CN115206421A - Drug repositioning method, and repositioning model training method and device - Google Patents
Drug repositioning method, and repositioning model training method and device Download PDFInfo
- Publication number
- CN115206421A CN115206421A CN202210848003.5A CN202210848003A CN115206421A CN 115206421 A CN115206421 A CN 115206421A CN 202210848003 A CN202210848003 A CN 202210848003A CN 115206421 A CN115206421 A CN 115206421A
- Authority
- CN
- China
- Prior art keywords
- data
- vector representation
- vector
- drug
- cytomic
- 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.)
- Granted
Links
- 238000012549 training Methods 0.000 title claims abstract description 171
- 238000000034 method Methods 0.000 title claims abstract description 136
- 238000009511 drug repositioning Methods 0.000 title claims description 26
- 239000003814 drug Substances 0.000 claims abstract description 371
- 229940079593 drug Drugs 0.000 claims abstract description 369
- 239000011159 matrix material Substances 0.000 claims abstract description 208
- 238000012545 processing Methods 0.000 claims abstract description 62
- 230000003993 interaction Effects 0.000 claims abstract description 28
- 230000004044 response Effects 0.000 claims description 75
- 230000006870 function Effects 0.000 claims description 61
- 238000013528 artificial neural network Methods 0.000 claims description 41
- 108090000623 proteins and genes Proteins 0.000 claims description 28
- 230000008569 process Effects 0.000 claims description 26
- 230000008447 perception Effects 0.000 claims description 21
- 238000004590 computer program Methods 0.000 claims description 14
- 238000011176 pooling Methods 0.000 claims description 14
- 239000000126 substance Substances 0.000 claims description 12
- 238000012163 sequencing technique Methods 0.000 claims description 10
- 238000010276 construction Methods 0.000 claims description 7
- 230000002380 cytological effect Effects 0.000 claims description 5
- 230000007246 mechanism Effects 0.000 abstract description 19
- 238000004364 calculation method Methods 0.000 abstract description 12
- 238000013473 artificial intelligence Methods 0.000 abstract description 6
- 238000013135 deep learning Methods 0.000 abstract description 6
- 210000004027 cell Anatomy 0.000 description 49
- 238000010586 diagram Methods 0.000 description 30
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 15
- 201000010099 disease Diseases 0.000 description 14
- 238000004891 communication Methods 0.000 description 8
- 206010028980 Neoplasm Diseases 0.000 description 6
- 238000005070 sampling Methods 0.000 description 6
- 230000009471 action Effects 0.000 description 5
- 201000011510 cancer Diseases 0.000 description 5
- 230000008827 biological function Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000010365 information processing Effects 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 238000012795 verification Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 238000010200 validation analysis Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- DQLATGHUWYMOKM-UHFFFAOYSA-L cisplatin Chemical compound N[Pt](N)(Cl)Cl DQLATGHUWYMOKM-UHFFFAOYSA-L 0.000 description 2
- 229960004316 cisplatin Drugs 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000006907 apoptotic process Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000857 drug effect Effects 0.000 description 1
- 230000001973 epigenetic effect Effects 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 102000054999 human core Human genes 0.000 description 1
- 108700026469 human core Proteins 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000011987 methylation Effects 0.000 description 1
- 238000007069 methylation reaction Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 229910052697 platinum Inorganic materials 0.000 description 1
- BASFCYQUMIYNBI-UHFFFAOYSA-N platinum Substances [Pt] BASFCYQUMIYNBI-UHFFFAOYSA-N 0.000 description 1
- 230000006916 protein interaction Effects 0.000 description 1
- 102000004169 proteins and genes Human genes 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000001953 sensory effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B15/00—ICT specially adapted for analysing two-dimensional or three-dimensional molecular structures, e.g. structural or functional relations or structure alignment
- G16B15/30—Drug targeting using structural data; Docking or binding prediction
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
-
- 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
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biophysics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Chemical & Material Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Biotechnology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioethics (AREA)
- Software Systems (AREA)
- Public Health (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Epidemiology (AREA)
- Databases & Information Systems (AREA)
- Medicinal Chemistry (AREA)
- Pharmacology & Pharmacy (AREA)
- Crystallography & Structural Chemistry (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The disclosure provides a medicine repositioning method, a repositioning model training method and a repositioning model training device, and relates to the technical field of deep learning and biological calculation in the technical field of artificial intelligence. The method comprises the following steps: obtaining drug molecule and cytomic data; obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data; performing multi-headed attention processing on the first vector representation and the second vector representation to obtain a predictive vector matrix, wherein the predictive vector matrix is used to characterize the interaction relationship between the drug molecule and the cytomic data; the drug molecules are repositioned according to the predictive vector matrix and the first vector representation. The method captures the relation between the neutron structure of the drug molecule and the cytomic data based on a multi-head attention mechanism, improves the prediction accuracy and enhances the relocation accuracy.
Description
Technical Field
The disclosure relates to the technical field of deep learning and biological computation in the technical field of artificial intelligence, in particular to a medicine repositioning method, a repositioning model training method and a device.
Background
Drug relocation can be applied to the scene of finding new indications for existing drugs, in the biomedical industry, drugs have the characteristics of difficulty in finding, long approval period and the like, and if new applications of approved drugs can be found, the approval period can be shortened, the market capacity of drug enterprises is expanded, and meanwhile, patients are benefited. At present, a new potential indication of a drug can be searched by predicting the response degree (such as an IC50 value) of the drug to omics characterization of a certain cell line, so that how to improve the prediction accuracy and enhance the accuracy of drug relocation become a problem to be solved urgently.
Disclosure of Invention
A method for repositioning drugs, a method for training a repositioning model and a device are provided.
According to a first aspect, there is provided a method of drug relocation comprising: obtaining drug molecule and cytomic data; obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data; performing multi-headed attention processing on the first and second vector representations to obtain a predictive vector matrix, wherein the predictive vector matrix is used to characterize the interaction relationship between the drug molecule and the cytomic data; relocating the drug molecules according to the predictive vector matrix and the first vector representation.
According to a second aspect, there is provided a method of drug relocation based on a relocation model, comprising: acquiring drug molecules and cytomic data, and inputting the drug molecules and the cytomic data into a trained target relocation model; obtaining, from a coding network in the target relocation model, a first vector representation of the drug molecule and a second vector representation of the cytomic data; performing multi-head attention processing on the first vector representation and the second vector representation by a multi-head attention network in the target relocation model to obtain a prediction vector matrix, wherein the prediction vector matrix is used for characterizing the interaction relationship between the drug molecules and the omics data; relocating, by a predictive network in the target relocation model, the drug molecule according to the predictive vector matrix and the first vector representation.
According to a third aspect, there is provided a training method of a repositioning model, comprising: obtaining samples corresponding to omics data of the same drug molecule and different cell lines and samples corresponding to omics data of different drug molecules and the same cell line from a training set as training samples, wherein the training set further comprises real values of response parameters between the drug molecules and the omics data; training a repositioning model to be trained according to the training sample; and adjusting the model parameters of the repositioning model according to the predicted values of the response parameters and the real values of the response parameters output by training, and continuing to train the adjusted repositioning model by using the next training sample until a trained target repositioning model is obtained.
According to a fourth aspect, there is provided a medication repositioning device, comprising: the first acquisition module is used for acquiring drug molecule and cytomic data; a second acquisition module for acquiring a first vector representation of the drug molecule and a second vector representation of the cytomic data; a third obtaining module configured to perform multi-point attention processing on the first vector representation and the second vector representation to obtain a predictive vector matrix, wherein the predictive vector matrix is configured to characterize an interaction relationship between the drug molecule and the cytological data; a prediction module for relocating the drug molecule according to the predictive vector matrix and the first vector representation.
According to a fifth aspect, a drug repositioning device based on a repositioning model, comprising: the first acquisition module is used for acquiring drug molecules and cytomics data and inputting the drug molecules and the cytomics data into a trained target relocation model; a second obtaining module for obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data from a coding network in the target relocation model; a third obtaining module, configured to perform multi-head attention processing on the first vector representation and the second vector representation by a multi-head attention network in the target relocation model to obtain a prediction vector matrix, wherein the prediction vector matrix is used for characterizing an interaction relationship between the drug molecule and the omics data; and the prediction module is used for performing relocation on the drug molecules according to the prediction vector matrix and the first vector representation by a prediction network in the target relocation model.
According to a sixth aspect, there is provided a training apparatus for a relocation model, comprising: the acquisition module is used for acquiring samples corresponding to omics data of the same drug molecule and different cell lines and samples corresponding to omics data of different drug molecules and the same cell line from a training set as training samples, wherein the training set also comprises real values of response parameters between the drug molecules and the omics data; the training module is used for training the repositioning model to be trained according to the training sample; and the updating module is used for adjusting the model parameters of the repositioning model according to the predicted values of the response parameters and the real values of the response parameters, which are output by training, and continuing to train the adjusted repositioning model by using the next training sample until the trained target repositioning model is obtained.
According to a seventh aspect, there is provided an electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method of drug relocation according to the first aspect of the disclosure, or a method of drug relocation based on a relocation model according to the second aspect of the disclosure, or a method of training a relocation model according to the third aspect of the disclosure.
According to an eighth aspect, there is provided a non-transitory computer readable storage medium having stored thereon computer instructions for causing a computer to execute the method of drug relocation according to the first aspect of the present disclosure, or the method of drug relocation based on a relocation model according to the second aspect of the present disclosure, or the method of training a relocation model according to the third aspect of the present disclosure.
According to a ninth aspect, there is provided a computer program product comprising a computer program which, when executed by a processor, carries out the steps of the method of drug relocation according to the first aspect of the present disclosure, or the steps of the method of drug relocation based on a relocation model according to the second aspect of the present disclosure, or the steps of the method of training of a relocation model according to the third aspect of the present disclosure.
