network model, and sets the uncertainty index of the yield prediction neural network model on the eigenvectors of each target CN coupled reaction.

[0036] The recommended index calculation module is used to calculate the recommended index parameters for each CN coupling reaction condition feature in the CN coupling reaction condition feature set based on the predicted CN coupling reaction yield data and uncertainty index.

[0037] The optimal condition screening module is used to screen the optimal CN coupling reaction condition features from the CN coupling reaction condition feature set based on the recommended index parameters, and set the CN coupling reaction conditions corresponding to the optimal CN coupling reaction conditions of the target CN coupling reactant corresponding to the target CN coupling reactant features as the optimal recommended CN coupling reaction conditions of the target CN coupling reactant.

[0038] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0039] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0040] The aforementioned CN-coupled reaction condition intelligent recommendation method and system concatenates the features of the target reactant with the features of each candidate condition and inputs the result into a yield prediction neural network. It then uses Monte Carlo estimation to quantify the uncertainty index predicted by the model, and combines the predicted yield and uncertainty to calculate a recommendation index, ultimately selecting the optimal reaction condition. This method not only provides yield predictions with uncertainty assessment, helping researchers identify high-risk, unfamiliar reaction spaces, but also, by introducing an uncertainty-guided reinforcement learning mechanism, enables continuous self-optimization and knowledge expansion of the model during application. This effectively overcomes the technical limitations of traditional static recommendation systems, such as low reliability and inability to adaptively evolve. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart of a method for intelligent recommendation of CN coupling reaction conditions in one embodiment of the present invention;

[0043] Figure 2 This is a schematic diagram of the intelligent recommendation system for CN coupling reaction conditions in one embodiment of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0045] In one embodiment, such as Figure 1 As shown, a CN-coupled reaction condition intelligent recommendation method is provided. This embodiment illustrates the application of this method to a reaction condition recommendation terminal. It is understood that this method can also be applied to a server, and further to a system including both a reaction condition recommendation terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0046] S1. Obtain the target CN coupling reactant features and CN coupling reaction condition feature set, and concatenate the target CN coupling reactant features with each CN coupling reaction condition feature in the CN coupling reaction condition feature set to obtain the target CN coupling reactant feature vector corresponding to each target CN coupling reaction feature.

[0047] Specifically, the characteristics of CN-coupled reactants can include molecular structure, electronic properties, and physicochemical properties. The reaction condition recommendation terminal can extract the molecular fingerprint by parsing the SMILES string to obtain the molecular structure, calculate the electronic parameters using density functional theory to obtain the electronic properties, and obtain the physicochemical properties by querying the database or obtaining experimentally measured physicochemical parameters.

[0048] For example, the reaction condition recommendation terminal can remove outliers from the CN-coupled reactant features through robust normalization to obtain the robustly normalized reactant features. The robust normalization expression can be:

[0049]

[0050] In the formula, To identify the characteristics of the reactants after robust normalization. Characteristic of CN-coupled reactants This represents the median characteristic of CN-coupled reactants. The interquartile range is characteristic of CN-coupled reactants.

[0051] Specifically, the CN coupling reaction condition feature set can be constructed based on screened historical experimental data. The CN coupling reaction condition feature set includes seven dimensions: catalyst, ligand, base, solvent, reaction temperature, reaction time, and reactant molar ratio. The reaction condition recommendation terminal can obtain the CN coupling reaction condition features in each dimension of the CN coupling reaction condition feature set after quantitative processing.

[0052] For example, the response condition recommendation terminal can remove outliers from the CN-coupled response condition feature set using box plots; subsequently, the response condition recommendation terminal can calculate the interpolation results for missing values ​​using a weighted interpolation method based on feature similarity. The expression for the weighted interpolation method can be:

[0053]

[0054] In the formula, Interpolation results representing missing values, Representing the Feature similarity between samples with no missing data and samples with missing data Representing the The feature values ​​corresponding to each sample without missing data. This represents the number of samples without missing values. The reaction condition recommendation terminal can supplement the missing values ​​into the CN-coupled reaction condition feature set, forming a complete CN-coupled reaction condition feature set. The reaction condition recommendation terminal can then use an element-wise concatenation method to combine the complete CN-coupled reaction condition feature set with the robustly standardized reactant features to obtain the target CN-coupled reaction feature vectors corresponding to the target CN-coupled reactant features.

[0055] S2. Input the feature vectors of each target CN coupling reaction into the yield prediction neural network model to obtain the predicted CN coupling reaction yield data corresponding to the feature vectors of each target CN coupling reaction.

[0056] Specifically, the yield prediction neural network model can include three layers: an input layer, a hidden layer, and an output layer. The input dimension of the input layer can be the same as the dimension of the feature vector of the target CN-coupled reaction; multiple hidden layers can be set, and each layer uses a modified ReLU activation function to optimize the feature mapping and obtain the feature mapping value; the output layer can be used to output the original predicted value of the yield prediction neural network model.

[0057] For example, the reaction condition recommendation terminal can input the feature vectors of each target CN-coupled reaction into the input layer of the yield prediction neural network model for model input preprocessing to eliminate dimensional interference and obtain the standard input target CN-coupled reaction feature vector. The reaction condition recommendation terminal can then input the standard input target CN-coupled reaction feature vector into the hidden layer, and calculate the high-dimensional feature mapping value using the activation function built into the hidden layer. The expression for the activation function can be:

[0058]

[0059] In the formula, For high-dimensional feature mapping values, For adaptive adjustment coefficient, This is the preprocessed feature vector of the target CN-coupled reaction. The reaction condition recommendation terminal can input the high-dimensional feature mapping value into the output layer, calculate the original predicted value of the model through the Sigmoid activation function, and use the Platt scaling method to calibrate the accuracy of the original predicted value of the model, correct the prediction bias, and obtain the predicted CN-coupled reaction yield data corresponding to the feature vector of the target CN-coupled reaction.

