Drug interaction intelligent prediction system and method based on multi-source drug characteristics
Through the intelligent prediction system of multi-source drug characteristics, the feature information in the drug database is integrated, the feature matrix is constructed and mapping and regulation is carried out, and the problem of insufficient and complexity of drug interaction prediction in the existing technology is solved, achieving more efficient and accurate drug interaction prediction provides a more reliable reference for drug research and development.
Patent Information
- Application Number
- CN202410873620.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-01
- Publication Date
- 2024-11-08
AI Technical Summary
The existing drug interaction prediction methods are based on similarity calculations, and cannot effectively capture the complex nonlinear relationships of drug interactions, and insufficient feature information extraction, resulting in limited prediction accuracy and scope of application.
Using an intelligent prediction system with multi-source drug characteristics, the characteristic matrix of consistent scale is constructed by integrating characteristic information such as pathways, enzymes, targets and chemical substructures, and combining mapping regulation and regularization processing to determine the drug pairing relationship to achieve more accurate and efficient drug interaction prediction.
It improves the accuracy and system performance of drug interaction prediction, provides more comprehensive input information, enhances the expression ability of the model, has a wide range of application, and provides a more reliable reference for drug research and development.
Smart Images

Figure CN118918982A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of drug interaction prediction in bioinformatics, and in particular to a drug interaction intelligent prediction system based on multi-source drug features and a method thereof. Background Art
[0002] Drug interaction occurs when the effect of one drug is altered by the co-use of another drug; predicting drug interactions is an important part of drug development, helping to reduce unexpected drug interactions, lower drug development costs, and optimize the drug design process; however, traditional drug interaction prediction methods, such as testing cytochrome P450, face challenges such as high costs and long cycles.
[0003] With the development of machine learning technology, researchers can collect data from public drug databases, including drug interaction information, targets, enzymes, pathways, etc., which can provide valuable information for identifying potential drug interactions. Currently, most of the existing drug interaction prediction methods are based on similarity calculations, which infer the interaction relationship between drugs by calculating the structural, functional, pharmacological and other similarities between them. However, different similarity measures may lead to different prediction results and cannot capture the complex nonlinear relationship of drug interactions. In addition, although many public databases provide relevant information on various drugs, there is still a lack of a comprehensive summary of drug interactions, resulting in limited interaction prediction capabilities for new drugs.
[0004] Therefore, existing drug interaction prediction methods still need to be improved. Summary of the invention
[0005] The purpose of the present invention is to propose a drug interaction intelligent prediction system and method based on multi-source drug features to solve the problem of insufficient feature information extraction in current drug interaction prediction methods; the system and processing method comprehensively consider the diversity of drug features, and construct a feature matrix of consistent scale by integrating feature information in multiple drug databases, such as pathways, enzymes, targets and chemical substructures, to provide a more comprehensive solution for drug interaction prediction; at the same time, through mapping regulation and regularization processing, the drug pairing relationship in the interaction space is determined to achieve more accurate and efficient drug interaction prediction analysis, and the system has a reasonable structure, high prediction accuracy, strong performance and a wide range of applications.
[0006] To achieve the above object, the present invention is achieved through the following technical solutions:
[0007] A drug interaction intelligent prediction system based on multi-source drug features, characterized by comprising:
[0008] The drug information collection and preprocessing module performs data collection, data cleaning and data preprocessing, and constructs a data set containing four types of characteristic information, namely pathway, enzyme, target and chemical substructure, and a data set containing drug interaction information. The module includes a data collection device and a data preprocessing device;
[0009] The drug feature matching module calculates the similarity between every two drugs under the four features of X, thereby converting the four feature matrices into four similarity matrices, using the similarity matrices to determine the matching relationship between the drugs, and iteratively normalizing the similarity matrices. The module includes a similarity calculation device and a normalization processing device;
[0010] The drug structure information enhancement module learns the intrinsic structure of drugs from their nearest neighbors, which can be used to predict new interactions. It finds the nearest neighbors of each drug based on distance calculation, and learns the local structural similarities between drugs from the nearest neighbors to enhance the structural information that can be expressed in the drug similarity matrix, and iteratively normalizes the similarity matrix. This module includes a nearest neighbor learning device and a normalization processing device;
[0011] The drug interaction prediction module inputs the drug similarity matrix training model, obtains the weight matrix of each feature, accumulates the product of the four feature matrices and their corresponding weight matrices, obtains the drug interaction prediction matrix, subtracts it from the drug interaction observation matrix, and the position with a value of 1 represents a potential drug interaction, and outputs a potential drug interaction list.