It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.
Drawings
The drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:
FIG. 1 is a schematic flow chart diagram of a drug repositioning method according to a first embodiment of the present disclosure;
FIG. 2 is a schematic flow chart diagram of a drug repositioning method according to a second embodiment of the present disclosure;
FIG. 3 is a schematic flow chart diagram of a drug repositioning method according to a third embodiment of the present disclosure;
FIG. 4 is a schematic flow chart diagram of a drug repositioning method according to a fourth embodiment of the present disclosure;
FIG. 5 is a schematic flow chart diagram of a drug repositioning method according to a fifth embodiment of the present disclosure;
FIG. 6 is a schematic flow chart diagram of a method of drug relocation based on a relocation model according to a first embodiment of the present disclosure;
FIG. 7 is a schematic flow chart diagram of a method of drug relocation based on a relocation model according to a second embodiment of the present disclosure;
FIG. 8 is an exemplary block diagram of a drug repositioning method based on a repositioning model according to an embodiment of the disclosure;
FIG. 9 is a schematic flow chart diagram of a method of training a relocation model according to a first embodiment of the present disclosure;
FIG. 10 is a schematic cross-sampling in a method of training a repositioning model according to an embodiment of the present disclosure;
FIG. 11 is a flow chart diagram of a method of training a relocation model according to a second embodiment of the present disclosure;
FIG. 12 is a diagram illustrating the results of validation of the effects of the trained repositioning model;
fig. 13 is a block diagram of a medication repositioning device according to a first embodiment of the disclosure;
FIG. 14 is a block diagram of a medication repositioning device according to a second embodiment of the present disclosure;
FIG. 15 is a block diagram of a drug repositioning device based on a repositioning model according to a first embodiment of the present disclosure;
FIG. 16 is a block diagram of a drug repositioning device based on a repositioning model according to a second embodiment of the present disclosure;
FIG. 17 is a block diagram of a training apparatus for a repositioning model according to a first embodiment of the present disclosure;
FIG. 18 is a block diagram of a training apparatus for a repositioning model according to a second embodiment of the present disclosure;
FIG. 19 is a block diagram of an electronic device used to implement methods of embodiments of the present disclosure.
Detailed Description
Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of embodiments of the present disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
Artificial Intelligence (AI) is a technical science that studies and develops theories, methods, techniques and application systems for simulating, extending and expanding human intelligence. At present, the AI technology has the advantages of high automation degree, high accuracy and low cost, and is widely applied.
Deep Learning (DL) is a new research direction in the field of Machine Learning (ML), and learns the intrinsic rules and representation levels of sample data, and information obtained in the Learning process is helpful for interpreting data such as text, images, and sound. The final aim of the method is to enable a machine to have analysis and learning capabilities like a human, and to recognize data such as characters, images and sounds. As for specific research content, the method mainly comprises a neural network system based on convolution operation, namely a convolution neural network; a multilayer neuron based self-coding neural network; and pre-training in a multilayer self-coding neural network mode, and further optimizing the deep confidence network of the neural network weight by combining the identification information. Deep learning has achieved many achievements in search technology, data mining, machine learning, machine translation, natural language processing, multimedia learning, speech, recommendation and personalization technologies, and other related fields. The deep learning enables the machine to imitate human activities such as audio-visual and thinking, solves a plurality of complex pattern recognition problems, and makes great progress on the artificial intelligence related technology.
The biological calculation is a new calculation mode which is researched and developed by utilizing an information processing mechanism inherent to a biological system. A biological computer is understood to be a supercomputer consisting of biological components that can perform computational tasks such as data storage and logical operations. Biometrics are mainly classified into 3 types: protein calculations, RNA calculations and DNA calculations. E.g. the biological calculations present in our body: DNA stores human core genetic information, RNA input data, ribosome-based logical operations, and the like.
The method for repositioning drugs, the method for training a repositioning model, and the device according to the embodiments of the present disclosure will be described below with reference to the accompanying drawings.
Fig. 1 is a schematic flow diagram of a drug repositioning method according to a first embodiment of the present disclosure.
As shown in fig. 1, the drug relocation method of the embodiment of the present disclosure may specifically include the following steps:
and S101, obtaining drug molecule and cytomic data.
In particular, the executing body of the drug repositioning method according to the embodiment of the present disclosure may be the drug repositioning device provided by the embodiment of the present disclosure, and the drug repositioning device may be a hardware device with data information processing capability and/or necessary software for driving the hardware device to work. Alternatively, the execution subject may include a workstation, a server, a computer, a user terminal, and other devices.
In the embodiment of the present disclosure, the drug molecules to be relocated and the cytomic data corresponding to a preselected disease condition are obtained, for example, the cytomic data can be transcriptomic data, genomics data, epigenetic data of cancer cells, and the cytomic data in the embodiment of the present disclosure can be single omic data or multiple groups of cytomic data.
S102, a first vector representation of the drug molecule and a second vector representation of the cytomic data are obtained.
In the disclosed embodiments, the drug molecules and the cytomic data of a disease are separately feature extracted to obtain respective vector representations, i.e., a first vector representation of the drug molecules and a second vector representation of the cytomic data.
And S103, performing multi-head attention processing on the first vector representation and the second vector representation to obtain a prediction vector matrix, wherein the prediction vector matrix is used for representing the interaction relation between the drug molecules and the cytomic data.
In embodiments of the present disclosure, the relationship between drug molecules and cytomic data, such as the interaction between the two, is captured based on a multi-point attention mechanism.
And performing multi-head attention calculation on the first vector representation of the drug molecules and the second vector representation of the cytomics data to finally obtain a prediction vector matrix between the first vector representation of the drug molecules and the second vector representation of the cytomics data, wherein the matrix can be used for representing the interaction relationship between the drug molecules and the cytomics data.
S104, repositioning the drug molecules according to the prediction vector matrix and the first vector representation.
In the embodiment of the disclosure, the first vector represents the characteristics such as the structure of the drug molecule, the predictive vector matrix represents the interaction between the drug molecule and the cytomics, and the degree of response of the drug molecule to the disease cell can be predicted based on the predictive vector matrix and the first vector representation, so as to determine whether the drug molecule can be applied to treat the disease, thereby achieving drug relocation.
In conclusion, the drug relocation method of the embodiment of the disclosure obtains drug molecule and cytomic data; obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data; performing multi-point attention processing on the first vector representation and the second vector representation to obtain a predictive vector matrix, wherein the predictive vector matrix is used for characterizing the interaction relationship between the drug molecules and the cytological data; the drug molecules are repositioned according to the predictive vector matrix and the first vector representation. Based on the multi-head attention mechanism, the relation between the neutron structure of the drug molecule and the cytomics data is captured, the prediction accuracy is improved, and the relocation accuracy is enhanced.
Fig. 2 is a schematic flow diagram of a drug repositioning method according to a second embodiment of the present disclosure.
As shown in fig. 2, based on the embodiment shown in fig. 1, the method for repositioning a drug according to the embodiment of the present disclosure may specifically include the following steps:
s201, obtaining drug molecule and cytomic data.
S202, a first vector representation of the drug molecule and a second vector representation of the cytomic data are obtained.
S203, determining a query vector matrix corresponding to multi-head attention processing according to the first vector representation.
In the disclosed embodiment, one input matrix, i.e., a query vector (query vector) matrix Q, for constructing a multi-head attention mechanism is constructed according to the first vector representation of the drug molecule.
Multiplying the matrix X corresponding to the first vector representation by a weight matrix W i Q To get the ith attention headCorresponding query matrix Q i Therefore, query vector matrixes of h attention heads are obtained respectively, wherein i =1,2, … …, h.
And S204, determining a key vector matrix and a value vector matrix corresponding to multi-head attention processing according to the second vector representation.
In the embodiment of the present disclosure, another two input matrices, i.e., a Key Vector matrix K and a Value Vector matrix V, for constructing a multi-head attention mechanism are constructed according to the second Vector representation of the cytomics data.
Multiplying the matrix R corresponding to the second vector representation by the weight matrix W i K To obtain the key matrix K corresponding to the ith attention i Thus, a key matrix of h head attention is obtained.
Multiplying the matrix R corresponding to the second vector representation by the weight matrix W i V To obtain the value vector matrix V corresponding to the ith attention head i Thus, h value vector matrixes of the attention heads are obtained respectively.
Wherein the weight matrix W i Q 、W i K And W i V Can be obtained by presetting or pre-training.
In some embodiments, the above process of determining the query vector matrix from the first vector representation and determining the key vector matrix and the value vector matrix from the second vector representation may be implemented by linear layer mapping.
S205, multi-head attention processing is carried out according to the query vector matrix, the key vector matrix and the value vector matrix to obtain a prediction vector matrix.
In the embodiment of the present disclosure, the vector matrix Q is respectively queried based on the following calculation formula i Key vector matrix K i Sum vector matrix V i Calculate the ith attention head H i 。
Wherein softmax () is an activation function, d k Is a dimension of one key vector in the key vector matrix, K T Is the transpose of the key vector matrix.
The final output result of the multi-head attention is obtained by splicing or stacking the results of the h attention heads through a linear layer, and can be realized based on the following formula:
MHA(Q,K,V)=concatenation(H 1 ,H 2 ,……H h )W O
wherein, registration () is a splicing function for stacking the results of h attention heads, W O Is a weight matrix that needs to be preset or trained.
MHA (Q, K, V) and the first vector represent the result of the corresponding matrix X after being processed by Add & Norm layers for preventing network degradation and feedforward network as the prediction vector matrix, where Add represents Residual Connection (Layer Normalization) for normalizing the activation value of each Layer.
S206, repositioning the drug molecules according to the prediction vector matrix and the first vector representation.