[0060] S3. Monte Carlo estimation based on the yield prediction neural network model: set the uncertainty index of the yield prediction neural network model on the eigenvectors of each target CN coupled reaction.

[0061] For example, the reaction condition recommendation terminal can calculate the dispersion based on the predicted yield set corresponding to the target CN-coupled reaction feature vector, and obtain the weighted standard deviation. The formula for calculating the weighted standard deviation is as follows:

[0062]

[0063] in, For weighted standard deviation, For the predicted yield set, the first Secondary predicted yield, To predict the average yield, To predict the number of times, The weights for the predicted results are used. The response condition recommendation terminal can calculate the initial uncertainty index using the coefficient of variation algorithm based on the weighted standard deviation and the mean predicted yield. The expression for calculating the initial uncertainty index is as follows:

[0064]

[0065] In the formula, This is the initial uncertainty index. The maximum value of the predicted yield corresponding to the feature vector of the target CN coupled reaction is given; the reaction condition recommendation terminal performs min-max normalization on the initial uncertainty index to obtain the uncertainty index.

[0066] S4. Based on the predicted yield data and uncertainty index of CN coupling reaction, the recommended index parameters of each CN coupling reaction condition feature in the CN coupling reaction condition feature set are calculated.

[0067] For example, the reaction condition recommendation terminal can perform min-max normalization on the predicted CN coupling reaction yield data to eliminate dimensional interference and obtain the predicted yield normalized value. The calculation expression for the predicted yield normalized value can be:

[0068]

[0069] In the formula, To predict the normalized yield, To predict the yield data of CN coupling reaction, To predict the median yield data for CN coupling reaction, To predict the lower quartile of CN coupling reaction yield data, To predict the upper quartile of CN coupling reaction yield data, the feasibility coefficients corresponding to the characteristics of each CN coupling reaction condition are calculated at the recommended terminal stage. The expression for calculating the feasibility coefficient can be:

[0070]

[0071] In the formula, The feasibility coefficient, For the first The weight of each dimension in the overall evaluation For the first Standardized scores across multiple dimensions. The response condition recommendation terminal inputs the uncertainty index, the normalized predicted yield value, and the feasibility coefficient into a weighted fusion calculation formula to obtain the recommendation index parameters. The calculation formula for the recommendation index parameters can be:

[0072]

[0073] in, Recommended indicator parameters, To predict the normalized yield, It is an uncertainty index; , and For the weighting coefficients, satisfying + + =1 and ; The higher the value, the higher the overall recommended priority for the reaction conditions. Reaction conditions with values ​​below the preset recommended threshold are marked as low-priority reaction conditions.

[0074] S5. Based on the recommended index parameters, the optimal CN coupling reaction condition features are selected from the CN coupling reaction condition feature set, and the CN coupling reaction conditions corresponding to the optimal CN coupling reaction conditions are set as the optimal recommended CN coupling reaction conditions for the target CN coupling reactant corresponding to the target CN coupling reactant features.

[0075] For example, the reaction condition recommendation terminal can, based on the recommended index parameters and the corresponding reaction condition features, eliminate ineffective prediction reaction conditions that are difficult to meet normal reactions, low-priority reaction conditions, and reaction conditions with excessively low feasibility coefficients, thus obtaining a candidate set of effective reaction condition features. The reaction condition recommendation terminal sorts each reaction condition feature in the candidate set according to its corresponding recommended index parameter value from largest to smallest, selecting the reaction condition feature with the largest recommended index parameter value as the optimal CN-coupled reaction condition feature. The reaction condition recommendation terminal can parse the optimal CN-coupled reaction condition feature, extract the corresponding reaction condition parameters, convert them into experimental parameters, and organize them into standardized and operable expressions to obtain the optimal recommended CN-coupled reaction conditions. The reaction condition recommendation terminal can output auxiliary information such as the predicted yield and uncertainty index corresponding to the optimal CN-coupled reaction condition feature, record the entire screening process, and generate a report.

[0076] This application provides a CN-coupled reaction condition intelligent recommendation method. The method concatenates the features of the target reactant with the features of each candidate condition and inputs the result into a yield prediction neural network. It then uses Monte Carlo estimation to quantify the uncertainty index predicted by the model, and combines the predicted yield with the uncertainty to calculate a recommendation index, ultimately selecting the optimal reaction condition. This method not only provides yield predictions with uncertainty assessment, helping researchers identify high-risk, unfamiliar reaction spaces, but also, by introducing an uncertainty-guided reinforcement learning mechanism, enables continuous self-optimization and knowledge expansion of the model during application. This effectively overcomes the technical limitations of traditional static recommendation systems, such as low reliability and inability to adaptively evolve.

[0077] In one embodiment of the present invention, after inputting the feature vectors of each target CN-coupled reaction into a yield prediction neural network model to obtain the predicted CN-coupled reaction yield data corresponding to the feature vectors of each target CN-coupled reaction, the method further includes:

[0078] S11. Map the target CN coupling reaction feature vector corresponding to the target CN coupling reactant feature to the CN coupling reaction chemical space t-SNE diagram to obtain the target CN coupling reaction point.

