Drug recommendation method and system based on multi-source embedding

By constructing an isomerographic diagram of the operation-diagnosis-drug relationship and combining graph convolution and attention mechanism, the problem of inaccurate drug recommendation results in the prior art is solved, and the accuracy of drug recommendation is improved.

CN119092040APending Publication Date: 2024-12-06UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411262735.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-10
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Existing longitudinal-based drug recommendation methods have weak extraction capabilities in long or complex sequence features, resulting in inaccurate drug prediction results.

Method used

Using a multi-source embedding method, by obtaining electronic medical record information and drug-drug interaction information, constructing operation-diagnosis-drug relationship isomerographic diagrams, enhancing information, encoding feature information, and combining graph convolution and attention mechanisms to perform drugs recommend.

Benefits of technology

Improve the accuracy of drug recommendations, enable more efficient extraction of features in long time series or complex time series, enhance patient characterization representation, and improve the accuracy of drug recommendation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119092040A_ABST
    Figure CN119092040A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of drug recommendation, and relates to a multi-source embedding-based drug recommendation method and system, the method comprises the steps of obtaining first information and second information, the first information comprising electronic medical record information, and the second information comprising drug-drug interaction information; constructing an operation-diagnosis-drug relation heterogeneous graph according to the first information and the second information; enhancing the first information to obtain enhanced first information; the enhanced first information is encoded to obtain encoded feature information, and the encoded feature information comprises diagnosis feature information, operation feature information and drug feature information; predicting a drug recommendation result according to the encoded feature information and the operation-diagnosis-drug relationship heterogeneous graph, and effectively improving the accuracy of drug recommendation by combining the feature information based on the instance and the feature information based on the longitudinal direction as embedded information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of drug recommendation, and in particular to a drug recommendation method and system based on multi-source embedding. Background Art

[0002] Existing technologies usually adopt a longitudinal drug recommendation method, specifically using natural language processing (NLP) methods to process patients' historical medical records, and make drug recommendations for current diagnostic information based on the empirical rules learned from historical records. This method converts the drug recommendation problem into a prediction problem of serialized input and output, but this type of network has a weak ability to extract long or complex sequence features, resulting in inaccurate drug prediction results. Summary of the invention

[0003] The purpose of the present invention is to provide a drug recommendation method and system based on multi-source embedding to improve the above problems.

[0004] In order to achieve the above objectives, the present application provides the following technical solutions:

[0005] On the one hand, an embodiment of the present application provides a drug recommendation method based on multi-source embedding, the method comprising:

[0006] Acquiring first information and second information, the first information comprising electronic medical record information, and the second information comprising drug-drug interaction information;

[0007] constructing an operation-diagnosis-drug relationship heterogeneous graph according to the first information and the second information;

[0008] enhancing the first information to obtain enhanced first information;

[0009] Encoding the enhanced first information to obtain encoded characteristic information, wherein the encoded characteristic information includes diagnosis characteristic information, operation characteristic information and drug characteristic information;

[0010] The drug recommendation result is predicted based on the encoded feature information and the operation-diagnosis-drug relationship heterogeneous graph.

[0011] In a second aspect, an embodiment of the present application provides a drug recommendation system based on multi-source embedding, the system comprising:

[0012] an acquisition module, configured to acquire first information and second information, wherein the first information includes electronic medical record information, and the second information includes drug-drug interaction information;

[0013] A first processing module, configured to construct an operation-diagnosis-drug relationship heterogeneous graph according to the first information and the second information;

[0014] A second processing module, used for enhancing the first information to obtain enhanced first information;

[0015] A third processing module, used for encoding the enhanced first information to obtain encoded characteristic information, wherein the encoded characteristic information includes diagnosis characteristic information, operation characteristic information and drug characteristic information;

[0016] A prediction module is used to predict drug recommendation results based on the encoded feature information and the operation-diagnosis-drug relationship heterogeneous graph.

[0017] In a third aspect, an embodiment of the present application provides a drug recommendation device based on multi-source embedding, the device comprising a memory and a processor. The memory is used to store a computer program; the processor is used to implement the steps of the drug recommendation method based on multi-source embedding when executing the computer program.