Specifically, steps S201 to S202 are the same as steps S101 to S102, and step S206 is the same as step S104, and are not described again here.
On the basis of the above embodiment, as shown in fig. 3, an embodiment of the present disclosure further includes an obtaining process of the first vector representation, which may include the following steps:
s301, generating a corresponding graph of the drug molecules according to the structural information of the drug molecules, wherein nodes in the graph correspond to atoms in the drug molecules, and edges correspond to chemical bonds in the drug molecules.
In some embodiments, the structural information of the drug molecules may be determined according to the chemical formula of the drug molecules, the connection relationship between atoms therein may be determined, and a corresponding graph may be generated based on the structural information of the drug molecules, where each node corresponds to an atom in the graph, and the edges between the nodes correspond to chemical bonds between atoms.
S302, coding the graph corresponding to the drug molecules based on the graph neural network to obtain vector representation of the nodes as first vector representation.
Inputting the graph corresponding to the drug molecules into a graph neural network to encode the nodes in the graph, namely extracting the characteristics of the nodes to obtain the vector representation of each node, thereby realizing the extraction of the characteristics of the structure and the like of the drug molecules.
On the basis of the above embodiment, as shown in fig. 4, an embodiment of the present disclosure further includes an obtaining process of the second vector representation, which may include the following steps:
s401, constructing a graph corresponding to the cytomic data according to the relation between genes in the cytomic data, wherein the nodes in the graph correspond to the genes, and the edges in the graph are used for representing the relation between the genes.
In the embodiment of the disclosure, the relationship between genes in the cytomic data can be determined based on a protein interaction network (PPI), and a graph corresponding to the cytomic data, such as an interaction network graph, is constructed, wherein nodes in the graph correspond to the genes, and edges in the graph are used for characterizing the relationship between the genes.
S402, encoding a map corresponding to the cytomic data based on a map neural network to obtain a vector representation of the cytomic data.
And inputting the map corresponding to the cytomic data into a map neural network, and encoding the map to extract the characteristics of the cytomic data, such as the corresponding characteristics of genome, transcriptome, methylation and the like, and using the characteristics as vector representation of the cytomic data.
And S403, performing multilayer perception processing on the vector representation of the cytomic data based on the multilayer perception network to obtain a second vector representation of the cytomic data.
In embodiments of the present disclosure, a vector representation of the cytomic data is input to a multi-layered sensing network, such as a multi-layered sensor MLP, and the output set of vector representations is taken as a second vector representation of the cytomic data.
In some embodiments, the characteristics of the cytomic data can also be extracted through convolutional neural networks and fully-connected networks to obtain a second vector representation, or the cytomic data can be directly encoded through a multilayer perceptron.
On the basis of the above embodiment, as shown in fig. 5, the step S206 of "relocating drug molecules according to the predictive vector matrix and the first vector representation" may include the following steps:
s501, the first vector representation of the drug molecules is subjected to global pooling to generate a vector matrix of the drug molecules.
In embodiments of the present disclosure, a first vector representation of a drug molecule is globally pooled (Global Pooling) to obtain a vector matrix characterizing the overall characteristics of the drug molecule.
S502, obtaining a predicted value of a response parameter between the drug molecule and the cytomics data according to the predicted vector matrix and the vector matrix of the drug molecule.
In the embodiment of the disclosure, after the vector matrix for prediction and the vector matrix for drug molecules are stacked, the response parameter between the drug molecules and the cytomic data is predicted through the multilayer perceptron, the response parameter can be IC50, and the IC50 value can be used for measuring the capacity of the drug for inducing the apoptosis of disease cells.
S503, repositioning the drug molecules according to the predicted value.
In the embodiment of the present disclosure, the response degree of the drug molecule to the disease cell involved in relocation can be determined according to the predicted value of IC50, so as to determine whether the drug molecule can be used for treating the disease cell corresponding disorder, thereby achieving drug relocation. When omics data of cell lines of different diseases (cancers) are used as input, IC50 predicted values of the same drug molecule in the cell lines of different diseases can be obtained, and therefore, the new application of the existing drug can be found.
In conclusion, the drug relocation method of the embodiment of the disclosure obtains drug molecule and cytomic data; obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data; performing multi-point attention processing on the first vector representation and the second vector representation to obtain a predictive vector matrix, wherein the predictive vector matrix is used for characterizing the interaction relationship between the drug molecules and the cytological data; the drug molecules are repositioned according to the predictive vector matrix and the first vector representation. The biological function differences of different substructures in the drug molecules are learned based on a multi-head attention mechanism, and the relationship between the substructures in the drug molecules and the cytomic data is captured, so that the action result of the drug molecules on a cell system is better predicted, the prediction accuracy is improved, and the repositioning accuracy is enhanced.
The above drug relocation process can be achieved by building a deep learning model as the relocation model.
Fig. 6 is a schematic flow diagram of drug relocation based on a relocation model according to a first embodiment of the present disclosure.
As shown in fig. 6, the method for drug relocation based on a relocation model according to an embodiment of the present disclosure may specifically include the following steps:
s601, obtaining drug molecule and cytomic data, and inputting the drug molecule and cytomic data into the trained target relocation model.
Specifically, the implementation subject of the relocation model-based drug relocation method according to the embodiments of the present disclosure may be the relocation model-based drug relocation apparatus provided in the embodiments of the present disclosure, and the relocation model-based drug relocation apparatus may be a hardware device having data information processing capability and/or necessary software for driving the hardware device to operate. Alternatively, the execution body may include a workstation, a server, a computer, a user terminal, and other devices.
In the embodiment of the disclosure, the acquired cytomic data of the drug molecules to be relocated and a preselected disease are input into a trained target relocation model, so that the drug molecules to be relocated are relocated based on the target relocation model.
In the disclosed embodiments, the target relocation model may include, but is not limited to, a coded network and a multi-head attention network.
S602, obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data from the coding network in the target relocation model.
In embodiments of the disclosure, features corresponding to the drug molecules and the cytomic data are extracted based on the coding network to obtain a first vector representation of the drug molecules and a second vector representation of the cytomic data.
And S603, performing multi-head attention processing on the first vector representation and the second vector representation by using a multi-head attention network in the target relocation model to obtain a prediction vector matrix, wherein the prediction vector matrix is used for representing the interaction relation between the drug molecules and the cytomics data.
The disclosed embodiments construct a multi-head attention network based on a multi-head attention mechanism, and extract a relationship between a first vector representation and a second vector representation based on the multi-head attention mechanism, so as to obtain a prediction vector matrix for characterizing an interaction relationship between a drug molecule and a cytomic data.
S604, the drug molecules are relocated by the prediction network in the target relocation model according to the prediction vector matrix and the first vector representation.
In an embodiment of the present disclosure, the relocation model may further include a prediction network for predicting a degree of response of the drug molecule to the disease cell based on the prediction vector matrix and the first vector representation, to thereby determine whether the drug molecule is applicable to treat the disease, thereby achieving drug relocation.
It should be noted that the above explanation of the embodiment of the drug relocation method is also applicable to the drug relocation method based on the relocation model in the embodiment of the present disclosure, and the specific process is not described herein again.
In conclusion, the drug relocation method based on the relocation model according to the embodiment of the present disclosure obtains drug molecule and cytomics data, and inputs the drug molecule and the cytomics data into the trained target relocation model; obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data from a coding network in the target relocation model; performing multi-head attention processing on the first vector representation and the second vector representation by a multi-head attention network in the target relocation model to obtain a prediction vector matrix, wherein the prediction vector matrix is used for representing the interaction relation between the drug molecules and the cytomics data; drug molecules are relocated by the predictive network in the target relocation model according to the predictive vector matrix and the first vector representation. The method is characterized in that a multi-head attention mechanism is used for modeling drug molecules and cytomics data, meanwhile, the characteristics of the drug molecules are extracted through a graph neural network, the characteristics of the cytomics data are extracted through the graph neural network and multi-layer perception, the method is applicable to various types of drug molecules, the universality of the model is enhanced, the relation between a neutron structure of the drug molecules and the cytomics data is learned through the multi-head attention mechanism, and the drug molecules are repositioned based on a trained target repositioning model, so that the prediction accuracy is improved, and the repositioning accuracy is enhanced.
Fig. 7 is a schematic flow diagram of a method of drug relocation based on a relocation model according to a second embodiment of the present disclosure.
As shown in fig. 7, based on the embodiment shown in fig. 6, the method for drug relocation of a relocation model according to an embodiment of the present disclosure may specifically include the following steps:
s701, obtaining drug molecule and cytomic data, and inputting the drug molecule and cytomic data into the trained target relocation model.
S702, obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data from the coding network in the target relocation model.
And S703, determining a query vector matrix corresponding to the multi-head attention processing according to the first vector representation, and determining a key vector matrix and a value vector matrix corresponding to the multi-head attention processing according to the second vector representation by the plurality of linear layers respectively.
In embodiments of the present disclosure, the multi-head attention network may include, but is not limited to, a plurality of linear layers and a corresponding plurality of attention layers. The linear layer corresponds to the attention layer, i.e. the linear layer corresponding to the ith attention layer is used for calculating the weight matrix W i Q Mapping a matrix X corresponding to the first vector representation to obtain an input matrix, namely a query vector matrix Q, corresponding to the first i attention layers, wherein i =1,2, … …, h; and according to the weight matrix W i K Mapping the matrix R corresponding to the second vector representation to obtain another input matrix corresponding to the first i attention layers, namely a key vector matrix K; and
according to a weight matrix W i V And mapping the matrix R corresponding to the second vector representation to obtain another input matrix corresponding to the first i attention layers, namely a value vector matrix V.
Wherein the weight matrix W i Q 、W i K And W i V Can be obtained by presetting or pre-training.
The specific process is the same as the above embodiment and is not described herein again.