[0079] Specifically, the CN-coupled reaction chemical space t-SNE diagram is constructed based on the CN-coupled reaction training set. The construction process is as follows: extract the reactant features of all samples in the training set, perform improved robust normalization, and then use the t-SNE algorithm to perform dimensionality reduction, mapping the high-dimensional reactant features to a 2-dimensional visualization space to generate the CN-coupled reaction chemical space t-SNE diagram. Each data point in the diagram corresponds to the reactant feature of a sample in the training set.

[0080] For example, the reaction condition recommendation terminal can extract the target CN coupling reactant features corresponding to the feature vectors of each target CN coupling reaction. The reaction condition recommendation terminal can standardize the target CN coupling reactant features using a robust standardization algorithm to obtain standardized target CN coupling reactant features, and map the standardized target CN coupling reactant features to the CN coupling reaction chemical space t-SNE diagram to obtain the target CN coupling reaction points.

[0081] S12. Calculate the minimum Euclidean distance between the target CN coupling reaction point and the nearest neighbor CN coupling reaction point in the t-SNE diagram of the chemical space of the target CN coupling reaction point.

[0082] For example, the reaction condition recommendation terminal can use a two-dimensional Euclidean distance calculation method to calculate the Euclidean distance between the target CN-coupled reaction point and the nearest neighbor CN-coupled reaction point in the CN-coupled reaction chemical space t-SNE diagram. The reaction condition recommendation terminal iterates through all calculated Euclidean distances, selects the Euclidean distance with the smallest value, and determines the smallest Euclidean distance as the target point's minimum Euclidean distance between the target CN-coupled reaction point and the nearest neighbor CN-coupled reaction point in the CN-coupled reaction chemical space t-SNE diagram. The expression for calculating the Euclidean distance can be:

[0083]

[0084] In the formula, For Euclidean distance, Let the coordinates of the target CN-coupled reaction point in the t-SNE two-dimensional visualization space be given. The coordinates of the coupled reaction points in a single training set CN in two-dimensional space.

[0085] S13. Input the minimum Euclidean distance of the target point into the uncertainty regression prediction model to generate the prediction error estimate of the yield prediction neural network model under the target CN coupled reaction eigenvector.

[0086] Specifically, the uncertainty regression prediction model can employ an improved linear regression structure. The reaction condition recommendation terminal can construct a CN-coupled reaction training set to train the uncertainty regression prediction model. This CN-coupled reaction training set can include the minimum Euclidean distance between training points and the actual absolute prediction error. The minimum Euclidean distance between training points can be the minimum Euclidean distance calculated between the reaction point in the CN-coupled reaction training set after mapping the reactant features of each sample to the t-SNE diagram and the reaction points in other CN-coupled reaction training sets in the t-SNE diagram. The actual absolute prediction error can refer to the actual absolute prediction error of the yield of the reaction sample in the CN-coupled reaction training set, i.e., the absolute value of the difference between the actual yield after the reaction sample correction and the yield predicted by the yield prediction neural network model. During the training of the uncertainty regression prediction model, the minimum Euclidean distance between training points can be used as the input feature, and the actual absolute prediction error as the training label to complete the parameter tuning and training of the uncertainty regression prediction model.

[0087] For example, the reaction condition recommendation terminal can input the minimum Euclidean distance of the target CN-coupled reaction point into the uncertainty regression prediction model. The uncertainty regression prediction model can perform standardized preprocessing on the minimum Euclidean distance of the target point, and then output the predicted value of the actual absolute error of the yield prediction corresponding to the target through preset linear mapping, parameter calibration and other processing procedures. The prediction error estimate is calculated through the reliability prediction function, and the expression of the reliability prediction function can be:

[0088]

[0089] In the formula, This is the estimated value of the prediction error. This is the predicted absolute error of the actual yield forecast. This represents the minimum actual prediction absolute error for all samples in the CN-coupled reaction training set. This represents the maximum actual prediction absolute error for all samples in the CN-coupled reaction training set. This is the minimum point.

[0090] S14. If the estimated prediction error is lower than the yield prediction reliability threshold, generate a reliability warning message for the target CN coupled reaction feature vector.

[0091] Specifically, the reliability early warning information includes the target CN coupling reaction feature vector identifier, the minimum Euclidean distance of the target point corresponding to the target CN coupling reaction feature vector, the prediction error estimate, and the early warning prompt.

[0092] For example, the reaction condition recommendation terminal can set a yield prediction reliability threshold by calculating the correspondence between the prediction error estimate of samples in the CN-coupled reaction training set and the actual yield prediction error. The reaction condition recommendation terminal can compare the prediction error estimate corresponding to each target CN-coupled reaction feature vector with the yield prediction reliability threshold. If the prediction error estimate corresponding to the target CN-coupled reaction feature vector is lower than the yield prediction reliability threshold, the reaction condition recommendation terminal can generate a reliability warning message for the target CN-coupled reaction feature vector. If the prediction error estimate is not lower than the yield prediction reliability threshold, no warning message is generated, and the prediction error estimate result is determined to be reliable.