[0012] The data acquisition device of the drug interaction intelligent prediction system based on multi-source drug features collects drug information including drug interactions, pathways, enzymes, targets and chemical substructures from the drug database, and constructs a data set X including four types of feature information including pathways, enzymes, targets and chemical substructures and a data set Y including drug interaction information;
[0013] The settings of the data acquisition device include:
[0014] (4) Collect drug information from DrugBank and KEGG DRUG drug databases, including drug interactions DDI, enzyme DE, target DT, and chemical substructure DS;
[0015] (5) Use the ID mapping system provided by the UniPort database to map the drug target to the KEGG DRUG database to obtain the drug pathway DP;
[0016] (6) Remove drug data without any interactive information to ensure that the number of drugs in X and Y is equal;
[0017] (7) Define data set X = {DP, DE, DT, DS}, Y = {DDI}.
[0018] The described intelligent prediction system for drug interactions based on multi-source drug features, the data preprocessing device, is used to process the data in the data sets X and Y, and construct the feature matrix and the drug interaction observation matrix;
[0019] The settings of the data preprocessing device include:
[0020] (4) Take DP, DE, DT, DS∈X so that they correspond to a set of descriptors respectively, and then each drug X in X n Represented as four binary eigenvectors, X n = {DP n ,DE n ,DT n ,DS n}, its length indicates the number of descriptors in each group, its value 1 or 0 indicates the presence or absence of the corresponding descriptor, n is the index of each drug in X, n∈[1,m], m is the number of drugs in X;
[0021] (5) Take X n ∈X, define the feature matrix of X i represents the i-th feature matrix of X, for H i , each row of which represents a drug feature vector, with a total of m rows, where m is the number of drugs in X;
[0022] (6) Take Y i ,Y j ∈Y, define the drug interaction observation matrix J of Y, if drug Y i and drug Y j If there is interaction between ij =1, otherwise, J ij =0, i,j∈[1,m], m is the number of drugs in Y.
[0023] The drug interaction intelligent prediction system based on multi-source drug features, similarity calculation device, each drug feature includes four aspects: pathway, enzyme, target and chemical substructure. The device calculates the similarity between every two drugs under the four features, thereby converting four feature matrices into four similarity matrices, and using the similarity matrices to determine the matching relationship between drugs;
[0024] The configuration contents of the similarity calculation device include:
[0025] (1) Take Calculate NUM = (H 1 ) T *H 1 , (H 1 ) T H 1The transposed matrix of
[0026] (2) Structure and (H 1 ) T The same shape of all 1 matrix, and add it to H 1 Multiply element by element to get the matrix E;
[0027] (3) Structure and H 1 The same shape of all 1 matrix, and H 1 Multiply it element by element to get the matrix F;
[0028] (4) Calculate DEN = E + F - NUM;
[0029] (5) Output H 1 The similarity matrix Z 1 , Z 1 =NUM / DEN;
[0030] (6) Similarly, we can calculate The similarity matrix Z 2 ,Z 3 ,Z 4 , output The similarity matrix Where Z i Each element (Z i ) xy Reflects H i The similarity between the two drugs under the i-th feature, Z i The main diagonal elements of are all 1, indicating that H i The similarity of each drug to itself.