S704, multi-head attention processing is carried out by the plurality of attention layers according to the query vector matrix, the key vector matrix and the value vector matrix to obtain a prediction vector matrix.
And inputting the acquired query vector matrix, key vector matrix and value vector matrix corresponding to each attention layer into the attention layer to obtain the attention head output by each layer, splicing or stacking the results of h attention heads, and obtaining the final output result MHA (Q, K, V) of the multi-head attention through the linear layer after the attention calculation.
In the embodiment of the present disclosure, the relocation model may further include an Add & Norm layer and a feed-forward network, and the MHA (Q, K, V) and the first vector represent a result of the corresponding matrix X after being processed by the Add & Norm layer and the feed-forward network as the prediction vector matrix.
The specific process is the same as the above embodiment, and is not described herein again.
S705, the drug molecules are relocated by the prediction network in the target relocation model according to the prediction vector matrix and the first vector representation.
Specifically, steps S701 to S702 are the same as steps S601 to S602, and step S705 is the same as step S604, and are not repeated here.
On the basis of the above embodiments, the embodiments of the present disclosure generate a graph corresponding to a drug molecule according to the structural information of the drug molecule before acquiring the first vector representation of the drug molecule from the coding network in the target relocation model, wherein the nodes in the graph correspond to atoms in the drug molecule, and the edges correspond to chemical bonds in the drug molecule.
On the basis of the embodiment, the coding network in the target relocation model comprises a graph neural network for extracting the characteristics of the drug molecules.
The graph corresponding to the drug molecule is encoded based on a graph neural network to obtain a vector representation of the nodes as a first vector representation.
On the basis of the above embodiments, the embodiments of the present disclosure further include, before obtaining the second vector representation of the cytomic data from the coding network in the target relocation model: constructing a map corresponding to the cytomic data according to the relationship between genes in the cytomic data, wherein nodes in the map correspond to the genes, and edges in the map are used for representing the relationship between the genes
On the basis of the above embodiment, the coding network in the target relocation model may include a graph neural network and a multilayer perception network for performing feature extraction on the cytomics data, and the graph corresponding to the cytomics data is coded based on the graph neural network to obtain vector representation of the cytomics data; the vector representation of the cytomic data is subjected to multi-tier perception processing based on a multi-tier perception network to obtain a second vector representation of the cytomic data.
In some embodiments, a convolutional neural network may be added to the target relocation model, and the cytomic data is directly input to the convolutional neural network for encoding, or a multi-layer sensing network, i.e., a multi-layer sensor or a multi-layer sensing layer, is added to the target relocation model, and the cytomic data is directly encoded by the multi-layer sensor.
On the basis of the above embodiment, the prediction network includes a pooling layer, a splicing layer and a multi-layer sensing layer, and the drug molecules are relocated by the prediction network in the target relocation model according to the prediction vector matrix and the first vector representation, including:
globally pooling, by a pooling layer, the first vector representations of the drug molecules to generate a vector matrix of drug molecules; stacking the predicted vector matrix and the vector matrix of the drug molecules by the splicing layer to generate a matrix to be processed; and obtaining a predicted value of a response parameter between the drug molecule and the cytomics data by the multilayer perception layer according to the matrix to be processed, and repositioning the drug molecule according to the predicted value.
It should be noted that the above explanation of the embodiment of the drug relocation method is also applicable to the drug relocation method based on the relocation model in the embodiment of the present disclosure, and the specific process is not described herein again.
In summary, the method for relocating a drug based on a relocation model according to the embodiment of the present disclosure obtains drug molecule and cytomic data, and inputs the drug molecule and the cytomic data into a trained target relocation model; obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data from a coding network in the target relocation model; performing multi-head attention processing on the first vector representation and the second vector representation by a multi-head attention network in the target relocation model to obtain a prediction vector matrix, wherein the prediction vector matrix is used for representing the interaction relation between the drug molecules and the cytomics data; drug molecules are relocated by the predictive network in the target relocation model according to the predictive vector matrix and the first vector representation. The method is characterized in that a multi-head attention mechanism is used for modeling drug molecules and cytomics data, the characteristics of the drug molecules are extracted through a graph neural network, the characteristics of the omics data are extracted through the graph neural network and multi-layer perception, the method is applicable to various types of drug molecules, the universality of the model is enhanced, the biological function differences of different substructures in the drug molecules and the relation between the substructures in the drug molecules and the cytomics data are learned through the multi-head attention mechanism, therefore, the action result of the drug molecules on a cell system is better predicted, the drug molecule repositioning is performed based on a trained target repositioning model, the prediction accuracy is improved, and the repositioning accuracy is enhanced.
To describe the drug relocation method based on the relocation model in detail, referring to fig. 8, as shown in fig. 8, a drug molecule is mapped, a map corresponding to the drug molecule is input into a map neural network to obtain a first vector representation, the mapped map of the configured omics data is input into a map neural network to extract features, the extracted features are input into a multilayer sensing layer to obtain a second vector representation of multiple sets of chemical data, the first vector representation and the second vector representation are input into a multi-head attention network after being stacked through a splicing layer to obtain attention between a substructure of the drug molecule and the multiple omics data, a final output result of the attention network and a matrix corresponding to the first vector representation are input into a multi-head attention network to obtain a predicted vector matrix through a residual error & normalization (Add & Norm) layer and a feed-forward network, the predicted vector matrix and a vector matrix obtained after all the first vector representations are globally pooled are input into a multilayer sensing layer MLP to obtain a predicted value of a response parameter IC50, so as to relocate the drug molecule according to the predicted value.
Fig. 9 is a flowchart illustrating a training method of a relocation model according to a first embodiment of the present disclosure.
As shown in fig. 9, the method for training a relocation model according to the embodiment of the present disclosure may specifically include the following steps:
s901, obtaining samples corresponding to omics data of the same drug molecule and different cell lines and samples corresponding to omics data of different drug molecules and the same cell line from the training set as training samples, wherein the training set further includes true values of response parameters between the drug molecules and the omics data.
Specifically, the execution subject of the training method for a relocation model according to the embodiment of the present disclosure may be a training device for a relocation model provided in the embodiment of the present disclosure, and the training device for a relocation model may be a hardware device having data information processing capability and/or necessary software for driving the hardware device to operate. Alternatively, the execution body may include a workstation, a server, a computer, a user terminal, and other devices.
In the embodiment of the disclosure, omics data of different cell lines and different drug molecules are used to construct a training set, a testing set and a verification set, omics data corresponding to one cell line and one drug molecule are used as a combination to construct a sample, and the response parameters, such as the value of IC50, corresponding to the omics data and the drug molecule are used as sample tags.
In order to make the model focus more on the effect of the same drug under different cell lines and the drug response degree of different drugs on the same cell line, which are more focused in the practical application scene. As shown in fig. 10, in the embodiment of the present disclosure, a cross-sampling manner is used to obtain samples composed of omics data of the same drug molecule and different cell lines and samples composed of omics data of different drug molecules and the same cell line from a training set, and a single training sample is formed by using the two types of samples, so as to obtain multiple training samples used in the model training process.
And S902, training the repositioning model to be trained according to the training samples.
In the embodiment of the present disclosure, the repositioning model to be trained is subjected to multiple rounds of training according to the multiple batches of training samples. The method comprises the step of training model parameters of a multi-head attention layer, a linear layer, a multi-layer sensing layer, a pooling layer and other structures in a repositioning model.
And S903, according to the real value of the response parameter and the predicted value of the response parameter output in the training process, adjusting the model parameter of the repositioning model and continuing to train the adjusted repositioning model by using the next training sample until the trained target repositioning model is obtained.
In the embodiment of the disclosure, the model parameters of the repositioning model are adjusted according to the predicted values of the response parameters output by the model in the training process and the real values of the response parameters in the training set, so that the real values are converged to the predicted values. And continuously using the next training sample after the parameters are adjusted, and training the adjusted repositioning model by using a plurality of training samples in the next round until the trained target repositioning model is obtained. The parameter adjustment can optimize the model parameters based on optimization methods such as a gradient descent method SGD and the like.
In summary, in the training method of the relocation model according to the embodiment of the present disclosure, samples corresponding to omics data of the same drug molecule and different cell lines and samples corresponding to omics data of different drug molecules and the same cell line are obtained from the training set as training samples, where the training set further includes true values of response parameters between the drug molecules and the omics data; training a repositioning model to be trained according to the training sample; and adjusting the model parameters of the repositioning model according to the predicted values of the response parameters and the real values of the response parameters output by training, and continuing to train the adjusted repositioning model by using the next training sample until the trained target repositioning model is obtained. Training samples are obtained based on cross sampling, the training model converges so that the model learns the relationship between the neutron structure and the cytomic data of the drug molecules in training, learns the response degree of the same drug under the condition of different cell lines and the response degree of different drugs on the same cell line, and enhances the accuracy of response parameter prediction between the drug molecules and the cytomic data by training the enhanced model so as to enhance the repositioning accuracy.
Fig. 11 is a flowchart illustrating a training method of a relocation model according to a second embodiment of the present disclosure.
As shown in fig. 11, on the basis of the embodiment shown in fig. 9, the method for training a relocation model according to the embodiment of the present disclosure may specifically include the following steps:
and S1101, obtaining samples corresponding to omics data of the same drug molecule and different cell lines and samples corresponding to omics data of different drug molecules and the same cell line from the training set as training samples, wherein the training set further comprises true values of response parameters between the drug molecules and the omics data.
And S1102, training the repositioning model to be trained according to the training samples.
And S1103, determining a sequencing loss function value and a regression loss function value according to the real value of the response parameter and the predicted value of the response parameter output in the training process.