[0093] This application provides an intelligent recommendation method for CN-coupled reaction conditions. This method maps the characteristics of the target CN-coupled reactant to the t-SNE chemical space, calculates the minimum Euclidean distance of the target point, and generates a prediction error estimate by combining the minimum Euclidean distance of the training points with the actual prediction absolute error of the uncertainty regression prediction model. Early warning is then provided through threshold comparison. This method can accurately quantify the reliability of the yield prediction neural network model's prediction results, effectively identify low-reliability prediction samples and suggest optimization directions, avoid interference from low-reliability predictions in subsequent processes, and ensure the accuracy and reliability of the entire CN-coupled reaction condition recommendation process.

[0094] In one embodiment of the present invention, the intelligent recommendation method for CN coupling reaction conditions may further include:

[0095] S21. Map the characteristics of each CN coupling reactant to the t-SNE diagram of CN coupling reaction chemical space to obtain each CN coupling reaction point.

[0096] Specifically, the CN-coupled reaction chemical space t-SNE diagram can be constructed from the CN-coupled reaction training set. The reaction condition recommendation terminal can extract the reactant features of all samples in the CN-coupled reaction training set, perform robust standardization on the reactant features, and obtain standardized reactant features. The reaction condition recommendation terminal can use the t-SNE algorithm to perform dimensionality reduction calculation on the standardized reactant features to obtain a two-dimensional CN-coupled reaction chemical space t-SNE diagram.

[0097] For example, the reaction condition recommendation terminal can extract the CN-coupled reactant features corresponding to all samples in the constructed CN-coupled reaction training set. The terminal can remove reaction condition features from the CN-coupled reactant features, retaining features related to molecular structure, electronic properties, and physicochemical properties, and then standardize them to obtain a high-dimensional reactant feature vector. The terminal can then map this high-dimensional reactant feature vector to a preset CN-coupled reaction chemical space t-SNE diagram using a t-SNE mapping function to obtain each CN-coupled reaction point. The expression for the t-SNE mapping function can be:

[0098]

[0099] In the formula, For the first The characteristics of each CN-coupled reactant are mapped to the two-dimensional coordinates of the reaction points in the t-SNE plot. These two-dimensional coordinates can be used to determine the CN-coupled reaction points. For the first A high-dimensional reactant feature vector For t-SNE mapping hyperparameters, The average eigenvector of high-dimensional reactants Norm mean.

[0100] S22. Calculate the minimum Euclidean distance between each CN-coupled reaction point and the training point of the nearest neighbor CN-coupled reaction point in the t-SNE diagram of the chemical space of CN-coupled reactions.

[0101] For example, the reaction condition recommendation terminal can calculate the training point Euclidean distance between the CN-coupled reaction point and all other CN-coupled reaction points in the t-SNE diagram based on the CN-coupled reaction chemical space. The reaction condition recommendation terminal can traverse all the calculated training point Euclidean distances and filter out the training point Euclidean distance with the smallest value to obtain the minimum training point Euclidean distance.

[0102] S23. Calculate the actual absolute prediction error of the yield prediction neural network model on the CN-coupled reaction training samples in the CN-coupled reaction training set corresponding to the CN-coupled reaction point. The actual absolute prediction error is used to characterize the uncertainty of the yield prediction neural network model on the CN-coupled reaction training samples.

[0103] For example, the reaction condition recommendation terminal can input the reaction feature vectors of the CN-coupled reaction training samples corresponding to each CN-coupled reaction point into the yield prediction neural network model to obtain the predicted yield data for each training sample; the reaction condition recommendation terminal can obtain the actual yield data for each training sample, and input the actual yield data and the predicted yield data for each training sample into the absolute error formula to calculate the actual prediction absolute error, wherein the expression of the absolute error formula can be:

[0104]

[0105] In the formula, The actual absolute error of the prediction. For the first The reliability weight factor for each training sample. For training sample true yield data, To predict yield data for training samples, This is a system correction item. This is a random error correction term. This is a yield-related correction factor.

[0106] S24. An uncertainty regression prediction model is constructed based on the minimum Euclidean distance between training points and the actual absolute error of prediction.

[0107] For example, the reaction condition recommendation terminal can use the minimum Euclidean distance as training input and the actual prediction absolute error as training label to construct a training dataset for the uncertainty regression prediction model. The reaction condition recommendation terminal can input the training dataset into the initial uncertainty regression prediction model and calculate the mean absolute error using the mean absolute error loss function. The expression for the mean absolute error loss function can be:

[0108]

[0109] In the formula, The mean absolute error, For the sample size, The first in the training dataset for the uncertainty regression prediction model The actual absolute error of the regression prediction model obtained by inputting a sample with initial uncertainty The actual absolute error of the prediction is used. The response condition recommendation terminal can use the Adam optimizer to iteratively update the uncertainty regression prediction model. Each iteration has a preset number of iterations, and the mean absolute error of the validation subset is calculated. If the mean absolute error does not decrease significantly after multiple iterations, the iteration is stopped, and a preliminary uncertainty regression prediction model is obtained. The response condition recommendation terminal can evaluate the preliminary uncertainty regression prediction model. If the evaluation index meets the preset threshold, the preliminary uncertainty regression prediction model is used as the uncertainty regression prediction model. If the evaluation index does not meet the preset threshold, the model training process is restarted until the preliminary uncertainty regression prediction model meets the preset threshold.

[0110] This application provides an intelligent recommendation method for CN-coupled reaction conditions. This method standardizes the reactant features of CN-coupled reaction training samples and maps them to a t-SNE diagram. It then calculates the minimum Euclidean distance between training points and the absolute error of actual yield prediction. Based on these two core data sets, a regularized improved uncertainty regression prediction model is trained. This achieves a precise correlation between the minimum Euclidean distance and the prediction error, providing qualified and efficient model support for the subsequent reliability assessment of the yield prediction neural network model. This ensures the reliability and accuracy of the entire CN-coupled reaction condition recommendation process.

