CN116543853A - Drug interaction prediction method, computer device, and medium - Google Patents
Drug interaction prediction method, computer device, and medium Download PDFInfo
- Publication number
- CN116543853A CN116543853A CN202210083440.2A CN202210083440A CN116543853A CN 116543853 A CN116543853 A CN 116543853A CN 202210083440 A CN202210083440 A CN 202210083440A CN 116543853 A CN116543853 A CN 116543853A
- Authority
- CN
- China
- Prior art keywords
- drug
- predicted
- medicine
- vector
- relationship
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 79
- 206010013710 Drug interaction Diseases 0.000 title claims abstract description 61
- 239000003814 drug Substances 0.000 claims abstract description 383
- 229940079593 drug Drugs 0.000 claims abstract description 214
- 239000013598 vector Substances 0.000 claims abstract description 195
- 201000010099 disease Diseases 0.000 claims abstract description 75
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims abstract description 75
- 238000013528 artificial neural network Methods 0.000 claims abstract description 29
- 238000013507 mapping Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims description 58
- 239000013604 expression vector Substances 0.000 claims description 40
- 239000011159 matrix material Substances 0.000 claims description 32
- 230000006870 function Effects 0.000 claims description 31
- 230000004913 activation Effects 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 10
- 239000000126 substance Substances 0.000 claims description 10
- 238000009472 formulation Methods 0.000 claims description 7
- 239000000203 mixture Substances 0.000 claims description 7
- 241000282326 Felis catus Species 0.000 claims description 6
- 230000015654 memory Effects 0.000 claims description 6
- 239000000523 sample Substances 0.000 claims 1
- 230000003993 interaction Effects 0.000 abstract description 19
- 239000002547 new drug Substances 0.000 abstract description 9
- 238000012827 research and development Methods 0.000 abstract description 2
- 125000004429 atom Chemical group 0.000 description 9
- 238000010586 diagram Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- 238000012545 processing Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 6
- 239000004973 liquid crystal related substance Substances 0.000 description 5
- 230000009471 action Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 229910052796 boron Inorganic materials 0.000 description 2
- 229910052729 chemical element Inorganic materials 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000008406 drug-drug interaction Effects 0.000 description 2
- 229910052760 oxygen Inorganic materials 0.000 description 2
- ZOXJGFHDIHLPTG-UHFFFAOYSA-N Boron Chemical compound [B] ZOXJGFHDIHLPTG-UHFFFAOYSA-N 0.000 description 1
- 208000024172 Cardiovascular disease Diseases 0.000 description 1
- 208000017667 Chronic Disease Diseases 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 230000001627 detrimental effect Effects 0.000 description 1
- 206010012601 diabetes mellitus Diseases 0.000 description 1
- 238000009509 drug development Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 125000004433 nitrogen atom Chemical group N* 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 230000005180 public health Effects 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000001988 toxicity Effects 0.000 description 1
- 231100000419 toxicity Toxicity 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/70—Machine learning, data mining or chemometrics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/50—Molecular design, e.g. of drugs
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Chemical & Material Sciences (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Data Mining & Analysis (AREA)
- Crystallography & Structural Chemistry (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Biophysics (AREA)
- Databases & Information Systems (AREA)
- Medicinal Chemistry (AREA)
- Pharmacology & Pharmacy (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Medical Treatment And Welfare Office Work (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention discloses a medicine interaction prediction method, computer equipment and a medium. In one embodiment, the method comprises: performing feature mapping on the molecular structure of the medicine to be predicted by using the trained graph neural network to obtain a representation vector of the medicine to be predicted, wherein the medicine to be predicted comprises a first medicine and a second medicine; predicting a relationship vector representing the relationship between the medicine to be predicted and the set disease type according to the representation vector of the medicine to be predicted by using the trained first network; and predicting the drug interactions of the first drug and the second drug according to the representation vectors of the first drug and the second drug and the relationship vectors of the relationship between the first drug and the second drug and the set disease types by using the trained second network. The implementation mode can improve coverage on the basis of ensuring accuracy, and can realize the prediction of drug interaction between a new drug and other drugs in research and development.
Description
Technical Field
The present invention relates to the field of artificial intelligence. And more particularly, to a drug interaction prediction method, a computer device, and a medium.
Background
Drug-Drug interactions (DDI) refers to the simultaneous or sequential administration of two or more drugs over a period of time, where the activity of one Drug may be altered by the presence of the other Drug. Drug interactions can cause many adverse effects in patients and have become one of the serious threats to public health. With the increasing spectrum of modern diseases and the increasing resistance of patients, multi-drug prescriptions have become a common treatment option, especially for patients with various chronic diseases such as diabetes, cardiovascular diseases and the like. This tends to increase the clinically relevant risk and presents new challenges for treatment management. The occurrence of drug interactions is often detrimental, exposing the patient to the risk of side effects and toxicity, and even worsening the patient's physical condition. At present, the mode of drug interaction is usually obtained through a drug instruction, however, due to the variety of drugs, the drug instruction cannot cover all the drugs that interact.
Disclosure of Invention
The present invention aims to provide a drug interaction prediction method, a computer device and a medium, which solve at least one of the problems existing in the prior art.
In order to achieve the above purpose, the invention adopts the following technical scheme:
the first aspect of the present invention provides a method of predicting drug interactions comprising:
performing feature mapping on the molecular structure of the medicine to be predicted by using the trained graph neural network to obtain a representation vector of the medicine to be predicted, wherein the medicine to be predicted comprises a first medicine and a second medicine;
predicting a relationship vector representing the relationship between the medicine to be predicted and a set disease type according to the representation vector of the medicine to be predicted by using a trained first network; and
and predicting the drug interaction of the first drug and the second drug according to the representation vector of the first drug, the representation vector of the second drug, the relationship vector of the relationship between the first drug and the set disease type and the relationship vector of the relationship between the second drug and the set disease type by using the trained second network.
Optionally, the performing feature mapping on the molecular structure of the drug to be predicted to obtain a representation vector of the drug to be predicted includes:
the atoms in the molecular structure of the medicine to be predicted are characterized as nodes of the graph structure, and the chemical bonds in the molecular structure of the medicine to be predicted are characterized as edges of the graph structure, so that the graph structure of the medicine to be predicted is obtained;
Performing iterative graph convolution operation on the graph structure of the medicine to be predicted to obtain a representation vector of each node of the medicine to be predicted; and
and obtaining the representation vector of the medicine to be predicted according to the representation vector of each node of the medicine to be predicted.