In the disclosed embodiment, the model is constrained using a rank penalty function and a regression penalty function, as shown in the following equations:
wherein L is rank As a function of rank penalty, y n The true value of the response parameter for the nth sample,is a predicted value of the response parameter of the nth sample, y m The true value of the response parameter for the mth sample,distance (y) being a predicted value of the response parameter of the m-th sample n ,y m ) Denotes y n And y m And margin is a settable offset.
And S1104, determining a loss function value of the model training according to the sequencing loss function value and the regression loss function value.
Based on the preset weight, summing the sequencing loss function value and the regression loss function value to obtain a loss function value of model training, as shown in the following formula:
L=β*L MSE +(1-β)*L rank
wherein beta is a hyper-parameter of the model, L MSE And L is a loss function of model training, and the preset weight can be determined according to the hyperparameter beta.
S1105, adjusting the model parameters of the repositioning model according to the loss function value, so as to continue training the adjusted repositioning model by using the next training sample until the trained target repositioning model is obtained.
The loss function L can be minimized using an SGD or other optimizer to obtain trained model parameters after the final model converges.
Specifically, steps S1101-S1102 are the same as steps S901-S902, and are not repeated here.
For example, a model training set is constructed by a Cancer-related Drug Sensitivity in Cancer (GDSC) database and a Cancer Cell Line Encyclopedia (CCLE), the input of the model is a Cell Line and Drug pair (i.e., a sample consisting of omics data of the Cell Line and Drug molecules), and the output of the model is a predicted value of IC 50. A verification set is constructed by collecting The medication data of a drug cis-platinum (Cisplatin) in a Cancer Genome Atlas (TCGA for short) database in different patients and multigroup chemical data of The patients.
Training is carried out on a training set by using a repositioning model to be trained, and hyperparameters such as beta =0.9 and batch size =256 are set. And (3) performing prediction verification on a verification set by using the trained model, and predicting the drug effect of the cell line and the drug pair (such as Cisplatin and patient transcriptomics pair), namely response parameters. We performed validation using different models, and as a result, as shown in fig. 12, the models involved in validation include Elastic network (Elastic Net), random forest model RF, a hybrid convolutional network deep cdr, and deep learning model TCR trained according to the training method of the present disclosure. The effect-size and the log (P _ value) are both metrics of predicted value of IC50 and true value of IC50, and the larger the value is, the better the value is, and the P _ value corresponds to a hypothesis value or a hypothesis probability in a hypothesis testing method, that is, the probability of occurrence of a sample result under the condition that a given original hypothesis is true. Based on the prediction results shown in fig. 12, it can be concluded that cissplatin is also a good predictor in patients with other diseases than the indicated one, and that cissplatin has the potential to treat this disease.
In summary, in the training method of the relocation model according to the embodiment of the present disclosure, samples corresponding to omics data of the same drug molecule and different cell lines and samples corresponding to omics data of different drug molecules and the same cell line are obtained from the training set as training samples, where the training set further includes true values of response parameters between the drug molecules and the omics data; training a repositioning model to be trained according to the training sample; and adjusting the model parameters of the repositioning model according to the predicted values of the response parameters and the real values of the response parameters output by training, and continuing to train the adjusted repositioning model by using the next training sample until the trained target repositioning model is obtained. Based on cross sampling training data, a sequencing loss function and a regression loss function training model are combined, the relationship between the neutron structure of the drug molecule and the cytomic data is learned, the response degree of the same drug under different cell lines and the drug response degree of different drugs on the same cell line are learned, and the accuracy of response parameter prediction between the drug molecule and the cytomic data is enhanced through training the enhancement model, so that the repositioning accuracy is enhanced.
Fig. 13 is a block diagram of a medication repositioning device according to a first embodiment of the present disclosure.
As shown in fig. 13, a drug repositioning device 1300 of an embodiment of the present disclosure includes: a first acquisition module 1301, a second acquisition module 1302, a third acquisition module 1303 and a prediction module 1304.
A first acquisition module 1301 for acquiring drug molecular and cytomic data.
A second obtaining module 1302 for obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data.
A third obtaining module 1303 for performing multi-point attention processing on the first vector representation and the second vector representation to obtain a prediction vector matrix, wherein the prediction vector matrix is used for characterizing an interaction relationship between the drug molecule and the cytomic data.
A prediction module 1304 for relocating the drug molecule based on the prediction vector matrix and the first vector representation.
It should be noted that the above explanation of the embodiment of the drug repositioning method is also applicable to the drug repositioning device according to the embodiment of the present disclosure, and the detailed process is not repeated here.
In summary, the drug relocation device of the embodiment of the present disclosure obtains drug molecule and cytomic data; obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data; performing multi-headed attention processing on the first vector representation and the second vector representation to obtain a predictive vector matrix, wherein the predictive vector matrix is used to characterize the interaction relationship between the drug molecule and the cytomic data; the drug molecules are repositioned according to the predictive vector matrix and the first vector representation. Based on the multi-head attention mechanism, the relation between the neutron structure of the drug molecule and the cytomics data is captured, the prediction accuracy is improved, and the repositioning accuracy is enhanced.
Fig. 14 is a block diagram of a medication repositioning device according to a second embodiment of the present disclosure.
As shown in fig. 14, a drug repositioning device 1400 according to an embodiment of the present disclosure includes: a first obtaining module 1401, a second obtaining module 1402, a third obtaining module 1403, and a prediction module 1404.
The first obtaining module 1401 has the same structure and function as the first obtaining module 1301 in the previous embodiment, the second obtaining module 1402 has the same structure and function as the second obtaining module 1302 in the previous embodiment, the third obtaining module 1403 has the same structure and function as the third obtaining module 1303 in the previous embodiment, and the predicting module 1404 has the same structure and function as the predicting module 1304 in the previous embodiment.
Further, the second obtaining module 1402 includes: the first construction unit is used for generating a graph corresponding to the drug molecules according to the structural information of the drug molecules, wherein nodes in the graph correspond to atoms in the drug molecules, and the edges correspond to chemical bonds in the drug molecules; the first coding unit is used for coding the graph corresponding to the drug molecules based on the graph neural network so as to obtain vector representation of the nodes as first vector representation.
Further, the second obtaining module 1402 includes: the second construction unit is used for constructing a map corresponding to the cytomic data according to the relationship among the genes in the cytomic data, wherein nodes in the map correspond to the genes, and edges in the map are used for representing the relationship among the genes; a second encoding unit for encoding a map corresponding to the cytomic data based on a map neural network to obtain a vector representation of the cytomic data; the first obtaining unit is used for carrying out multi-layer perception processing on the vector representation of the cytomic data based on the multi-layer perception network so as to obtain a second vector representation of the cytomic data.
Further, the third obtaining module 1403 includes: a first determining unit 14031, configured to determine a query vector matrix corresponding to multi-head attention processing according to the first vector representation; a second determining unit 14032, configured to determine, according to the second vector representation, a key vector matrix and a value vector matrix corresponding to multi-head attention processing; a second obtaining unit 14033, configured to perform multi-head attention processing according to the query vector matrix, the key vector matrix, and the value vector matrix to obtain a prediction vector matrix.
Further, the prediction module 1404 includes: a generating unit for global pooling of the first vector representations of the drug molecules to generate a vector matrix of drug molecules; the third obtaining unit is used for obtaining the predicted value of the response parameter between the drug molecule and the cytomic data according to the vector matrix corresponding to the predicted vector matrix and the drug molecule; and the repositioning unit is used for repositioning the drug molecules according to the predicted value.
In conclusion, the drug relocation device of the embodiment of the present disclosure obtains drug molecule and cytomic data; obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data; performing multi-point attention processing on the first vector representation and the second vector representation to obtain a predictive vector matrix, wherein the predictive vector matrix is used for characterizing the interaction relationship between the drug molecules and the cytological data; the drug molecules are repositioned according to the predictive vector matrix and the first vector representation. The biological function differences of different substructures in the drug molecules are learned based on a multi-head attention mechanism, and the relationship between the substructures in the drug molecules and the cytomic data is captured, so that the action result of the drug molecules on a cell system is better predicted, the prediction accuracy is improved, and the repositioning accuracy is enhanced.
Fig. 15 is a block diagram of a drug repositioning device based on a repositioning model according to a first embodiment of the present disclosure.
As shown in fig. 15, a relocation model-based drug relocation device 1500 of an embodiment of the present disclosure includes: a first acquisition module 1501, a second acquisition module 1502, a third acquisition module 1503, and a prediction module 1504.
The first obtaining module 1501 obtains the drug molecule and the cytomic data, and inputs the drug molecule and the cytomic data into the trained target relocation model.
A second obtaining module 1502 for obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data from the coding network in the target relocation model.
A third obtaining module 1503 for performing multi-head attention processing on the first vector representation and the second vector representation by a multi-head attention network in the target relocation model to obtain a prediction vector matrix, wherein the prediction vector matrix is used for characterizing an interaction relationship between the drug molecule and the cytomic data.
A prediction module 1504 for relocating the drug molecule according to the prediction vector matrix and the first vector representation by the prediction network in the target relocation model.
It should be noted that the above explanation of the embodiment of the drug repositioning method based on the repositioning model is also applicable to the drug repositioning device based on the repositioning model in the embodiment of the present disclosure, and the specific process is not repeated here.
In conclusion, the drug repositioning device based on the repositioning model of the embodiment of the disclosure obtains drug molecule and cytomics data, and inputs the drug molecule and the cytomics data into the trained target repositioning model; obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data from a coding network in the target relocation model; performing multi-head attention processing on the first vector representation and the second vector representation by a multi-head attention network in the target relocation model to obtain a predictive vector matrix, wherein the predictive vector matrix is used for representing the interaction relation between the drug molecules and the cytomics data; drug molecules are relocated by the predictive network in the target relocation model according to the predictive vector matrix and the first vector representation. The method is characterized in that a multi-head attention mechanism is used for modeling drug molecules and cytomics data, the characteristics of the drug molecules are extracted through a graph neural network, the characteristics of the omics data are extracted through the graph neural network and multi-layer perception, the method is applicable to various types of drug molecules, the universality of the model is enhanced, the relation between a neutron structure of the drug molecules and the cytomics data is learned through the multi-head attention mechanism, and the drug molecules are repositioned based on a trained target repositioning model, so that the prediction accuracy is improved, and the repositioning accuracy is enhanced.