[0111] In one embodiment of the present invention, the intelligent recommendation method for CN coupling reaction conditions may further include:

[0112] S31. Identify the target CN coupling reaction feature vector with an uncertainty index higher than the uncertainty index threshold as a high-uncertainty target CN coupling reaction feature.

[0113] Specifically, the uncertainty index threshold can be selected by calculating the correspondence between the uncertainty index of the CN-coupled reaction training set samples and the actual yield prediction error.

[0114] For example, the reaction condition recommendation terminal can compare the normalized uncertainty index and uncertainty index threshold corresponding to each target CN coupling reaction feature vector one by one. If the uncertainty index of a target CN coupling reaction feature vector is higher than the uncertainty index threshold, the target CN coupling reaction feature vector is identified as a high-uncertainty target CN coupling reaction feature.

[0115] S32. Analyze the characteristics of the high-uncertainty target CN coupling reaction and obtain the experimental verification parameters of the CN coupling reaction.

[0116] For example, the reaction condition recommendation terminal can perform structured analysis on the identified high-uncertainty target CN coupling reaction features, decomposing the high-uncertainty target CN coupling reaction features to obtain CN coupling reactant features and CN coupling reaction condition features. The reaction condition recommendation terminal can then organize the analyzed CN coupling reactant features and CN coupling reaction condition features into standardized CN coupling reaction experimental verification parameters.

[0117] S33. Obtain the actual CN coupling reaction yield data corresponding to the experimental verification parameters of CN coupling reaction characteristics of each high-uncertainty target, and construct each CN coupling reaction enhancement sample based on the CN coupling reaction characteristics of each high-uncertainty target and the actual CN coupling reaction yield data corresponding to the CN coupling reaction characteristics of each high-uncertainty target.

[0118] For example, the reaction condition recommendation terminal can obtain experimental verification parameters corresponding to the characteristics of each highly uncertain target CN coupling reaction. Based on the experimental verification parameters, the reaction condition recommendation terminal can obtain the corresponding experimental verification measured data. The reaction condition recommendation terminal can input the measured yield data of the real CN coupling reaction into the yield correction function for correction, and obtain the corrected real CN coupling reaction yield data. The expression of the yield correction function can be:

[0119]

[0120] In the formula, This is the corrected actual yield data for CN coupling reactions. These are actual yield data for real CN-coupled reactions. The yield variation coefficient is used for experimental verification; the recommended reaction conditions terminal can correlate the characteristics of high-uncertainty target CN coupling reaction with the corrected real CN coupling reaction yield data to obtain enhanced CN coupling reaction samples.

[0121] S34. Using the high-uncertainty target CN coupling reaction characteristics of each CN coupling reaction enhancement sample as the model optimization input of the yield prediction neural network model, and using the real CN coupling reaction yield data of each CN coupling reaction enhancement sample as the model optimization label of the yield prediction neural network model, the yield prediction neural network model is updated and optimized.

[0122] For example, the reaction condition recommendation terminal can integrate CN-coupled reaction enhancement samples into a model optimization dataset, and divide the optimization dataset into an optimization training subset and an optimization validation subset according to a preset ratio. The reaction condition recommendation terminal can use the high-uncertainty target CN-coupled reaction features of each enhancement sample as the model optimization input of the yield prediction neural network model, and use the corrected real CN-coupled reaction yield data corresponding to each enhancement sample as the model optimization label to update and optimize the weight parameters and bias parameters of the yield prediction neural network model. The reaction condition recommendation terminal can use the mean absolute error as the loss function, use the Adam optimizer for iterative parameter updates, and calculate the root mean square error of the validation subset to optimize the yield prediction neural network model. When the yield prediction neural network model reaches a preset number of iterations or the root mean square error of the validation subset does not decrease significantly, the iteration stops, resulting in an updated and optimized yield prediction neural network model.

[0123] This application provides an intelligent recommendation method for CN-coupled reaction conditions. By identifying the characteristics of CN-coupled reactions with high uncertainty, analyzing experimental verification parameters, constructing enhanced samples containing real yield data, and using them for model optimization, this method accurately compensates for the prediction shortcomings of yield prediction neural network models on high uncertainty samples, effectively improving the model's prediction accuracy for high uncertainty targets. At the same time, it enriches the amount of training set data, enhances the overall generalization ability and prediction reliability of the model, and ensures the accuracy and practicality of CN-coupled reaction condition recommendations.

[0124] In one embodiment of the present invention, obtaining the target CN coupling reactant characteristics and the CN coupling reaction condition characteristic set includes:

[0125] S41. Obtain structural information of the target reactant and candidate ligands.

[0126] Specifically, the target reactant structural information refers to the structural data of the reactants to be reacted in the CN coupling reaction. The target reactant structural information may include the structural data of the target aryl halide and the target amine compound. The candidate ligand structural information may be the structural data of the potential ligands that are compatible with the target reactant. The candidate ligand structural information may include parameters such as the molecular configuration, functional group types and distribution, and atomic composition ratio of the candidate ligands.

[0127] For example, the reaction condition recommendation terminal can receive standardized structural information input by the user through the terminal's interactive interface; the terminal can also retrieve the target reactant structure information by calling a preset CN-coupled reaction reactant structure database and matching the target reactant's corresponding structural information based on the user-input target reactant name and molecular formula, etc. Candidate ligand structure information can be obtained through user input or from the database.