Optionally, the performing iterative graph convolution operation on the graph structure of the to-be-predicted drug to obtain the representation vector of each node of the to-be-predicted drug includes: and carrying out graph convolution operation of set iteration times on the graph structure of the medicine to be predicted so as to obtain the expression vector of each node of the medicine to be predicted.
Alternatively, the formulation of the graph convolution operation is described as:
h t+1 (e i )=σ(W e ×h t (e i )+W m ×m t (e i ))
wherein h is t+1 (e i ) The ith node e obtained by carrying out the (t+1) th graph convolution operation on the graph structure of the medicine to be predicted i Is a natural number; sigma is a nonlinear activation function; w (W) e Is N 1 ×N 1 Is a first parameter matrix of (a); w (W) m Is N 1 ×N 1 Is a second parameter matrix of (a); ith node e for carrying out t+1st graph convolution operation on graph structure of medicine to be predicted i Is the first intermediate variable of (2)Initial value m of first intermediate variable 0 (e i )=0;N(e i ) Representing node e i Is a neighbor node set; type (k, i) represents the ith node e of the drug to be predicted i And the kth node e k The type of edge between; w (W) type(k,i) Is N 1 ×N 1 Is a third parameter matrix of (a); ith node e of drug to be predicted i Representing the vector initial value h 0 (e i )=σ(W 0 ×onehot(e i );W 0 Is N 1 ×N 2 A fourth parameter matrix of (2); onehot (e) i ) The ith node e for the drug to be predicted i N of the independent heat representation of (2) 2 And (5) maintaining the column vector.
Optionally, the obtaining the representation vector of the drug to be predicted according to the representation vector of each node of the drug to be predicted includes: and taking the average value, the maximum value or the linear weighted value of the calculated expression vector of each node of the medicine to be predicted as the expression vector of the medicine to be predicted.
Alternatively, the process may be carried out in a single-stage,
the calculated average value of the expression vectors of all the nodes of the medicine to be predicted is used as the expression vector of the medicine to be predicted to be formulated and described as follows:
wherein h is a representation vector of the drug to be predicted; the drug to be predicted comprises N nodes; h is a T (e i ) The ith node e of the medicine to be predicted, which is obtained by carrying out iterative graph convolution operation on graph structure of the medicine to be predicted i Is a representation vector of (1);
the calculated maximum value of the expression vector of each node of the medicine to be predicted is used as the expression vector of the medicine to be predicted to be formulated and described as follows:
h=max(h T (e i )),i=1,…,N;
The calculated linear weighted values of the expression vectors of the nodes of the medicine to be predicted are used as the expression vectors of the medicine to be predicted to be formulated and described as follows:
wherein the ith node e of the drug to be predicted i Weight alpha of (2) i The method comprises the following steps:
wherein θ is N 1 The first parameter column vector of the dimension.
Optionally, the formulation of the relation vector representing the relation between the drug to be predicted and the set disease category according to the vector prediction of the drug to be predicted is described as follows:
y=softmax(W 2 ×h 1 +b 2 )
wherein y is a relation vector representing the relation between the medicine to be predicted and the set disease type; second intermediate variable h 1 =σ(W 1 ×h+b 1 ) The method comprises the steps of carrying out a first treatment on the surface of the Sigma is a nonlinear activation function; h is the expression vector of the medicine to be predicted; w (W) 1 Is N 3 ×N 1 Is a fifth parameter matrix of (a); w (W) 2 Is N 4 ×N 3 N, N 4 The value of (2) is the number of the set disease types; b 1 Is N 3 A second parameter column vector of dimensions; b 2 Is N 4 A third parameter column vector of the dimension; softmax () represents a softmax sort function.
Optionally, the formulation of the drug interaction between the first drug and the second drug according to the expression vector of the first drug, the expression vector of the second drug, the relation vector of the relation between the first drug and the set disease type and the relation vector of the relation between the second drug and the set disease type is predicted as follows:
y label =softmax(W 4 ×h a,b +b 4 )
Wherein y is label Is a predicted outcome of a drug interaction of the first drug with the second drug; softmax () represents a softmax classification function; third intermediate variable h a,b The method comprises the following steps:
h a,b =σ(W 3 ×cat(h a ,h b ,h a -h b ,h a +h b ,h a ·h b ,y a +y b -y a ·y b )+b 3 );
sigma is a nonlinear activation function; cat () represents a splicing function; y is a A relation vector of N for the relation between the first medicine and the set disease type 4 A dimension column vector; y is b A relation vector of the relation between the second medicine and the set disease type is N 4 A dimension column vector; h is a a A representation vector of the first medicine, N 1 A dimension column vector; h is a b A representation vector of the second drug, N 1 A dimension column vector; w (W) 3 Is N 5 ×N 6 Is a seventh parameter matrix of (N) 6 =5*N 1 +N 4 ;b 3 Is N 5 A fourth parameter column vector of dimensions; w (W) 4 Is 2 XN 5 Eighth parameter matrix of b 4 A fifth parameter column vector of 2 dimensions.
Optionally, before the feature mapping is performed on the molecular structure of the drug to be predicted by using the trained neural network to obtain the representation vector of the drug to be predicted, the method further includes:
constructing a training data set comprising a plurality of first training data labeled with molecular structures of drugs having a relationship with a set disease category and a plurality of pairs of second training data labeled with molecular structures of drugs having drug interactions; a kind of electronic device with a high-performance liquid crystal display
And simultaneously training to obtain a graph neural network, a first network and a second network based on the loss value of the drug interaction and the loss value of the relation between the drug and the set disease type by using the training data set.
Optionally, the training the graph neural network, the first network and the second network simultaneously based on the loss value of the drug interaction and the loss value of the relationship between the drug and the set disease type includes:
and simultaneously training to obtain a graph neural network, a first network and a second network based on the loss value of the drug interaction and the loss value of the relation between the drug and the set disease type by using a cross entropy loss function and a random gradient descent algorithm.
A second aspect of the invention provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of the first aspect of the invention when executing the program.
A third aspect of the invention provides a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the method of the first aspect of the invention.
The beneficial effects of the invention are as follows:
According to the technical scheme, vector representation of the medicine is obtained through graph neural network learning, and then medicine interaction is predicted by combining the relation between the medicine and the disease type, so that the coverage can be improved on the basis of ensuring the accuracy, and the medicine interaction prediction between the new medicine in research and development and other medicines can be realized.
Drawings
The following describes the embodiments of the present invention in further detail with reference to the drawings.