Fig. 16 is a block diagram of a drug repositioning device based on a repositioning model according to a second embodiment of the present disclosure.
As shown in fig. 16, the repositioning device 1600 based on the repositioning model in the embodiment of the present disclosure includes: a first obtaining module 1601, a second obtaining module 1602, a third obtaining module 1603, and a prediction module 1604.
The first obtaining module 1601 has the same structure and function as the first obtaining module 1501 in the previous embodiment, the second obtaining module 1602 has the same structure and function as the second obtaining module 1502 in the previous embodiment, the third obtaining module 1603 has the same structure and function as the third obtaining module 1503 in the previous embodiment, and the predicting module 1604 has the same structure and function as the predicting module 1504 in the previous embodiment.
Further, the apparatus 1600 further comprises: the first building module is used for generating a graph corresponding to the drug molecules according to the structural information of the drug molecules, wherein nodes in the graph correspond to atoms in the drug molecules, and edges correspond to chemical bonds in the drug molecules.
Further, the encoding network includes a graph neural network, and the second obtaining module 1602 includes: the first coding unit is used for coding the graph corresponding to the drug molecules based on the graph neural network so as to obtain vector representation of the nodes as first vector representation.
Further, the apparatus 1600 further comprises: and the second construction module is used for constructing a corresponding map of the cytomic data according to the relation between the genes in the cytomic data, wherein the nodes in the map correspond to the genes, and the edges in the map are used for representing the relation between the genes.
Further, the encoding network includes a graph neural network and a multi-layer perception network, and the second obtaining module 1602 includes: a second encoding unit for encoding a map corresponding to the cytomic data based on a map neural network to obtain a vector representation of the cytomic data; the first obtaining unit is used for carrying out multi-layer perception processing on the vector representation of the cytomic data based on the multi-layer perception network so as to obtain a second vector representation of the cytomic data.
Further, the multi-head attention network includes a plurality of linear layers and a plurality of attention layers, and the third obtaining module 1603 includes: a determining unit 16031, configured to determine, according to the first vector representation, a query vector matrix corresponding to multi-head attention processing, and according to the second vector representation, a key vector matrix and a value vector matrix corresponding to multi-head attention processing, by the plurality of linear layers, respectively; a second obtaining unit 16032, configured to perform multi-head attention processing by multiple attention layers according to the query vector matrix, the key vector matrix, and the value vector matrix to obtain a prediction vector matrix.
Further, the prediction network includes a pooling layer, a splicing layer, and a multi-layer sensing layer, and the prediction module 1604 includes: a first generating unit for global pooling by the pooling layer of the first vector representation of the drug molecules to generate a vector matrix of the drug molecules; the second generation unit is used for stacking the prediction vector matrix and the vector matrix of the drug molecules by the splicing layer to generate a matrix to be processed; and the third acquisition unit is used for acquiring a predicted value of a response parameter between the drug molecule and the cytomic data by the multilayer sensing layer according to the matrix to be processed and relocating the drug molecule according to the predicted value.
The drug relocation device based on the relocation model of the embodiment of the disclosure obtains drug molecule and cytomic data, and inputs the drug molecule and the cytomic data into the trained target relocation model; obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data from a coding network in the target relocation model; performing multi-head attention processing on the first vector representation and the second vector representation by a multi-head attention network in the target relocation model to obtain a prediction vector matrix, wherein the prediction vector matrix is used for representing the interaction relation between the drug molecules and the cytomics data; drug molecules are relocated by the predictive network in the target relocation model according to the predictive vector matrix and the first vector representation. The method is characterized in that a multi-head attention mechanism is used for modeling drug molecules and cytomics data, the characteristics of the drug molecules are extracted through a graph neural network, the characteristics of the omics data are extracted through the graph neural network and multi-layer perception, the method is applicable to various types of drug molecules, the universality of the model is enhanced, the biological function differences of different substructures in the drug molecules and the relation between the substructures in the drug molecules and the cytomics data are learned through the multi-head attention mechanism, therefore, the action result of the drug molecules on a cell system is better predicted, the drug molecule repositioning is performed based on a trained target repositioning model, the prediction accuracy is improved, and the repositioning accuracy is enhanced.
Fig. 17 is a block diagram of a training apparatus of a relocation model according to a first embodiment of the present disclosure.
As shown in fig. 17, a training apparatus 1700 for a relocation model according to an embodiment of the present disclosure includes: an acquisition module 1701, a training module 1702, and an update module 1703.
An obtaining module 1701 is configured to obtain, from the training set, samples corresponding to omics data of the same drug molecule and different cell lines and samples corresponding to omics data of different drug molecules and the same cell line as training samples, where the training set further includes a true value of a response parameter between the drug molecule and the omics data.
A training module 1702, configured to train the relocation model to be trained according to the training sample.
And an updating module 1703, configured to adjust model parameters of the repositioning model according to the real values of the response parameters and the predicted values of the response parameters output in the training process, and continue to train the adjusted repositioning model using the next training sample until a trained target repositioning model is obtained.
It should be noted that the explanation of the above embodiment of the method for training a relocation model is also applicable to the device for training a relocation model in the embodiment of the present disclosure, and the specific process is not described herein again.
In summary, the training device of the repositioning model according to the embodiment of the present disclosure obtains samples corresponding to omics data of the same drug molecule and different cell lines from the training set, and samples corresponding to omics data of different drug molecules and the same cell line as training samples, where the training set further includes a true value of a response parameter between the drug molecule and the omics data; training the repositioning model to be trained according to the training sample; and adjusting the model parameters of the repositioning model according to the real values of the response parameters and the predicted values of the response parameters output in the training process, and continuing to train the adjusted repositioning model by using the next training sample until the trained target repositioning model is obtained. And obtaining a training sample based on cross sampling, and training the model to converge so that the model learns the relationship between the neutron structure of the drug molecule and the cytomic data in training, learns the response degree of the same drug under the condition of different cell lines and the response degree of different drugs on the same cell line, and enhances the accuracy of response parameter prediction between the drug molecule and the cytomic data by training the enhanced model so as to enhance the repositioning accuracy.
Fig. 18 is a block diagram of a training apparatus of a relocation model according to a second embodiment of the present disclosure.
As shown in fig. 18, the training apparatus 1800 for repositioning the model according to the embodiment of the present disclosure includes: an acquisition module 1801, a training module 1802, and an update module 1803.
The obtaining module 1801 has the same structure and function as the obtaining module 1701 in the previous embodiment, the training module 1802 has the same structure and function as the training module 1702 in the previous embodiment, and the updating module 1803 has the same structure and function as the updating module 1703 in the previous embodiment.
Further, the updating module 1803 includes: a first determining unit 18031, configured to determine a ranking loss function value and a regression loss function value according to a true value of the response parameter and a predicted value of the response parameter output in the training process; a second determining unit 18032, configured to determine a loss function value of the model training according to the sorted loss function values and the regression loss function values; an adjusting unit 18033, configured to adjust model parameters of the relocation model according to the loss function.
Further, the second determination unit includes: and the determining subunit is used for summing the sequencing loss function and the regression loss function based on the preset weight to obtain a loss function of the model training.
In summary, the training device of the repositioning model according to the embodiment of the present disclosure obtains samples corresponding to omics data of the same drug molecule and different cell lines from the training set, and samples corresponding to omics data of different drug molecules and the same cell line as training samples, where the training set further includes a true value of a response parameter between the drug molecule and the omics data; training a repositioning model to be trained according to the training sample; and adjusting the model parameters of the repositioning model according to the real values of the response parameters and the predicted values of the response parameters output in the training process, and continuing to train the adjusted repositioning model by using the next training sample until the trained target repositioning model is obtained. Based on cross sampling training data, a sequencing loss function and a regression loss function training model are combined, the relationship between the neutron structure of the drug molecule and the cytomic data is learned, the response degree of the same drug under different cell lines and the drug response degree of different drugs on the same cell line are learned, and the accuracy of response parameter prediction between the drug molecule and the cytomic data is enhanced through training the enhancement model, so that the repositioning accuracy is enhanced.
In the technical scheme of the disclosure, the collection, storage, use, processing, transmission, provision, disclosure and other processing of the personal information of the related user are all in accordance with the regulations of related laws and regulations and do not violate the good customs of the public order.
The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.
FIG. 19 shows a schematic block diagram of an example electronic device 1900 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.
As shown in fig. 19, the electronic apparatus 1900 includes a computing unit 1901, which can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 1902 or a computer program loaded from a storage unit 1908 into a Random Access Memory (RAM) 1903. In the RAM1903, various programs and data necessary for the operation of the electronic apparatus 1900 can be stored. The calculation unit 1901, ROM1902, and RAM1903 are connected to each other via a bus 1904. An input/output (I/O) interface 1905 is also connected to bus 1904.
A number of components in electronic device 1900 are connected to I/O interface 1905, including: an input unit 1906 such as a keyboard, a mouse, and the like; an output unit 1907 such as various types of displays, speakers, and the like; a storage unit 1908 such as a magnetic disk, optical disk, or the like; and a communication unit 1909 such as a network card, modem, wireless communication transceiver, and the like. The communication unit 1909 allows the electronic device 1900 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.