[0031] The described intelligent prediction system for drug interactions based on multi-source drug features, the normalization processing device, is used to accelerate model training and convergence, improve numerical stability and enhance the performance of the model;
[0032] The setting contents of the normalization processing device include:
[0033] (1) Take Z i The main diagonal elements of are set to 0, that is, the self-similarity is 0, which means that the similarity between each drug and itself is not considered;
[0034] (2) Calculate Z i The sum of each row constructs the diagonal matrix D i , where D i [n,n] represents the sum of the nth row, n∈[1,m], m is the number of drugs;
[0035] (3) Calculate D i The inverse square root of
[0036] (4) Iterative Update A total of 200 times, output the updated similarity matrix
[0037] The drug interaction intelligent prediction system based on multi-source drug features, the nearest neighbor learning device, learns the local structural similarities between drugs from the nearest neighbor matrix to enhance the structural information that can be expressed in the drug similarity matrix;
[0038] The configuration contents of the nearest neighbor learning device include:
[0039] (7) Take Calculate H i The distance between each two rows Among them, n1 and n2 are H i The two row vectors in r i is the length of the vector, corresponding to The number of descriptors under the i-th feature in is also H i The number of columns, and are the kth eigenvalues of row vectors n1 and n2 respectively, and we get Stored in the distance matrix L i In the i is symmetrical, so in the calculation When only the upper triangular part is calculated;
[0040] (8) To prevent the self-distance from being 0, set the distance matrix L i The main diagonal elements of are 100000, which means that the values of the main diagonal elements are infinite;
[0041] (9) For L i Each drug in Sort by distance in ascending order, marking the top 600 drugs as The nearest neighbor of the corresponding position is set to 1, and the remaining positions are set to 0 to form the nearest neighbor matrix C i ;
[0042] (10) Construct an m×m random matrix R, where m is the number of drugs, and define the initialized similarity matrix W i =R℃ i ,° is the Hadamard product;
[0043] (11) Define the matrix P i =H i *H i ) T +λ*e T, λ=6*e, the parameter λ is introduced to make the value of the matrix P smoother and avoid the influence of extreme values. λ is a hyperparameter, λ∈[0,1], e is a column vector with all values 1, and its length is equal to the number of drugs m;
[0044] (12) Iterative update W i =(C i °W i )°P i ) / M(C i °W i )*P i ) 50 times in total, output the updated similarity matrix
[0045] The drug interaction intelligent prediction system based on multi-source drug features, the normalization processing device, the normalized feature values are within a similar scale range, making the comparison between features more intuitive and more interpretable;
[0046] The setting contents of the normalization processing device include:
[0047] (1) Take W i The main diagonal elements of are set to 0, that is, the self-similarity is 0, which means that the similarity between each drug and itself is not considered;
[0048] (2) Calculate W i The sum of each row constructs the diagonal matrix D i , where D i [n,n] represents the sum of the nth row, n∈[1,m], m is the number of drugs;
[0049] (3) Calculate D i The inverse square root of
[0050] (4) Iterative Update A total of 200 times, output the updated similarity matrix
[0051] The drug interaction intelligent prediction system based on multi-source drug features, the prediction model training device, trains the prediction model, and outputs four feature matrices and their corresponding weight matrices;
[0052] The settings of the prediction model training device include:
[0053] (1) Initialization list Ns = [S i ],S i =δ*(IZ i ), is the similarity matrix after normalization, δ is the hyperparameter, δ∈[0,1], I is the m×m unit matrix, m is Z i The number of rows indicates the number of drugs, S i Used to construct the regularization term S, so as to adjust the prediction weight matrix G of the model i To impose restraints;
[0054] (2) Initialize the prediction weight matrix G i , is the similarity matrix for each input Generate an m×m random weight matrix G i ;
[0055] (3) Initialize an m×m all-zero matrix LIM, using the formula LIM=LIM+δ*S i Update LIM;
[0056] (4) Initialize the matrix U = J, where J is the drug interaction observation matrix of Y, using the formula Update U, μ is a hyperparameter, μ∈[0,1];
[0057] (5) Define the matrix V = (Ns + (1 + 3μ) * I) -1 ;
[0058] (6) Define the matrix F mat =U*V, and iteratively update F using the semi-non-negative matrix factorization strategy mat and G i ;
[0059] (7) Calculate the loss function · is the L2 norm, Tr is the trace of the matrix, when Loss < 0.005, the iteration terminates, when the input is When, its calculation method is the same as Similarly, the updated prediction weight matrix G is finally output i .
[0060] The drug interaction intelligent prediction system based on multi-source drug features, the drug interaction prediction device, accumulates the product of four feature matrices and their corresponding weight matrices to obtain a drug interaction prediction matrix, subtracts it from the drug interaction observation matrix, and the position with a value of 1 represents a potential drug interaction, and outputs a potential drug interaction list;
[0061] The settings of the drug interaction prediction device include:
[0062] (1) Initialize an m×m matrix Pred of all zeros. or Calculate and accumulate the predicted value Pred = Pred + ((Zi ) T *G i ) / (∑(Z i ) T *G i ) or Pred=Pred+((W i ) T *G i ) / (∑(W i ) T *G i ), i∈[1,4], and obtain the drug interaction prediction matrix Pred;
[0063] (2) In the drug interaction prediction matrix Pred, if its element value is greater than the threshold α=0.9, it is set to 1; if its element value is less than or equal to the threshold α, it is set to 0, generating the drug interaction prediction matrix Pred new ;
[0064] (3) Define the matrix Result = Pred new -J, J is the drug interaction observation matrix. The position with a value of 1 in Result indicates the existence of potential drug interaction. Mark the position coordinates of all positions with a value of 1 in Result and output a potential drug interaction list.