FIG. 1 illustrates an exemplary system architecture diagram in which an embodiment of the present invention may be applied.
Fig. 2 shows a flow chart of a drug interaction prediction method according to an embodiment of the present invention.
Fig. 3 is a schematic structural diagram of a network model used in the drug interaction prediction method according to the embodiment of the present invention.
Fig. 4 is a schematic diagram of a computer system for performing the drug interaction prediction method according to the embodiment of the present invention.
Detailed Description
In order to more clearly illustrate the present invention, the present invention will be further described with reference to examples and drawings. Like parts in the drawings are denoted by the same reference numerals. It is to be understood by persons skilled in the art that the following detailed description is illustrative and not restrictive, and that this invention is not limited to the details given herein.
In the process of implementing the present invention, the inventor finds that at least the following problems exist in the prior art: at present, the mode of drug interaction is usually obtained through a drug instruction, however, due to the variety of drugs, the drug instruction cannot cover all the drugs that interact.
In addition, the inventor discovers that in the process of developing a new drug, the existing drug interaction acquiring mode cannot acquire the drug interaction between the new drug and other drugs, and if the effective prediction of the drug interaction between the new drug and other drugs can be realized, a coping strategy can be prepared in advance, which is significant for drug development.
In view of this, an embodiment of the present invention provides a method for predicting drug interactions, the method comprising:
performing feature mapping on a molecular structure of a drug to be predicted by using a trained graph neural network (GNN, graph Nerual Network) to obtain a representation vector of the drug to be predicted, wherein the drug to be predicted comprises a first drug and a second drug;
predicting a relationship vector representing the relationship between the medicine to be predicted and a set disease type according to the representation vector of the medicine to be predicted by using a trained first network; a kind of electronic device with a high-performance liquid crystal display
And predicting the drug interaction of the first drug and the second drug according to the representation vector of the first drug, the representation vector of the second drug, the relationship vector of the relationship between the first drug and the set disease type and the relationship vector of the relationship between the second drug and the set disease type by using the trained second network.
According to the drug interaction prediction method provided by the embodiment, vector representation of the drug is obtained through graph neural network learning, and then drug interaction is predicted by combining the relationship between the drug and the disease type. Therefore, the medicine interaction prediction method provided by the embodiment can improve medicine interaction prediction coverage on the basis of ensuring medicine interaction prediction accuracy, and can perform effective medicine interaction prediction as long as the molecular structure of the medicine can be obtained through channels such as medicine specifications. Even for a new drug under development, although the drug specification does not exist clinically, the molecular structure of the new drug exists, that is, it can be predicted by the drug interaction prediction method provided in this example.
The drug interaction prediction method provided in this embodiment may be used in multiple scenarios, such as scenarios where a drug enterprise develops a new drug, and then, for example, a doctor-patient does not see that a drug interaction exists between a certain drug and another drug through a drug instruction, but suspects that there is a drug interaction, but the drug instruction is not covered, but a scenario where further verification is desired, and so on, which are not specifically limited herein.
The method for predicting drug interactions provided in this embodiment may be implemented by a computer device with data processing capability, and specifically, the computer device may be a computer with data processing capability, including a personal computer (PC, personal Computer), a mini-computer or a mainframe, or may be a server or a server cluster with data processing capability, which is not limited in this embodiment.
In order to facilitate understanding of the technical solution of the present embodiment, a scene of the above method provided in the present embodiment in practice will be described below with reference to fig. 1. Referring to fig. 1, the scenario includes a training server 10 and a prediction server 20. In this embodiment, the training server 10 first trains the overall network model including the neural network by using the training data set to obtain a trained network model. Subsequently, the prediction server 20 may perform drug interaction prediction using the network model trained by the training server 10. The molecular structures of the two medicines to be predicted are input into the prediction server 20, and then the medicine interaction prediction result of the two medicines can be obtained.
In practical applications, the training server 10 and the prediction server 20 in fig. 1 may be two independent servers, or may be a server integrated with a model training function and a drug interaction prediction function. When separate servers, the two servers may communicate over a network that may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.
Next, a drug interaction prediction method provided by the present embodiment will be described from the viewpoint of a processing apparatus having data processing capability.
One embodiment of the present invention provides a drug interaction prediction method, as shown in FIG. 2, comprising steps S210-S240. In the whole, step S210 belongs to the training phase, and steps S220 to S240 belong to the prediction phase, and although the description is made in the order of S210 to S240, it is not necessarily meant to be executed in such order, and some steps may be adjusted as needed without violating the logic, or may be appropriately reduced.
Specifically, as shown in fig. 2, the method for predicting drug interactions provided in this embodiment includes the following steps:
and S210, training the network model by using the training data set to obtain a trained network model.
Wherein, as shown in fig. 3, the trained network model includes a graph neural network having a capability of feature-mapping a molecular structure of a drug to be predicted including a first drug and a second drug to obtain a representation vector of the drug to be predicted, a first network having a capability of prediction of a relationship vector representing a relationship between the drug to be predicted and a set disease type from the representation vector of the drug to be predicted by training the learning parameters, and a second network having a capability of prediction of a drug interaction of the first drug and the second drug from the representation vector of the first drug, the representation vector of the second drug, the relationship vector of the relationship between the first drug and the set disease type (first relationship vector), and the relationship vector of the second drug and the set disease type (second relationship vector) by training the learning parameters.
It should be noted that, for ease of understanding, the network model shown in fig. 3 includes two graph neural networks and two first networks, where the two graph neural networks shown in fig. 3 may be twin networks having the same network structure and shared parameters, and the two first networks shown in fig. 3 may be twin networks having the same network structure and shared parameters. In addition, the network model may include only one graph neural network for feature mapping the molecular structures of the first and second drugs, respectively, and one first network for predicting a relationship vector characterizing a relationship between the first drug and a set disease category and a relationship vector capability of the second drug and a set disease category, respectively, based on the representation vectors of the first and second drugs, respectively. The above-described feature mapping and prediction method is specifically described in the subsequent prediction stage.
In one possible implementation, step S210 further includes:
constructing a training data set, wherein the training data set comprises a plurality of first training data and a plurality of pairs of second training data, the first training data is marked with a molecular structure of a drug which is related to a set disease type, and the second training data is marked with a drug molecular structure of drug interaction; a kind of electronic device with a high-performance liquid crystal display
And simultaneously training to obtain a graph neural network, a first network and a second network based on the loss value of the drug interaction and the loss value of the relation between the drug and the set disease type by using the training data set.
