Relative retention time prediction model construction method, prediction method and device

Abstract

The present invention provides a method for constructing a relative retention time prediction model, a prediction method and a device, the method for constructing the relative retention time prediction model comprising: obtaining a sample data set, and calculating a plurality of molecular descriptors of each sample compound in the sample data set; optimizing the plurality of molecular descriptors to obtain a target molecular descriptor related to the relative retention time among the plurality of molecular descriptors; optimizing and training the relative retention time prediction model according to the target molecular descriptors and relative retention time labels of each sample compound in the sample data set to obtain an optimal relative retention time prediction model; the optimal relative retention time prediction model is used to predict the relative retention time of the compound to be tested according to the target molecular descriptor of the compound to be tested. The present invention realizes the rapid and accurate construction and acquisition of the relative retention time prediction model, and can effectively improve the interpretability of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of chemical analysis, and in particular to a method for constructing a relative retention time prediction model, a prediction method and a device. Background Art

[0002] In recent years, emerging environmental pollutants (EPS) have gradually replaced traditional pollutants and become a new focus in the field of environmental analysis; EPS refer to pollutants that have been newly discovered in the environment or have been discovered for a long time but have only recently received attention, including medicines, personal care products, pesticides, cleaning products, fragrances and hormones. EPS are widely present in water bodies and pose a huge threat to human health and environmental safety. Therefore, how to screen and identify emerging environmental pollutants is one of the technical problems that the industry is currently facing.

[0003] Liquid chromatography-mass spectrometry is the main method for screening and identifying a large number of pollutants in water environments. However, it has disadvantages such as complex analytical methods and inability to analyze new compounds that have not been synthesized. In contrast, constructing a quantitative structure-chromatographic retention relationship (QSRR) model based on the structure and experimental data of pollutants can provide a faster and more efficient way to predict the chromatographic retention behavior of compounds.

[0004] The QSRR model is widely used in various fields. Relevant scholars have trained the deep learning neural network model based on the METLIN small molecule database (containing 80,038 small molecule RT data, 75% as training set and 25% as test set) with RT (Retention Time) as the response value and extended connectivity fingerprints (ECFP) as the variable to obtain the QSRR model and achieve accurate RT prediction.

[0005] Most of the existing technologies use molecular fingerprints as variables to construct QSRR models. The QSRR models constructed based on molecular fingerprints have poor interpretability and cannot obtain the correlation between the physicochemical properties of compound molecules and the relative retention time. Summary of the invention

[0006] The present invention provides a relative retention time prediction model construction method, prediction method and device, which are used to solve the defect that the QSRR model is constructed by using molecular fingerprints in the prior art, resulting in poor model interpretability, and improve the interpretability of the model while ensuring the accuracy of relative retention time prediction.

[0007] The present invention provides a method for constructing a relative retention time prediction model, comprising:

[0008] Acquire a sample data set, and calculate a plurality of molecular descriptors of each sample compound in the sample data set;

[0009] Optimizing the multiple molecular descriptors to obtain a target molecular descriptor related to the relative retention time among the multiple molecular descriptors;

[0010] According to the target molecule descriptor and relative retention time label of each sample compound in the sample data set, the relative retention time prediction model is optimized and trained to obtain an optimal relative retention time prediction model; the optimal relative retention time prediction model is used to predict the relative retention time of the compound to be tested according to the target molecule descriptor of the compound to be tested.

[0011] According to a method for constructing a relative retention time prediction model provided by the present invention, the optimization process of the multiple molecular descriptors to obtain a target molecular descriptor related to the relative retention time among the multiple molecular descriptors includes:

[0012] Deleting from the plurality of molecular descriptors molecular descriptors having a correlation coefficient greater than a first preset value, molecular descriptors with missing values, and molecular descriptors whose mean, standard deviation, and relative standard deviation do not meet preset conditions;

[0013] Based on Lasso regression analysis and multiple regression linear analysis, target molecular descriptors related to relative retention time were selected from the remaining molecular descriptors.

[0014] According to a method for constructing a relative retention time prediction model provided by the present invention, the step of calculating a plurality of molecular descriptors of each sample compound in the sample data set includes:

[0015] Optimizing the sample data set;

[0016] The optimization process includes deleting sample compounds whose relative retention time is less than a second preset value, and deleting sample compounds belonging to abnormal outliers based on the Cook distance;

[0017] Calculate multiple molecular descriptors for each sample compound in the optimized sample data set.

[0018] A method for constructing a relative retention time prediction model provided by the present invention further includes:

[0019] generating each target molecule descriptor sequence according to each target molecule descriptor of all sample compounds in the sample data set;

[0020] Based on the kernel density function, the distribution characteristics of each target molecule descriptor sequence are determined;

[0021] According to the distribution characteristics, the applicable scope of the optimal relative retention time prediction model is determined.

[0022] According to a method for constructing a relative retention time prediction model provided by the present invention, the relative retention time prediction model is generated based on a deep learning neural network.

[0023] According to a method for constructing a relative retention time prediction model provided by the present invention, the categories of sample compounds in the sample data set at least include drugs, pesticides, hormones and biochemical reagents.

[0024] The present invention also provides a prediction method of a relative retention time prediction model, comprising:

[0025] Obtaining the optimal relative retention time prediction model;

[0026] Inputting the target molecule descriptor of the compound to be tested into the optimal relative retention time prediction model to obtain the relative retention time of the compound to be tested output by the optimal relative retention time prediction model;

[0027] The optimal relative retention time prediction model is constructed and generated based on any of the relative retention time prediction model construction methods described above.

[0028] The present invention also provides a device for constructing a relative retention time prediction model, comprising:

[0029] A calculation module, used for acquiring a sample data set and calculating a plurality of molecular descriptors of each sample compound in the sample data set;

[0030] An optimization module, used for optimizing the multiple molecular descriptors to obtain a target molecular descriptor related to the relative retention time among the multiple molecular descriptors;

[0031] A construction module is used to optimize and train the relative retention time prediction model according to the target molecule descriptor and relative retention time label of each sample compound in the sample data set to obtain an optimal relative retention time prediction model; the optimal relative retention time prediction model is used to predict the relative retention time of the compound to be tested according to the target molecule descriptor of the compound to be tested.