The computing unit 1901 may be a variety of general purpose and/or special purpose processing components with processing and computing capabilities. Some examples of the computation unit 1901 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computation chips, various computation units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The calculation unit 1901 performs the various methods and processes described above, such as the drug relocation method shown in fig. 1-5, or the drug relocation method based on a relocation model shown in fig. 6-8, or the training method of a relocation model shown in fig. 9-12. For example, in some embodiments, the above-described methods may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage unit 1908. In some embodiments, part or all of the computer program can be loaded and/or installed onto the electronic device 1900 via the ROM1902 and/or the communication unit 1909. When the computer program is loaded into the RAM1903 and executed by the computing unit 1901, one or more steps of the semantic parsing method described above may be performed. Alternatively, in other embodiments, the computing unit 1901 may be configured to perform the speech processing method by any other suitable means (e.g., by means of firmware).
Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), system on a chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.
Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program code, when executed by the processor or controller, causes the functions/acts specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.
In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic, speech, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), the internet, and blockchain networks.
The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The Server can be a cloud Server, also called a cloud computing Server or a cloud host, and is a host product in a cloud computing service system, so as to solve the defects of high management difficulty and weak service expansibility in the traditional physical host and VPS service ("Virtual Private Server", or simply "VPS"). The server may also be a server of a distributed system, or a server incorporating a blockchain.
According to an embodiment of the present disclosure, there is also provided a computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements the steps of the speech processing method according to the above-mentioned embodiment of the present disclosure.
It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present disclosure may be executed in parallel or sequentially or in different orders, and are not limited herein as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved.
The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the scope of protection of the present disclosure.
Claims (30)
1. A method of drug relocation comprising:
obtaining drug molecule and cytomic data;
obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data;
performing multi-point attention processing on the first vector representation and the second vector representation to obtain a predictive vector matrix, wherein the predictive vector matrix is used to characterize the interaction relationship between the drug molecule and the cytological data;
repositioning the drug molecule according to the predictive vector matrix and the first vector representation.
2. The relocation method of claim 1, wherein the obtaining of the first vector representation comprises:
generating a graph corresponding to the drug molecules according to the structural information of the drug molecules, wherein the nodes in the graph correspond to atoms in the drug molecules, and the edges correspond to chemical bonds in the drug molecules;
encoding the graph corresponding to the drug molecule based on a graph neural network to obtain a vector representation of the node as the first vector representation.
3. The relocation method of claim 1, wherein the obtaining of the second vector representation comprises:
constructing a map corresponding to the cytomic data according to the relation between genes in the cytomic data, wherein the nodes in the map correspond to the genes, and the edges in the map are used for representing the relation between the genes;
encoding a map corresponding to the cytomic data based on a map neural network to obtain a vector representation of the cytomic data;
performing multi-tier perception processing on the vector representation of the cytomic data based on a multi-tier perception network to obtain the second vector representation of the cytomic data.
4. The relocation method of any one of claims 1-3, wherein the multi-head attention processing the first vector representation and the second vector representation to obtain a prediction vector matrix comprises:
determining a query vector matrix corresponding to multi-head attention processing according to the first vector representation;
determining a key vector matrix and a value vector matrix corresponding to multi-head attention processing according to the second vector representation;
and performing multi-head attention processing according to the query vector matrix, the key vector matrix and the value vector matrix to obtain the prediction vector matrix.
5. The method of repositioning according to claim 1 wherein said repositioning the drug molecule according to the predictive vector matrix and the first vector representation comprises:
globally pooling the first vector representation of the drug molecules to generate a vector matrix of the drug molecules;
obtaining a predicted value of a response parameter between the drug molecule and the omics data according to the vector matrix corresponding to the drug molecule and the prediction vector matrix;
and repositioning the drug molecules according to the predicted value.
6. A method of drug relocation based on a relocation model comprising:
acquiring drug molecules and cytomic data, and inputting the drug molecules and the cytomic data into a trained target relocation model;
obtaining, from a coding network in the target relocation model, a first vector representation of the drug molecule and a second vector representation of the cytomic data;
performing multi-headed attention processing on the first vector representation and the second vector representation by a multi-headed attention network in the target relocation model to obtain a predictive vector matrix, wherein the predictive vector matrix is used to characterize the interaction relationship between the drug molecules and the omics data;
relocating, by a predictive network in the target relocation model, the drug molecule according to the predictive vector matrix and the first vector representation.
7. The relocation method of claim 6, wherein, prior to obtaining the first vector representation of the drug molecule from the coding network in the target relocation model, further comprising:
and generating a graph corresponding to the drug molecules according to the structural information of the drug molecules, wherein nodes in the graph correspond to atoms in the drug molecules, and edges correspond to chemical bonds in the drug molecules.
8. The relocation method of claim 7, wherein the encoding network comprises a graph neural network, the first vector representing an acquisition process comprising:
encoding a graph corresponding to the drug molecule based on the graph neural network to obtain a vector representation of the node as the first vector representation.
9. The relocation method of claim 6, wherein prior to obtaining the second vector representation of the cytomic data from the encoded network in the target relocation model, further comprising:
and constructing a map corresponding to the cytomic data according to the relation between the genes in the cytomic data, wherein the nodes in the map correspond to the genes, and the edges in the map are used for representing the relation between the genes.
10. The relocation method of claim 9, wherein the coding network includes a neural network and a multi-layer perceptual network, and the obtaining of the second vector representation includes:
encoding a map corresponding to the cytomic data based on a map neural network to obtain a vector representation of the cytomic data;
performing multi-tier perception processing on the vector representation of the cytomic data based on a multi-tier perception network to obtain the second vector representation of the cytomic data.
11. The relocation method of claim 6, wherein the multi-head attention network includes a plurality of linear layers and a plurality of attention layers, the multi-head attention processing, by the multi-head attention network in the target relocation model, the first vector representation and the second vector representation to obtain a prediction vector matrix, comprising:
determining a query vector matrix corresponding to multi-head attention processing according to the first vector representation and determining a key vector matrix and a value vector matrix corresponding to multi-head attention processing according to the second vector representation by a plurality of the linear layers respectively;
and performing multi-head attention processing by the plurality of attention layers according to the query vector matrix, the key vector matrix and the value vector matrix to obtain the prediction vector matrix.
12. The relocation method of claim 6, wherein the prediction network includes a pooling layer, a stitching layer and a multi-layer perception layer, the relocation of the drug molecule by the prediction network in the target relocation model according to the prediction vector matrix and the first vector representation includes:
globally pooling, by the pooling layer, the first vector representation of the drug molecule to generate a vector matrix of the drug molecule;
stacking the prediction vector matrix and the vector matrix of the drug molecules by the splicing layer to generate a matrix to be processed;
and obtaining a predicted value of a response parameter between the drug molecule and the omics data by the multilayer perception layer according to the matrix to be processed, and repositioning the drug molecule according to the predicted value.
13. A method of training a repositioning model, comprising:
obtaining samples corresponding to omics data of the same drug molecule and different cell lines and samples corresponding to omics data of different drug molecules and the same cell line from a training set as training samples, wherein the training set also comprises real values of response parameters between the drug molecules and the omics data;
training a repositioning model to be trained according to the training sample;
and adjusting the model parameters of the repositioning model according to the real values of the response parameters and the predicted values of the response parameters output in the training process, and continuing to train the adjusted repositioning model by using the next training sample until a trained target repositioning model is obtained.
14. The method for training a repositioning model according to claim 13, wherein the adjusting the model parameters of the repositioning model according to the real values of the response parameters and the predicted values of the response parameters output in the training process comprises:
determining a sequencing loss function value and a regression loss function value according to the real value of the response parameter and the predicted value of the response parameter output in the training process;
determining a loss function value of the model training according to the sequencing loss function value and the regression loss function value;
and adjusting the model parameters of the repositioning model according to the loss function values.
15. The training method of claim 14, wherein said determining a loss function value for said model training based on said rank order loss function values and said regression loss function values comprises:
and summing the sequencing loss function value and the regression loss function value based on a preset weight to obtain a loss function value of the model training.
16. A medication repositioning device comprising:
the first acquisition module is used for acquiring drug molecule and cytomic data;
a second acquisition module for acquiring a first vector representation of the drug molecule and a second vector representation of the cytomic data;
a third obtaining module, configured to perform multi-point attention processing on the first vector representation and the second vector representation to obtain a predictive vector matrix, wherein the predictive vector matrix is used to characterize an interaction relationship between the drug molecule and the omics data;
a prediction module for relocating the drug molecules according to the prediction vector matrix and the first vector representation.
17. The relocating device as claimed in claim 16 wherein the second obtaining means includes:
the first construction unit is used for generating a graph corresponding to the drug molecules according to the structural information of the drug molecules, wherein the nodes in the graph correspond to atoms in the drug molecules, and the edges correspond to chemical bonds in the drug molecules;
a first encoding unit, configured to encode the graph corresponding to the drug molecule based on a graph neural network to obtain a vector representation of the node as the first vector representation.
18. The relocating device as claimed in claim 16 wherein the second obtaining means includes:
a second construction unit, configured to construct a map corresponding to the cytomic data according to the relationship between the genes in the cytomic data, wherein the nodes in the map correspond to the genes, and the edges in the map are used for characterizing the relationship between the genes;
a second encoding unit, configured to encode a map corresponding to the cytomic data based on a map neural network to obtain a vector representation of the cytomic data;
a first obtaining unit, configured to perform multi-layer sensing processing on the vector representation of the cytomic data based on a multi-layer sensing network to obtain the second vector representation of the cytomic data.
19. The relocating device as claimed in any one of claims 16 to 18 wherein the third obtaining means includes:
the first determining unit is used for determining a query vector matrix corresponding to multi-head attention processing according to the first vector representation;
a second determining unit, configured to determine, according to the second vector representation, a key vector matrix and a value vector matrix corresponding to multi-head attention processing;
a second obtaining unit, configured to perform multi-head attention processing according to the query vector matrix, the key vector matrix, and the value vector matrix, so as to obtain the predictive vector matrix.