[0128] S42. Based on the target reactant structure information and the candidate ligand structure information, construct the target reactant structure Morgan fingerprint vector and the candidate ligand structure Morgan fingerprint vector respectively, and concatenate the target reactant structure Morgan fingerprint vector and the candidate ligand structure Morgan fingerprint vector to obtain the target CN-coupled reactant features.

[0129] For example, the reaction condition recommendation terminal can convert the structural information in SMILES string format into molecular structure objects to eliminate format differences; the reaction condition recommendation terminal can construct the Morgan fingerprint vector of the target reactant structure and the Morgan fingerprint vector of the candidate ligand structure respectively; the reaction condition recommendation terminal can concatenate the Morgan fingerprint vector of the target reactant structure first and the Morgan fingerprint vector of the candidate ligand structure last, and then perform normalization preprocessing on the obtained feature vector using a robust normalization algorithm to obtain the features of the target CN-coupled reactant.

[0130] For example, the fingerprint radius of the target reactant structure Morgan fingerprint vector and the candidate ligand structure Morgan fingerprint vector can be 2 and the number of fingerprint bits can be 1024. In the process of constructing the target reactant structure Morgan fingerprint vector and the candidate ligand structure Morgan fingerprint vector, the key reactive site features such as heteroatoms, double bonds, and triple bonds in the molecule can be retained, while the inactive inert structural features can be eliminated.

[0131] Specifically, the reaction condition recommendation terminal can construct a Morgan fingerprint vector of the target reactant structure based on the target reactant structure information. This Morgan fingerprint vector can be used to characterize the molecular structure and reactive sites of the target reactant. The Morgan fingerprint vector of the candidate ligand structure is a feature vector constructed based on the candidate ligand structure information. This Morgan fingerprint vector can be used to characterize the coordination activity, spatial structure, and other core features of the candidate ligand. The target CN-coupled reactant feature is a feature vector obtained by concatenating and normalizing the target reactant structure Morgan fingerprint vector and the candidate ligand structure Morgan fingerprint vector. This target CN-coupled reactant feature can be used for subsequent yield prediction and reaction condition recommendation. The SMILES string, or simplified linear input canonical string, is a standardized input format for the target reactant and candidate ligand structure information. The molecular structure object is standardized molecular structure data converted from the SMILES string. The SMILES string can be used to eliminate differences between different input formats.

[0132] S43. Obtain the set of CN coupling reaction conditions, perform one-hot encoding on each CN coupling reaction condition in the set of CN coupling reaction conditions to obtain the characteristics of each CN coupling reaction condition, and summarize the characteristics of each CN coupling reaction condition to construct the CN coupling reaction condition feature set.

[0133] For example, the reaction condition recommendation terminal can obtain a set of CN coupling reaction conditions. This set can include seven dimensions of reaction conditions required for the CN coupling reaction: catalyst, ligand, base, solvent, reaction temperature, reaction time, and reactant molar ratio. The terminal can perform one-heat encoding on each CN coupling reaction condition in the set, obtaining a one-heat encoded feature vector corresponding to each condition, i.e., the feature of each CN coupling reaction condition. The terminal can then aggregate and concatenate all categories of reaction condition feature vectors in a preset order to obtain a CN coupling reaction condition feature set. Each CN coupling reaction condition feature vector in the feature set corresponds to a specific combination of CN coupling reaction conditions, and these feature vectors can be used for screening and recommending target CN coupling reaction conditions.

[0134] Optionally, the reaction condition recommendation terminal can assign an independent feature dimension to each specific option under a reaction condition category based on multiple reaction condition categories. If a reaction condition belongs to that option, it is encoded as 1; otherwise, it is encoded as 0.

[0135] This application provides an intelligent recommendation method for CN-coupled reaction conditions. By standardizing the acquisition of target reactant and candidate ligand structural information, constructing standardized Morgan fingerprint vectors, performing one-hot encoding of reaction conditions, and constructing a feature set, this method achieves accurate extraction, standardized characterization, and structured integration of target CN-coupled reactant features and reaction condition features. This provides standardized and high-quality feature input support for subsequent yield prediction and reaction condition recommendation.

[0136] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0137] Based on the same inventive concept, this application also provides a CN-coupled reaction condition intelligent recommendation system for implementing the CN-coupled reaction condition intelligent recommendation method described above. The solution provided by this system is similar to the implementation scheme described in the above method; therefore, the specific limitations of one or more CN-coupled reaction condition intelligent recommendation system embodiments provided below can be found in the limitations of the CN-coupled reaction condition intelligent recommendation method described above, and will not be repeated here.

[0138] In one exemplary embodiment, such as Figure 2 As shown, a CN-coupled reaction condition intelligent recommendation system 500 is provided to implement the methods in the above-described method embodiments, including:

[0139] The feature acquisition and splicing module 501 can be used to acquire the target CN coupling reactant features and the CN coupling reaction condition feature set, and splice the target CN coupling reactant features with each CN coupling reaction condition feature in the CN coupling reaction condition feature set to obtain each target CN coupling reaction feature vector corresponding to the target CN coupling reactant features.

[0140] The yield prediction module 502 can be used to input the feature vectors of each target CN coupling reaction into the yield prediction neural network model to obtain the predicted CN coupling reaction yield data corresponding to the feature vectors of each target CN coupling reaction.