[0032] The present invention also provides a prediction device for a relative retention time prediction model, comprising:

[0033] An acquisition module, used for acquiring an optimal relative retention time prediction model;

[0034] A prediction module, used for inputting the target molecule descriptor of the compound to be tested into the optimal relative retention time prediction model to obtain the relative retention time of the compound to be tested output by the optimal relative retention time prediction model;

[0035] The optimal relative retention time prediction model is constructed and generated based on any of the relative retention time prediction model construction methods described above.

[0036] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for constructing a relative retention time prediction model as described in any one of the above, or the steps of a prediction method for a relative retention time prediction model are implemented.

[0037] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for constructing a relative retention time prediction model as described in any one of the above, or the steps of a relative retention time prediction model prediction method.

[0038] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned methods for constructing a relative retention time prediction model or a relative retention time prediction model prediction method.

[0039] The relative retention time prediction model construction method, prediction method and device provided by the present invention, on the one hand, calculate multiple molecular descriptors of each sample compound in a sample data set, and optimize the multiple molecular descriptors to obtain target molecular descriptors related to relative retention time among the multiple molecular descriptors, so that the input vector of the relative retention time prediction model contains abundant target molecular descriptors related to relative retention time, and thus the relative retention time prediction model can be constructed and obtained quickly and accurately; on the other hand, the relative retention time prediction model is obtained by using the target molecular descriptors that can characterize the physicochemical properties of the compound molecules and the relative retention time label training, so that the trained relative retention time prediction model can be used to accurately characterize the correlation between the physicochemical properties and the relative retention time, and the interpretability of the model is effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 It is a schematic diagram of a process for constructing a relative retention time prediction model provided by the present invention;

[0042] Figure 2 It is one of the structural schematic diagrams of the simulation results in the method for constructing the relative retention time prediction model provided by the present invention;

[0043] Figure 3 This is the second structural schematic diagram of the simulation results in the method for constructing the relative retention time prediction model provided by the present invention;

[0044] Figure 4 This is the third structural schematic diagram of the simulation results in the method for constructing the relative retention time prediction model provided by the present invention;

[0045] Figure 5 This is the fourth structural schematic diagram of the simulation results in the method for constructing the relative retention time prediction model provided by the present invention;

[0046] Figure 6 It is a schematic flow chart of a prediction method of a relative retention time prediction model provided by the present invention;

[0047] Figure 7 It is a schematic diagram of the structure of a device for constructing a relative retention time prediction model provided by the present invention;

[0048] Figure 8 It is a structural schematic diagram of a prediction device of a relative retention time prediction model provided by the present invention;

[0049] Fig. 9 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] The QSRR model is widely used in the field of chemistry. Some scholars have constructed a QSRR model based on 86 drug compounds included in the drug testing list of the Olympic Games and Paralympic Games through an artificial neural network (ANN) algorithm. For 93% of the doping-related compounds, the retention time (RT) prediction error of this model is less than 0.5 min.

[0052] Some scholars constructed a QSRR model for the relative retention time (RRT) and molecular distance-edge vector (MEDV) index of polybrominated diphenyl ethers (PBDEs) using the radial basis function artificial neural network (RBF-ANN) algorithm. The data set contained a total of 126 polybrominated diphenyl ether compounds; the training set contained 88 PBDEs and the test set contained 36 PBDEs. The root mean square relative error of the model for predicting the relative retention time of the test set compounds was 12.58.

[0053] Some scholars have also constructed a QSRR model for a data set containing 444 emerging pollutants RT using the ANN method, and predicted 100 randomly selected compounds, with 95% of the compounds having an error within 2 minutes. Some scholars have also constructed a QSRR model for 95 veterinary drugs using multiple linear regression (MLR), and found that there is a good correlation between drug RT and molecular descriptors, with the correlation index R 2 It can reach 0.966.

[0054] Although the above-mentioned prior art has constructed QSRR models based on various methods, most of these methods are only applicable to a single category of compounds and have a limited scope of application. In addition, the prior art usually adopts simple models to construct QSRR models, such as exponential functions, multivariate linear regression or simple neural network models. However, the physicochemical properties that usually affect the relative retention time of compounds are complex and diverse, and it is difficult to accurately obtain the relative retention time of compounds based on the above-mentioned simple models. In addition, some scholars have constructed QSRR models based on extended connection fingerprints, but the model has poor interpretability and cannot obtain the correlation between the physicochemical properties of compound molecules and the relative retention time.

[0055] In summary, the QSRR model constructed in the prior art has the problems of limited adaptability and poor interpretability.

[0056] In order to solve the above problems, this embodiment proposes a method for constructing a relative retention time prediction model, so that the constructed relative retention time prediction model can be applicable to relative retention time prediction scenarios of various categories of compounds, has a wide range of adaptability, can accurately obtain the relative retention time of the compound according to the molecular descriptor of the compound, and has strong interpretability.

[0057] It should be noted that the execution subject of the construction method and prediction method in this embodiment can be an electronic device, a component in an electronic device, an integrated circuit, or a chip. The electronic device can be a mobile electronic device or a non-mobile electronic device. For example, the mobile electronic device can be a laptop computer, a PDA, a super mobile personal computer, etc., and the non-mobile electronic device can be a server, a network attached storage, etc., which is not specifically limited in the present invention.

[0058] Combine the following Figure 1 The method for constructing the relative retention time prediction model of the present invention is described, and the method comprises:

[0059] Step 101, obtaining a sample data set, and calculating a plurality of molecular descriptors of each sample compound in the sample data set;

[0060] Optionally, the sample data set in this embodiment includes multiple types of compounds, including but not limited to drugs, pesticides, hormones, biochemical reagents, and veterinary drugs, etc., which are not specifically limited in this embodiment.

[0061] The sample data set may be obtained by transmitting or downloading through the Internet, which is not specifically limited in this embodiment.

[0062] For example, the structural information of 490 compounds including drugs, pesticides, hormones and biochemical reagents, as well as their relative retention time RRT, can be obtained through the MassBank website to form a sample data set.

[0063] Then, multiple molecular descriptors of each sample compound in the sample data set are directly calculated; or after processing the sample data set, multiple molecular descriptors of each sample compound in the processed sample data set are calculated, which is not specifically limited in this embodiment.

[0064] The method of calculating the multiple molecular descriptors of each sample compound is to calculate the molecular descriptors of each sample compound through the Calculate Descriptors module of the MOE (Molecular Operating Environment) software and the PaDEL-Descriptor software to obtain the multiple molecular descriptors of each sample compound;

[0065] The number of molecular descriptors for each sample compound can be set according to actual needs, such as 2311.