20. A drug repositioning device based on a repositioning model, comprising:
the first acquisition module is used for acquiring drug molecules and cytomics data and inputting the drug molecules and the cytomics data into a trained target relocation model;
a second obtaining module for obtaining a first vector representation of the drug molecule and a second vector representation of the cytomic data from a coding network in the target relocation model;
a third obtaining module, configured to perform multi-head attention processing on the first vector representation and the second vector representation by a multi-head attention network in the target relocation model to obtain a prediction vector matrix, wherein the prediction vector matrix is used for characterizing an interaction relationship between the drug molecule and the omics data;
and the prediction module is used for performing relocation on the drug molecules according to the prediction vector matrix and the first vector representation by a prediction network in the target relocation model.
21. The relocating device as claimed in claim 20 wherein the device further comprises:
the first building module is used for generating a graph corresponding to the drug molecules according to the structural information of the drug molecules, wherein nodes in the graph correspond to atoms in the drug molecules, and edges of the nodes in the graph correspond to chemical bonds in the drug molecules.
22. The relocating device as claimed in claim 21 wherein the encoding network comprises a graph neural network and the second obtaining module comprises:
a first encoding unit, configured to encode the graph corresponding to the drug molecule based on the graph neural network to obtain a vector representation of the node as the first vector representation.
23. The relocating device as claimed in claim 20 wherein the device further comprises:
a second construction module for constructing a map corresponding to the cytomic data according to relationships between genes in the cytomic data, wherein nodes in the map correspond to the genes, and edges in the map are used for characterizing the relationships between the genes.
24. The relocating device as claimed in claim 23 wherein the encoded network includes a graph neural network and a multi-layer aware network, the second obtaining module includes:
a second encoding unit, configured to encode a map corresponding to the cytomic data based on a map neural network to obtain a vector representation of the cytomic data;
a first obtaining unit configured to perform multi-tier awareness processing on the vector representation of the cytomic data based on a multi-tier awareness network to obtain the second vector representation of the cytomic data.
25. The relocating device as claimed in claim 20 wherein the multi-head attention network includes a plurality of linear layers and a plurality of attention layers, the third obtaining module including:
a determining unit, configured to determine, by the plurality of linear layers, a query vector matrix corresponding to multi-head attention processing according to the first vector representation, and determine a key vector matrix and a value vector matrix corresponding to multi-head attention processing according to the second vector representation, respectively;
a second obtaining unit, configured to perform multi-head attention processing by the plurality of attention layers according to the query vector matrix, the key vector matrix, and the value vector matrix to obtain the prediction vector matrix.
26. A training apparatus for a repositioning model, comprising:
the acquisition module is used for acquiring samples corresponding to omics data of the same drug molecule and different cell lines and samples corresponding to omics data of different drug molecules and the same cell line from a training set as training samples, wherein the training set also comprises real values of response parameters between the drug molecules and the omics data;
the training module is used for training the repositioning model to be trained according to the training sample;
and the updating module is used for adjusting the model parameters of the repositioning model according to the predicted values of the response parameters and the real values of the response parameters, which are output by training, and continuing to train the adjusted repositioning model by using the next training sample until the trained target repositioning model is obtained.
27. The training apparatus of a repositioning model according to claim 26, wherein the updating module comprises:
the first determining unit is used for determining a sequencing loss function value and a regression loss function value according to the real value of the response parameter and the predicted value of the response parameter output in the training process;
a second determining unit, configured to determine a loss function value of the model training according to the sorting loss function value and the regression loss function value;
and the adjusting unit is used for adjusting the model parameters of the repositioning model according to the loss function values.
28. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-15.
29. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1-15.
30. A computer program product comprising a computer program which, when being executed by a processor, carries out the steps of the method according to any one of claims 1-15.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210848003.5A CN115206421B (en) | 2022-07-19 | 2022-07-19 | Drug repositioning method, and repositioning model training method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210848003.5A CN115206421B (en) | 2022-07-19 | 2022-07-19 | Drug repositioning method, and repositioning model training method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115206421A true CN115206421A (en) | 2022-10-18 |
CN115206421B CN115206421B (en) | 2023-04-18 |
Family
ID=83582045
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210848003.5A Active CN115206421B (en) | 2022-07-19 | 2022-07-19 | Drug repositioning method, and repositioning model training method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115206421B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117012403A (en) * | 2023-09-27 | 2023-11-07 | 武汉纺织大学 | Medicine repositioning method and system based on bilinear attention network |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110232978A (en) * | 2019-06-14 | 2019-09-13 | 西安电子科技大学 | Cancer cell system therapeutic agent prediction technique based on multidimensional network |
CN112927765A (en) * | 2021-03-29 | 2021-06-08 | 天士力国际基因网络药物创新中心有限公司 | Method for repositioning medicine |
CN113707341A (en) * | 2021-08-27 | 2021-11-26 | 上海商汤智能科技有限公司 | Training method of prediction model of drug and cell line reaction and related device |
CN113782089A (en) * | 2021-11-15 | 2021-12-10 | 浙江大学 | Drug sensitivity prediction method and device based on multigroup chemical data fusion |
CN114067905A (en) * | 2021-11-08 | 2022-02-18 | 大连大学 | Drug-target interaction prediction method fusing multilayer drug structure information |
CN114420309A (en) * | 2021-09-13 | 2022-04-29 | 北京百度网讯科技有限公司 | Method for establishing drug synergy prediction model, prediction method and corresponding device |
CN114496303A (en) * | 2022-01-06 | 2022-05-13 | 湖南大学 | Anticancer drug screening method based on multichannel neural network |
CN114530258A (en) * | 2022-01-28 | 2022-05-24 | 华南理工大学 | Deep learning drug interaction prediction method, device, medium and equipment |
-
2022
- 2022-07-19 CN CN202210848003.5A patent/CN115206421B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110232978A (en) * | 2019-06-14 | 2019-09-13 | 西安电子科技大学 | Cancer cell system therapeutic agent prediction technique based on multidimensional network |
CN112927765A (en) * | 2021-03-29 | 2021-06-08 | 天士力国际基因网络药物创新中心有限公司 | Method for repositioning medicine |
CN113707341A (en) * | 2021-08-27 | 2021-11-26 | 上海商汤智能科技有限公司 | Training method of prediction model of drug and cell line reaction and related device |
CN114420309A (en) * | 2021-09-13 | 2022-04-29 | 北京百度网讯科技有限公司 | Method for establishing drug synergy prediction model, prediction method and corresponding device |
CN114067905A (en) * | 2021-11-08 | 2022-02-18 | 大连大学 | Drug-target interaction prediction method fusing multilayer drug structure information |
CN113782089A (en) * | 2021-11-15 | 2021-12-10 | 浙江大学 | Drug sensitivity prediction method and device based on multigroup chemical data fusion |
CN114496303A (en) * | 2022-01-06 | 2022-05-13 | 湖南大学 | Anticancer drug screening method based on multichannel neural network |
CN114530258A (en) * | 2022-01-28 | 2022-05-24 | 华南理工大学 | Deep learning drug interaction prediction method, device, medium and equipment |
Non-Patent Citations (2)
Title |
---|
张永祥等: "药物重定位――网络药理学的重要应用领域", 《中国药理学与毒理学杂志》 * |
李叙潼: "基于深度学习的抗肿瘤药物作用机制研究", 《中国博士学位论文全文数据库医药卫生科技辑》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117012403A (en) * | 2023-09-27 | 2023-11-07 | 武汉纺织大学 | Medicine repositioning method and system based on bilinear attention network |
Also Published As
Publication number | Publication date |
---|---|
CN115206421B (en) | 2023-04-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Ma et al. | Using deep learning to model the hierarchical structure and function of a cell | |
US11651860B2 (en) | Drug efficacy prediction for treatment of genetic disease | |
CN113535984B (en) | Knowledge graph relation prediction method and device based on attention mechanism | |
CN109376222B (en) | Question-answer matching degree calculation method, question-answer automatic matching method and device | |
JP2023082017A (en) | computer system | |
Cai et al. | An CNN-LSTM attention approach to understanding user query intent from online health communities | |
JP7262539B2 (en) | Conversation recommendation method, device and equipment | |
CN114048331A (en) | Knowledge graph recommendation method and system based on improved KGAT model | |
CN112149414B (en) | Text similarity determination method, device, equipment and storage medium | |
US11651841B2 (en) | Drug compound identification for target tissue cells | |
CN111191002A (en) | Neural code searching method and device based on hierarchical embedding | |
CN114564593A (en) | Completion method and device of multi-mode knowledge graph and electronic equipment | |
EP3862891A2 (en) | Method and apparatus for retrieving multi-turn dialogue, storage medium, and electronic device | |
CN113707299A (en) | Auxiliary diagnosis method and device based on inquiry session and computer equipment | |
Chitty-Venkata et al. | Neural architecture search benchmarks: Insights and survey | |
CN115206421B (en) | Drug repositioning method, and repositioning model training method and device | |
CN111160049B (en) | Text translation method, apparatus, machine translation system, and storage medium | |
CN115374771A (en) | Text label determination method and device | |
CN109597881B (en) | Matching degree determination method, device, equipment and medium | |
CN108475346A (en) | Neural random access machine | |
WO2023116572A1 (en) | Word or sentence generation method and related device | |
CN115905518B (en) | Emotion classification method, device, equipment and storage medium based on knowledge graph | |
CN115714002B (en) | Training method for depression risk detection model, depression symptom early warning method and related equipment | |
CN116109449A (en) | Data processing method and related equipment | |
CN115795025A (en) | Abstract generation method and related equipment thereof |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |