ADMET property prediction method and system for anticancer drug candidates

Abstract

The present invention belongs to the technical field of ADMET property data processing of anticancer candidate drugs and provides a method and system for predicting the ADMET properties of anticancer candidate drugs. The method includes obtaining molecular descriptor information of anticancer candidate drug compounds; performing multi-feature extraction on the molecular descriptor information of anticancer candidate drug compounds to screen out key features that determine the properties of ADMET compounds; performing ADMET property classification prediction on the screened key features to obtain ADMET property prediction results of anticancer candidate drugs.

Description

Technical Field

[0001] The present invention belongs to the technical field of ADMET property data processing of anticancer molecule candidate anticancer drugs, and in particular relates to a method and system for predicting the ADMET property of anticancer candidate drugs. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the development of anticancer drugs, potential drug compounds are screened by establishing an effective expression model of the bioactivity of the cancer target protein, and new compound molecules with better inhibition of the target bioactivity are predicted. At the same time, in addition to paying attention to the bioactivity of the candidate drug anticancer molecules to inhibit the target, it is more necessary to consider the impact of the ADMET properties of the compounds that make up the anticancer drugs on human health in the further drug development process. The ADMET properties mainly refer to the need to have good pharmacokinetic properties and safety in the human body. Among them, A stands for absorption, D stands for distribution, M stands for metabolism, E stands for excretion, and T stands for toxicity. ADME mainly refers to the pharmacokinetic properties of the compound, which describes the law of changes in the concentration of the compound in the organism over time, and T mainly refers to the toxic and side effects that the compound may produce in the human body.

[0004] Therefore, a successful therapeutic drug not only needs to rely on its compound anticancer molecules to antagonize the expression of the biological activity of cancer target proteins, but also needs to meet the pharmacokinetic properties and biological toxicity of the drug, such as the permeability of the drug compound in the small intestinal epithelial cells, the degree of compound metabolism, the oral bioavailability of the compound, and whether the compound has cardiotoxicity, genotoxicity, etc. However, due to the high data dimensionality of the compounds and their molecular descriptors, the limited amount of data, and the similarity of the compound structures, it is very challenging to establish an accurate ADMET property prediction model for anticancer drug candidates.

[0005] The inventors found that in terms of the prediction of the properties of anticancer candidate drugs, the traditional anticancer candidate drug feature representation is high-dimensional and sparse, the feature expression ability is weak, and the traditional algorithm needs to consume a lot of manpower and material resources in feature engineering to find an effective method to predict the ADMET properties of drug compounds. Therefore, the prediction of drug compound properties is a complex and lengthy process. In addition, for anticancer drugs, due to their strong domain, large amount of data and complex content, the existing feature extraction method is not accurate. Summary of the invention

[0006] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a method and system for predicting the ADMET properties of anticancer candidate drugs, which models the correlation between the descriptors of a substance and its physicochemical properties, thereby revealing the influence of certain characteristics of the substance compound molecules on the properties, thereby improving the accuracy of the ADMET property prediction results of the anticancer candidate drugs.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] The first aspect of the present invention provides a method for predicting the ADMET properties of an anticancer drug candidate, comprising:

[0009] Obtain molecular descriptor information of anticancer drug candidate compounds;

[0010] Perform multi-feature extraction on the molecular descriptor information of anticancer drug candidate compounds to screen out the key features that determine the properties of ADMET compounds;

[0011] The ADMET properties of the screened key features are classified and predicted to obtain the ADMET property prediction results of the anticancer drug candidates.

[0012] A second aspect of the present invention provides an ADMET property prediction system for anticancer drug candidates, comprising:

[0013] A data acquisition module, which is used to obtain molecular descriptor information of anticancer candidate drug compounds;

[0014] The variable screening module is used to extract multiple features from the molecular descriptor information of candidate anticancer drug compounds and screen out the key features that determine the properties of ADMET compounds;

[0015] The property prediction module is used to classify and predict the ADMET properties of the screened key features to obtain the ADMET property prediction results of the anticancer candidate drugs.

[0016] The third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for predicting the ADMET properties of anticancer drug candidates as described above.

[0017] The fourth aspect of the present invention 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 steps in the method for predicting the ADMET properties of candidate anticancer drugs as described above are implemented.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] The present invention provides a method and system for predicting the ADMET properties of anticancer candidate drugs, which extracts multiple features from molecular descriptor information of anticancer candidate drug compounds, screens out key features that determine the properties of ADMET compounds, and then classifies and predicts the ADMET properties of the screened key features, performs efficient and reliable dimensionality reduction processing on the ADMET properties, combines feature and sensitivity parameter analysis, and embeds them into a machine learning model, thereby improving the accuracy of the prediction model and the accuracy of the ADMET property prediction results of the anticancer candidate drugs;

[0020] The present invention aims at the classification prediction of the ADMET properties of candidate anticancer drug compounds, establishes an effective model based on the traditional machine learning method, mines and analyzes the anticancer drug candidate related compounds and molecular descriptor information and ADMET property data, studies its molecular mechanism, and uses the model to predict the ADMET properties of candidate anticancer drugs, thereby developing new compound molecules with better drug properties, or guiding the structural optimization of existing active compounds, accelerating the research and development process of cancer candidate drugs, and providing a theoretical reference for clinical cancer treatment research.

[0021] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0023] Figure 1 is a flow chart of a method for predicting the ADMET properties of an anticancer drug candidate according to an embodiment of the present invention;

[0024] FIG. 2( a ) is a visualization diagram of Caco-2 properties and correlation coefficients among the properties of five types of drug compounds in an embodiment of the present invention;

[0025] FIG2( b ) is a visualization diagram of the CYP3A4 properties and correlation coefficients among the properties of the five types of drug compounds in the embodiments of the present invention;

[0026] Figure 3 : is a confusion matrix diagram of the ADMET property prediction model of the support vector machine SVM in an embodiment of the present invention;

[0027] Figure 4 Schematic diagram of n_estimators evaluation in the random forest model in an embodiment of the present invention;

[0028] Figure 5Schematic diagram of the structure of the ADMET property prediction system for anticancer drug candidates according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0030] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0031] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0032] Embodiment 1

[0033] The biological activity of compounds for successful cancer treatment is extremely important. That is, if the biological activity of a compound molecule that makes up a drug is very weak, the corresponding drug will have a poor effect in treating the disease and will not achieve the goal of preventing or treating the disease. However, in the process of predicting anticancer drugs, it is far from enough to only consider the strength of its inhibition of target biological activity. The chemical properties of the drug molecules must also be considered, and different compound properties are affected by different key molecular descriptors.

[0034] Reference Figure 1 This embodiment provides a method for predicting the ADMET properties of anticancer drug candidates, which specifically includes the following steps:

[0035] S101: Acquire molecular descriptor information of candidate anticancer drug compounds.

[0036] S102: Perform multi-feature extraction on the molecular descriptor information of candidate anticancer drug compounds to screen out the key features that determine the properties of ADMET compounds.

[0037] The properties of the ADMET compound include properties of Caco-2, CYP3A4, hERG, HOB and MN; wherein Caco-2 represents the permeability of the compound to the small intestinal epithelial cells, '1' represents good permeability of the compound to the small intestinal epithelial cells, and '0' represents poor permeability of the compound to the small intestinal epithelial cells.

[0038] CYP3A4 indicates whether a compound can be metabolized by CYP3A4, wherein '1' represents that the compound can be metabolized by CYP3A4, and '0' represents that the compound cannot be metabolized by CYP3A4.

[0039] hERG indicates whether the compound is cardiotoxic, '1' means the compound is cardiotoxic, and '0' means the compound is not cardiotoxic.

[0040] HOB represents the oral bioavailability of the compound, where '1' represents that the compound has good oral bioavailability and '0' represents that the compound has poor oral bioavailability.

[0041] MN indicates whether the compound is genotoxic, where '1' means the compound is genotoxic and '0' means the compound is not genotoxic.

[0042] For each property of compound ADMET, the selection of data variables may be the same or different, and the same is true for the selection of the number of variables. Based on the provided sample information, before predicting the five types of ADMET properties of the compound, the data information is first cleaned to screen key variables, thereby providing more accurate services for compound property prediction.

[0043] In the specific implementation process, multi-feature extraction is performed on the molecular descriptor information of anti-cancer candidate drug compounds based on the random forest ensemble model.

[0044] In applications such as medical genetic engineering diagnosis and biological cell identification, the number of data or features is often several orders of magnitude greater than the number of instance categories. A large number of features will increase the noise of the data, thereby increasing the error of the learning algorithm. Therefore, feature selection techniques are used to remove features that are irrelevant or problematic to the classification data, thereby reducing the data dimension.

[0045] Integration algorithms represented by the random forest algorithm can not only classify data, but also evaluate the importance of each variable. As a supervised machine learning paradigm, ensemble learning uses multiple models to solve the same problem, solving the problem of limited predictive performance of a single classifier system.

[0046] This embodiment uses a random forest ensemble model to fit the compound sample data and the feature subset after molecular descriptor feature screening, evaluates all features from the original data set, retains relevant features that are efficient and reliable for data classification, and generates a diverse and high-information feature set, which is the key to obtaining good integrated classification results. The optimal feature subset helps to improve classification accuracy.

[0047] The random forest ensemble model (RF) has the characteristics of replacement sampling and random feature selection. RF consists of a series of classifiers {h(x, θk), k=1, 2...n}, where {θk} is an independent and identically distributed random variable, which is used to control the growth of each classifier; the variable n represents the number of classifiers; the independent variable x represents the input data set sample, and the classification label with the most votes from all classifiers is assigned to x.

[0048] When selecting node attributes, the feature selection algorithm randomly selects F (F≤N) attributes from all attributes, and compares the attributes with the best splitting rules on these F attributes to split the nodes. The value of F is the random feature variable. The random variable selection of this embodiment adopts the encapsulated selection algorithm to obtain a relatively simplified set. When processing the feature data set, each classifier of RF will randomly select X sample training sets with replacement through the Bootstrap method, and the remaining unselected data is the out-of-bag data (Out-Of-Bag, OOB).

[0049] Specifically, the key features of the properties of ADMET compounds are determined based on the correlation between the various molecular descriptors of the anticancer drug candidate compounds and the extracted features.

[0050] The process of screening out the key features that determine the properties of ADMET compounds is as follows:

[0051] Calculate the importance of each feature based on the set feature importance formula and sort them in descending order;

[0052] According to the feature importance, the features with a preset elimination ratio are eliminated to obtain a new feature set;

[0053] Repeat the above importance ranking and feature elimination process for the new feature set until a preset number of features remain;

[0054] According to the obtained feature sets and the out-of-bag error rates corresponding to the feature sets, the feature set with the lowest out-of-bag error rate is selected as the key feature for determining the properties of the ADMET compound.

[0055] The feature importance formula is set as:

[0056]

[0057] in, represents OOB, B represents the selected data, I(·) represents the representation function, yi represents The i-th classification label in It represents the classification label of xj after RF performs feature permutation after fetching the data, and hk(i) represents the classification label of i predicted for the sample set Bk.

[0058] After fitting the data, the random forest model will evaluate the importance of the given training attribute columns after fitting. In the variable importance measurement array, the attribute columns with larger values ​​are more important for the accuracy of the prediction.

[0059] It should be noted that those skilled in the art may also use other existing neural network models to perform multi-feature extraction on the molecular descriptor information of candidate anti-cancer drug compounds, which will not be described in detail here.

[0060] In the process of training the random forest ensemble model, sample data with existing labels are obtained. Among them, an effective drug compound for treating cancer must not only inhibit the expression of the biological activity of the cancer target to a certain extent but also have good ADMET properties.

[0061] This example requires that according to the sample data information provided, there are 1974 compound samples, each sample has 729 molecular descriptor variables, and five types of ADMET property data. In this example, breast cancer is used as an example: Figure 4 An evaluation of n_estimators (n estimators) in the random forest model in this embodiment is given.

[0062] Combined with the fact that the data sample contains a large number of features but a small sample size, it is necessary to use feature selection technology to purify the data processing. At the same time, the 729 molecular features of the compounds provided by the data are different for the five types of ADMET properties. Therefore, the feature selection of the ensemble learning random forest is used to automatically analyze the molecular descriptor features, evaluate all features from the original data set, and retain the relevant features that are efficient and reliable in reducing the dimensions of the five properties of ADMET.

[0063] S103: Classify and predict the ADMET properties of the screened key features to obtain the ADMET property prediction results of the anticancer drug candidate.

[0064] In this embodiment, a property prediction model is used to perform ADMET property classification prediction on the screened key features; wherein the property prediction model is a support vector machine.

[0065] It should be noted that those skilled in the art may also use other existing neural network models to perform ADMET property classification prediction on the screened key features, which will not be described in detail here.

[0066] In this embodiment, it is known that different drug molecule descriptors have different properties, and compounds are composed of a large number of different molecules. Therefore, based on the known ADMET properties of 1974 compounds, a property prediction model is constructed from the perspective of molecular descriptors, and corresponding parameters are trained to predict the ADMET properties of the compounds respectively.

[0067] First, for the five different characteristics of drug compound ADMET, the key features are different, that is, the key variables are different, and the input variables of the model are different. Therefore, this embodiment first constructs a random forest model, extracts key features according to different properties of ADMET, and obtains the correlation coefficient values ​​of different molecular descriptors, as shown in Tables 1 and 2 below, and the visualization effect is shown in Figures 2 (a) and 2 (b), and the larger the correlation coefficient value, the more obvious the positive correlation between the two.

[0068] With regard to the five types of Caco-2, CYP3A4, hERG, HOB, and MN properties described in the ADMET data of the compound, it can be seen that Caco-2, CYP3A4, and HOB represent the intestinal epithelial cell permeability, metabolic capacity, and oral bioavailability of the compound, respectively. When these three indicators are all 1, it means that the compound has good performance; hERG and MN represent whether the compound has cardiotoxicity and whether it has genotoxicity, respectively. When these two indicators are both 0, it means that the compound has good performance.

[0069] The molecular descriptors were positively correlated with the properties of Caco-2, CYP3A4, and HOB, and negatively correlated with the properties of hERG and MN, respectively, and the training parameters of the model were optimized.

[0070] Table 1 Molecular descriptors related to the three properties of Caco-2, CYP3A4, and HOB

[0071]

[0072]

[0073] Table 2 Molecular descriptors related to two properties of MN and hREG

[0074]

[0075] From the analysis of Table 1 and Table 2, it can be seen that for genotoxicity (MN), the molecular descriptors with higher correlation coefficients are ETA_BetaP_s, ETA_EtaP_F, ETA_EtaP_F_L, ETA_EtaP_L, ETA_Epsilon_1, ETA_BetaP, ETA_dEpsilon_A, etc. From the structure of the molecular descriptors, it can be seen that this type of molecular descriptors have a common chemical group ETA. Therefore, it can be predicted that the chemical group ETA has the possibility of genotoxicity.

[0076] In this embodiment, the random forest ensemble model feature selection is used to obtain the key features related to the five properties of ADMET, and then prediction models are established for Caco-2, CYP3A4, hERG, HOB, and MN of the compound.

[0077] This embodiment applies the machine learning support vector machine (SVM) algorithm to the classification prediction of drug compound-ADMET properties. According to the different properties of drug compounds, the key variables of each influencing factor are extracted, and a support vector machine (SVM) classifier prediction model is constructed to overcome the main problems existing in the prior art and provide new ideas for the prediction of drug compound properties.

[0078] The basic idea of ​​support vector machine (SVM) is to find the separating hyperplane that can correctly divide the training data set and has the largest geometric interval. Its algorithm has been widely used in image recognition, text classification and other fields. The algorithm is also widely used in the field of drug molecules.

[0079] When using support vector machines to solve classification problems, the basic idea is to find a partitioning hyperplane based on the training set D to separate samples of different categories. The training set of the two-class support vector machine is:

[0080] D={(x1, y1), (x2, y2),..., (x m ,y m )},y i ∈{-1, ,1}

[0081] where x i Represents the vector of the i-th input, y i represents the value of the ith output, and m represents the number of samples. The purpose of building a model is to provide a regression function y = f(x) that can be used to calculate the regression function of a new input x. i To predict y i A support vector regression is expressed as follows:

[0082] There exists the following partitioning hyperplane:

[0083] ω Tx+b=0

[0084] Where ω=(w1;w2;...;w d ) is the normal vector, d is the dimension, and b is the offset. The maximum interval of classification is:

[0085]

[0086] sty i (ω T x i +b)≥1, i=1, 2,..., m.

[0087] Among them, ω and b are the hyperplane parameters that need to be solved, ω represents the coefficient vector, b represents the displacement, x i represents the vector of the i-th input, and m represents the number of samples.

[0088] In practical applications, there are a large number of linear inseparable problems. There is no reasonable hyperplane in the original sample space to ensure the correct division of the samples. Then, by selecting the appropriate kernel function K(x,x′) and the appropriate parameter C, we build and solve the optimization problem.

[0089]

[0090]

[0091] 0≤α i ≤C, i=1, 2, ..., m.

[0092] Where α is the kernel hyperparameter, which defines the characteristic length scale of the similarity between learning samples, that is, the ratio of the distance between samples before and after feature space mapping from the perspective of weight space, and x i Represents the vector of the i-th input, y i represents the value of the i-th output, and m represents the number of samples.

[0093] The optimal solution obtained: According to the selection of α * A positive component of Calculate the threshold:

[0094]

[0095] In the formula, represents the optimal kernel parameter, K(·) represents the kernel function, and x i Represents the vector of the i-th input, y i represents the value of the i-th output, l represents the number of samples, b * Represents the calculation threshold.

[0096] Construct a decision function:

[0097]

[0098] Where sgn(·) is a step function, represents the current optimal kernel parameter, y i represents the output value of the i-th sample, K(·) represents the kernel function, b * represents the calculation threshold, and l represents the number of samples.

[0099] Based on the test data, the ADMET property model of drug compounds obtained through training is used for prediction and evaluation;

[0100] Accuracy is usually used as a measure of classification effect, but accuracy is not applicable to the evaluation of all classification situations. This problem is obvious in unbalanced classification. Take medical testing as an example. A large number of people in the hospital test samples are disease-free, and only a small number of people are sick. If accuracy is used as the basis for the test results, it may be found that the accuracy of the test results is as high as 99% or more, but this 99% accuracy shows that the patients are disease-free, and those who are sick may also be tested as disease-free. Such test results cause immeasurable losses to patients.

[0101] In this example, the classification prediction of the ADMET characteristics of drug compounds is also an unbalanced classification problem, so the accuracy is not used as the classification evaluation index. This example uses confusion matrix, Precision, Recall, F1 _scores The ADMET properties of the compounds were evaluated by using indicators such as

[0102] The accuracy ACC refers to the ratio of the number of samples correctly classified by the classifier to the total number of samples for a given test data set.

[0103]

[0104] Among them, TP: true positives, the positive class is judged as the positive class; FP, false positives, the negative class is judged as the positive class, "keeping the false"; FN, false negatives, the positive class is judged as the negative class, "removing the true"; TN, true negatives, the negative class is judged as the negative class.

[0105] The precision is based on the prediction results and refers to the proportion of samples with correct compound properties predicted to the total number of all predicted samples.

[0106]

[0107] The recall rate Recall refers to the ratio of the data of the correctly predicted compound to the number of samples of the corresponding properties of the actual compound.

[0108]

[0109] The F1 _scores (Balanced F-score) refers to the harmonic mean of precision and recall.

[0110]

[0111] Using the 729 molecular descriptors provided in this example and the ADMET data of the corresponding 1974 compounds, after obtaining the candidate features, the classification accuracy of the compounds for Caco-2, CYP3A4, hERG, HOB, and MN was calculated by the support vector machine SVM, and the 50 compounds in the test table were given for corresponding predictions.

[0112] SVM was trained with 1700 sample data and the validation set was 274 compound data. The relevant molecular features corresponding to the five properties of ADMET were used as input training samples of the binary support vector machine in turn. The training and test sample points were extracted through 10 rounds of cross-validation, and the experimental results in Table 3 were obtained.

[0113] In this embodiment, the five properties are evaluated using four evaluation indicators, namely Accuracy, Precision, Recall and F1_scores.

[0114] Table 3 Classification experiment results

[0115] Accuracy Precision Recall F1-score Caco-2 94.16％ 98.53％ 93.22％ 96.50％ CYP3A4 92.34％ 78.35％ 86.36％ 84.30％ HkDJ 88.33％ 82.46％ 74.42％ 78.48％ HOB 90.88％ 95.34％ 94.14％ 94.73％ MN 89.94％ 95.21％ 71.79％ 92.33％

[0116] In this embodiment, as shown in Table 3, the small intestinal epithelial cell permeability of "Caco-2" measuring the ability of a compound to be absorbed by the human body achieved 94.16% in the Accuracy index, 98.53% in the Precision index, 93.22% in the Recall index, and the F1_scores index used for balance also achieved an accuracy of 96.50%. For the other four properties, the classification model used in this embodiment also achieved an accuracy of more than 88%. Although the support vector machine algorithm is relatively simple, it can effectively solve the classification problem for the case of a large number of feature sets and a small number of samples.

[0117] In this embodiment, in addition to providing an analysis of quantitative evaluation indicators, detailed information on correct and incorrect sample classification is further provided. Therefore, a confusion matrix is ​​drawn using tools such as python as shown in Tables 4 to 8, where positive refers to the number of samples predicted correctly, and negative refers to the number of samples predicted incorrectly. Since this embodiment classifies and predicts the ADMET data attributes of the molecular descriptors of compounds, the goal of the model is to maximize the accuracy of predicting the ADMET attributes of compounds.

[0118] Table 4 Confusion matrix of Caco-2 property prediction by support vector machine

[0119]

[0120] Table 5 Confusion matrix of CYP3A4 property prediction by support vector machine

[0121]

[0122] Table 6 Confusion matrix of hERG property prediction by support vector machine

[0123]

[0124] Table 7 Confusion matrix of HOB property prediction by support vector machine

[0125]

[0126] Table 8 Confusion matrix of MN property prediction of support vector machine

[0127]

[0128] In this embodiment, the support vector machine (SVM) kernel function is proposed to solve the problem of linear inseparability of data, and maps data from the source space to the target space (linearly separable space). The linear kernel function and the Gaussian kernel function are used as two indicators for parameter analysis. Table 9 shows the accuracy of the five properties using the two kernel functions. In order to intuitively see the difference in accuracy between the two, a bar chart comparing the two is drawn according to Table 9. Figure 3 .

[0129] Table 9 Support vector machine SVM kernel function parameter analysis

[0130] Linear kernel function Gaussian kernel function Caco-2 94.16％ 86.50％ CYP3A4 92.34％ 86.50％ HkDJ 85.40％ 78.10％ HOB 90.88％ 89.42％ MN 83.94％ 82.48％

[0131] In this embodiment, combined with Table 9 and Figure 3The experimental data show that the linear kernel function is more effective in solving the binary ADMET property prediction problem. The accuracy of the linear kernel function in the classification prediction of the five properties is 6%-8% higher than that of the Gaussian kernel function on average. This also indirectly shows that the molecular descriptor variables, biological activity and ADMET properties given by the experiment present a complex linear relationship, which provides theoretical and experimental support for further solutions.

[0132] It should be noted here that the ADMET property prediction method of the anticancer drug candidate of this embodiment is applicable not only to breast cancer but also to other cancers, which will not be described in detail here.

[0133] Embodiment 2

[0134] like Figure 5 As shown, this embodiment provides an ADMET property prediction system for anticancer drug candidates, which specifically includes the following modules:

[0135] (1) a data acquisition module, which is used to obtain molecular descriptor information of anticancer candidate drug compounds;

[0136] (2) a variable screening module, which is used to extract multiple features from the molecular descriptor information of anticancer drug candidate compounds and screen out the key features that determine the properties of ADMET compounds;

[0137] Specifically, in the variable screening module, multi-feature extraction is performed on the molecular descriptor information of candidate anticancer drug compounds based on a random forest ensemble model.

[0138] Based on the correlation between the molecular descriptors of the anticancer drug candidate compounds and the extracted features, the key features of the ADMET compound properties were determined.

[0139] (3) A property prediction module, which is used to classify and predict the ADMET properties of the screened key features to obtain the ADMET property prediction results of the anticancer drug candidates.

[0140] Specifically, in the property prediction module, a property prediction model is used to perform ADMET property classification prediction on the screened key features; wherein the property prediction model is a support vector machine.

[0141] It should be noted here that each module in this embodiment corresponds to each step in Example 1 one by one, and the specific implementation process is the same, which will not be repeated here.

[0142] Embodiment 3

[0143] This embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for predicting the ADMET properties of anticancer drug candidates as described above.

[0144] Embodiment 4

[0145] This embodiment provides an electronic device, including 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 steps in the method for predicting the ADMET properties of candidate anticancer drugs as described above are implemented.

[0146] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products of the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of the processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0147] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the ADMET properties of an anticancer drug candidate, characterized in that: include: Obtain molecular descriptor information of anticancer drug candidate compounds; Perform multi-feature extraction on the molecular descriptor information of candidate anticancer drug compounds to screen out the key features that determine the properties of ADMET compounds; Classify and predict the ADMET properties of the selected key features to obtain the ADMET property prediction results of the anticancer drug candidates; Based on the correlation between each molecular descriptor of the anticancer drug candidate compound and each extracted feature, the key features of the ADMET compound properties are determined; The process of screening out the key features that determine the properties of ADMET compounds is: Calculate the importance of each feature based on the set feature importance formula and sort them in descending order; According to the feature importance, the features with a preset elimination ratio are eliminated to obtain a new feature set; Repeat the above importance ranking and feature elimination process for the new feature set until a preset number of features remain; According to the obtained feature sets and the out-of-bag error rates corresponding to the feature sets, the feature set with the lowest out-of-bag error rate is selected as the key feature for determining the properties of the ADMET compound; The feature importance formula is set as: in, represents the out-of-bag data, B represents the selected data, I(·) represents the representation function, y i represent The i-th classification label in Represents RF after taking data and performing feature permutation x j The classification label, h k (i) represents the sample set B k , the predicted classification label of i; The properties of the ADMET compound include properties of Caco-2, CYP3A4, hERG, HOB and MN; wherein Caco-2 indicates the permeability of the compound to the small intestinal epithelial cells, CYP3A4 indicates whether the compound can be metabolized by CYP3A4, hERG indicates whether the compound has cardiotoxicity, HOB indicates the oral bioavailability of the compound, and MN indicates whether the compound has genotoxicity; Multi-feature extraction of molecular descriptor information of anticancer candidate drug compounds was performed based on the random forest ensemble model; firstly, a random forest model was constructed to extract key features according to different properties of ADMET, and the correlation coefficient values ​​of different molecular descriptors were obtained. The larger the correlation coefficient value, the more obvious the positive correlation between the two; the molecular descriptors were positively correlated with the properties of Caco-2, CYP3A4, and HOB, and negatively correlated with the properties of hERG and MN, and the training parameters of the model were optimized; The random forest ensemble model feature selection was used to obtain the key features related to the five properties of ADMET, and then prediction models were established for Caco-2, CYP3A4, hERG, HOB, and MN of the compounds respectively; A property prediction model is used to classify and predict the ADMET properties of the selected key features; wherein the property prediction model is a support vector machine; Based on the training set D, a dividing hyperplane is found to separate samples of different categories. The training set of the binary support vector machine is: D={(x1,y1),(x2,y2),...,(x m ,y m )},y i ∈{-1,,1} where x i Represents the vector of the i-th input, y i represents the value of the ith output, and m represents the number of samples. The purpose of building a model is to provide a regression function y = f(x) that can be used to calculate the regression function of a new input x. i To predict y i ; Support vector regression is expressed by the following formula: There exists the following partitioning hyperplane: oh T x+b=0 Where ω=(w1;w2;...;w d ) is the normal vector, d is the dimension, and b is the offset. The maximum interval of classification is: cool t (ω T xi+b)≥1,i=1,2,...,m. Among them, ω and b are the hyperplane parameters that need to be solved, ω represents the coefficient vector, b represents the displacement, x i represents the vector of the i-th input, and m represents the number of samples; By selecting an appropriate kernel function K(x,x′) and appropriate parameters C, we can build and solve the optimization problem. 0≤a t ≤C,i=1,2...,m Where α is the kernel hyperparameter, which defines the characteristic length scale of the similarity between learning samples, that is, the ratio of the distance between samples before and after feature space mapping from the perspective of weight space, and x i Represents the vector of the i-th input, y i represents the value of the i-th output, and m represents the number of samples; The optimal solution obtained: According to the selection of α * A positive component of Calculate the threshold: In the formula, represents the optimal kernel parameter, K(·) represents the kernel function, and x i Represents the vector of the i-th input, y i represents the value of the i-th output, l represents the number of samples, b * represents the calculation threshold; Construct a decision function: Where sgn(·) is a step function, represents the current optimal kernel parameter, y i represents the output value of the i-th sample, K(·) represents the kernel function, b * represents the calculation threshold, and l represents the number of samples.

2. A system for predicting the ADMET properties of an anticancer drug candidate, using the method for predicting the ADMET properties of an anticancer drug candidate as claimed in claim 1, characterized in that: include: A data acquisition module, which is used to obtain molecular descriptor information of anticancer candidate drug compounds; The variable screening module is used to extract multiple features from the molecular descriptor information of candidate anticancer drug compounds and screen out the key features that determine the properties of ADMET compounds; The property prediction module is used to classify and predict the ADMET properties of the screened key features to obtain the ADMET property prediction results of the anticancer candidate drugs.

3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the method for predicting the ADMET properties of anticancer drug candidates as claimed in claim 1 are implemented.

4. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the method for predicting the ADMET properties of anticancer drug candidates as claimed in claim 1 are implemented.

Images