[0018] In a fourth aspect, an embodiment of the present application provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned drug recommendation method based on multi-source embedding are implemented.

[0019] The beneficial effects of the present invention are:

[0020] The present invention constructs an operation-diagnosis-drug relationship heterogeneous graph based on the first information and the second information, obtains instance-based feature information, enhances the first information and encodes the enhanced first information to obtain the encoded feature information, obtains the vertical-based feature information, and effectively improves the accuracy of drug recommendations by combining the instance-based feature information and the vertical-based feature information.

[0021] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0023] Figure 1Schematic diagram of the process of the drug recommendation method based on multi-source embedding described in an embodiment of the present invention.

[0024] Figure 2 Schematic diagram of the structure of the drug recommendation system based on multi-source embedding described in an embodiment of the present invention.

[0025] Figure 3 It is a schematic diagram of the structure of a drug recommendation device based on multi-source embedding described in an embodiment of the present invention.

[0026] Labels in the figure: 901, acquisition module; 902, first processing module; 903, second processing module; 904, third processing module; 905, prediction module; 800, drug recommendation device based on multi-source embedding; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0029] Embodiment 1:

[0030] This embodiment provides a drug recommendation method based on multi-source embedding. It can be understood that a scenario can be laid out in this embodiment, for example: a scenario of recommending drugs to patients.

[0031] See also Figure 1 , the figure shows that the method includes steps S1, S2, S3, S4 and S5, which specifically include:

[0032] Step S1, obtaining first information and second information, wherein the first information includes electronic medical record information, and the second information includes drug-drug interaction information;

[0033] In this step, the top 50 most common drug-drug interactions were extracted from the database and converted into a DDI map for subsequent processing.

[0034] Step S2, constructing an operation-diagnosis-drug relationship heterogeneous graph according to the first information and the second information;

[0035] Some operations and drugs are usually used in combination in clinical practice to enhance the therapeutic effect. Therefore, this information is saved in the patient's electronic medical record for modeling, so that when the model obtains partial recommendation results, it can make more complete drug recommendations according to the combination relationship of treatment plans in the current results. The drug-drug interaction information represents the adverse reactions between drugs that are opposite to the co-occurrence relationship of the drug combination, which needs to be avoided when recommending drugs. Therefore, in this step, a heterogeneous graph is constructed based on the first information and the second information, which can accurately obtain instance features.

[0036] The step S2 also includes step S21, step S22, step S23 and step S24, which specifically include:

[0037] Step S21, determining diagnostic vector information according to the first information, wherein the diagnostic vector information includes whether the operation or drug has a therapeutic effect on the patient;

[0038] In this step, the electronic medical record information mainly includes three types of information: patient diagnosis, medication, and operation. The patient's diagnosis information can be obtained based on the first information, thereby constructing the diagnosis vector information. The diagnosis vector information is recorded as N = (n 1 , n 2 ,…,n h ), where n h =1 indicates that the corresponding h-th operation or drug has a therapeutic effect on the diagnosis. h =0 indicates that the corresponding h-th operation or drug has no therapeutic effect on the diagnosis.

[0039] Step S22: establishing at least two nodes of heterogeneous graphs based on the diagnostic vector information;

[0040] Step S23, calculating the similarity between the nodes of each heterogeneous graph to obtain similarity information;

[0041] In this step, there is no limitation on the method for calculating the similarity.

[0042] Step S24: Establish edges between nodes in the heterogeneous graph according to the similarity information.

[0043] In this step, a threshold is set. When the cosine similarity between nodes is greater than the threshold, the nodes are considered to have a relationship, which is represented as a vector to establish an edge between the diagnostic nodes in the heterogeneous graph.

[0044] In this embodiment, firstly, the connection relationship between operation and diagnosis and the connection relationship between drug and diagnosis can be established according to the diagnosis vector information, thereby establishing a heterogeneous graph node with diagnosis as the core; secondly, the connection relationship between drugs is established based on the existing DDI relationship, and external knowledge is introduced; finally, the connection relationship between diagnosis and diagnosis is screened according to the set threshold, thereby establishing an operation-diagnosis-drug relationship heterogeneous graph. It should be noted that when n in the diagnosis vector information h =1, an edge is established between the operation and the diagnosis or between the drug and the diagnosis.

[0045] Step S3, enhancing the first information to obtain enhanced first information;

[0046] The step S3 also includes step S31, step S32, step S33, step S34 and step S35, which specifically include:

[0047] Step S31, converting the first information into at least three one-hot vectors, wherein the one-hot vectors include a diagnosis one-hot vector, an operation one-hot vector, and a drug one-hot vector;

[0048] In this step, a tuple is used To represent the tth visit record of patient i, where One-hot vectors representing diagnosis, operation, and drug, respectively.

[0049] Step S32: determining a connection set of relationships between drugs according to the second information to obtain first connection relationship information;

[0050] Step S33: determining the adjacency matrix corresponding to the first connection relationship information according to the first connection relationship information, and obtaining first adjacency matrix information;

[0051] Step S34, performing calculations based on the drug one-hot vector and the first adjacency matrix information to obtain drug relationship feature information;

[0052] In this step, a two-layer graph convolutional neural network is used to calculate the drug relationship feature information as follows:

[0053] G d =GCN(tanh(GCN(E m ,A D ))W d ,A D )

[0054] In the above formula, G d is the drug relationship feature information, E m is the embedding representation of the drug’s one-hot vector, A D represents the adjacency matrix of known drug-drug interactions from external knowledge, that is, the first adjacency matrix information, W d represents the learnable parameter matrix.

[0055] Step S35: Calculate based on the drug relationship feature information and the drug unique hot vector to obtain an enhanced drug unique hot vector.

[0056] In this embodiment, the specific calculation process of the enhanced drug unique heat vector is:

[0057]

[0058] In the above formula, represents the enhanced drug one-hot vector, E m is the embedding representation of the drug’s unique hot vector, σ is a learnable parameter, G d It is the drug relationship characteristic information.

[0059] The step S3 also includes step S36, step S37, step S38 and step S39, which specifically include:

[0060] Step S36: Determine a connection set of relationships between different operations according to the first information to obtain second connection relationship information;

[0061] Step S37: determining the adjacency matrix corresponding to the second connection relationship information according to the second connection relationship information, and obtaining second adjacency matrix information;

[0062] Step S38, performing calculation according to the operation one-hot vector and the second adjacency matrix information to obtain operation relationship feature information;

[0063] Step S39: Calculate according to the operation relationship feature information and the operation one-hot vector to obtain an enhanced operation one-hot vector.

[0064] In this embodiment, the principle of enhancing the operation one-hot vector is the same as the principle of enhancing the drug one-hot vector, so it is not repeated here.

[0065] Step S4, encoding the enhanced first information to obtain encoded characteristic information, wherein the encoded characteristic information includes diagnosis characteristic information, operation characteristic information and drug characteristic information;

[0066] The step S4 also includes step S41, step S42, step S43, step S44, step S45 and step S46, which specifically include:

[0067] Step S41, obtaining a preset time window;

[0068] Step S42, sending the enhanced first information to the long short-time neural network to obtain an output result, wherein the output result includes an output result of the last iteration of the hidden layer of the long short-time neural network;

[0069] After obtaining the enhanced first information of the patient, in order to obtain the time series information therein, it is sent to a long short time neural network (LSTM) for time series information extraction.

[0070] Step S43, determining a hidden state matrix according to the preset time window and the output result;

[0071] In this step, we select w as a time window, which represents the selected historical data range, and select the w hidden states before the output result to obtain the hidden state matrix, which is recorded as H. H is specifically expressed as: H = h t-w ,h t-w+1 ,…,h t-1 .h t Represents the output result; the row vector of H represents the state of a single variable at all time steps, that is, the vector composed of all time steps of the same variable; the column vector of H represents the state of a single time step, that is, the vector composed of all variables at the same time step.

[0072] Step S44, using one-dimensional convolution to extract features from the row vector of the hidden state matrix to obtain first feature information;

[0073] In this step, the specific calculation process of the first characteristic information is:

[0074]

[0075] In the above formula, H i,j represents the result value of the i-th row vector and the j-th convolution kernel, i.e., the first feature information; H i represents the i-th row vector; It represents the specific convolution operation, l is used to select the sliding value, and the size of the convolution kernel is adjusted by changing l. (tw, tw-1+l) indicates which time steps the hidden state is calculated each time. This formula represents the part of the convolution kernel that is selected to convolve the hidden state of the corresponding time step.

[0076] Step S45: Calculate the attention weight using the first feature information and weight each row in the first feature information to obtain second feature information;

[0077] Step S46: perform calculation according to the second feature information and the output result to obtain encoded feature information.

[0078] In this step, the idea of ​​residual is used to add the second feature information and the output result to obtain the encoded feature information.

[0079] The existing technology usually uses a multi-layer neural network to simulate the patient's long-term medical dependence, but more layers will lead to an increase in the mutual information between the obtained patient representation and the input patient information, and even the forgetting of key information reflecting the patient's disease status. At the same time, the traditional attention mechanism focuses on the weight distribution of different time points, and has a better effect when there is only one variable in each time step. This attention ignores the information characteristics that may be possessed between multiple hidden layers in LSTM. It is especially difficult to process in the case of multiple time series and each time step contains multiple variables, and there may be complex nonlinear internal connections between the variables. This embodiment processes the enhanced first information of the input through time window convolution combined with the attention mechanism, which can extract features in long time series or complex time series and more fully represent the patient representation.

[0080] Step S5: predicting drug recommendation results based on the encoded feature information and the operation-diagnosis-drug relationship heterogeneous graph.

[0081] The step S5 also includes step S51, step S52, step S53 and step S54, which specifically include:

[0082] Step S51, performing graph convolution on the operation-diagnosis-drug relationship heterogeneous graph to obtain third feature information;

[0083] In this step, graph convolution is performed on the operation-diagnosis-drug relationship heterogeneous graph to extract the feature information of the instance.

[0084] Step S52, calculating the third feature information, the enhanced operation unique hot vector and the enhanced drug unique hot vector to obtain fourth feature information;

[0085] In this step, the calculation process of the fourth characteristic information is specifically as follows:

[0086] K=h mix +α 1 G d +α 2 G e

[0087] In the above formula, K represents the fourth feature information, h mix Represents the third feature information, α 1 and α 2 Respectively represent the configurable parameters, Gd and G e They represent drug relationship feature information and operation relationship feature information respectively.

[0088] Step S53: sending the encoded feature information to the attention mechanism to obtain fifth feature information;

[0089] In this step, in order to select relatively important visits during the patient's disease process and further improve the accuracy of predicting drug recommendation results, visit-level attention is designed to obtain the weight β corresponding to each visit. The specific calculation formula is as follows:

[0090]

[0091] In the above formula, represents the importance weight ratio of the jth visit of the i-th patient, w h represents the learnable parameter matrix, The output result of sending the encoded feature information corresponding to the jth visit of the i-th patient to the recurrent neural network.

[0092] Step S54: perform prediction based on the fifth characteristic information and the fourth characteristic information to obtain the drug recommendation result.

[0093] Embodiment 2:

[0094] like Figure 2 As shown, this embodiment provides a drug recommendation system based on multi-source embedding, the system includes an acquisition module 901, a first processing module 902, a second processing module 903, a third processing module 904 and a prediction module 905, which specifically include:

[0095] An acquisition module 901 is used to acquire first information and second information, wherein the first information includes electronic medical record information, and the second information includes drug-drug interaction information;

[0096] A first processing module 902, configured to construct an operation-diagnosis-drug relationship heterogeneous graph according to the first information and the second information;

[0097] A second processing module 903 is used to enhance the first information to obtain enhanced first information;

[0098] A third processing module 904 is used to encode the enhanced first information to obtain encoded feature information, where the encoded feature information includes diagnosis feature information, operation feature information and drug feature information;

[0099] The prediction module 905 is used to predict the drug recommendation result according to the encoded feature information and the operation-diagnosis-drug relationship heterogeneous graph.

[0100] In a specific embodiment of the present disclosure, the first processing module further includes a first processing unit, a second processing unit, a first calculation unit and a third processing unit, which specifically include:

[0101] a first processing unit, configured to determine diagnosis vector information according to the first information, wherein the diagnosis vector information includes whether the operation or the drug has a therapeutic effect on the patient;

[0102] A second processing unit, configured to establish at least two nodes of a heterogeneous graph based on the diagnostic vector information;

[0103] A first calculation unit, used for calculating the similarity between the nodes of each heterogeneous graph to obtain similarity information;

[0104] The third processing unit is used to establish edges between nodes of the heterogeneous graph according to the similarity information.

[0105] In a specific embodiment of the present disclosure, the second processing module further includes a fourth processing unit, a fifth processing unit, a sixth processing unit, a second computing unit and a third computing unit, which specifically include:

[0106] a fourth processing unit, configured to convert the first information into at least three one-hot vectors, wherein the one-hot vectors include a diagnosis one-hot vector, an operation one-hot vector, and a drug one-hot vector;

[0107] A fifth processing unit, configured to determine a connection set of relationships between drugs according to the second information, to obtain first connection relationship information;

[0108] A sixth processing unit, configured to determine an adjacency matrix corresponding to the first connection relationship information according to the first connection relationship information, to obtain first adjacency matrix information;

[0109] A second calculation unit, used for performing calculation according to the drug one-hot vector and the first adjacency matrix information to obtain drug relationship feature information;

[0110] The third calculation unit is used to perform calculation according to the drug relationship feature information and the drug unique hot vector to obtain an enhanced drug unique hot vector.

[0111] In a specific implementation of the present disclosure, the third computing unit further includes a seventh processing unit, an eighth processing unit, a fourth computing unit, and a fifth computing unit, which specifically include:

[0112] A seventh processing unit, configured to determine a connection set of relationships between different operations according to the first information, and obtain second connection relationship information;

[0113] an eighth processing unit, configured to determine an adjacency matrix corresponding to the second connection relationship information according to the second connection relationship information, and obtain second adjacency matrix information;

[0114] a fourth calculation unit, configured to perform calculation according to the operation one-hot vector and the second adjacency matrix information to obtain operation relationship feature information;

[0115] A fifth calculation unit is used to perform calculation according to the operation relationship feature information and the operation one-hot vector to obtain an enhanced operation one-hot vector.

[0116] In a specific implementation of the present disclosure, the third processing module further includes an acquisition unit, a ninth processing unit, a tenth processing unit, an eleventh processing unit, a twelfth processing unit and a thirteenth processing unit, which specifically include:

[0117] An acquisition unit, used for acquiring a preset time window;

[0118] A ninth processing unit, configured to send the enhanced first information to the long short-time neural network to obtain an output result, wherein the output result includes an output result of the last iteration of the hidden layer of the long short-time neural network;

[0119] a tenth processing unit, configured to determine a hidden state matrix according to the preset time window and the output result;

[0120] an eleventh processing unit, configured to perform feature extraction on a row vector of the hidden state matrix using a one-dimensional convolution to obtain first feature information;

[0121] a twelfth processing unit, configured to calculate an attention weight using the first feature information and weight each row in the first feature information to obtain second feature information;

[0122] A thirteenth processing unit is used to perform calculation according to the second feature information and the output result to obtain encoded feature information.

[0123] In a specific implementation of the present disclosure, the prediction module further includes a fourteenth processing unit, a sixth calculation unit, a fifteenth processing unit, and a sixteenth processing unit, which specifically include:

[0124] A fourteenth processing unit, configured to perform graph convolution on the operation-diagnosis-drug relationship heterogeneous graph to obtain third feature information;

[0125] a sixth calculation unit, configured to calculate the third feature information, the enhanced operation unique hot vector, and the enhanced drug unique hot vector to obtain fourth feature information;

[0126] A fifteenth processing unit, configured to send the encoded feature information to an attention mechanism to obtain fifth feature information;

[0127] The sixteenth processing unit is used to make a prediction based on the fifth characteristic information and the fourth characteristic information to obtain the drug recommendation result.

[0128] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0129] Embodiment 3:

[0130] Corresponding to the above method embodiment, this embodiment also provides a drug recommendation device based on multi-source embedding. The drug recommendation device based on multi-source embedding described below and the drug recommendation method based on multi-source embedding described above can refer to each other.

[0131] Figure 3 is a block diagram of a drug recommendation device 800 based on multi-source embedding according to an exemplary embodiment. Figure 3 As shown, the drug recommendation device 800 based on multi-source embedding may include: a processor 801 and a memory 802. The drug recommendation device 800 based on multi-source embedding may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0132] The processor 801 is used to control the overall operation of the drug recommendation device 800 based on multi-source embedding to complete all or part of the steps in the drug recommendation method based on multi-source embedding. The memory 802 is used to store various types of data to support the operation of the drug recommendation device 800 based on multi-source embedding. These data may include, for example, instructions for any application or method operating on the drug recommendation device 800 based on multi-source embedding, as well as application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk. The multimedia component 803 can include a screen and an audio component. The screen can be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component can include a microphone, and the microphone is used to receive external audio signals. The received audio signal can be further stored in the memory 802 or sent through the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the multi-source embedded drug recommendation device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, near field communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0133] In an exemplary embodiment, the drug recommendation device 800 based on multi-source embedding can be implemented by one or more application-specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned drug recommendation method based on multi-source embedding.

[0134] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned drug recommendation method based on multi-source embedding are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 802 including program instructions, and the above-mentioned program instructions can be executed by the processor 801 of the drug recommendation device based on multi-source embedding 800 to complete the above-mentioned drug recommendation method based on multi-source embedding.

[0135] Embodiment 4:

[0136] Corresponding to the above method embodiment, a readable storage medium is also provided in this embodiment. The readable storage medium described below and the drug recommendation method based on multi-source embedding described above can refer to each other.

[0137] A readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the drug recommendation method based on multi-source embedding of the above method embodiment.

[0138] The readable storage medium may specifically be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or other readable storage medium that can store program codes.

[0139] 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.

[0140] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A drug recommendation method based on multi-source embedding, characterized in that: include: Acquiring first information and second information, the first information comprising electronic medical record information, and the second information comprising drug-drug interaction information; constructing an operation-diagnosis-drug relationship heterogeneous graph according to the first information and the second information; enhancing the first information to obtain enhanced first information; Encoding the enhanced first information to obtain encoded characteristic information, wherein the encoded characteristic information includes diagnosis characteristic information, operation characteristic information and drug characteristic information; The drug recommendation result is predicted based on the encoded feature information and the operation-diagnosis-drug relationship heterogeneous graph.

2. The drug recommendation method based on multi-source embedding according to claim 1, characterized in that: Constructing an operation-diagnosis-drug relationship heterogeneous graph according to the first information and the second information, including: Determine diagnostic vector information according to the first information, the diagnostic vector information including whether the operation or the drug has a therapeutic effect on the patient; Establishing at least two nodes of heterogeneous graphs based on the diagnostic vector information; Calculating the similarity between the nodes of each heterogeneous graph to obtain similarity information; Edges between nodes of the heterogeneous graph are established based on the similarity information.

3. The drug recommendation method based on multi-source embedding according to claim 1, characterized in that: The first information is enhanced to obtain enhanced first information, including: Convert the first information into at least three one-hot vectors, wherein the one-hot vectors include a diagnosis one-hot vector, an operation one-hot vector, and a drug one-hot vector; Determine a connection set of relationships between drugs according to the second information to obtain first connection relationship information; Determine the adjacency matrix corresponding to the first connection relationship information according to the first connection relationship information, and obtain first adjacency matrix information; Calculating according to the drug one-hot vector and the first adjacency matrix information to obtain drug relationship feature information; Calculation is performed based on the drug relationship feature information and the drug unique hot vector to obtain an enhanced drug unique hot vector.

4. The drug recommendation method based on multi-source embedding according to claim 3, characterized in that: After calculating according to the drug relationship feature information and the drug unique hot vector, it also includes: Determine a connection set of relationships between different operations according to the first information, and obtain second connection relationship information; Determine the adjacency matrix corresponding to the second connection relationship information according to the second connection relationship information, and obtain second adjacency matrix information; Calculating according to the operation one-hot vector and the second adjacency matrix information to obtain operation relationship feature information; Calculation is performed according to the operation relationship feature information and the operation one-hot vector to obtain an enhanced operation one-hot vector.

5. The drug recommendation method based on multi-source embedding according to claim 1, characterized in that: Encoding the enhanced first information to obtain encoded feature information includes: Get the preset time window; Sending the enhanced first information to the long-short time neural network to obtain an output result, wherein the output result includes an output result of the last iteration of the hidden layer of the long-short time neural network; Determine a hidden state matrix according to the preset time window and the output result; Performing feature extraction on the row vector of the hidden state matrix using one-dimensional convolution to obtain first feature information; Calculating an attention weight using the first feature information and weighting each row in the first feature information to obtain second feature information; Calculation is performed based on the second feature information and the output result to obtain encoded feature information.

6. A drug recommendation system based on multi-source embedding, characterized in that: include: an acquisition module, configured to acquire first information and second information, wherein the first information includes electronic medical record information, and the second information includes drug-drug interaction information; A first processing module, configured to construct an operation-diagnosis-drug relationship heterogeneous graph according to the first information and the second information; A second processing module, used for enhancing the first information to obtain enhanced first information; A third processing module, used for encoding the enhanced first information to obtain encoded characteristic information, wherein the encoded characteristic information includes diagnosis characteristic information, operation characteristic information and drug characteristic information; A prediction module is used to predict drug recommendation results based on the encoded feature information and the operation-diagnosis-drug relationship heterogeneous graph.

7. The drug recommendation system based on multi-source embedding according to claim 6, characterized in that: The first processing module comprises: a first processing unit, configured to determine diagnosis vector information according to the first information, wherein the diagnosis vector information includes whether the operation or the drug has a therapeutic effect on the patient; A second processing unit, configured to establish at least two nodes of a heterogeneous graph based on the diagnostic vector information; A first calculation unit, used for calculating the similarity between the nodes of each heterogeneous graph to obtain similarity information; The third processing unit is used to establish edges between nodes of the heterogeneous graph according to the similarity information.

8. The drug recommendation system based on multi-source embedding according to claim 6, characterized in that: The second processing module comprises: a fourth processing unit, configured to convert the first information into at least three one-hot vectors, wherein the one-hot vectors include a diagnosis one-hot vector, an operation one-hot vector, and a drug one-hot vector; A fifth processing unit, configured to determine a connection set of relationships between drugs according to the second information, to obtain first connection relationship information; A sixth processing unit, configured to determine an adjacency matrix corresponding to the first connection relationship information according to the first connection relationship information, to obtain first adjacency matrix information; A second calculation unit, used for performing calculation according to the drug one-hot vector and the first adjacency matrix information to obtain drug relationship feature information; The third calculation unit is used to perform calculation according to the drug relationship feature information and the drug unique hot vector to obtain an enhanced drug unique hot vector.

9. The drug recommendation method based on multi-source embedding according to claim 8, characterized in that: After the third calculation unit, it also includes: A seventh processing unit, configured to determine a connection set of relationships between different operations according to the first information, and obtain second connection relationship information; an eighth processing unit, configured to determine an adjacency matrix corresponding to the second connection relationship information according to the second connection relationship information, and obtain second adjacency matrix information; a fourth calculation unit, configured to perform calculation according to the operation one-hot vector and the second adjacency matrix information to obtain operation relationship feature information; A fifth calculation unit is used to perform calculation according to the operation relationship feature information and the operation one-hot vector to obtain an enhanced operation one-hot vector.

10. The drug recommendation method based on multi-source embedding according to claim 6, characterized in that: The third processing module comprises: An acquisition unit, used for acquiring a preset time window; A ninth processing unit, configured to send the enhanced first information to the long short-time neural network to obtain an output result, wherein the output result includes an output result of the last iteration of the hidden layer of the long short-time neural network; a tenth processing unit, configured to determine a hidden state matrix according to the preset time window and the output result; an eleventh processing unit, configured to perform feature extraction on a row vector of the hidden state matrix using a one-dimensional convolution to obtain first feature information; a twelfth processing unit, configured to calculate an attention weight using the first feature information and weight each row in the first feature information to obtain second feature information; A thirteenth processing unit is used to perform calculation according to the second feature information and the output result to obtain encoded feature information.