[0141] The uncertainty prediction module 503 can be used for Monte Carlo estimation based on the yield prediction neural network model, and sets the uncertainty index of the yield prediction neural network model on the feature vector of each target CN coupled reaction.

[0142] The recommended index calculation module 504 can be used to calculate the recommended index parameters of each CN coupling reaction condition feature in the CN coupling reaction condition feature set based on the predicted CN coupling reaction yield data and uncertainty index.

[0143] The optimal condition screening module 505 can be used to screen the optimal CN coupling reaction condition features from the CN coupling reaction condition feature set based on the recommended index parameters, and set the CN coupling reaction conditions corresponding to the optimal CN coupling reaction conditions of the target CN coupling reactant corresponding to the target CN coupling reactant features as the optimal recommended CN coupling reaction conditions of the target CN coupling reactant.

[0144] In one embodiment of the present invention, the CN-coupled reaction condition intelligent recommendation system 500 can also be used for:

[0145] S11. Map the target CN coupling reaction feature vector to the target CN coupling reaction chemical space t-SNE diagram to obtain the target CN coupling reaction point;

[0146] S12. Calculate the minimum Euclidean distance between the target CN coupling reaction point and the nearest neighbor CN coupling reaction point in the t-SNE diagram of the chemical space of the target CN coupling reaction point;

[0147] S13. Input the minimum Euclidean distance of the target point into the uncertainty regression prediction model to generate the prediction error estimate of the yield prediction neural network model under the target CN coupled reaction feature vector.

[0148] S14. If the estimated prediction error is lower than the yield prediction reliability threshold, generate a reliability warning message for the target CN coupled reaction feature vector.

[0149] In one embodiment of the present invention, the CN-coupled reaction condition intelligent recommendation system 500 can also be used for:

[0150] S21. Map the characteristics of each CN coupling reactant to the t-SNE diagram of CN coupling reaction chemical space to obtain the CN coupling reaction sites;

[0151] S22. Calculate the minimum Euclidean distance between each CN-coupled reaction point and the nearest neighbor CN-coupled reaction point in the t-SNE diagram of the chemical space of CN-coupled reactions.

[0152] S23. Calculate the actual absolute error of the yield prediction neural network model on the CN-coupled reaction training samples in the CN-coupled reaction training set corresponding to the CN-coupled reaction point. The actual absolute error of the prediction is used to characterize the uncertainty of the yield prediction neural network model on the CN-coupled reaction training samples.

[0153] S24. An uncertainty regression prediction model is constructed based on the minimum Euclidean distance between training points and the actual absolute error of prediction.

[0154] In an optional embodiment of this application, the CN-coupled reaction condition intelligent recommendation system 500 can also be used for:

[0155] S31. Identify the target CN coupling reaction feature vector with an uncertainty index higher than the uncertainty index threshold as a high-uncertainty target CN coupling reaction feature.

[0156] S32. Analyze the characteristics of the highly uncertain target CN coupling reaction and obtain the experimental verification parameters of the CN coupling reaction;

[0157] S33. Obtain the actual CN coupling reaction yield data corresponding to the experimental verification parameters of CN coupling reaction characteristics of each high-uncertainty target, and construct the CN coupling reaction enhancement samples based on the CN coupling reaction characteristics of each high-uncertainty target and the actual CN coupling reaction yield data corresponding to the CN coupling reaction characteristics of each high-uncertainty target.

[0158] S34. Using the high-uncertainty target CN coupling reaction characteristics of each CN coupling reaction enhancement sample as the model optimization input of the yield prediction neural network model, and using the real CN coupling reaction yield data of each CN coupling reaction enhancement sample as the model optimization label of the yield prediction neural network model, the yield prediction neural network model is updated and optimized.

[0159] In one embodiment of the present invention, the feature acquisition and splicing module 501 can also be used for:

[0160] S41. Obtain structural information of the target reactant and candidate ligands;

[0161] S42. Based on the target reactant structure information and the candidate ligand structure information, construct the target reactant structure Morgan fingerprint vector and the candidate ligand structure Morgan fingerprint vector respectively, and splice the target reactant structure Morgan fingerprint vector and the candidate ligand structure Morgan fingerprint vector to obtain the target CN-coupled reactant features.

[0162] S43. Obtain the set of CN coupling reaction conditions, perform one-hot encoding on each CN coupling reaction condition in the set of CN coupling reaction conditions to obtain the characteristics of each CN coupling reaction condition, and summarize the characteristics of each CN coupling reaction condition to construct the CN coupling reaction condition feature set.

[0163] In one embodiment, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0164] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0165] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0166] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for intelligently recommending CN coupling reaction conditions, characterized in that, The method includes: S1. Obtain the target CN coupling reactant features and the CN coupling reaction condition feature set, and concatenate the target CN coupling reactant features with each CN coupling reaction condition feature in the CN coupling reaction condition feature set to obtain each target CN coupling reaction feature vector corresponding to the target CN coupling reactant features. S2. Input the feature vectors of each target CN coupling reaction into the yield prediction neural network model to obtain the predicted CN coupling reaction yield data corresponding to the feature vectors of each target CN coupling reaction; S3. Based on the Monte Carlo estimation of the yield prediction neural network model, set the uncertainty index of the yield prediction neural network model on the characteristic vector of each target CN coupled reaction; S4. Based on the predicted CN coupling reaction yield data and the uncertainty index, calculate the recommended index parameters for each CN coupling reaction condition feature in the CN coupling reaction condition feature set. S5. Based on the recommended index parameters, the optimal CN coupling reaction condition feature is obtained from the CN coupling reaction condition feature set, and the CN coupling reaction condition corresponding to the optimal CN coupling reaction feature is set as the optimal recommended CN coupling reaction condition for the target CN coupling reactant corresponding to the target CN coupling reactant feature.