In one possible implementation manner, the training to obtain the graph neural network, the first network and the second network simultaneously based on the loss value of the drug interaction and the loss value of the relationship between the drug and the set disease type includes:
and simultaneously training to obtain a graph neural network, a first network and a second network based on the loss value of the drug interaction and the loss value of the relation between the drug and the set disease type by using a cross entropy loss function and a random gradient descent algorithm.
In order to solve the problems that the training data amount is small and the network model is easy to be fitted excessively, the network model is trained by adopting a multi-task learning strategy in the implementation mode, and the generalization capability of the network model can be effectively improved. The multi-task learning is realized by simultaneously learning the three networks, namely the graph neural network, the first network and the second network through the back propagation of two loss values, and the multi-task learning realizes the migration of knowledge through the commonality of other tasks and target tasks, so that the problems of small training data quantity and easy overfitting of a network model can be overcome.
In one specific example, the training data set may be constructed from a drug specification of an existing drug from which drug d is extracted i Molecular structure of (2) and diseases p treated by drugs i Is denoted as (d) i ,p i ) The method comprises the steps of carrying out a first treatment on the surface of the And extracting the medicine d from the medicine instruction i Molecular structure of (d) and drug d i Drug d with interaction j And e.g. by means of the drug d j Drug instruction acquisition of drug d j The molecular structure of (d) i ,d j ). Thus, given a training dataset, including k1= { (d) i ,p i ) Sum k2= { (d) i ,d j ) And then training by using a cross entropy loss function and a random gradient descent algorithm, and optimizing the accuracy of the relationship between the prediction medicine of the network model and the set disease type and the interaction of the medicine to learn parameters.
Continuing the above example, the cross entropy loss function is:
wherein y' i Representing the actual probability of the drug treatable i-th disease, e.g., y 'for the drug treatable disease, obtained from the training data set K1' i 1, y 'is the number of diseases which can not be treated by the medicine' i Is 0.y is i A predicted probability value indicating that the drug can treat the ith disease.Actual values of probability indicating whether there is an interaction between a pair of drugs obtained from the training data set K2, e.g./there is an interaction >1, in the absence of interaction +.>Is 0./>A predicted probability value indicating whether there is an interaction between a pair of drugs. Both of the partial losses can update the parameters of the network model, i.e., train each network in the network model simultaneously, by a back-propagation algorithm, such as a random gradient descent algorithm.
S220, performing feature mapping on the molecular structure of the medicine to be predicted by using the trained graph neural network to obtain a representation vector of the medicine to be predicted, wherein the medicine to be predicted comprises a first medicine and a second medicine. That is, the trained graph neural network is utilized to perform feature mapping on the molecular structure of the first drug to obtain a representation vector of the first drug, and to perform feature mapping on the molecular structure of the second drug to obtain a representation vector of the second drug.
In one possible implementation manner, the feature mapping the molecular structure of the drug to be predicted to obtain the representation vector of the drug to be predicted includes:
the atoms in the molecular structure of the medicine to be predicted are characterized as nodes of the graph structure, and the chemical bonds in the molecular structure of the medicine to be predicted are characterized as edges of the graph structure, so that the graph structure of the medicine to be predicted is obtained;
Performing iterative graph convolution operation on the graph structure of the medicine to be predicted to obtain a representation vector of each node of the medicine to be predicted; and
and obtaining the representation vector of the medicine to be predicted according to the representation vector of each node of the medicine to be predicted.
That is, feature mapping the molecular structure of the first drug to obtain a representation vector of the first drug, and feature mapping the molecular structure of the second drug to obtain a representation vector of the second drug comprises:
the method comprises the steps of characterizing atoms in a molecular structure of a first medicament as nodes of a graph structure, characterizing chemical bonds in the molecular structure of the first medicament as edges of the graph structure to obtain the graph structure of the first medicament, characterizing atoms in a molecular structure of a second medicament as nodes of the graph structure, and characterizing chemical bonds in the molecular structure of the second medicament as edges of the graph structure to obtain the graph structure of the second medicament;
performing iterative graph convolution operation on the graph structure of the first medicament to obtain a representation vector of each node of the first medicament, and performing iterative graph convolution operation on the graph structure of the second medicament to obtain a representation vector of each node of the second medicament; a kind of electronic device with a high-performance liquid crystal display
And obtaining the representation vector of the first medicament according to the representation vector of each node of the first medicament, and obtaining the representation vector of the second medicament according to the representation vector of each node of the second medicament.
In one specific example, the molecular structure of the drug is presented as a chemical formula of the drug, such as the chemical formula of tavalborole:
the present implementation characterizes the molecular structure of the drug as a graph structure, wherein atoms in the molecular structure of the drug are characterized as nodes (or vertices) of the graph structure, chemical bonds in the molecular structure of the drug are characterized as edges of the graph structure, and the graph structure of the drug to be predicted is obtained, for example, when the molecular structure represented by the chemical molecular formula of the tavaboro is characterized as the graph structure, atoms B (boron) and atoms O (oxygen) characterize the nodes in the graph structure, and single bonds between atoms B and O are characterized as edges in the graph structure. Illustratively, since the known chemical elements are 118, the types of nodes in this implementation are 118, while the types of sides include 4 types corresponding to single bond, double bond, triple bond, and aromatic bond, respectively. In this implementation, based on the molecular structure of the drug, the molecular structure features of the drug are mapped to a representation vector of the set dimension of the first drug through the graph neural network. Wherein, if the molecular structure of the drug contains N atoms and M chemical bonds, the graph structure G obtained by characterizing the molecular structure of the drug contains N nodes { e } i I=1, …, N } and M edges { r } j ,j=1,…,M}。
It is verified that the convergence of the iterative convolution operation may require more than ten thousands of iterative operations, and the variation of the representation vector of each node of the drug to be predicted is small after several tens of iterative operations, so, in one possible implementation, to improve the efficiency, the performing the iterative convolution operation on the graph structure of the drug to be predicted to obtain the representation vector of each node of the drug to be predicted includes: and carrying out graph convolution operation of set iteration times on the graph structure of the medicine to be predicted so as to obtain the expression vector of each node of the medicine to be predicted.