[0066] Step 102, optimizing the plurality of molecular descriptors to obtain a target molecular descriptor associated with a relative retention time among the plurality of molecular descriptors;

[0067] Optionally, after obtaining multiple molecular descriptors of each sample compound, some of the molecular descriptors have the same meaning of physical and chemical properties represented by them, are not applicable to all sample compounds (the molecular descriptors cannot be calculated for some compounds), have low discrimination between different compounds (the values ​​of the same molecular descriptor for different compounds do not vary much), or are not related to the relative retention time. These molecular descriptors are variables irrelevant to the construction of the relative retention time prediction model, which will affect the efficiency and accuracy of the construction of the relative retention time prediction model.

[0068] Therefore, before constructing a relative retention time prediction model, it is necessary to optimize multiple molecular descriptors and delete the molecular descriptors that are irrelevant to the construction of the relative retention time prediction model from the multiple molecular descriptors, so as to retain the target molecular descriptors related to the relative retention time among the multiple molecular descriptors, thereby improving the construction efficiency and accuracy of the relative retention time prediction model.

[0069] Step 103, optimizing and training the relative retention time prediction model according to the target molecule descriptor and relative retention time label of each sample compound in the sample data set to obtain an optimal relative retention time prediction model; the optimal relative retention time prediction model is used to predict the relative retention time of the test compound according to the target molecule descriptor of the test compound.

[0070] Among them, the relative retention time prediction model can be based on other deep networks such as deep convolutional neural networks and recurrent neural networks. This embodiment does not limit the specific structure of the relative retention time prediction model.

[0071] Optionally, after obtaining the target molecule descriptor of each sample compound in the sample data set, the sample data set is divided into a training set and a test set;

[0072] The division method may be to first obtain the categories of the sample compounds, and then randomly divide the sample compounds under each category into a training set and a test set according to a preset ratio, so that each category of sample compounds is in the training set;

[0073] Among them, the preset ratio can be determined according to actual needs, and this example does not make a specific limitation on this; for example, according to the ratio of 7:3, the 490 sample compounds in the sample data set are divided into a training set containing 343 sample compounds and a test set containing 147 compounds.

[0074] Then, the target molecular descriptor of each sample compound in the training set is used as an input variable, and the relative retention time label of each sample compound in the training set is used as a response value, and the relative retention time prediction model is iteratively trained to obtain an optimal relative retention time prediction model, so that the optimal relative retention time prediction model obtained by training can accurately predict the relative retention time of the test compound according to the target molecular descriptor of the test compound;

[0075] Among them, the training parameters of the relative retention time prediction model can be set according to actual needs, such as the optimization method, loss function, metric function, number of iterations and number of samples in each batch can all be set according to actual needs.

[0076] For example, the optimization method of the relative retention time prediction model can be set to Adam optimizer, with mean square error and mean absolute error as loss function and metric function, the number of iterations epoch is 250 times, and the number of samples in each batch batchsize is 1.

[0077] In order to verify the effectiveness of the method for constructing the relative retention time prediction model in this embodiment, the performance of the relative retention time prediction model obtained through training was evaluated using a test set. The metric function may be the mean absolute error and the mean square error. The calculation formula is:

[0078]

[0079]

[0080] Among them, MSE is the mean square error of the test set, MAE is the mean absolute error of the test set, is the relative retention time label of the i-th sample compound in the test set, is the relative retention time of the i-th sample compound in the test set predicted by the relative retention time prediction model; n is the number of sample compounds in the test set.

[0081] For example, this implementation uses the constructed DNN model to predict a test set containing 116 compounds. The simulation results show that the DNN model has a good prediction effect on the test set; the mean square error MSE of the test set is 0.0143, the mean absolute error MAE of the test set is 0.0860; the square of the correlation coefficient R between the RRT label value and the RRT prediction value in the test set is 2 It is 0.913.

[0082] like Figure 2As shown, there is an obvious linear relationship between the RRT label value and the RRT predicted value in the test set, indicating that the relative retention time predicted by the optimal relative retention time prediction model is almost close to the true relative retention time. The performance of the optimal relative retention time prediction model is good and can accurately predict the relative retention time of the compound.

[0083] On the one hand, this example calculates multiple molecular descriptors of each sample compound in the sample data set and optimizes the multiple molecular descriptors to obtain target molecular descriptors related to relative retention time among the multiple molecular descriptors, so that the input vector of the relative retention time prediction model contains abundant target molecular descriptors related to relative retention time, and thus the relative retention time prediction model can be constructed quickly and accurately; on the other hand, the relative retention time prediction model is obtained by training the target molecular descriptors that can characterize the physicochemical properties of the compound molecules and the relative retention time labels, so that the trained relative retention time prediction model can be used to accurately characterize the correlation between the physicochemical properties and the relative retention time, effectively improving the interpretability of the model.

[0084] On the basis of the above embodiments, the optimization process of the multiple molecular descriptors is performed in this embodiment to obtain target molecular descriptors related to the relative retention time from the multiple molecular descriptors, including: deleting molecular descriptors whose correlation coefficients are greater than a first preset value, molecular descriptors with missing values, and molecular descriptors whose means, standard deviations, and relative standard deviations do not meet preset conditions from the multiple molecular descriptors; and then selecting target molecular descriptors related to the relative retention time from the remaining molecular descriptors based on Lasso regression analysis and multivariate regression linear analysis.

[0085] Optionally, each molecular descriptor corresponding to all sample compounds in the sample data set is taken as a sequence; a correlation coefficient between any two molecular descriptor sequences is calculated; if the correlation coefficient between any two molecular descriptor sequences is greater than a first preset value, any descriptor of the any two molecular descriptors is deleted from the multiple molecular descriptors.

[0086] Furthermore, if the value of any descriptor of any compound is missing, it indicates that the description is not applicable to all sample compounds, and the descriptor is deleted from the multiple molecular descriptors.

[0087] In order to ensure the discrimination of descriptors for different sample compounds and avoid the small differences between the same molecular descriptors of different sample compounds, which will affect the prediction performance of the relative retention time prediction model, the molecular descriptors whose mean, standard deviation and relative standard deviation do not meet the preset conditions are deleted from the multiple molecular descriptors, and only the molecular descriptors that meet the preset conditions are retained.

[0088] Then, Lasso regression analysis is performed on the remaining molecular descriptors and relative retention time labels, and parameter reduction is performed on the remaining molecular descriptors to achieve the purpose of dimensionality reduction.

[0089] Finally, multiple linear regression was performed on the molecular descriptors and relative retention time labels retained after the Lasso regression analysis, and the significance level P-value between each molecular descriptor retained after the Lasso regression analysis and the relative retention time label was obtained; since the larger the significance level, the more significant the effect of the molecular descriptor on the relative retention time; therefore, only the target molecular descriptors with a P-value less than the significance threshold were retained.

[0090] Among them, Lasso regression analysis and multivariate regression linear analysis can be implemented through the Lars toolkit in R language, etc., and this example does not make specific limitations on this.

[0091] The first preset value can be set according to actual needs, such as 0.9; the preset conditions can also be set according to actual needs, such as mean>0.01, standard deviation>0.1 and relative standard deviation>0.1; the significance threshold can also be set according to actual needs, which is not specifically limited in this embodiment.

[0092] The following describes the optimization process of the molecular descriptor in this embodiment by taking the parameter setting of the above example as an example;

[0093] Optionally, the Calculate Descriptors module of the MOE software and the PaDEL-Descriptor software are used to calculate molecular descriptors for each sample compound in the sample data set, for a total of 2311.

[0094] First, if the correlation coefficient between any two molecular descriptors was greater than 0.9, any of the descriptors was deleted from the multiple molecular descriptors; the molecular descriptors with missing values ​​were deleted from the multiple molecular descriptors; the molecular descriptors that did not meet the mean>0.01, standard deviation>0.1 or relative standard deviation>0.1 were deleted from the multiple molecular descriptors, and only 306 descriptors with mean>0.01, standard deviation>0.1 and relative standard deviation>0.1 were retained;

[0095] Then, the Lars toolkit in R language was used to perform lasso regression analysis on the 306 molecular descriptors and RRT, and 29 molecular descriptors were obtained. Finally, multiple regression linear analysis was performed on the 29 molecular descriptors and RRT, and 10 target molecular descriptors with a significance level P-value less than 0.001 were retained, namely E_nb, XLogP, PEOE_VSA_FPOL (P_V_FPOL for short), Q_VSA_FPNEG (Q_V_FPNEG for short), GCUT_SLOGP_0 (G_SLOGP_0 for short), SpMAD_Dzp, PEOE_VSA_FNEG (P_V_FNEG for short), SpMax8_Bhm, MIC0 and TDB4v.

[0096] Finally, the model was optimized and trained based on the above 10 target molecular descriptors and relative retention time labels of each sample compound in the sample data set to obtain a relative retention time prediction model that can quickly and accurately predict the relative retention time based on the target molecular descriptors.

[0097] like Figure 3 As shown, it is effectively verified that there is a strong correlation between the 10 target molecule descriptors optimized in this embodiment and the relative retention time, and the optimal relative retention time prediction model can be quickly and accurately constructed.

[0098] exist Figure 3 In the data, the sample compounds are divided into two parts according to the size of RRT, one part is the sample compounds with RRT greater than or equal to 1, and the other part is the sample compounds with RRT less than 1; by observing the value distribution of the 10 target molecular descriptors in each part, it can be seen that the values ​​of the five target descriptors XLogP, G_S_LOGP_0, P_V_FNEG, SpMax8_Bhm and TDBv4 are significantly different, and when RRT is greater than or equal to 1, the values ​​of these five descriptors are greater than, and when RRT is less than 1, the values ​​of these five descriptors are just the opposite, while the value rules of the two target molecular descriptors P_V_FPOL and MIC0 are just the opposite.

[0099] Moreover, the target molecule descriptors XLogP and P_V_FENG increase with the increase of RRT, while the target molecule descriptor P_V_FPOL decreases with the increase of RRT. The target molecule XLogP is related to the lipid-water partition coefficient LogP, which is the criterion for judging hydrophobicity and one of the chemical descriptors covering most aspects of liquid chromatography; while P_V_FENG and P_V_FPOL are related to the van der Waals surface area, which is associated with the charge and energy of the molecule, affecting the interaction between the molecule and the stationary phase and the mobile phase, thereby affecting the retention behavior.

[0100] In summary, the optimized target molecule descriptors in this embodiment are closely related to the relative retention time, which makes the constructed relative retention time prediction model more accurate and more interpretable.

[0101] On the basis of the above-mentioned embodiment, the calculation of multiple molecular descriptors of each sample compound in the sample data set described in this embodiment includes: optimizing the sample data set; wherein the optimization includes deleting sample compounds whose relative retention time is less than a second preset value, and deleting sample compounds that are abnormal outliers based on the Cook distance; and calculating multiple molecular descriptors of each sample compound in the sample data set after optimization.

[0102] The second preset value is a preset relative retention time threshold for optimizing the sample data set, and can be set according to actual needs, such as 0.085.

[0103] Cook's distance is a common distance in statistical analysis, which is used to diagnose whether there are abnormal data in various regression analyses.

[0104] Optionally, before executing the step of calculating the molecular descriptors in step 101, it is necessary to optimize the sample data set to quickly and accurately construct a relative retention time prediction model.

[0105] Optionally, since some sample compounds are non-retained molecules, that is, molecules that will not be retained in the chromatographic column and will be eluted before the start of gradient elution (usually within the first minute), such sample compounds will interfere with the construction of the relative retention time prediction model, and therefore, they need to be deleted; specifically, it is determined whether the relative retention time of each sample compound is less than a second preset value, and all sample compounds in the sample data set whose relative retention time is less than the second preset value are deleted; at the same time, the Cook distance is calculated by the Car toolkit in the R language to check whether each sample compound is an abnormal outlier, and the abnormal sample compounds are deleted from the sample data set.

[0106] The method of testing whether a sample compound is an abnormal outlier is to determine whether the Cook distance of each sample compound is greater than a distance threshold, and to regard the sample compound whose distance is greater than the distance threshold as an abnormal outlier.

[0107] The distance threshold can be determined according to the number of sample compounds in the sample data set and the number of features of each sample compound. The specific formula is:

[0108] T = [4 / (MN-1)];

[0109] Among them, T is the distance threshold, M is the number of sample compounds, and N is the number of features of the sample compounds.

[0110] For example, there are 490 sample compounds in the sample data set, and 391 sample compounds can be obtained after optimization;

[0111] Then, multiple molecular descriptors of each sample compound in the optimized sample data set are calculated, and the target molecular descriptor of each sample compound is obtained; according to the target molecular descriptors and relative retention time labels of each sample compound in the optimized sample data set, a relative retention time prediction model is quickly and accurately constructed.

[0112] On the basis of the above embodiments, the present embodiment also includes: generating each target molecule descriptor sequence according to each target molecule descriptor of all sample compounds in the sample data set; determining the distribution characteristics of each target molecule descriptor sequence based on the kernel density function; and determining the applicable scope of the optimal relative retention time prediction model according to the distribution characteristics.

[0113] Optionally, forming a sequence of each identical target molecule descriptor of all sample compounds in the data set to generate each target molecule descriptor sequence;

[0114] Then, the distribution characteristics of each target molecule descriptor sequence are calculated based on the kernel density function, and the applicable scope of the optimal relative retention time prediction model is determined according to the part with the largest distribution density of target molecule descriptor values ​​in each target molecule descriptor sequence.

[0115] In order to verify the accuracy of the applicable range of the optimal relative retention time prediction model determined in this embodiment, the applicable range determined according to the distribution characteristics of each target molecule descriptor sequence is compared with the distribution of each target molecule descriptor sequence generated by all sample compounds with a prediction error of less than 10% of the optimal relative retention time prediction model for comparison analysis.

[0116] like Figure 4 As shown in , it is the distribution of molecular descriptor values ​​in each target molecular descriptor sequence determined based on the kernel density function; Figure 5 As shown in FIG. 1 , the box plot shows the distribution of descriptor values ​​of sample compounds that can be accurately predicted by the optimal relative retention time prediction model (relative error is less than 10%). Figure 4 and Figure 5 It can be seen that the place where the molecular descriptor value distribution density in the kernel density function diagram is the largest corresponds to the distribution of the molecular descriptor values ​​of the accurately predicted sample compounds. Therefore, through the distribution characteristics of the molecular descriptor sequence of the sample compound, the applicable range of the model can be accurately obtained, so that the model can accurately predict the compounds within the range to obtain accurate relative retention times.

[0117] Taking the 10 selected target molecule descriptors as an example, the applicable scope of the optimal relative retention time prediction model in this embodiment is described in detail;

[0118] The applicable range of the model can be limited by the distribution characteristics of the selected 10 target molecular descriptors: the applicable range of XLogP is [-0.871, 4.574]; the applicable range of TDB4v is [520.36, 1010.28]; the applicable range of SpMax8_Bhm is [2.27, 3.18]; the applicable range of SpMAD_Dzp is [7.26, 13.66]; the applicable range of Q_VSA_FPNEG is [0.041, 0.37]; P The applicable range of EOE_VSA_FPOL is [0.028, 0.39]; the applicable range of PEOE_VSA_FNEG is [0.18, 0.61]; the applicable range of MIC0 is [9.64, 15.36]; GCUT_SLOGP_0 has three applicable ranges, namely [-2.26, -2.25], [-1.49, -1.43] and [-1.17, -0.78]; the applicable range of E_nb is [-36.21, 68.665].

[0119] On the basis of the above embodiments, the relative retention time prediction model described in this embodiment is generated based on a deep learning neural network.

[0120] Optionally, the relative retention time prediction model is generated based on deep neural networks (DNN).

[0121] Among them, structural parameters such as the number of hidden layers of the deep neural network, the number of node units in each hidden layer, the activation function of each hidden layer, and the language and tools for constructing the deep neural network can all be set according to actual needs, and this embodiment does not make specific limitations on this.

[0122] For example, a deep neural network can be constructed by using the Keras toolkit in R language or python language, with the RRT labels of the sample compounds in the training set as response values ​​and the target molecular descriptors screened out from multiple molecular descriptors as independent variables; the deep neural network is based on two fully connected hidden layers, with 256 node units and 123 node units respectively; the activation function is a nonlinear function, such as the Relu activation function; and the output layer does not contain an activation function.

[0123] This embodiment generates a relative retention time prediction model by constructing a deep learning neural network, so that the relative retention time prediction model can characterize the complex nonlinear relationship between the target molecule descriptor and the relative retention time, so as to quickly and accurately predict the relative retention time according to the target molecule descriptor.

[0124] On the basis of the above-mentioned embodiment, the categories of sample compounds in the sample data set in this embodiment at least include drugs, pesticides, hormones and biochemical reagents.

[0125] It should be noted that the categories of compounds in the sample data set in this embodiment are not limited to the above categories, and may also include veterinary drug compounds, etc., which is not specifically limited in this example.

[0126] In this embodiment, sample compounds with rich categories are used, so that the optimal relative retention time prediction model after training can be applied to fields including but not limited to drugs, pesticides, hormones, and biochemical reagents, etc., and has a wide range of applicability.

[0127] The prediction method of the relative retention time prediction model provided by the present invention is described below. The prediction method of the relative retention time prediction model described below and the construction method of the relative retention time prediction model described above can be referred to each other;

[0128] like Figure 6 As shown, a prediction method of a relative retention time prediction model provided in this embodiment comprises:

[0129] Step 601, obtaining an optimal relative retention time prediction model; the optimal relative retention time prediction model is constructed and generated based on the relative retention time prediction model construction method;

[0130] Optionally, based on Figure 1 The relative retention time prediction model construction steps in the embodiment of the present invention are used to construct and generate the optimal relative retention time prediction model; specifically, multiple molecular descriptors of each sample compound in the sample data set are first obtained, and then the multiple molecular descriptors are optimized to obtain the target molecular descriptors related to the relative retention time; finally, according to the target molecular descriptors and relative retention time labels of each sample compound in the sample data set, the relative retention time prediction model is optimized and trained to obtain the optimal relative retention time prediction model.

[0131] Step 602: input the target molecule descriptor of the compound to be tested into the optimal relative retention time prediction model to obtain the relative retention time of the compound to be tested output by the optimal relative retention time prediction model.

[0132] The test compound is a compound for which relative retention time prediction is required, and may be a compound of the type of medicine, pesticide, hormone, biochemical reagent, veterinary drug, etc., which is not specifically limited in this embodiment.

[0133] Optionally, when the relative retention time of the test compound needs to be predicted, the target molecule descriptor of the test compound can be obtained first. For specific acquisition steps, see Figure 1 Steps 101 and 102 in the above are not described in detail here.

[0134] Then, after the target molecule descriptor of the compound to be tested is input into the optimal relative retention time prediction model, the relative retention time prediction model learns the target molecule descriptor and can accurately output the relative retention time of the compound to be tested.

[0135] On the one hand, this embodiment calculates multiple molecular descriptors of each sample compound in the sample data set and optimizes the multiple molecular descriptors to obtain target molecular descriptors related to relative retention time from the multiple molecular descriptors, so that the input vector of the relative retention time prediction model contains abundant target molecular descriptors related to relative retention time, and thus the relative retention time prediction model can be constructed and obtained quickly and accurately; on the other hand, the relative retention time prediction model is obtained by training the target molecular descriptors that can characterize the physicochemical properties of the compound molecules and the relative retention time labels, so that the trained relative retention time prediction model can be used to accurately characterize the correlation between the physicochemical properties and the relative retention time, and the interpretability of the model is effectively improved; and the relative retention time prediction model can accurately predict the relative retention time of the test compound according to the target molecular descriptors of the test compound input in real time.

[0136] The following is a description of a relative retention time prediction model construction device provided by the present invention. The relative retention time prediction model construction device described below and the relative retention time prediction model construction method described above can be referenced to each other.

[0137] like Figure 7 As shown, this example provides a device for constructing a relative retention time prediction model, the device comprising a calculation module 701, an optimization module 702 and a construction module 703, wherein:

[0138] The calculation module 701 is used to obtain a sample data set and calculate a plurality of molecular descriptors of each sample compound in the sample data set;

[0139] Optionally, the sample data set in this embodiment includes multiple types of compounds, including but not limited to drugs, pesticides, hormones, biochemical reagents, and veterinary drugs, etc., which are not specifically limited in this embodiment.

[0140] The sample data set may be obtained by transmitting or downloading through the Internet, which is not specifically limited in this embodiment.

[0141] Then, multiple molecular descriptors of each sample compound in the sample data set are directly calculated; or after processing the sample data set, multiple molecular descriptors of each sample compound in the processed sample data set are calculated, which is not specifically limited in this embodiment.

[0142] The method of calculating the multiple molecular descriptors of each sample compound is to calculate the molecular descriptors of each sample compound through the CalculateDescriptors module of the MOE software and the PaDEL-Descriptor software to obtain the multiple molecular descriptors of each sample compound;

[0143] The number of molecular descriptors for each sample compound can be set according to actual needs.

[0144] The optimization module 702 is used to optimize the multiple molecular descriptors to obtain a target molecular descriptor related to the relative retention time among the multiple molecular descriptors;

[0145] Optionally, after obtaining multiple molecular descriptors of each sample compound, some of the molecular descriptors have the same meaning of physical and chemical properties represented by them, are not applicable to all sample compounds (the molecular descriptors cannot be calculated for some compounds), have low discrimination between different compounds (the values ​​of the same molecular descriptor for different compounds do not vary much), or are not related to the relative retention time. These molecular descriptors are variables irrelevant to the construction of the relative retention time prediction model, which will affect the efficiency and accuracy of the construction of the relative retention time prediction model.

[0146] Therefore, before constructing a relative retention time prediction model, it is necessary to optimize multiple molecular descriptors and delete the molecular descriptors that are irrelevant to the construction of the relative retention time prediction model from the multiple molecular descriptors, so as to retain the target molecular descriptors related to the relative retention time among the multiple molecular descriptors, thereby improving the construction efficiency and accuracy of the relative retention time prediction model.

[0147] The construction module 703 is used to optimize and train the relative retention time prediction model according to the target molecule descriptor and relative retention time label of each sample compound in the sample data set to obtain the optimal relative retention time prediction model; the optimal relative retention time prediction model is used to predict the relative retention time of the compound to be tested according to the target molecule descriptor of the compound to be tested.

[0148] Among them, the relative retention time prediction model can be based on other deep networks such as deep convolutional neural networks and recurrent neural networks. This embodiment does not limit the specific structure of the relative retention time prediction model.

[0149] Optionally, after obtaining the target molecule descriptor of each sample compound in the sample data set, the sample data set is divided into a training set and a test set;

[0150] The division method may be to first obtain the categories of the sample compounds, and then randomly divide the sample compounds under each category into a training set and a test set according to a preset ratio, so that each category of sample compounds is in the training set;

[0151] Then, the target molecular descriptor of each sample compound in the training set is used as an input variable, and the relative retention time label of each sample compound in the training set is used as a response value, and the relative retention time prediction model is iteratively trained to obtain an optimal relative retention time prediction model, so that the optimal relative retention time prediction model obtained by training can accurately predict the relative retention time of the test compound according to the target molecular descriptor of the test compound;

[0152] Among them, the training parameters of the relative retention time prediction model can be set according to actual needs, such as the optimization method, loss function, metric function, number of iterations and number of samples in each batch can all be set according to actual needs.

[0153] On the one hand, this example calculates multiple molecular descriptors of each sample compound in the sample data set and optimizes the multiple molecular descriptors to obtain target molecular descriptors related to relative retention time among the multiple molecular descriptors, so that the input vector of the relative retention time prediction model contains abundant target molecular descriptors related to relative retention time, and thus the relative retention time prediction model can be constructed quickly and accurately; on the other hand, the relative retention time prediction model is obtained by training the target molecular descriptors that can characterize the physicochemical properties of the compound molecules and the relative retention time labels, so that the trained relative retention time prediction model can be used to accurately characterize the correlation between the physicochemical properties and the relative retention time, effectively improving the interpretability of the model.

[0154] On the basis of the above embodiments, the optimization module in this embodiment is specifically used to: delete molecular descriptors whose correlation coefficients are greater than a first preset value, molecular descriptors with missing values, and molecular descriptors whose means, standard deviations, and relative standard deviations do not meet preset conditions from the multiple molecular descriptors; and then obtain target molecular descriptors related to relative retention times from the remaining molecular descriptors based on Lasso regression analysis and multivariate regression linear analysis.

[0155] On the basis of the above embodiments, the calculation module in this embodiment is specifically used to: optimize the sample data set; wherein the optimization process includes deleting sample compounds whose relative retention time is less than a second preset value, and deleting sample compounds that are abnormal outliers based on the Cook distance; and calculating multiple molecular descriptors of each sample compound in the sample data set after optimization.

[0156] On the basis of the above embodiments, this embodiment also includes a determination module, which is specifically used to: generate each target molecule descriptor sequence according to each target molecule descriptor of all sample compounds in the sample data set; determine the distribution characteristics of each target molecule descriptor sequence based on the kernel density function; and determine the applicable scope of the optimal relative retention time prediction model according to the distribution characteristics.

[0157] On the basis of the above-mentioned embodiments, the relative retention time prediction model described in this embodiment is generated based on a deep learning neural network.

[0158] On the basis of the above-mentioned embodiment, the categories of sample compounds in the sample data set in this embodiment at least include drugs, pesticides, hormones and biochemical reagents.

[0159] The prediction device of the relative retention time prediction model provided by the present invention is described below. The prediction device of the relative retention time prediction model described below and the prediction method of the relative retention time prediction model described above can be referred to each other.

[0160] like Figure 8 As shown, this example provides a prediction device for a relative retention time prediction model, the device includes an acquisition module 801 and a prediction module 802, wherein:

[0161] The acquisition module 801 is used to acquire the optimal relative retention time prediction model; the optimal relative retention time prediction model is constructed and generated based on the relative retention time prediction model construction method;

[0162] Optionally, based on Figure 1 The relative retention time prediction model construction steps in the embodiment of the present invention are used to construct and generate the optimal relative retention time prediction model; specifically, multiple molecular descriptors of each sample compound in the sample data set are first obtained, and then the multiple molecular descriptors are optimized to obtain the target molecular descriptors related to the relative retention time; finally, according to the target molecular descriptors and relative retention time labels of each sample compound in the sample data set, the relative retention time prediction model is optimized and trained to obtain the optimal relative retention time prediction model.

[0163] The prediction module 802 is used to input the target molecule descriptor of the test compound into the optimal relative retention time prediction model to obtain the relative retention time of the test compound output by the optimal relative retention time prediction model.

[0164] The test compound is a compound for which relative retention time prediction is required, and may be a compound of the type of medicine, pesticide, hormone, biochemical reagent, veterinary drug, etc., which is not specifically limited in this embodiment.

[0165] Optionally, when the relative retention time of the test compound needs to be predicted, the target molecule descriptor of the test compound can be obtained first. For specific acquisition steps, see Figure 1 Steps 101 and 102 in the above are not described in detail here.

[0166] Then, after the target molecule descriptor of the compound to be tested is input into the optimal relative retention time prediction model, the relative retention time prediction model learns the target molecule descriptor and can accurately output the relative retention time of the compound to be tested.

[0167] On the one hand, this embodiment calculates multiple molecular descriptors of each sample compound in the sample data set and optimizes the multiple molecular descriptors to obtain target molecular descriptors related to relative retention time from the multiple molecular descriptors, so that the input vector of the relative retention time prediction model contains abundant target molecular descriptors related to relative retention time, and thus the relative retention time prediction model can be constructed and obtained quickly and accurately; on the other hand, the relative retention time prediction model is obtained by training the target molecular descriptors that can characterize the physicochemical properties of the compound molecules and the relative retention time labels, so that the trained relative retention time prediction model can be used to accurately characterize the correlation between the physicochemical properties and the relative retention time, and the interpretability of the model is effectively improved; and the relative retention time prediction model can accurately predict the relative retention time of the test compound according to the target molecular descriptors of the test compound input in real time.

[0168] The device provided by the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific processes and detailed contents, which will not be repeated here.

[0169] Fig. 9 An example of a physical structure diagram of an electronic device is shown in FIG. Fig. 9As shown, the electronic device may include: a processor 901, a communication interface 902, a memory 903 and a communication bus 904, wherein the processor 901, the communication interface 902 and the memory 903 communicate with each other through the communication bus 904. The processor 901 may call the logic instructions in the memory 903 to execute a method for constructing a relative retention time prediction model or a method for predicting a relative retention time prediction model, the method comprising: obtaining a sample data set, and calculating a plurality of molecular descriptors of each sample compound in the sample data set; optimizing the plurality of molecular descriptors to obtain a target molecular descriptor related to the relative retention time in the plurality of molecular descriptors; optimizing and training the relative retention time prediction model according to the target molecular descriptors and relative retention time labels of each sample compound in the sample data set to obtain an optimal relative retention time prediction model; the optimal relative retention time prediction model is used to predict the relative retention time of the compound to be tested according to the target molecular descriptor of the compound to be tested;

[0170] Or obtaining an optimal relative retention time prediction model; inputting the target molecule descriptor of the compound to be tested into the optimal relative retention time prediction model to obtain the relative retention time of the compound to be tested output by the optimal relative retention time prediction model; the optimal relative retention time prediction model is constructed and generated based on the relative retention time prediction model construction method.

[0171] In addition, the logic instructions in the above-mentioned memory 903 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0172] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the relative retention time prediction model construction method or the relative retention time prediction model prediction method provided by the above methods, the method comprising: obtaining a sample data set, and calculating multiple molecular descriptors of each sample compound in the sample data set; optimizing the multiple molecular descriptors to obtain target molecular descriptors related to the relative retention time in the multiple molecular descriptors; optimizing and training the relative retention time prediction model according to the target molecular descriptors and relative retention time labels of each sample compound in the sample data set to obtain an optimal relative retention time prediction model; the optimal relative retention time prediction model is used to predict the relative retention time of the compound to be tested according to the target molecular descriptors of the compound to be tested;

[0173] Or obtaining an optimal relative retention time prediction model; inputting the target molecule descriptor of the compound to be tested into the optimal relative retention time prediction model to obtain the relative retention time of the compound to be tested output by the optimal relative retention time prediction model; the optimal relative retention time prediction model is constructed and generated based on the relative retention time prediction model construction method.

[0174] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when executed by a processor to execute the relative retention time prediction model construction method or relative retention time prediction model prediction method provided by the above methods, the method comprising: obtaining a sample data set, and calculating a plurality of molecular descriptors of each sample compound in the sample data set; optimizing the plurality of molecular descriptors to obtain a target molecular descriptor related to the relative retention time in the plurality of molecular descriptors; optimizing and training the relative retention time prediction model according to the target molecular descriptors and relative retention time labels of each sample compound in the sample data set to obtain an optimal relative retention time prediction model; the optimal relative retention time prediction model is used to predict the relative retention time of the compound to be tested according to the target molecular descriptor of the compound to be tested;

[0175] Or obtaining an optimal relative retention time prediction model; inputting the target molecule descriptor of the compound to be tested into the optimal relative retention time prediction model to obtain the relative retention time of the compound to be tested output by the optimal relative retention time prediction model; the optimal relative retention time prediction model is constructed and generated based on the relative retention time prediction model construction method.

[0176] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0177] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution 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 can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a relative retention time prediction model, characterized in that: include: Acquire a sample data set, and calculate a plurality of molecular descriptors of each sample compound in the sample data set; Optimizing the multiple molecular descriptors to obtain a target molecular descriptor related to the relative retention time among the multiple molecular descriptors; According to the target molecule descriptor and relative retention time label of each sample compound in the sample data set, the relative retention time prediction model is optimized and trained to obtain an optimal relative retention time prediction model; the optimal relative retention time prediction model is used to predict the relative retention time of the test compound according to the target molecule descriptor of the test compound; The calculating of a plurality of molecular descriptors of each sample compound in the sample data set comprises: Optimizing the sample data set; The optimization process includes deleting sample compounds whose relative retention time is less than a second preset value, and deleting sample compounds belonging to abnormal outliers based on the Cook distance; Calculating multiple molecular descriptors of each sample compound in the sample data set after optimization processing; The method of deleting sample compounds belonging to abnormal outliers based on Cook distance includes: Determine whether the Cook distance of each sample compound is greater than a distance threshold, regard the sample compounds greater than the distance threshold as sample compounds belonging to abnormal outliers, and delete the sample compounds belonging to abnormal outliers; The distance threshold is determined according to the number of sample compounds in the sample data set and the number of features of each sample compound. The calculation formula of the distance threshold is as follows: T = [4 / (MN-1)]; Wherein, T is the distance threshold, M is the number of sample compounds in the sample data set, and N is the number of features of each sample compound; The step of optimizing the plurality of molecular descriptors to obtain a target molecular descriptor related to the relative retention time among the plurality of molecular descriptors includes: Deleting from the plurality of molecular descriptors molecular descriptors having a correlation coefficient greater than a first preset value, molecular descriptors with missing values, and molecular descriptors whose mean, standard deviation, and relative standard deviation do not meet preset conditions; Then, Lasso regression analysis was performed on the remaining molecular descriptors and relative retention time labels, and parameter reduction was performed on the remaining molecular descriptors; Then, multiple linear regression was performed on the molecular descriptors and relative retention time labels retained after the Lasso regression analysis, and the significance level between each molecular descriptor retained after the Lasso regression analysis and the relative retention time label was obtained, and the target molecular descriptors with significance levels less than the significance threshold were retained.

2. The method for constructing a relative retention time prediction model according to claim 1, characterized in that: Also includes: generating each target molecule descriptor sequence according to each target molecule descriptor of all sample compounds in the sample data set; Based on the kernel density function, the distribution characteristics of each target molecule descriptor sequence are determined; According to the distribution characteristics, the applicable scope of the optimal relative retention time prediction model is determined.

3. The method for constructing a relative retention time prediction model according to claim 1, characterized in that: The relative retention time prediction model is generated based on a deep learning neural network.

4. The method for constructing a relative retention time prediction model according to claim 1, characterized in that: The categories of sample compounds in the sample data set at least include medicines, pesticides, hormones and biochemical reagents.

5. A prediction method for a relative retention time prediction model, characterized in that: include: Obtaining the optimal relative retention time prediction model; Inputting the target molecule descriptor of the compound to be tested into the optimal relative retention time prediction model to obtain the relative retention time of the compound to be tested output by the optimal relative retention time prediction model; The optimal relative retention time prediction model is constructed and generated based on the relative retention time prediction model construction method according to any one of claims 1 to 4.

6. A device for constructing a relative retention time prediction model, characterized in that: include: A calculation module, used for acquiring a sample data set and calculating a plurality of molecular descriptors of each sample compound in the sample data set; An optimization module, used for optimizing the multiple molecular descriptors to obtain a target molecular descriptor related to the relative retention time among the multiple molecular descriptors; A construction module is used to optimize and train the relative retention time prediction model according to the target molecule descriptor and relative retention time label of each sample compound in the sample data set to obtain an optimal relative retention time prediction model; the optimal relative retention time prediction model is used to predict the relative retention time of the test compound according to the target molecule descriptor of the test compound; The computing module is specifically used for: Optimizing the sample data set; The optimization process includes deleting sample compounds whose relative retention time is less than a second preset value, and deleting sample compounds belonging to abnormal outliers based on the Cook distance; Calculating multiple molecular descriptors of each sample compound in the sample data set after optimization processing; The method of deleting sample compounds belonging to abnormal outliers based on Cook distance includes: Determine whether the Cook distance of each sample compound is greater than a distance threshold, regard the sample compounds greater than the distance threshold as sample compounds belonging to abnormal outliers, and delete the sample compounds belonging to abnormal outliers; The distance threshold is determined according to the number of sample compounds in the sample data set and the number of features of each sample compound. The calculation formula of the distance threshold is as follows: T = [4 / (MN-1)]; Wherein, T is the distance threshold, M is the number of sample compounds in the sample data set, and N is the number of features of each sample compound; The optimization module is specifically used for: Deleting from the plurality of molecular descriptors molecular descriptors having a correlation coefficient greater than a first preset value, molecular descriptors with missing values, and molecular descriptors whose mean, standard deviation, and relative standard deviation do not meet preset conditions; Then, Lasso regression analysis was performed on the remaining molecular descriptors and relative retention time labels, and parameter reduction was performed on the remaining molecular descriptors; Then, multiple linear regression was performed on the molecular descriptors and relative retention time labels retained after the Lasso regression analysis, and the significance level between each molecular descriptor retained after the Lasso regression analysis and the relative retention time label was obtained, and the target molecular descriptors with significance levels less than the significance threshold were retained.

7. A prediction device for a relative retention time prediction model, characterized in that: include: An acquisition module, used for acquiring an optimal relative retention time prediction model; A prediction module, used for inputting the target molecule descriptor of the compound to be tested into the optimal relative retention time prediction model to obtain the relative retention time of the compound to be tested output by the optimal relative retention time prediction model; The optimal relative retention time prediction model is constructed and generated based on the relative retention time prediction model construction method according to any one of claims 1 to 4.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the method for constructing the relative retention time prediction model as described in any one of claims 1 to 4, or implements the steps of the prediction method of the relative retention time prediction model as described in claim 5.

Images