2. The method according to claim 1, characterized in that, After inputting the feature vectors of each target CN-coupled reaction into the yield prediction neural network model to obtain the predicted CN-coupled reaction yield data corresponding to the feature vectors of each target CN-coupled reaction, the method further includes: S11. Map the target CN coupling reaction feature vector to the target CN coupling reaction chemical space t-SNE diagram to obtain the target CN coupling reaction point. S12. Calculate the minimum Euclidean distance between the target CN coupling reaction point and the nearest neighbor CN coupling reaction point in the t-SNE diagram of the CN coupling reaction chemical space; S13. Input the minimum Euclidean distance of the target point into the uncertainty regression prediction model to generate the prediction error estimate of the yield prediction neural network model under the target CN coupled reaction feature vector; S14. If the estimated prediction error is lower than the yield prediction reliability threshold, generate a reliability warning message for the target CN coupled reaction feature vector.

3. The method according to claim 2, characterized in that, The method further includes: S21. Map the characteristics of each CN coupling reactant onto the CN coupling reaction chemical space t-SNE diagram to obtain each CN coupling reaction point; S22. Calculate the minimum Euclidean distance between each CN coupling reaction point and the training point of the nearest neighbor CN coupling reaction point in the t-SNE diagram of the CN coupling reaction chemical space. S23. Calculate the actual absolute prediction error of the yield prediction neural network model on the CN-coupled reaction training samples in the CN-coupled reaction training set corresponding to the CN-coupled reaction point. The actual absolute prediction error is used to characterize the uncertainty of the yield prediction neural network model on the CN-coupled reaction training samples. S24. The uncertainty regression prediction model is constructed based on the minimum Euclidean distance between the training points and the actual prediction absolute error.

4. The method according to claim 1, characterized in that, The method further includes: S31. Identify the target CN coupling reaction feature vectors whose uncertainty index is higher than the uncertainty index threshold as high uncertainty target CN coupling reaction features; S32. Analyze the characteristics of the highly uncertain target CN coupling reaction to obtain the experimental verification parameters of the CN coupling reaction; S33. Obtain the actual CN coupling reaction yield data corresponding to the experimental verification parameters of the CN coupling reaction of each of the high uncertainty target CN coupling reaction characteristics, and construct each CN coupling reaction enhancement sample based on the CN coupling reaction characteristics of each of the high uncertainty target CN coupling reaction and the actual CN coupling reaction yield data corresponding to the CN coupling reaction characteristics of each of the high uncertainty target CN coupling reaction characteristics. S34. Using the high-uncertainty target CN coupling reaction features of each CN coupling reaction enhancement sample as the model optimization input of the yield prediction neural network model, and using the real CN coupling reaction yield data of each CN coupling reaction enhancement sample as the model optimization label of the yield prediction neural network model, the yield prediction neural network model is updated and optimized.

5. The method according to claim 1, characterized in that, The acquisition of the target CN coupling reactant characteristics and CN coupling reaction condition characteristic set includes: S41. Obtain structural information of the target reactant and candidate ligands; S42. Based on the target reactant structure information and the candidate ligand structure information, construct the target reactant structure Morgan fingerprint vector and the candidate ligand structure Morgan fingerprint vector respectively, and concatenate the target reactant structure Morgan fingerprint vector and the candidate ligand structure Morgan fingerprint vector to obtain the characteristics of the target CN-coupled reactant. S43. Obtain the set of CN coupling reaction conditions, perform one-hot encoding on each CN coupling reaction condition in the set of CN coupling reaction conditions to obtain the features of each CN coupling reaction condition, and summarize the features of each CN coupling reaction condition to construct the feature set of CN coupling reaction conditions.

6. A CN-coupled reaction condition intelligent recommendation system, used to implement the method according to any one of claims 1 to 5, characterized in that, The system includes: The feature acquisition and splicing module is used to acquire the target CN coupling reactant features and the CN coupling reaction condition feature set, and splice the target CN coupling reactant features with each CN coupling reaction condition feature in the CN coupling reaction condition feature set to obtain each target CN coupling reaction feature vector corresponding to the target CN coupling reactant features. The yield prediction module is used to input the feature vectors of each target CN coupling reaction into the yield prediction neural network model to obtain the predicted CN coupling reaction yield data corresponding to the feature vectors of each target CN coupling reaction. An uncertainty prediction module is used to set the uncertainty index of the yield prediction neural network model on each of the target CN coupled reaction feature vectors based on the Monte Carlo estimation of the yield prediction neural network model. The recommendation index calculation module is used to calculate the recommendation index parameters of each CN coupling reaction condition feature in the CN coupling reaction condition feature set based on the predicted CN coupling reaction yield data and the uncertainty index. The optimal condition screening module is used to screen the optimal CN coupling reaction condition features from the CN coupling reaction condition feature set based on the recommended index parameters, and set the CN coupling reaction conditions corresponding to the optimal CN coupling reaction conditions of the target CN coupling reactant corresponding to the target CN coupling reactant features as the optimal recommended CN coupling reaction conditions of the target CN coupling reactant.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images