[0065] A drug interaction intelligent prediction method based on multi-source drug features, the steps of which are:
[0066] Step 1: Collect drug information, organize and clean it, and construct a dataset X containing four types of feature information: pathway, enzyme, target, and chemical substructure, and a dataset Y containing drug interaction information. The drug feature matrix is constructed from datasets X and Y respectively. and drug interaction observation matrix J;
[0067] Step 2: Calculate the similarity between each drug in X and convert the feature matrix Convert to similarity matrix
[0068]
[0069] Step 3: Calculate the distance between each drug in X, find the nearest neighbor of each drug, learn the local structural similarity between drugs from the nearest neighbors, and convert the feature matrix Convert to similarity matrix
[0070] Step 4: Use or Train the drug interaction prediction model and output the prediction weight matrix G i ;
[0071] Step 5: By Gi The drug interaction prediction matrix Pred is calculated and converted into a binary matrix containing only 0 and 1. Then, it is subtracted from the drug interaction observation matrix J. The position with a value of 1 indicates the existence of potential drug interaction, and a potential drug interaction list is output.
[0072] A computer program product comprises: computer instructions; when part or all of the computer instructions are executed on a computer, the aforementioned intelligent prediction method for drug interaction based on multi-source drug features is executed.
[0073] A computing device comprises: a processor and a memory; the memory is used to store computer programs; the processor is used to execute the aforementioned intelligent prediction method for drug interaction based on multi-source drug features.
[0074] A computer-readable storage medium stores computer instructions or programs. When the computer instructions or programs are executed on a computer, the aforementioned intelligent prediction method for drug interaction based on multi-source drug features is executed.
[0075] In summary, the beneficial effects of the present invention include:
[0076] The system and processing method of the present invention provide more comprehensive input information for the system by processing multi-source drug features.
[0077] The system and processing method of the present invention realize dimensionality reduction processing of the feature matrix, thereby improving system efficiency while retaining key information.
[0078] The system and processing method of the present invention learn the local structural similarities between drugs from the nearest neighbors to enhance the structural information that can be expressed in the drug similarity matrix and improve the performance of the model.
[0079] The system and processing method of the present invention realize the prediction of drug interactions, output a potential drug interaction list, and provide a more reliable reference for drug research and development.
[0080] The computer program product, computing device, and computer-readable storage medium of the present invention realize efficient automation and improve efficiency and accuracy by automatically executing the intelligent prediction method of drug interaction based on multi-source drug characteristics through computer instructions. Secondly, the instruction-based characteristics make the above method highly consistent and reliable in complex experimental scenarios.
[0081] The foregoing is an overview of the present application, and therefore there are necessarily simplifications, generalizations, and omissions of details; those skilled in the art should recognize that the overview is merely an illustration of the present application and should not be regarded as any limitation of the present application. Other aspects, features, and advantages of the apparatus and / or method and / or other subject matter described in this specification will become clear as a result of the description of this specification. The overview is used to introduce a number of concepts that will be further described in the following detailed description in a simplified manner. This overview is neither intended to determine the key features or necessary features of the claimed subject matter, nor is it intended to be used as an auxiliary means for determining the scope of the claimed subject matter. BRIEF DESCRIPTION OF THE DRAWINGS
[0082] The above and other features of the present application will be more fully understood through the following description and the attached claims in conjunction with the accompanying drawings. It should be understood that these drawings are only descriptions of several embodiments of the present application and should not be considered as limiting the scope of the present application. The contents of the present application will be more clearly and detailed through the accompanying drawings.
[0083] Figure 1 It is a system block diagram of the drug interaction intelligent prediction system and method based on multi-source drug characteristics of the present invention.
[0084] Figure 2 It is a method flow chart of the drug interaction intelligent prediction system and method based on multi-source drug characteristics of the present invention;
[0085] Figure 3 It is an electronic device diagram of the drug interaction intelligent prediction system based on multi-source drug characteristics and method thereof of the present invention. DETAILED DESCRIPTION
[0086] In the following detailed description, the drawings that form part of the specification are described. In the drawings, the same / similar marks generally represent the same / similar parts, unless otherwise specified in the specification. The embodiments described in the specific embodiments, drawings and claims for illustration should not be considered as limitations on the present application. Other embodiments of the present application may be adopted, and other changes may be made to the present application without departing from the spirit or scope of the subject matter described in the present application. It should be easily understood that various aspects of the present application generally described in this specification and illustrated in the drawings may be configured, replaced, combined, and designed in various different configurations, and all of these changes are obviously expected and constitute part of the present application.
[0087] The present invention will be further described below using the accompanying drawings and specific implementations.
[0088] Reference Figure 1, a drug interaction intelligent prediction system based on multi-source drug features of the present invention includes: a drug information collection and preprocessing module, a drug feature matching module, a drug structure information enhancement module and a drug interaction prediction module. Among them, the main function of the drug information collection and preprocessing module is to collect drug feature information including pathways, enzymes, targets, chemical substructures and known drug interaction information from the drug database, and respectively construct a data set containing four types of feature information of pathways, enzymes, targets, and chemical substructures and a data set containing drug interaction information, and correspond each feature to a set of descriptors, whose values 1 or 0 represent the presence or absence of the corresponding descriptors, so that the four types of feature information in X are converted into four feature matrices, and then, a drug interaction observation matrix is constructed according to the drug interaction information in Y, whose values 1 or 0 represent the presence or absence of interaction between every two drugs in Y;
[0089] The data acquisition device in this module is used to obtain drug feature information including drug interactions, enzymes, targets and chemical substructures from the DrugBank database, and use the ID mapping system provided by the UniPort database to map the drug target to the KEGG DRUG database to obtain the drug pathway information. Finally, all the information is stored in the database for unified management, and a data set containing four feature information including pathways, enzymes, targets and chemical substructures and a data set containing drug interaction information are constructed respectively;
[0090] The data preprocessing device in this module is used to construct the feature matrix And the drug interaction observation matrix J, each feature corresponds to a set of descriptors, whose value 1 or 0 indicates the presence or absence of the corresponding descriptor, so that the four feature information in X is converted into four feature matrices. Taking the chemical substructure feature as an example, PubChem defines 881 substructure descriptors, so 881-dimensional feature vectors can be used to represent drugs, that is, each drug can be represented by four binary feature matrices. Then, the drug interaction observation matrix is constructed based on the drug interaction information in Y, and its value 1 or 0 indicates the presence or absence of interaction between each two drugs in Y.
[0091] The main function of the drug feature matching module is to calculate the similarity between every two drugs under the four features of X, thereby converting the four feature matrices into four similarity matrices, using the similarity matrices to determine the matching relationship between drugs, and iteratively normalizing the similarity matrices;
[0092] The similarity calculation device in this module is used to calculate the similarity between each two drugs under four characteristics, thereby converting the four feature matrices into four similarity matrices.
[0093] The normalization processing device in this module is used to accelerate model training and convergence, improve numerical stability and improve model performance.
[0094] The main function of the drug structure information enhancement module is to learn the local structural similarities between drugs from the nearest neighbors to enhance the structural information that can be expressed in the drug similarity matrix and iteratively normalize the similarity matrix;
[0095] The nearest neighbor learning device in this module is used to learn the local structural similarities between drugs from the nearest neighbor matrix to enhance the structural information that can be expressed in the drug similarity matrix;
[0096] The normalization processing device in this module is used to accelerate model training and convergence, making the comparison between features more intuitive and more interpretable.
[0097] The main function of the drug interaction prediction module is to input the drug similarity matrix training model, obtain the weight matrix of each feature, accumulate the product of the four feature matrices and their corresponding weight matrices, obtain the drug interaction prediction matrix, subtract it from the drug interaction observation matrix, and the position with a value of 1 indicates the existence of potential drug interaction, and output a potential drug interaction list;
[0098] The prediction model training device in this module is used to train the prediction model and output four feature matrices and their corresponding weight matrices;
[0099] The drug interaction prediction device in this module is used to accumulate the products of four feature matrices and their corresponding weight matrices to obtain a drug interaction prediction matrix, which is subtracted from the drug interaction observation matrix. Positions with a value of 1 indicate the existence of potential drug interactions, and a potential drug interaction list is output.
[0100] Reference Figure 2 , an intelligent prediction method for drug interaction based on multi-source drug features, as follows:
[0101] Step 1: Collect drug information, organize and clean it, and construct a dataset X containing four types of feature information: pathway, enzyme, target, and chemical substructure, and a dataset Y containing drug interaction information. The drug feature matrix is constructed from datasets X and Y respectively. and drug interaction observation matrix J;
[0102] Step 2: Calculate the similarity between each drug in X and convert the feature matrix Convert to similarity matrix
[0103]
[0104] Step 3: Calculate the distance between each drug in X, find the nearest neighbor of each drug, learn the local structural similarity between drugs from the nearest neighbors, and convert the feature matrix Convert to similarity matrix
[0105] Step 4: Use or Train the drug interaction prediction model and output the prediction weight matrix G i ;
[0106] Step 5: By G i The drug interaction prediction matrix Pred is calculated and converted into a binary matrix containing only 0 and 1. Then, it is subtracted from the drug interaction observation matrix J. The position with a value of 1 indicates the existence of potential drug interaction, and a potential drug interaction list is output.
[0107] The system and processing method of the present invention realize the scientific processing of multi-source drug features, providing the system with more comprehensive input information; by reducing the dimension of the feature matrix, the system performance is improved while retaining key information; by enhancing the drug structure information, the expression ability of the drug similarity matrix is improved, and the performance of the system is further improved; through the drug similarity matrix training model, a potential drug interaction list is output, thereby realizing the prediction of drug interaction, which has a reasonable structure, high prediction accuracy, strong performance, and a wide range of applications, providing a more reliable reference for drug research and development.
[0108] A computer program product, a computing device, and a computer-readable storage medium. Figure 3 , taking electronic equipment as an example, a detailed explanation is given.
[0109] The so-called electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Computing devices may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are intended to be examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein.
[0110] like Figure 3As shown, the electronic device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 502 or a computer program loaded from a storage unit 508 to a RAM (Random Access Memory) 503. In the RAM 503, various programs and data required for the operation of the device 500 can also be stored. The computing unit 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An I / O (Input / Output) interface 505 is also connected to the bus 504.
[0111] A number of components in the device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.
[0112] The computing unit 501 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above, such as a drug interaction intelligent prediction method based on multi-source drug features. For example, in some embodiments, the drug interaction intelligent prediction method based on multi-source drug features may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the computing unit 501, one or more steps of the method described above may be performed. Alternatively, in other embodiments, the computing unit 501 may be configured to execute the aforementioned intelligent prediction method for drug interaction based on multi-source drug features in any other appropriate manner (eg, by means of firmware).
[0113] The foregoing has been described in detail through block diagrams, flow charts and / or implementation examples to illustrate different implementation methods of the device and / or method of the present application. When these block diagrams, flow charts and / or embodiments include one or more functions and / or operations, those skilled in the art will understand that each function and / or operation in these block diagrams, flow charts and / or embodiments can be implemented separately and / or together through various hardware, software, firmware or substantially any combination thereof. Those skilled in the art will recognize that it is common in the art to describe the device and / or method in the manner described in this specification and then perform engineering practices to integrate the described device and / or method into a data processing system, that is, at least a portion of the device and / or method described in this specification can be integrated into a data processing system through a reasonable number of experiments. For substantially any plural and / or singular terms used in this specification, those skilled in the art may interpret the plural as the singular and / or the singular as the plural, as long as it is appropriate to do so from the context and / or application. For the sake of clarity, various singular / plural combinations may be explicitly stated in this specification.
[0114] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0115] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention without departing from the principles and intent of the present invention.
Claims
1. A drug interaction intelligent prediction system based on multi-source drug features, characterized in that: include: The drug information collection and preprocessing module is used to collect drug information from the drug database, and construct a data set X containing four feature information of pathway, enzyme, target, and chemical substructure, and a data set Y containing drug interaction information. Four feature matrices are constructed from the four feature information in X, and a drug interaction observation matrix is constructed from the drug interaction information in Y. The drug feature matching module is used to calculate the similarity between every two drugs under the four features of X, thereby converting the four feature matrices into four similarity matrices, using the similarity matrices to determine the matching relationship between the drugs, and iteratively normalizing the similarity matrices; The drug structure information enhancement module finds the nearest neighbor of each drug based on distance calculation, and then iteratively updates its similarity matrix according to the nearest neighbor information of all drugs in X, so that it can learn the local structural similarity between drugs from the nearest neighbors to enhance the structural information that can be expressed in the drug similarity matrix, and iteratively normalizes the similarity matrix; The drug interaction prediction module inputs the drug similarity matrix training model, obtains the weight matrix of each feature, accumulates the product of the four feature matrices and their corresponding weight matrices, obtains the drug interaction prediction matrix, and subtracts it from the drug interaction observation matrix. The position with a value of 1 indicates the existence of potential drug interaction, and outputs a potential drug interaction list.
2. The prediction system according to claim 1, characterized in that The drug information collection and preprocessing module includes: a data collection device and a data preprocessing device; 1) The settings of the data acquisition device include: (1) Collect drug information from the drug database, including drug interaction DDI, enzyme DE, target DT, chemical substructure DS, and pathway DP; (2) Remove drug data without any interactive information and ensure that the number of drugs in X and Y is equal; (3) Define data set X = {DP, DE, DT, DS}, Y = {DDI}; 2) The settings of the data preprocessing device include: (1) Take DP, DE, DT, DS∈X and make them correspond to a set of descriptors respectively, then each drug X in X n Represented as four binary eigenvectors, X n = {DP n ,DE n ,DT n ,DS n }, its length indicates the number of descriptors in each group, its value 1 or 0 indicates the presence or absence of the corresponding descriptor, n is the index of each drug in X, n∈[1,m], m is the number of drugs in X; (2) Take X n ∈X, define the feature matrix of X i represents the i-th feature matrix of X, for H i , each row of which represents a drug feature vector, with a total of m rows, where m is the number of drugs in X; (3) Take Y i ,Y j ∈Y, define the drug interaction observation matrix J of Y, if drug Y i and drug Y j If there is interaction between ij =1, otherwise, J ij =0, i,j∈[1,m], m is the number of drugs in Y.
3. The prediction system according to claim 1, characterized in that The drug feature matching module includes: a similarity calculation device and a normalization processing device; 1) The configuration contents of the similarity calculation device include: (1) Take Calculate NUM = (H 1 ) T *H 1 , (H 1 ) T H 1 The transposed matrix of (2) Structure and (H 1 ) T The same shape of all 1 matrix, and add it to H 1 Multiply element by element to get the matrix E; (3) Structure and H 1 The same shape of all 1 matrix, and H 1 Multiply it element by element to get the matrix F; (4) Calculate DEN = E + F - NUM; (5) Output H 1 The similarity matrix Z 1 , Z 1 =NUM / DEN; (6) Similarly, we can calculate The similarity matrix Z 2 ,Z 3 ,Z 4 , output The similarity matrix Where Z i Each element (Z i ) xy Reflects H i The similarity between the two drugs under the i-th feature, Z i The main diagonal elements of are all 1, indicating that H i The similarity of each drug to itself; 2) The setting contents of the normalization processing device include: (1) Take Z i The main diagonal elements of are set to 0, that is, the self-similarity is 0, which means that the similarity between each drug and itself is not considered; (2) Calculate Z i The sum of each row constructs the diagonal matrix D i , where D i [n,n] represents the sum of the nth row, n∈[1,m], m is the number of drugs; (3) Calculate D i The inverse square root of (4) Iterative Update A total of 200 times, output the updated similarity matrix 4. The prediction system according to claim 1, characterized in that The drug structure information enhancement module includes: a nearest neighbor learning device and a normalization processing device; 1) The configuration of the nearest neighbor learning device includes: (1) Take Calculate H i The distance between each two rows Among them, n1 and n2 are H i The two row vectors in r i is the length of the vector, corresponding to The number of descriptors under the i-th feature in is also H i The number of columns, and are the kth eigenvalues of row vectors n1 and n2 respectively, and we get Stored in the distance matrix L i In the i is symmetrical, so in the calculation When only the upper triangular part is calculated; (2) To prevent the self-distance from being 0, set the distance matrix L i The main diagonal elements of are 100000, which means that the values of the main diagonal elements are infinite; (3) For L i Each drug in Sort by distance in ascending order, marking the top 600 drugs as The nearest neighbor of the corresponding position is set to 1, and the remaining positions are set to 0 to form the nearest neighbor matrix C i ; (4) Construct an m×m random matrix R, where m is the number of drugs, and define the initialized similarity matrix is the Hadamard product; (5) Define the matrix P i =H i *(H i ) T +λ*e T , λ=6*e, the parameter λ is introduced to make the value of the matrix P smoother and avoid the influence of extreme values. λ is a hyperparameter, λ∈[0,1], e is a column vector with all values 1, and its length is equal to the number of drugs m; (6) Iterative Update A total of 50 times, output the updated similarity matrix 2) The setting contents of the normalization processing device include: (1) Take W i The main diagonal elements of are set to 0, that is, the self-similarity is 0, which means that the similarity between each drug and itself is not considered; (2) Calculate W i The sum of each row constructs the diagonal matrix D i , where D i [n,n] represents the sum of the nth row, n∈[1,m], m is the number of drugs; (3) Calculate D i The inverse square root of (4) Iterative Update A total of 200 times, output the updated similarity matrix 5. The prediction system according to claim 1, characterized in that The drug interaction prediction module includes a prediction model training device and a drug interaction prediction device; 1) The settings of the prediction model training device include: (1) Initialization list Ns = [S i ],S i =δ*(IZ i ), is the similarity matrix after normalization, δ is the hyperparameter, δ∈[0,1], I is the m×m unit matrix, m is Z i The number of rows indicates the number of drugs, S i Used to construct the regularization term S, so as to adjust the prediction weight matrix G of the model i To impose restraints; (2) Initialize the prediction weight matrix G i , is the similarity matrix for each input Generate an m×m random weight matrix G i ; (3) Initialize an m×m all-zero matrix LIM, using the formula LIM=LIM+δ*δ i Update LIM; (4) Initialize the matrix U = J, where J is the drug interaction observation matrix of Y, using the formula Update U, μ is a hyperparameter, μ∈[0,1]; (5) Define the matrix V = (Ns + (1 + 3μ) * I) -1 ; (6) Define the matrix F mat =U*V, and iteratively update F using the semi-non-negative matrix factorization strategy mat and G i ; (7) Calculate the loss function · is the L2 norm, Tr is the trace of the matrix, when Loss < 0.005, the iteration terminates, when the input is When, its calculation method is the same as Similarly, the updated prediction weight matrix G is finally output i ; 2) The settings of the drug interaction prediction device include: (1) Initialize an m×m matrix Pred of all zeros. or Calculate and accumulate the predicted value Pred = Pred + ((Z i ) T *G i ) / (∑(Z i ) T *G i ) or Pred=Pred+((W i ) T *G i ) / (∑(W i ) T *G i ), i∈[1,4], and obtain the drug interaction prediction matrix Pred; (2) In the drug interaction prediction matrix Pred, if its element value is greater than the threshold α=0.9, it is set to 1; if its element value is less than or equal to the threshold α, it is set to 0, generating the drug interaction prediction matrix Pred new ; (3) Define the matrix Result = Pred new -J, J is the drug interaction observation matrix. The position with a value of 1 in Result indicates the existence of potential drug interaction. Mark the position coordinates of all positions with a value of 1 in Result and output a potential drug interaction list.
6. A drug interaction intelligent prediction method based on multi-source drug features, characterized in that: include: Step 1: Collect drug information, organize and clean it, and construct a dataset X containing four types of feature information: pathway, enzyme, target, and chemical substructure, and a dataset Y containing drug interaction information. The drug feature matrix is constructed from datasets X and Y respectively. and drug interaction observation matrix J; Step 2: Calculate the similarity between each drug in X and convert the feature matrix Convert to similarity matrix Step 3: Calculate the distance between each drug in X, find the nearest neighbor of each drug, learn the local structural similarity between drugs from the nearest neighbors, and convert the feature matrix Convert to similarity matrix Step 4: Use or Train the drug interaction prediction model and output the prediction weight matrix G i ; Step 5: By G i The drug interaction prediction matrix Pred is calculated and converted into a binary matrix containing only 0 and 1. Then, it is subtracted from the drug interaction observation matrix J. The position with a value of 1 indicates the existence of potential drug interaction, and a potential drug interaction list is output.
7. A computer program product, characterized in that include: Computer instructions; When part or all of the computer instructions are executed on a computer, the intelligent prediction method for drug interaction based on multi-source drug features as claimed in claim 6 is executed.
8. A computing device, characterized in that include: Processor and memory; The memory is used to store computer programs; The processor is used to execute the computer program to implement the drug interaction intelligent prediction method based on multi-source drug features as described in claim 6.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions or programs, and when the computer instructions or programs are executed on a computer, the intelligent prediction method for drug interaction based on multi-source drug features as claimed in claim 6 is executed.