In one possible implementation, the formulation of the graph convolution operation is described as:
h t+1 (e i )=σ(W e ×h t (e i )+W m ×m t (e i ))
wherein operator x represents a matrix multiplication operation; h is a t+1 (e i ) The ith node e obtained by carrying out the (t+1) th graph convolution operation on the graph structure of the medicine to be predicted i Is a representation vector of (1); t is a positive integer, and when performing a graph convolution operation for a set number of iterations, t=0, …, T-1, T is the set number of iterations; h is a t+1 (e i ) Is N 1 Column vectors of dimensions; sigma is a nonlinear activation function; w (W) e Is N 1 ×N 1 A first parameter matrix W e The method comprises the steps of obtaining through training and learning; w (W) m Is N 1 ×N 1 A second parameter matrix W m The method comprises the steps of obtaining through training and learning; ith node e for carrying out t+1st graph convolution operation on graph structure of medicine to be predicted i Is the first intermediate variable of (2)Initial value m of first intermediate variable 0 (e i )=0;N(e i ) Representing node e i Is a neighbor node set; type (k, i) represents the ith node e of the drug to be predicted i And the kth node e k The type of edge between (corresponding to one of single, double, triple and aromatic bonds);W type(k,i) Is N 1 ×N 1 A third parameter matrix W type(k,i) The method comprises the steps of obtaining through training and learning; ith node e of drug to be predicted i Representing the vector initial value h 0 (e i )=σ(W 0 ×onehot(e i );W 0 Is N 1 ×N 2 Fourth parameter matrix W of (2) 0 The method comprises the steps of obtaining through training and learning; onehot (e) i ) The ith node e for the drug to be predicted i N of the one-hot representation (onehot representation) 2 And (5) maintaining the column vector.
According to the implementation mode, the information of the neighbor nodes is gathered through iterative graph convolution operation, so that the molecular structural features of the medicine can be accurately and effectively mapped into the expression vector of the medicine.
In a specific example, the nonlinear activation function σ employs a leak ReLU activation function, and other nonlinear activation functions such as Sigmoid, reLU, and the like may also be employed.
In a specific example, the set iteration number T set in this embodiment has a value of 50, and the graph neural network obtains the representation vector h of each node through the graph convolution operation of 50 iterations 50 (e i )(i=1,…,N)。
In one possible implementation manner, the obtaining the representation vector of the to-be-predicted medicine according to the representation vector of each node of the to-be-predicted medicine includes: and taking the average value, the maximum value or the linear weighted value of the calculated expression vector of each node of the medicine to be predicted as the expression vector of the medicine to be predicted.
In one possible implementation of the present invention,
the calculated average value of the expression vectors of all the nodes of the medicine to be predicted is used as the expression vector of the medicine to be predicted to be formulated and described as follows:
wherein h is a representation vector of the drug to be predicted; to be treatedPredicting that the drug comprises N nodes; h is a T (e i ) The ith node e of the medicine to be predicted, which is obtained by carrying out iterative graph convolution operation on graph structure of the medicine to be predicted i In the case of performing a graph convolution operation for a set number of iterations, T represents the set number of iterations, h T (e i ) The ith node e of the medicine to be predicted is obtained by carrying out T times of graph convolution operation on the graph structure of the medicine to be predicted i Is a representation vector of (1);
the calculated maximum value of the expression vector of each node of the medicine to be predicted is used as the expression vector of the medicine to be predicted to be formulated and described as follows:
h=max(h T (e i )),i=1,…,N;
The calculated linear weighted values of the expression vectors of the nodes of the medicine to be predicted are used as the expression vectors of the medicine to be predicted to be formulated and described as follows:
wherein the ith node e of the drug to be predicted i Weight alpha of (2) i The method comprises the following steps:
wherein the operator<·>Representing a dot product (inner product) operation; θ is N 1 First parameter column vector of dimension, N 1 The first parameter column vector θ of the dimension is obtained by training learning. Wherein each dimension element in θ represents the weight of the corresponding dimension element of the representation vector h of the drug to be predicted, and the ith node e of the drug to be predicted is obtained i Weight alpha of (2) i In the formula of (2), the ith node e of the medicine to be predicted is calculated molecularly i The denominator is used for normalization.
If the average value or the maximum value of the expression vector of each node is used as the drugThe method of representing the vector does not include parameters to be obtained by training, and the method of using the linear weighted value of each node representing the vector as the representing vector of the medicine includes N to be obtained by training and learning 1 The first parameter column vector θ of the dimension.
In one possible implementation, the N 1 The value of the N is in the range of 200 to 300 2 And 118. Wherein: regarding N 1 The value of (2) is 200 to 300N 1 Namely, the dimension of the representation vector of the self-defined medicine to be predicted is too simple to describe the information if the value is too small, and too complex to fit if the value is too large, so that the implementation mode is implemented by N 1 The value range of the product is set between 200 and 300, and various requirements can be met. Regarding N 2 Has a value of 118, and since the number of chemical elements known to date is 118, N is 2 Fetch 118 may cover all atom types.
In one particular example, N 1 =256,N 2 =118, then, h t+1 (e i ) A column vector of 256 dimensions; w (W) e A first parameter matrix of 256×256; w (W) m A second parameter matrix of 256×256; w (W) type(k,i) A third parameter matrix of 256×256; w (W) 0 A fourth parameter matrix of 256×118; onehot (e) i ) 118-dimensional column vectors; θ is the 256-dimensional first parameter column vector.
S230, predicting a relation vector representing the relation between the medicine to be predicted and the set disease type according to the representation vector of the medicine to be predicted by using the trained first network.
That is, a relationship vector representing the relationship between the first drug and the set disease type is predicted from the representation vector of the first drug using the trained first network, and a relationship vector representing the relationship between the second drug and the set disease type is predicted from the representation vector of the second drug.
In one possible implementation, the formulation of the relationship vector representing the relationship between the drug to be predicted and the set disease category predicted from the representation vector of the drug to be predicted is described as:
y=softmax(W 2 ×h 1 +b 2 )
wherein y is a relation vector representing the relation between the medicine to be predicted and the set disease type, and y is N 4 Column vector of dimension, N 4 The value of (1) is the number of the set disease types, and each element in the column vector y respectively represents the probability value that the medicine to be predicted can treat each disease type; second intermediate variable h 1 =σ(W 1 ×h+b 1 ) Second intermediate variable h 1 Is N 3 Column vectors of dimensions; sigma is a nonlinear activation function; h is the representation vector of the medicine to be predicted obtained in the step S220, which is N 1 Column vectors of dimensions; w (W) 1 Is N 3 ×N 1 A fifth parameter matrix W 1 The method comprises the steps of obtaining through training and learning; w (W) 2 Is N 4 ×N 3 A sixth parameter matrix W 2 The method comprises the steps of obtaining through training and learning; b 1 Is N 3 Second parameter column vector of dimension, second parameter column vector b 1 The method comprises the steps of obtaining through training and learning; b 2 Is N 4 Third parameter column vector of dimension, third parameter column vector b 2 The method comprises the steps of obtaining through training and learning; softmax () represents a softmax sort function. In the above formula for obtaining the relation vector y of the relation between the medicine to be predicted and the set disease type, the input variable is the second intermediate variable h obtained based on the expression vector h of the medicine to be predicted 1 The output relation vector y of the relation between the medicine to be predicted and the set disease types represents the probability value of each disease type in the set disease types which can be treated by the medicine to be predicted.
In one possible implementation, the N 3 The value of (2) is in the range of 450 to 550. And N 1 The reason for the value range of 200 to 300 is similar to that of N 3 The value range of (2) is set between 450 and 550, and various requirements can be met.
In one specific example, the present implementation uses the 500 diseases most frequently counted from the existing drug instructions as the set disease typeThen N 4 500, i.e., y is a 500-dimensional column vector, 500 probability values for which the drug to be predicted can treat the 500 diseases constitute a 500-dimensional column vector y. For example N 3 =512, and continuing with the previous example N 1 If 256, y is a 500-dimensional column vector, the second intermediate variable h 1 A column vector of 512 dimensions; h is a 256-dimensional column vector; w (W) 1 A fifth parameter matrix of 512 x 256; w (W) 2 A sixth parameter matrix of 500 x 512; b 1 A second parameter column vector of 512 dimensions; b 2 A third parameter column vector of 500 dimensions.
S240, predicting the drug interaction of the first drug and the second drug according to the representation vector of the first drug, the representation vector of the second drug, the relationship vector of the relationship between the first drug and the set disease type and the relationship vector of the relationship between the second drug and the set disease type by using the trained second network.
In one possible implementation, the formulation of the drug interaction between the first drug and the second drug according to the expression vector of the first drug, the expression vector of the second drug, the relation vector of the relation between the first drug and the set disease type, and the relation vector of the relation between the second drug and the set disease type is predicted is described as follows:
y label =softmax(W 4 ×h a,b +b 4 )
wherein y is label Y, the predicted result of drug interaction of the first drug and the second drug label For a 2-dimensional vector, for example, 0 indicates no drug interaction and 1 indicates drug interaction; softmax () represents a softmax classification function; third intermediate variable h a,b The method comprises the following steps:
h a,b =σ(W 3 ×cat(h a ,h b ,h a -h b ,h a +h b ,h a ·h b ,y a +y b -y a ·y b )+b 3 );
sigma is a nonlinear activation function; cat () represents a stitching function, see above, and the stitched object includes six, the first being h a The second is h b Third is h a -h b Fourth is h a +h b Fifth is h a ·h b The sixth is y a +y b -y a ·y b ;y a A relation vector of N for the relation between the first medicine and the set disease type 4 A dimension column vector; y is b A relation vector of the relation between the second medicine and the set disease type is N 4 A dimension column vector; h is a a Is the expression vector of the first medicine, h a Is N 1 A dimension column vector; h is a b Is the expression vector of the second medicine, h b Is N 1 A dimension column vector; w (W) 3 Is N 5 ×N 6 Is a seventh parameter matrix of (N) 6 =5*N 1 +N 4 Seventh parameter matrix W 3 The method comprises the steps of obtaining through training and learning; b 3 Is N 5 A fourth parameter column vector of the dimension is obtained through training and learning; w (W) 4 Is 2 XN 5 Eighth parameter matrix W of (2) 4 The method comprises the steps of obtaining through training and learning; b 4 A fifth parameter column vector b which is 2-dimensional 4 Is obtained through training and learning.
In one possible implementation, the N 5 The value of (2) is in the range of 200 to 300. And N 1 The reason for the value range of 200 to 300 is similar to that of N 5 The value range of the product is set between 200 and 300, and various requirements can be met.
In one particular example, N 5 =256, and continuing with the previous example N 1 =256,N 4 =500, then y a Is a 500-dimensional column vector; y is b Is a 500-dimensional column vector; h is a a 256-dimensional column vectors; h is a b 256-dimensional column vectors; w (W) 3 A seventh parameter matrix of 256×1780, where 1780=5×256+500 (since the first five splice objects are all 256-dimensional column vectors, 256×5 would be straightforward); b 3 A fourth parameter column vector of 256 dimensions; w (W) 4 An eighth parameter matrix of 2 x 256; b 4 A fifth parameter column vector of 2 dimensions.
In summary, the method for predicting drug interactions provided in this embodiment is a prediction method based on molecular representation learning, and based on the complete molecular structure of a drug, vector representations of the drug are obtained through graph neural network learning, and then drug interactions are predicted by combining the relationship between the drug and the disease type. Therefore, the medicine interaction prediction method provided by the embodiment can improve medicine interaction prediction coverage on the basis of ensuring medicine interaction prediction accuracy, and can perform effective medicine interaction prediction as long as the molecular structure of the medicine can be obtained through channels such as medicine specifications. Even for a new drug under development, although the drug specification does not exist clinically, the molecular structure of the new drug exists, that is, it can be predicted by the drug interaction prediction method provided in this example. In order to solve the problems that the training data amount is small and the network model is easy to be over-fitted, the medicine interaction prediction method provided by the embodiment adopts a multi-task learning strategy to train the network model, so that the generalization capability of the network model can be effectively improved.
As shown in fig. 4, a computer system suitable for performing the drug interaction prediction method provided in the above-described embodiment includes a central processing module (CPU) that can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) or a program loaded from a storage section into a Random Access Memory (RAM). In the RAM, various programs and data required for the operation of the computer system are also stored. The CPU, ROM and RAM are connected by a bus. An input/output (I/O) interface is also connected to the bus.
The following components are connected to the I/O interface, including the input part of the keyboard, mouse, etc.; an output section including a display such as a Liquid Crystal Display (LCD) and a speaker; a storage section including a hard disk or the like; and a communication section including a network interface card such as a LAN card, a modem, and the like. The communication section performs communication processing via a network such as the internet. The drives are also connected to the I/O interfaces as needed. Removable media such as magnetic disks, optical disks, magneto-optical disks, semiconductor memories, and the like are mounted on the drive as needed so that a computer program read therefrom is mounted into the storage section as needed.
In particular, according to the present embodiment, the procedure described in the above flowcharts may be implemented as a computer software program. For example, the present embodiments include a computer program product comprising a computer program tangibly embodied on a computer-readable medium, the computer program containing program code for performing the method shown in the flowchart. In such embodiments, the computer program may be downloaded and installed from a network via a communication portion, and/or installed from a removable medium.
The flowcharts and diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to the present embodiments. In this regard, each block in the flowchart or schematic diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the diagrams and/or flowchart illustration, and combinations of blocks in the diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
On the other hand, the present embodiment also provides a nonvolatile computer storage medium, which may be the nonvolatile computer storage medium included in the apparatus in the above embodiment or may be a nonvolatile computer storage medium existing separately and not incorporated in the terminal. The non-volatile computer storage medium stores one or more programs that, when executed by an apparatus, cause the apparatus to: performing feature mapping on the molecular structure of the medicine to be predicted by using the trained graph neural network to obtain a representation vector of the medicine to be predicted, wherein the medicine to be predicted comprises a first medicine and a second medicine; predicting a relationship vector representing the relationship between the medicine to be predicted and a set disease type according to the representation vector of the medicine to be predicted by using a trained first network; and predicting, by using the trained second network, a drug interaction of the first drug and the second drug based on the representation vector of the first drug, the representation vector of the second drug, the relationship vector of the relationship between the first drug and the set disease type, and the relationship vector of the relationship between the second drug and the set disease type.
In the description of the present invention, it should be noted that the azimuth or positional relationship indicated by the terms "upper", "lower", etc. are based on the azimuth or positional relationship shown in the drawings, and are merely for convenience of describing the present invention and simplifying the description, and are not indicative or implying that the apparatus or element in question must have a specific azimuth, be constructed and operated in a specific azimuth, and thus should not be construed as limiting the present invention. Unless specifically stated or limited otherwise, the terms "mounted," "connected," and "coupled" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
It is further noted that in the description of the present invention, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
It should be understood that the foregoing examples of the present invention are provided merely for clearly illustrating the present invention and are not intended to limit the embodiments of the present invention, and that various other changes and modifications may be made therein by one skilled in the art without departing from the spirit and scope of the present invention as defined by the appended claims.
Claims (12)
1. A method of predicting drug interactions, comprising:
performing feature mapping on the molecular structure of the medicine to be predicted by using the trained graph neural network to obtain a representation vector of the medicine to be predicted, wherein the medicine to be predicted comprises a first medicine and a second medicine;
predicting a relationship vector representing the relationship between the medicine to be predicted and a set disease type according to the representation vector of the medicine to be predicted by using a trained first network; and
and predicting the drug interaction of the first drug and the second drug according to the representation vector of the first drug, the representation vector of the second drug, the relationship vector of the relationship between the first drug and the set disease type and the relationship vector of the relationship between the second drug and the set disease type by using the trained second network.
2. The method of claim 1, wherein the feature mapping the molecular structure of the drug to be predicted to obtain a representation vector of the drug to be predicted comprises:
the atoms in the molecular structure of the medicine to be predicted are characterized as nodes of the graph structure, and the chemical bonds in the molecular structure of the medicine to be predicted are characterized as edges of the graph structure, so that the graph structure of the medicine to be predicted is obtained;
performing iterative graph convolution operation on the graph structure of the medicine to be predicted to obtain a representation vector of each node of the medicine to be predicted; and
and obtaining the representation vector of the medicine to be predicted according to the representation vector of each node of the medicine to be predicted.
3. The method of claim 2, wherein performing an iterative graph convolution operation on the graph structure of the drug to be predicted to obtain a representation vector of each node of the drug to be predicted comprises: and carrying out graph convolution operation of set iteration times on the graph structure of the medicine to be predicted so as to obtain the expression vector of each node of the medicine to be predicted.
4. The method of claim 2, wherein the formulating of the graph convolution operation is described as:
h t+1 (e i )=σ(W e ×h t (e i )+W m ×m t (e i ))
Wherein h is t+1 (e i ) The ith node e obtained by carrying out the (t+1) th graph convolution operation on the graph structure of the medicine to be predicted i Is a natural number; sigma is a nonlinear activation function; w (W) e Is N 1 ×N 1 Is a first parameter matrix of (a); w (W) m Is N 1 ×N 1 Is a second parameter matrix of (a); ith node e for carrying out t+1st graph convolution operation on graph structure of medicine to be predicted i Is the first intermediate variable of (2)Initial value m of first intermediate variable 0 (e i )=0;N(e i ) Representing node e i Is a neighbor node set; type (k, i) represents the ith node e of the drug to be predicted i And the kth node e k The type of edge between; w (W) type(k,i) Is N 1 ×N 1 Is a third parameter matrix of (a); ith node e of drug to be predicted i Representing the vector initial value h 0 (e i )=σ(W 0 ×onehot(e i );W 0 Is N 1 ×N 2 A fourth parameter matrix of (2); onehot (e) i ) The ith node e for the drug to be predicted i N of the independent heat representation of (2) 2 And (5) maintaining the column vector.
5. The method of claim 2, wherein the obtaining the representation vector of the drug to be predicted from the representation vector of each node of the drug to be predicted comprises: and taking the average value, the maximum value or the linear weighted value of the calculated expression vector of each node of the medicine to be predicted as the expression vector of the medicine to be predicted.
6. The method of claim 5, wherein the step of determining the position of the probe is performed,
The calculated average value of the expression vectors of all the nodes of the medicine to be predicted is used as the expression vector of the medicine to be predicted to be formulated and described as follows:
wherein h is a representation vector of the drug to be predicted; the drug to be predicted comprises N nodes; h is a T (e i ) The ith node e of the medicine to be predicted, which is obtained by carrying out iterative graph convolution operation on graph structure of the medicine to be predicted i Is a representation vector of (1);
the calculated maximum value of the expression vector of each node of the medicine to be predicted is used as the expression vector of the medicine to be predicted to be formulated and described as follows:
h=max(h T (e i )),i=1,...,N;
the calculated linear weighted values of the expression vectors of the nodes of the medicine to be predicted are used as the expression vectors of the medicine to be predicted to be formulated and described as follows:
wherein the ith node e of the drug to be predicted i Weight alpha of (2) i The method comprises the following steps:
wherein θ is N 1 The first parameter column vector of the dimension.
7. The method of claim 1, wherein the predicting from the representation vector of the drug to be predicted yields a formulated description of a relationship vector characterizing a relationship between the drug to be predicted and a set disease category as:
y=softmax(W 2 ×h 1 +b 2 )
wherein y is a relation vector representing the relation between the medicine to be predicted and the set disease type; second intermediate variable h 1 =σ(W 1 ×h+b 1 ) The method comprises the steps of carrying out a first treatment on the surface of the Sigma is a nonlinear activation function; h is the expression vector of the medicine to be predicted; w (W) 1 Is N 3 ×N 1 Is a fifth parameter matrix of (a); w (W) 2 Is N 4 ×N 3 N, N 4 The value of (2) is the number of the set disease types; b 1 Is N 3 A second parameter column vector of dimensions; b 2 Is N 4 A third parameter column vector of the dimension; softmax () represents a softmax sort function.
8. The method of claim 1, wherein the formulation for predicting drug interactions of the first drug with the second drug based on the expression vector of the first drug, the expression vector of the second drug, the relationship vector of the relationship between the first drug and the set disease category, and the relationship vector of the relationship between the second drug and the set disease category is described as:
y label =softmax(W 4 ×h a,b +b 4 )
wherein y is label Is a predicted outcome of a drug interaction of the first drug with the second drug; softmax () represents a softmax classification function; third intermediate variable h a,b The method comprises the following steps:
h a,b =σ(W 3 ×cat(h a ,h b ,h a -h b ,h a +h b ,h a ·h b ,y a +y b -y a ·y b )+b 3 );
sigma is a nonlinear activation function; cat () represents a splicing function; y is a A relation vector of N for the relation between the first medicine and the set disease type 4 A dimension column vector; y is b A relation vector of the relation between the second medicine and the set disease type is N 4 A dimension column vector; h is a a A representation vector of the first medicine, N 1 A dimension column vector; h is a b A representation vector of the second drug, N 1 A dimension column vector; w (W) 3 Is N 5 ×N 6 Is a seventh parameter matrix of (N) 6 =5*N 1 +N 4 ;b 3 Is N 5 A fourth parameter column vector of dimensions; w (W) 4 An eighth parameter matrix of 2 XN 5, b 4 A fifth parameter column vector of 2 dimensions.
9. The method according to any one of claims 1-8, wherein prior to said feature mapping of the molecular structure of the drug to be predicted using the trained graph neural network to obtain a representation vector of the drug to be predicted, the method further comprises:
constructing a training data set comprising a plurality of first training data labeled with molecular structures of drugs having a relationship with a set disease category and a plurality of pairs of second training data labeled with molecular structures of drugs having drug interactions; and
and simultaneously training to obtain a graph neural network, a first network and a second network based on the loss value of the drug interaction and the loss value of the relation between the drug and the set disease type by using the training data set.
10. The method of claim 9, wherein simultaneously training the graph neural network, the first network, and the second network based on the loss values of the drug interactions and the loss values of the relationships between the drugs and the set disease categories comprises:
And simultaneously training to obtain a graph neural network, a first network and a second network based on the loss value of the drug interaction and the loss value of the relation between the drug and the set disease type by using a cross entropy loss function and a random gradient descent algorithm.
11. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of claims 1-10 when the program is executed by the processor.
12. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any one of claims 1-10.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210083440.2A CN116543853A (en) | 2022-01-25 | 2022-01-25 | Drug interaction prediction method, computer device, and medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210083440.2A CN116543853A (en) | 2022-01-25 | 2022-01-25 | Drug interaction prediction method, computer device, and medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116543853A true CN116543853A (en) | 2023-08-04 |
Family
ID=87445744
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210083440.2A Pending CN116543853A (en) | 2022-01-25 | 2022-01-25 | Drug interaction prediction method, computer device, and medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116543853A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117976244A (en) * | 2024-04-01 | 2024-05-03 | 天津理工大学 | Medicine interaction prediction method and device based on multidimensional characteristics |
-
2022
- 2022-01-25 CN CN202210083440.2A patent/CN116543853A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117976244A (en) * | 2024-04-01 | 2024-05-03 | 天津理工大学 | Medicine interaction prediction method and device based on multidimensional characteristics |
CN117976244B (en) * | 2024-04-01 | 2024-06-07 | 天津理工大学 | Medicine interaction prediction method and device based on multidimensional characteristics |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Che et al. | Interpretable deep models for ICU outcome prediction | |
Ma et al. | Loss odyssey in medical image segmentation | |
Dias et al. | Artificial intelligence in clinical and genomic diagnostics | |
Nagaraj et al. | Artificial flora algorithm-based feature selection with gradient boosted tree model for diabetes classification | |
US10542940B2 (en) | Active patient risk prediction | |
CN109637669B (en) | Deep learning-based treatment scheme generation method, device and storage medium | |
US11075008B2 (en) | Generating drug repositioning hypotheses based on integrating multiple aspects of drug similarity and disease similarity | |
CN109785928A (en) | Diagnosis and treatment proposal recommending method, device and storage medium | |
Alahmar et al. | Application of data mining techniques to predict the length of stay of hospitalized patients with diabetes | |
US10535424B2 (en) | Method for proactive comprehensive geriatric risk screening | |
CN113808693A (en) | Medicine recommendation method based on graph neural network and attention mechanism | |
Gupta et al. | Concurrent imputation and prediction on EHR data using bi-directional GANs: Bi-GANs for EHR imputation and prediction | |
WO2023155441A1 (en) | Medical resource recommendation method and apparatus, device, and storage medium | |
CN116543853A (en) | Drug interaction prediction method, computer device, and medium | |
Tahiri et al. | Optimized quaternion radial Hahn Moments application to deep learning for the classification of diabetic retinopathy | |
CN116072298B (en) | Disease prediction system based on hierarchical marker distribution learning | |
Nelakurthi et al. | Finding cut from the same cloth: Cross network link recommendation via joint matrix factorization | |
US20230253122A1 (en) | Systems and methods for generating a genotypic causal model of a disease state | |
CN116129988B (en) | Model construction method, device, equipment and medium | |
US20130253892A1 (en) | Creating synthetic events using genetic surprisal data representing a genetic sequence of an organism with an addition of context | |
Eshmawi et al. | An ensemble of deep learning enabled brain stroke classification model in magnetic resonance images | |
Rekha Sivakumar et al. | Deep Neural Artificial Intelligence for IoT Based Tele Health Data Analytics. | |
US20240203569A1 (en) | Intelligent triage method and device, storage medium and electronic device | |
Saad et al. | Novel extreme regression-voting classifier to predict death risk in vaccinated people using VAERS data | |
Singh et al. | Deep learning in content-based medical image retrieval |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |