CN117976245A - Asymmetric drug interaction prediction method, system and storage medium - Google Patents

Asymmetric drug interaction prediction method, system and storage medium Download PDF

Info

Publication number
CN117976245A
CN117976245A CN202410391016.3A CN202410391016A CN117976245A CN 117976245 A CN117976245 A CN 117976245A CN 202410391016 A CN202410391016 A CN 202410391016A CN 117976245 A CN117976245 A CN 117976245A
Authority
CN
China
Prior art keywords
relationship
drug
target
role
asymmetric
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202410391016.3A
Other languages
Chinese (zh)
Inventor
张德海
王正武
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yunnan University YNU
Original Assignee
Yunnan University YNU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Yunnan University YNU filed Critical Yunnan University YNU
Priority to CN202410391016.3A priority Critical patent/CN117976245A/en
Publication of CN117976245A publication Critical patent/CN117976245A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H70/00—ICT specially adapted for the handling or processing of medical references
    • G16H70/40—ICT specially adapted for the handling or processing of medical references relating to drugs, e.g. their side effects or intended usage
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • 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
    • G16B40/30—Unsupervised data analysis
    • 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
    • G16B50/00—ICT programming tools or database systems specially adapted for bioinformatics
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/70—Machine learning, data mining or chemometrics
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/90—Programming languages; Computing architectures; Database systems; Data warehousing
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Public Health (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Biology (AREA)
  • Epidemiology (AREA)
  • Biomedical Technology (AREA)
  • Chemical & Material Sciences (AREA)
  • Molecular Biology (AREA)
  • Bioethics (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Computational Linguistics (AREA)
  • Biotechnology (AREA)
  • Mathematical Physics (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Primary Health Care (AREA)
  • Pathology (AREA)
  • Medicinal Chemistry (AREA)
  • Pharmacology & Pharmacy (AREA)
  • Toxicology (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

The invention provides an asymmetric medicine interaction prediction method, an asymmetric medicine interaction prediction system and a storage medium, and relates to the technical field of deep learning. By extracting medicine data information in a medicine database, taking initial characteristics corresponding to medicines described in the medicine data information as medicine nodes, taking interactions between the medicines described in the medicine data information as directed edges to construct a multi-relationship directed graph, determining relationship source role embedding, relationship target role embedding and relationship self-role embedding of medicine node pairs according to the multi-relationship directed graph, determining the invasiveness of the relationship source role embedding, and the vulnerability of the relationship target role embedding, determining asymmetric interaction prediction probability values between the medicine node pairs according to the relationship source role embedding, the relationship target role embedding, the relationship self-role embedding, the invasiveness and the vulnerability of the relationship target role embedding, and determining interaction prediction results of the medicine node pairs according to the magnitude of the asymmetric interaction prediction probability values.

Description

Asymmetric drug interaction prediction method, system and storage medium
Technical Field
The invention relates to the technical field of deep learning, in particular to an asymmetric medicine interaction prediction method, an asymmetric medicine interaction prediction system and a storage medium.
Background
With the increase of complex diseases and the increase of drug resistance of patients, multi-drug combination therapy has become an emerging treatment method. multi-Drug therapy is a good treatment, but when one Drug is taken together with another Drug or drugs, it may cause a change in the activity of one Drug, the effect of the Drug may be increased or decreased, this interaction is known as Drug-Drug interaction (Drug-Drug Interactions, DDIs), and this unexpected DDIs may cause adverse Drug reactions (Adverse Drug Reactions, ADRs) in the patient, thereby damaging the patient's body with the Drug.
To solve this problem, researchers have proposed a calculation method based on deep learning to perform DDI prediction, and currently, DDI prediction methods based on deep learning are roughly classified into two categories: the prediction is based on the molecular structural characteristics of the drug and the prediction is based on the network structure. Methods based on the structural features of drug molecules rely on the assumption that drugs with similar features will have similar DDI, which learns the feature vector of each drug using a graph neural network with the drug molecular structure, or learns the different-sized substructures of the drug molecules to encode their functional feature vectors, and performs training prediction by fusing these drug features with other features (e.g., side effects, targets, proteins, etc.) into pairs of input deep neural network species; while network structure-based methods are more focused on the topology of the network, they construct DDI information as a drug interaction network, where nodes represent drugs and edges represent specific types of interactions. The method generally utilizes a graph neural network to directly aggregate the characteristics of neighbor nodes to learn the characteristic vectors of the medicine nodes, and finally outputs the probability value of whether interaction exists between the nodes.
However, the inventors have discovered in the course of conception and implementation of the present application that: calculation methods based on molecular structural features of drugs typically learn each drug feature only individually and perform interaction modeling only in the final prediction process, ignoring the correlation between drug pairs and the dependence between interactions between specific interaction information, which always handle interactions between each pair of drugs individually; the calculation method based on the network structure often directly utilizes a graph neural network or an embedding method to directly learn the embedding of the drug entity, ignores the inequality among drugs caused by asymmetric interaction relations and the influence of the different interaction relations on the neighborhood information propagation of the drug entity, and leads to incomplete embedded learning information. Thus, a new method of predicting drug interactions is needed to remedy the shortcomings of the above approaches.
The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present invention and is not intended to represent an admission that the foregoing is prior art.
Disclosure of Invention
The invention mainly aims to provide an asymmetric medicine interaction prediction method, which aims to solve the problem of inaccurate medicine interaction prediction.
In order to achieve the above object, the present invention provides an asymmetric drug interaction prediction method, which includes:
extracting medicine data information in a medicine database, taking initial characteristics corresponding to medicines described in the medicine data information as medicine nodes, and taking interactions among the medicines described in the medicine data information as directed edges to construct a multi-relation directed graph;
selecting any two target drug nodes to form a drug node pair, and determining the relation source role embedding, the relation target role embedding and the relation self role embedding of the drug node pair according to the multi-relation directed graph;
Determining the invasiveness of the embedding of the relationship source roles and the vulnerability of the embedding of the relationship target roles;
Determining an asymmetric interaction prediction probability value between the pair of drug nodes according to the relationship source role embedding, the relationship target role embedding, the relationship self role embedding, the invasiveness and the vulnerability;
and determining an interaction prediction result of the drug node pair according to the magnitude of the asymmetric interaction prediction probability value.
Optionally, the step of extracting the drug data information in the drug database, taking the initial feature corresponding to the drug described in the drug data information as a drug node, and taking the interaction between the drugs described in the drug data information as a directed edge to construct a multi-relation directed graph includes:
converting chemical structures of medicines corresponding to all medicine nodes in the multi-relation directed graph into SMILES character strings by using an SMILES representation method;
generating a molecular fingerprint corresponding to the drug node based on the SMILES character string;
determining structural similarity characteristics between the molecular fingerprints of the drug nodes and the molecular fingerprints of other adjacent drug nodes based on the Tanimoto coefficients;
And reducing the structural similarity characteristic to a specified dimension by utilizing principal component analysis to obtain the initial characteristic.
Optionally, the step of determining a relationship source role embedding, a relationship target role embedding and a relationship self role embedding of the drug node pair according to the multi-relationship directed graph and the initial feature includes:
Randomly selecting one target drug node from the drug node pair as a relationship source role, and the other target drug node as a relationship target role, wherein the target drug node as the relationship source role is characterized as an attacker in asymmetric interaction, and the target drug node as the relationship target role is characterized as a victim in asymmetric interaction;
Acquiring a first neighborhood corresponding to the relationship source role and a second neighborhood corresponding to the relationship target role from the multi-relationship directed graph, wherein the first neighborhood represents a first-order outgoing neighbor of the relationship source role, and the second neighborhood represents a first-order incoming neighbor of the relationship target role;
Adopting a first relationship graph attention network, and acquiring a relationship source embedding corresponding to the relationship source role according to the first neighborhood of the relationship source role to aggregate first target neighborhood information in the multi-relationship directed graph; adopting a second relationship graph attention network, and acquiring a relationship target embedding corresponding to the relationship target role according to the second neighborhood of the relationship target role for aggregating second target neighborhood information in the multi-relationship directed graph;
Determining a relationship self-character embedding corresponding to the relationship source character according to the initial characteristics of the relationship target character added in the relationship source character by adopting a first relationship perception network; and determining the embedding of the self-relation roles corresponding to the relation target roles according to the initial characteristics of the additional relation source roles in the relation target roles by adopting a second relation perception network.
Optionally, the step of determining the invasiveness of the embedding of the relationship source character and the vulnerability of the embedding of the relationship target character includes:
Taking the last bit embedded in the relationship source role obtained by aggregation as the invasiveness of the relationship source role; and taking the last bit embedded in the relation target role obtained by aggregation as the vulnerability of the relation target role.
Optionally, the step of determining the asymmetric interaction prediction probability value between the pair of drug nodes based on the relationship source role embedding, the relationship target role embedding, the relationship self role embedding, the invasiveness and the vulnerability comprises:
Embedding the relationship source roles into the relationship self-role corresponding to the relationship source roles to perform inner product calculation, and obtaining a first inner product result;
Embedding the relationship target roles into the relationship self-role corresponding to the relationship target roles to perform inner product calculation, and obtaining a second inner product result;
and carrying out summation calculation on the first inner product result, the second inner product result, the invasiveness of the relation source role and the damage degree of the relation target role to obtain an asymmetric interaction prediction probability value between the medicine pair nodes.
Optionally, the step of determining the interaction prediction result of the drug node pair according to the magnitude of the asymmetric interaction prediction probability value includes:
Acquiring a first target drug node in the drug node pair as a relationship source role, and acquiring a first asymmetric interaction prediction probability value when a second target drug node is a relationship target role; acquiring a second asymmetric interaction prediction probability value obtained when a first target drug node in the drug pair node is used as a relationship target role and a second target drug node is used as a relationship source role;
and determining an interaction prediction result of the drug node pair according to the first asymmetric interaction prediction probability value and the second asymmetric interaction prediction probability value.
Optionally, the step of determining the interaction prediction result of the drug node pair according to the first asymmetric interaction prediction probability value and the second asymmetric interaction prediction probability value comprises:
When the first asymmetric interaction prediction probability value is greater than or equal to a first preset value and the second asymmetric interaction prediction probability value is smaller than the first preset value, determining that asymmetric interaction exists between the drug node pairs, wherein the asymmetric interaction is represented as an attacker by the first target drug node and a victim by the second target drug node;
When the first asymmetric interaction prediction probability value is smaller than a first preset value and the second asymmetric interaction prediction probability value is larger than or equal to the first preset value, determining that asymmetric interaction exists between the drug node pairs, wherein the asymmetric interaction is represented as a victim to the first target drug node and an attacker to the second target drug node;
And when the first asymmetric interaction prediction probability value is smaller than a first preset value and the second asymmetric interaction prediction probability value is smaller than the first preset value, judging that no asymmetric interaction exists between the medicine node pairs.
In addition, to achieve the above object, the present invention also provides a terminal system including: the device comprises a memory, a processor and an asymmetric drug interaction prediction program stored on the memory and capable of running on the processor, wherein the asymmetric drug interaction prediction program realizes the steps of the asymmetric drug interaction prediction method when being executed by the processor.
In addition, in order to achieve the above object, the present invention also provides a computer-readable storage medium including a memory, a processor, and an asymmetric drug interaction prediction program stored on the memory and executable on the processor, the asymmetric drug interaction prediction program implementing the steps of the asymmetric drug interaction prediction method as described above when executed by the processor.
The invention provides an asymmetric medicine interaction prediction method, a system and a computer readable storage medium, which are characterized in that medicine data information in a medicine database is extracted, initial characteristics corresponding to medicines described in the medicine data information are used as medicine nodes, interactions between medicines described in the medicine data information are used as directed edges to construct a multi-relation directed graph, then a medicine node pair is selected between any two target medicine nodes, according to the multi-relation directed graph, relation source role embedding, relation target role embedding and relation self role embedding of the medicine node pair are determined, then the invasiveness of the relation source role embedding and the vulnerability of the relation target role embedding are determined, and then asymmetric interaction prediction probability values between the medicine node pairs are determined according to the relation source role embedding, the relation target role embedding, the relation self role embedding and the invasiveness and the vulnerability, and the interaction prediction result of the medicine node pair is determined according to the size of the asymmetric interaction prediction probability values. According to the asymmetric medicine interaction prediction method, asymmetric interaction information between medicines and entity and asymmetric relation characteristics in the neighborhood of any medicine entity can be captured, and abundant semantic information is extracted by utilizing network structure information of each medicine entity, so that the performance of predicting potential asymmetric interaction of medicine nodes is improved.
The technical proposal of the invention has the beneficial effects that
1. A new method for predicting asymmetric DDI is established, the architecture consisting of two modules, an encoder and a decoder. It can capture asymmetric interaction information between drugs and entities and asymmetric relationship features of any drug entity neighborhood.
2. And providing asymmetric interaction information dominated by the relationship information, and aggregating adjacent node information and asymmetric information by utilizing a relationship graph attention mechanism integrating the relationship information to obtain asymmetric drug feature vectors, namely relationship source node embedding and relationship target role embedding.
3. In order to refine the asymmetric information, a relationship self role is introduced to enhance the asymmetric interaction information, the structural characteristics of the asymmetric interaction information are encoded by taking the similarity characteristics of the medicine structure as initial characteristics, the network structure characteristics of the asymmetric interaction information are also encoded by embedding a relationship sensing network structure, and the effects of different relationships are considered in the process of extracting the network structure information.
4. The relevance of the nodes and other nodes is reflected by the utilization center, and how the number of interaction relations affects the interaction degree of the medicines is reflected by the infringement of the relation source roles and the vulnerability of the relation target roles.
5. The roles of different drugs in different asymmetric interaction relations are different, and the method can capture how the asymmetric interaction drugs are caused or affected according to the relation information, so that the prediction performance of the method is improved.
Drawings
FIG. 1 is a flow chart of a first embodiment of an asymmetric drug interaction prediction method of the present invention;
FIG. 2 is a flow chart of a second embodiment of an asymmetric drug interaction prediction method according to the present invention;
FIG. 3 is a diagram of a predictive model framework of the asymmetric drug interaction prediction method of the present invention;
FIG. 4 is a diagram of a relationship diagram attention network framework in a model framework of the asymmetric drug interaction prediction method of the present invention;
FIG. 5 is a schematic illustration of a relationship diagram attention network learning embedded representation of the present invention;
FIG. 6 is a diagram of a relational awareness network in a model diagram of an asymmetric drug interaction prediction method of the present invention;
FIG. 7 is a flow chart of a third embodiment of an asymmetric drug interaction prediction method according to the present invention;
FIG. 8 is a flow chart of the initial feature calculation of the asymmetric drug interaction prediction method of the present invention;
Fig. 9 is a schematic architecture diagram of a hardware operating environment of an end system according to an embodiment of the present invention.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
In order to better understand the above technical solution, exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
Referring to fig. 1, the present invention proposes a flow chart of a first embodiment of an asymmetric drug interaction prediction method, which comprises the following steps:
Step S10: extracting medicine data information in a medicine database, taking initial characteristics corresponding to medicines described in the medicine data information as medicine nodes, and taking interactions among the medicines described in the medicine data information as directed edges to construct a multi-relation directed graph;
DrugBank is a comprehensive biological database, described in great detail with respect to drugs, drugBank records at least drug molecular level data and drug clinical data, wherein the drug molecular level data includes, but is not limited to, chemical structure of the drug, target sequence, protein identifier, etc.; the drug clinical data includes, but is not limited to, drug-drug interactions, drug-food interactions, side effects, and the like.
In this embodiment, the drug data information in the DrugBank database may be extracted, so that initial features corresponding to the drugs described in the drug data information are used as drug nodes, and interactions between the drugs described in the drug data information are used as directed edges to construct a multi-relationship directed graph.
Step S20: selecting any two target drug nodes to form a drug node pair, and determining the relation source role embedding, the relation target role embedding and the relation self role embedding of the drug node pair according to the multi-relation directed graph;
The asymmetric interactions between pairs of drug nodes are manifested as how one target drug node affects other drug nodes in an asymmetric relationship (target drug nodes as victims) when acting as an attacker, and how the other target drug node is affected by other drug nodes in an asymmetric relationship (target drug nodes as aggressors) when acting as victims.
In this embodiment, the drug node pair is defined by a first target drug nodeAnd a second target drug node/>Composition, drug node pair may be expressed as/>. At least two predictions are required to predict the asymmetric interactions of a pair of drug nodes, the first prediction being aimed at taking a first target drug node as an aggressor and a second target drug node as a victim, and studying how the first target drug node affects the second target drug node as the victim when it is taken as an aggressor; the second prediction aims at taking a second target drug node as an attacker, taking a first target drug node as a victim, researching how the first target drug node is influenced by the second target drug node as the attacker when the first target drug node is taken as the victim, and further determining an interaction prediction result of the drug node pair according to the asymmetric interaction prediction probability value obtained by the two predictions.
Optionally, the invention randomly selects one target drug node from the drug node pair as a relationship source role, and the other target drug node as a relationship target role. A first prediction is performed. And then taking the target drug node which is taken as the relationship target role in the first prediction as the relationship source role in the second prediction, and taking the target drug node which is taken as the relationship source role as the relationship target role. Wherein the target drug node that is the role of the relationship source is characterized as an attacker in the asymmetric interaction and the target drug node that is the role of the relationship target is characterized as a victim in the asymmetric interaction.
The prediction flow steps of the two predictions are completely identical, and only the target drug nodes corresponding to the relationship source roles and the relationship target roles are exchanged.
Optionally, after the relationship source role and the relationship target role are selected, different neighbor propagation information and relationship characteristics of the target drug node are acquired according to different asymmetric relationships among the drug entities recorded in the multi-relationship directed graph, so that the relationship source role embedding and relationship target role embedding corresponding to the two asymmetric relationship source roles of the drug pair node are respectively generated.
Optionally, the relationship source role embedding and the relationship target role embedding are determined, and simultaneously, the relationship self role embedding corresponding to the relationship target role added in the relationship source role embedding is also learned, and the relationship self role embedding corresponding to the relationship source role added in the relationship target role embedding is learned.
Step S30: determining the invasiveness of the embedding of the relationship source roles and the vulnerability of the embedding of the relationship target roles;
In a multi-relational directed graph, a drug has more interactions, indicating that it is more likely to interact with more other drugs, and if the penetration of one drug node is large, this would mean that it attracts the attention of many other drug nodes, and is more likely to be affected. On the other hand, a drug node with a large number of degrees can directly affect many other drug nodes, thus representing that the drug node is easier to affect other drug nodes, in short, the greater the degree of ingress of the node is, the greater the vulnerability of the node is, the greater the degree of egress of the node is, and the greater the invasiveness of the node is. We refer to both cases as invasiveness of the relationship source role and vulnerability of the relationship target role. These two types of information will show how the number of interactions affects the extent of drug interactions.
In this embodiment, the last bit of the relationship source character embedding obtained by aggregation is added to the degree centrality of the relationship source character as the invasiveness of the relationship source character, and the last bit of the relationship target character embedding obtained by aggregation is added to the degree centrality of the relationship target character as the vulnerability of the relationship target character.
Degree of invasiveness of relationship source rolesAnd vulnerability/>, related to target roleThe expression of (2) can be expressed as follows:
Wherein the method comprises the steps of Is the relationship source role/>Is the number of first order outbound neighbor nodes. /(I)Is the relationship target role/>Is the number of first order ingress neighbor nodes. /(I)Is the number of all nodes in the multi-relation directed graph. /(I)Is node/>Is embedded in relation source role representing neighborhood aggregation of its outgoing first-order neighbor nodes,/>Is node/>Is embedded in the relationship target role representing the neighborhood aggregation of its incoming first-order neighbor nodes. /(I)Indicating the last element to be fetched for calculation.
Step S40: determining an asymmetric interaction prediction probability value between the pair of drug nodes according to the relationship source role embedding, the relationship target role embedding, the relationship self role embedding, the invasiveness and the vulnerability;
in this embodiment, drug node pairs are predicted Whether or not there is asymmetric interaction/>The invention relates to source role/>Similarity of relationship source role embedding and relationship self role embedding, relationship target role/>Similarity of relationship target role embedding and relationship self role embedding, relationship source role/>Is the invasiveness and relationship of target roles/>Four kinds of information on vulnerability of (a) together indicate/>Whether or not to pair/>Causing an effect.
Optionally, the invention obtains a first inner product result by performing inner product calculation on the relation self-character embedding corresponding to the relation source character, performs inner product calculation on the relation self-character embedding corresponding to the relation target character, obtains a second inner product result, and finally performs summation calculation on the first inner product result, the second inner product result, the invasiveness of the relation source character and the damage degree of the relation target character, so as to obtain the asymmetric interaction prediction probability value between the medicine pair nodes.
The expression for predicting probability values for asymmetric interactions between drug pair nodes can be expressed as:
Wherein, ，/>，/>And/>Is an alignment matrix because the relationship source (target) character embedding and the relationship self-embedding are not directly computationally similar in the same embedding space, and we need to map them into the corresponding space. /(I)Is node/>Is embedded in the relationship self-character. /(I)Is node/>Is embedded in relation to self-roles,/>Is node/>Is embedded in the relationship source role. /(I)Is node/>Is embedded in the relationship target role. /(I)Is node/>Degree of invasiveness of/(I)Is node/>Is a vulnerability of (2). /(I)Is a Sigmoid function. /(I)，/>Is the similarity of embedded roles between control nodes,/>Is the invasiveness and easy damage degree of the control node.
Step S50: and determining an interaction prediction result of the drug node pair according to the magnitude of the asymmetric interaction prediction probability value.
In this embodiment, the present invention obtains two-time predicted asymmetric interaction prediction probability values, and determines an interaction prediction result of the drug node pair according to the magnitude of the two-time asymmetric interaction prediction probability values. Namely, the invention obtains the first asymmetric interaction prediction probability value obtained when the first target medicine node in the medicine node pair is used as a relation source role and the second target medicine node is used as a relation target role, and obtains the second asymmetric interaction prediction probability value obtained when the first target medicine node in the medicine node pair is used as a relation target role and the second target medicine node is used as a relation source role. And determining an interaction prediction result of the drug node pair according to the first asymmetric interaction prediction probability value and the second asymmetric interaction prediction probability value.
Optionally, the step of determining the interaction prediction result of the drug node pair according to the first asymmetric interaction prediction probability value and the second asymmetric interaction prediction probability value comprises: when the first asymmetric interaction prediction probability value is greater than or equal to a first preset value and the second asymmetric interaction prediction probability value is smaller than the first preset value, determining that asymmetric interaction exists between the drug node pairs, wherein the asymmetric interaction is represented by the first target drug node being an attacker and the second target drug node being a victim; when the first asymmetric interaction prediction probability value is smaller than a first preset value and the second asymmetric interaction prediction probability value is larger than or equal to the first preset value, determining that asymmetric interaction exists between the drug node pairs, wherein the asymmetric interaction is represented by the fact that the first target drug node is a victim and the second target drug node is an attacker; and when the first asymmetric interaction prediction probability value is smaller than a first preset value and the second asymmetric interaction prediction probability value is smaller than the first preset value, judging that no asymmetric interaction exists between the drug node pairs.
In this embodiment, the first preset value may be set to 0.5.
It will be appreciated that when the first preset value is set, there is no bi-directional interaction between the drugs, corresponding to the three cases described above. When (when)And/>Then there is/>. When/>And/>Then there is. When/>And/>Then/>And/>There is no asymmetric interaction between them.
In the technical scheme provided by the embodiment, the initial characteristics corresponding to medicines described in the medicine data information are taken as medicine nodes, interactions between the medicines described in the medicine data information are taken as directed edges to construct a multi-relationship directed graph, then a medicine node pair is selected between any two target medicine nodes, according to the multi-relationship directed graph, the relationship source role embedding, the relationship target role embedding and the relationship self-role embedding of the medicine node pair are determined, the invasiveness of the relationship source role embedding and the vulnerability of the relationship target role embedding are determined, and then the asymmetric interaction prediction probability value between the medicine node pair is determined according to the relationship source role embedding, the relationship target role embedding, the relationship self-role embedding, the invasiveness and the vulnerability of the relationship target role embedding, and the interaction prediction result of the medicine node pair is determined according to the magnitude of the asymmetric interaction prediction probability value. According to the asymmetric medicine interaction prediction method, asymmetric interaction information between medicines and entity and asymmetric relation characteristics in the neighborhood of any medicine entity can be captured, and abundant semantic information is extracted by utilizing network structure information of each medicine entity, so that the performance of predicting potential asymmetric interaction of medicine nodes is improved.
Referring to fig. 2, in a second embodiment, based on the first embodiment, the step S20 includes:
Step S21: randomly selecting one target drug node from the drug node pair as a relationship source role, and the other target drug node as a relationship target role, wherein the target drug node as the relationship source role is characterized as an attacker in asymmetric interaction, and the target drug node as the relationship target role is characterized as a victim in asymmetric interaction;
In the embodiment, during the first prediction, randomly selecting one target medicine node from the medicine node pair as a relationship source role, and the other target medicine node as a relationship target role; and in the second prediction, exchanging the target drug nodes corresponding to the relationship source roles and the relationship target roles, and continuing the prediction.
Step S22: acquiring a first neighborhood corresponding to the relationship source role and a second neighborhood corresponding to the relationship target role from the multi-relationship directed graph, wherein the first neighborhood represents a first-order outgoing neighbor of the relationship source role, and the second neighborhood represents a first-order incoming neighbor of the relationship target role;
The invention builds the prediction model in advance and trains the prediction model until convergence. And then inputting the multi-relation directed graph into a pre-trained prediction model, so that the prediction model can automatically predict the asymmetric interaction prediction probability value of the drug node pair according to the input multi-relation directed graph.
Alternatively, when training the prediction model, a binary cross entropy loss function may be employed as the loss function of the prediction model, and the expression of the binary cross entropy loss function may be expressed as:
Wherein the method comprises the steps of Is a drug node and/>，/>Is the interaction relationship between drugs,/>。/>Is a real tag,/>Indicating that there is asymmetric interaction between nodes,/>Indicating that there is no asymmetric interaction between the nodes. /(I)Is the collection of drug pairs and their interactions. /(I)Is the drug pair node/>And/>The asymmetric interactions between them predict probability values.
Referring to fig. 3, fig. 3 is a diagram showing a prediction model framework of an asymmetric drug interaction prediction method according to the present invention. The predictive model adopts the structural design of Encoder-Decoder model framework, including encoder and Decoder. The encoder consists of two relationship diagram attention networks (Relational Graph Attention Network, RGAT for short) and two relationship Aware networks (Relation-Aware networks, RAN for short). The relationship diagram attention network is composed of N relationship diagram attention machine layers which are connected in series. The relation sensing network is formed by connecting M relation diagram sensing layers in series. Wherein N may be set to 16, m may be set to 1, and the symmetry of the embodiment is not particularly limited. Relationship graph attention network one is used to determine the relationship source role embedment of the relationship source role and the other is used to determine the relationship target role embedment of the relationship target role. The relationship awareness network is used for determining the relationship self-character embedding corresponding to the relationship target character attached to the relationship source character, and is used for determining the relationship self-character embedding corresponding to the relationship source character attached to the relationship target character. The decoder consists of 2 inner product modules and 3 summation modules. One inner product module is used for carrying out inner product calculation on the relation source role embedding and the relation self role embedding corresponding to the relation source role embedding, and the other inner product module is used for carrying out inner product calculation on the relation target role embedding and the relation self role embedding corresponding to the relation target role embedding. And the summation module is used for respectively summing the result of the inner product calculation and the corresponding data to finally obtain the asymmetric interaction prediction probability value of the drug node pair.
In this embodiment, after the multi-relationship directed graph is input to the encoder, the relationship attention network passes through the neighborhood corresponding to the role of the target drug node in the multi-relationship directed graph. The first neighborhood corresponding to the relationship source role and the second neighborhood corresponding to the relationship target role are obtained from the multi-relationship directed graph, so that the embedded representation corresponding to the target medicine node is further learned according to the neighborhood.
For example, assume that in a multi-relational directed graphIn/>And/>Two target drug nodes of a drug node pair, respectively. /(I)Is a collection of medication nodes. /(I)Is a set of interaction relationships. Now give the target drug node/>Neighborhood representation when acting as a relationship source role and as a relationship target role, respectively. For target drug node/>The corresponding neighborhood of the relationship source role as an attacker: /(I)It represents a relationship source role/>To the first order (outbound) outgoing neighbor. Corresponding target drug node/>The neighborhood corresponding to the relationship target role as the victim: It represents a relational target role/> Is an order of first-order (incoming degree) incoming neighbor. For target drug node/>The corresponding neighborhood of the relationship source role as an attacker: /(I)It represents a relationship source roleTo the first order (outbound) outgoing neighbor. Corresponding target drug node/>The neighborhood corresponding to the relationship target role as the victim: It represents a relational target role/> Is an order of first-order (incoming degree) incoming neighbor.
Thus, assume a target drug nodeAs a relationship source role of an attacker, target drug node/>Relationship target roles as victims, and further obtain relationship source roles/>, in a multi-relationship directed graphThe corresponding first neighborhood may be expressed as: ; acquiring relationship target roles/>, from a multi-relationship directed graph The corresponding second neighborhood may be expressed as: /(I)。
Step S23: adopting a first relationship graph attention network, and acquiring a relationship source embedding corresponding to the relationship source role according to the first neighborhood of the relationship source role to aggregate first target neighborhood information in the multi-relationship directed graph; adopting a second relationship graph attention network, and acquiring a relationship target embedding corresponding to the relationship target role according to the second neighborhood of the relationship target role for aggregating second target neighborhood information in the multi-relationship directed graph;
In this embodiment, two relationship diagram attention networks, namely a first relationship diagram attention network and a second relationship diagram attention network, are provided in the encoder, where the first relationship diagram attention network is specifically configured to determine first target neighborhood information to be aggregated in the multi-relationship directed graph according to a first neighborhood of a relationship source role, and the first target neighborhood information includes node information and asymmetric information adjacent to the relationship source role determined in the multi-relationship directed graph according to the first neighborhood. The second relationship graph attention network is specifically configured to determine second target neighborhood information to be aggregated in the multi-relationship directed graph according to a second neighborhood of the relationship target role, where the second target neighborhood information includes node information and asymmetric information adjacent to the domain relationship target role determined in the multi-relationship directed graph according to the second neighborhood.
Optionally, referring to fig. 4, fig. 4 is a diagram of a relationship diagram attention network framework in a model framework of the asymmetric drug interaction prediction method of the present invention. The relationship graph attention network model of the present invention takes into account the relationship between each relationshipFor each node/>Assigning an intermediate representation to effect different relationships to convey different information, and for nodesAnd node/>Each relation/>Attention logic value/>At a given linear transformation, and each relationship/>Is independent and can be shown in relation/>Lower node/>Intermediate representation pair node/>The importance of the intermediate representation of (c). Thus, for node/>Attention coefficient/>Only at node/>Is node/>In relation/>The next neighbor node exists. Others not belonging to the relationship/>The attention of the lower neighbor node will be masked. Attention logic value/>Two schemes are proposed in the graph attention network model, namely additional attention and multiplicative attention. Attention coefficient/>Application of softmax (activation function) to any logical value/>To achieve this, two variants are internal relationship graph attention (WIRGAT) and cross relationship graph attention (ARGAT), respectively, by encoding different prior beliefs on the importance of different relationships. WIRGAT independent calculation of each relationship/>ARGAT consider cross-node neighborhood and not relation/>A node neighborhood below. The method fully considers the importance of the relationship, and can improve the accuracy of asymmetric interaction prediction of the drug node pairs.
Alternatively, according to the attention calculation method of the attention network using the relationship diagram, the attention mechanism pair node can be obtainedFor simplicity we pair/>Neighbor node different from it/>With respect to relationship roles/>Attention coefficient/>The calculation mode of the internal relation diagram attention (WIRGAT) is selected, and the calculation mode of the additional attention is realized. The formula is defined as follows:
Wherein the method comprises the steps of Relationship roles representing nodes,/>Is a relationship source role,/>Is a relational target role. /(I)Is node/>As a relational role/>In relation/>Is described herein). /(I)Is a nodeIs a feature of the initial feature of (a). /(I)Is corresponding to each relation/>Is a matrix of linearly translated learnable parameters.And/>The intermediate representation is mapped to the query core and the key-value core, respectively. /(I)A nonlinear activation function. Attention coefficient/>Aggregation of nodes/>Different neighbor nodes/>, under the same relational rolesAnd different relationship importance information. /(I)Is an exponential function with e low.
Optionally, referring to fig. 5, fig. 5 is a schematic diagram of a relationship diagram attention network learning embedded representation of the present invention. The relation source role embedding and the relation target role embedding are obtained according to the combination mode of the neighborhood aggregation step and the attention mechanism. Relational source role embeddingAnd relational target role embedding/>The expression of (2) is as follows:
Wherein the method comprises the steps of Is node/>Is embedded in relation source role representing neighborhood aggregation of outgoing nodes,/>Is node/>Is embedded in relation to the target role representing the neighborhood aggregation of the incoming node. /(I)And/>Is the corresponding attention coefficient. /(I)Is node/>As relationship source role in relationship/>The next level passes out the neighbor node set. /(I)Is node/>As relationship target role in relationship/>The next-order set of incoming neighbor nodes. /(I)Is node/>Is a feature of the initial feature of (a). /(I)Is in outgoing relation/>And (3) the corresponding learnable weight matrix. /(I)Is in the afferent relation/>And (3) the corresponding learnable weight matrix.
Step S24: determining a relationship self-character embedding corresponding to the relationship source character according to the initial characteristics of the relationship target character added in the relationship source character by adopting a first relationship perception network; and determining the embedding of the self-relation roles corresponding to the relation target roles according to the initial characteristics of the additional relation source roles in the relation target roles by adopting a second relation perception network.
In this embodiment, two relationship-aware networks, namely a first relationship-aware network and a second relationship-aware network, are provided in the encoder, where the first relationship-aware network is specifically configured to learn, according to initial characteristics of a relationship target role attached in a relationship source role, a relationship self-role embedding corresponding to the relationship source role, where the relationship self-role embedding represents a relationship self-role embedding of the relationship target role attached in the relationship source role. The second relationship awareness network is specially used for learning the relationship self-role embedding corresponding to the relationship target role according to the initial characteristics of the relationship source roles attached to the relationship target role, and the relationship self-role embedding table is used for the relationship self-role embedding of the relationship source roles attached to the relationship target role.
Alternatively, referring to fig. 6, fig. 6 is a diagram of a relational awareness network in a model diagram of the asymmetric drug interaction prediction method of the present invention. For asymmetric actionWe use/>Representing nodes/>In relation/>Next-order neighbor and/>Relationship self role embedding/>The expression of (2) can be expressed as:
Wherein the method comprises the steps of Is node/>Is a set of first-order neighbors of a group. /(I)Is node/>Is a feature of the initial feature of (a).Is a parameter matrix of the corresponding relationship. /(I)Is an updatable weight parameter assigned to each node.
In the technical solution provided in this embodiment, one target drug node is randomly selected from a drug node pair as a relationship source role, and the other target drug node is then used as a relationship target role, where the target drug node serving as the relationship source role is characterized as an attacker in asymmetric interaction, the target drug node serving as the relationship target role is characterized as a victim in asymmetric interaction, and then a first neighborhood corresponding to the relationship source role and a second neighborhood corresponding to the relationship target role are obtained in a multi-relationship directed graph, where the first neighborhood represents a first-order outgoing neighbor of the relationship source role, the second neighborhood represents a first-order incoming neighbor of the relationship target role, a first relationship graph attention network is adopted, and the first target neighborhood information in the multi-relationship directed graph is aggregated according to the first neighborhood of the relationship source role, obtaining a relationship source embedding corresponding to a relationship source role, adopting a second relationship graph attention network, obtaining a relationship target embedding corresponding to the relationship target role according to second neighborhood-to-multiple relationship directed graph second target neighborhood information aggregation of the relationship target role, adopting a first relationship perception network, determining a relationship self role embedding corresponding to the relationship source role according to initial characteristics of the relationship target role added in the relationship source role, adopting a second relationship perception network, and determining a relationship self role embedding corresponding to the relationship target role according to initial characteristics of the relationship source role added in the relationship target role, thereby enabling the follow-up evaluation of whether asymmetric interaction exists in a drug node pair according to the relationship source role embedding, the relationship target role embedding and the relationship self role embedding together, and improving the prediction accuracy.
Referring to fig. 7, in a third embodiment, based on any of the above embodiments, the step S10 includes:
Step S11: converting chemical structures of medicines corresponding to all medicine nodes in the multi-relation directed graph into SMILES character strings by using an SMILES representation method;
step S12: generating a molecular fingerprint corresponding to the drug node based on the SMILES character string;
step S13: determining structural similarity characteristics between the molecular fingerprints of the drug nodes and the molecular fingerprints of other adjacent drug nodes based on the Tanimoto coefficients;
In this embodiment, the Structural Similarity Profile (SSP) of the drug pair is calculated as a structural similarity feature.
Step S14: and reducing the structural similarity characteristic to a specified dimension by utilizing principal component analysis to obtain the initial characteristic.
A SMILES (SIMPLIFIED MOLECULAR INPUT LINE ENTRY specification) string is a specification that explicitly describes a molecular structure with an ASCII string. The SMILES string can be interpreted and converted by the molecular editing software into a two-dimensional graphic or a three-dimensional model of the molecule. Molecular fingerprinting is an abstract representation of a molecule that converts (encodes) the molecule into a series of bit strings, typically by extracting structural features of the molecule and then Hashing (Hashing) to generate bit vectors.
In this embodiment, the SMILES string may be hashed to generate a molecular fingerprint.
For example, referring to fig. 8, fig. 8 is a flow chart of an initial feature calculation of an asymmetric drug interaction prediction method of the present invention. Morphine and amphetamine are two drugs, respectively.
In the technical scheme provided by the embodiment, chemical structures of medicines corresponding to all medicine nodes in the multi-relation directed graph are converted into SMILES character strings by using an SMILES representation method, molecular fingerprints corresponding to the medicine nodes are generated based on the SMILES character strings, structural similarity characteristics between the molecular fingerprints of the medicine nodes and molecular fingerprints of other adjacent medicine nodes are determined based on Tanimoto coefficients, the structural similarity characteristics are reduced to specified dimensions by utilizing principal component analysis, initial characteristics are obtained, the medicine structures of the medicine nodes are considered, and the prediction accuracy of asymmetric interaction of the medicine node pairs is improved.
Referring to fig. 9, fig. 9 is a schematic architecture diagram of a hardware running environment of a terminal system according to an embodiment of the present invention.
As shown in fig. 9, the terminal may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, a communication bus 1002. Wherein the communication bus 1002 is used to enable connected communication between these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), a mouse, etc., and the optional user interface 1003 may further include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also optionally be a storage device separate from the processor 1001 described above.
It will be appreciated by those skilled in the art that the terminal structure shown in fig. 9 is not limiting of the terminal and may include more or fewer components than shown, or may combine certain components, or a different arrangement of components.
As shown in fig. 9, an operating system, a network communication module, a user interface module, and an asymmetric drug interaction prediction program may be included in a memory 1005 as one type of computer storage medium.
In the terminal shown in fig. 9, the network interface 1004 is mainly used for connecting to a background server and performing data communication with the background server; the processor 1001 may be configured to invoke the asymmetric drug interaction prediction program stored in the memory 1005 and perform the following operations:
extracting medicine data information in a medicine database, taking initial characteristics corresponding to medicines described in the medicine data information as medicine nodes, and taking interactions among the medicines described in the medicine data information as directed edges to construct a multi-relation directed graph;
selecting any two target drug nodes to form a drug node pair, and determining the relation source role embedding, the relation target role embedding and the relation self role embedding of the drug node pair according to the multi-relation directed graph;
Determining the invasiveness of the embedding of the relationship source roles and the vulnerability of the embedding of the relationship target roles;
Determining an asymmetric interaction prediction probability value between the pair of drug nodes according to the relationship source role embedding, the relationship target role embedding, the relationship self role embedding, the invasiveness and the vulnerability;
and determining an interaction prediction result of the drug node pair according to the magnitude of the asymmetric interaction prediction probability value.
Further, the processor 1001 may call an asymmetric drug interaction prediction program stored in the memory 1005, and further perform the following operations:
converting chemical structures of medicines corresponding to all medicine nodes in the multi-relation directed graph into SMILES character strings by using an SMILES representation method;
generating a molecular fingerprint corresponding to the drug node based on the SMILES character string;
determining structural similarity characteristics between the molecular fingerprints of the drug nodes and the molecular fingerprints of other adjacent drug nodes based on the Tanimoto coefficients;
And reducing the structural similarity characteristic to a specified dimension by utilizing principal component analysis to obtain the initial characteristic.
Further, the processor 1001 may call an asymmetric drug interaction prediction program stored in the memory 1005, and further perform the following operations:
Randomly selecting one target drug node from the drug node pair as a relationship source role, and the other target drug node as a relationship target role, wherein the target drug node as the relationship source role is characterized as an attacker in asymmetric interaction, and the target drug node as the relationship target role is characterized as a victim in asymmetric interaction;
Acquiring a first neighborhood corresponding to the relationship source role and a second neighborhood corresponding to the relationship target role from the multi-relationship directed graph, wherein the first neighborhood represents a first-order outgoing neighbor of the relationship source role, and the second neighborhood represents a first-order incoming neighbor of the relationship target role;
Adopting a first relationship graph attention network, and acquiring a relationship source embedding corresponding to the relationship source role according to the first neighborhood of the relationship source role to aggregate first target neighborhood information in the multi-relationship directed graph; adopting a second relationship graph attention network, and acquiring a relationship target embedding corresponding to the relationship target role according to the second neighborhood of the relationship target role for aggregating second target neighborhood information in the multi-relationship directed graph;
Determining a relationship self-character embedding corresponding to the relationship source character according to the initial characteristics of the relationship target character added in the relationship source character by adopting a first relationship perception network; and determining the embedding of the self-relation roles corresponding to the relation target roles according to the initial characteristics of the additional relation source roles in the relation target roles by adopting a second relation perception network.
Further, the processor 1001 may call an asymmetric drug interaction prediction program stored in the memory 1005, and further perform the following operations:
Taking the last bit embedded in the relationship source role obtained by aggregation as the invasiveness of the relationship source role; and taking the last bit embedded in the relation target role obtained by aggregation as the vulnerability of the relation target role.
Further, the processor 1001 may call an asymmetric drug interaction prediction program stored in the memory 1005, and further perform the following operations:
Embedding the relationship source roles into the relationship self-role corresponding to the relationship source roles to perform inner product calculation, and obtaining a first inner product result;
Embedding the relationship target roles into the relationship self-role corresponding to the relationship target roles to perform inner product calculation, and obtaining a second inner product result;
and carrying out summation calculation on the first inner product result, the second inner product result, the invasiveness of the relation source role and the damage degree of the relation target role to obtain an asymmetric interaction prediction probability value between the medicine pair nodes.
Further, the processor 1001 may call an asymmetric drug interaction prediction program stored in the memory 1005, and further perform the following operations:
Acquiring a first target drug node in the drug node pair as a relationship source role, and acquiring a first asymmetric interaction prediction probability value when a second target drug node is a relationship target role; acquiring a second asymmetric interaction prediction probability value obtained when a first target drug node in the drug pair node is used as a relationship target role and a second target drug node is used as a relationship source role;
and determining an interaction prediction result of the drug node pair according to the first asymmetric interaction prediction probability value and the second asymmetric interaction prediction probability value.
Further, the processor 1001 may call an asymmetric drug interaction prediction program stored in the memory 1005, and further perform the following operations:
When the first asymmetric interaction prediction probability value is greater than or equal to a first preset value and the second asymmetric interaction prediction probability value is smaller than the first preset value, determining that asymmetric interaction exists between the drug node pairs, wherein the asymmetric interaction is represented as an attacker by the first target drug node and a victim by the second target drug node;
When the first asymmetric interaction prediction probability value is smaller than a first preset value and the second asymmetric interaction prediction probability value is larger than or equal to the first preset value, determining that asymmetric interaction exists between the drug node pairs, wherein the asymmetric interaction is represented as a victim to the first target drug node and an attacker to the second target drug node;
And when the first asymmetric interaction prediction probability value is smaller than a first preset value and the second asymmetric interaction prediction probability value is smaller than the first preset value, judging that no asymmetric interaction exists between the medicine node pairs.
In addition, in order to achieve the above object, the present invention further provides a terminal system, including: the system comprises a memory, a processor and an asymmetric drug interaction prediction program stored in the memory and capable of running on the processor, wherein the asymmetric drug interaction prediction program realizes the steps of the control method of the terminal system when being executed by the processor.
In addition, in order to achieve the above object, the present invention also provides a computer-readable storage medium having stored thereon an asymmetric drug interaction prediction program which, when executed by a processor, implements the steps of the control method of the terminal system as described above.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) as described above, comprising instructions for causing a terminal device (which may be a mobile phone, a computer) to perform the method according to the embodiments of the present invention.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.

Claims (9)

1. A method of asymmetric drug interaction prediction, the method comprising the steps of:
extracting medicine data information in a medicine database, taking initial characteristics corresponding to medicines described in the medicine data information as medicine nodes, and taking interactions among the medicines described in the medicine data information as directed edges to construct a multi-relation directed graph;
selecting any two target drug nodes to form a drug node pair, and determining the relation source role embedding, the relation target role embedding and the relation self role embedding of the drug node pair according to the multi-relation directed graph;
Determining the invasiveness of the embedding of the relationship source roles and the vulnerability of the embedding of the relationship target roles;
Determining an asymmetric interaction prediction probability value between the pair of drug nodes according to the relationship source role embedding, the relationship target role embedding, the relationship self role embedding, the invasiveness and the vulnerability;
and determining an interaction prediction result of the drug node pair according to the magnitude of the asymmetric interaction prediction probability value.
2. The method of claim 1, wherein the step of extracting the drug data information in the drug database, using the initial feature corresponding to the drug described in the drug data information as a drug node, and using the interaction between the drugs described in the drug data information as a directed edge to construct the multi-relationship directed graph comprises:
converting chemical structures of medicines corresponding to all medicine nodes in the multi-relation directed graph into SMILES character strings by using an SMILES representation method;
generating a molecular fingerprint corresponding to the drug node based on the SMILES character string;
determining structural similarity characteristics between the molecular fingerprints of the drug nodes and the molecular fingerprints of other adjacent drug nodes based on the Tanimoto coefficients;
And reducing the structural similarity characteristic to a specified dimension by utilizing principal component analysis to obtain the initial characteristic.
3. The method of claim 1, wherein determining the relationship source role embedding, relationship target role embedding, and relationship self role embedding for the drug node pair from the multi-relationship directed graph comprises:
Randomly selecting one target drug node from the drug node pair as a relationship source role, and the other target drug node as a relationship target role, wherein the target drug node as the relationship source role is characterized as an attacker in asymmetric interaction, and the target drug node as the relationship target role is characterized as a victim in asymmetric interaction;
Acquiring a first neighborhood corresponding to the relationship source role and a second neighborhood corresponding to the relationship target role from the multi-relationship directed graph, wherein the first neighborhood represents a first-order outgoing neighbor of the relationship source role, and the second neighborhood represents a first-order incoming neighbor of the relationship target role;
Adopting a first relationship graph attention network, and acquiring a relationship source embedding corresponding to the relationship source role according to the first neighborhood of the relationship source role to aggregate first target neighborhood information in the multi-relationship directed graph; adopting a second relationship graph attention network, and acquiring a relationship target embedding corresponding to the relationship target role according to the second neighborhood of the relationship target role for aggregating second target neighborhood information in the multi-relationship directed graph;
Determining a relationship self-character embedding corresponding to the relationship source character according to the initial characteristics of the relationship target character added in the relationship source character by adopting a first relationship perception network; and determining the embedding of the self-relation roles corresponding to the relation target roles according to the initial characteristics of the additional relation source roles in the relation target roles by adopting a second relation perception network.
4. The method of claim 3, wherein the step of determining the invasiveness of the embedding of the relationship source character and the vulnerability of the embedding of the relationship target character comprises:
Taking the last bit embedded in the relationship source role obtained by aggregation as the invasiveness of the relationship source role; and taking the last bit embedded in the relation target role obtained by aggregation as the vulnerability of the relation target role.
5. The method of claim 1, wherein determining the asymmetric interaction prediction probability value between the pair of drug nodes based on the relationship source role embedding, the relationship target role embedding, the relationship self role embedding, the invasiveness and the vulnerability comprises:
Embedding the relationship source roles into the relationship self-role corresponding to the relationship source roles to perform inner product calculation, and obtaining a first inner product result;
Embedding the relationship target roles into the relationship self-role corresponding to the relationship target roles to perform inner product calculation, and obtaining a second inner product result;
and carrying out summation calculation on the first inner product result, the second inner product result, the invasiveness of the relation source role and the damage degree of the relation target role to obtain an asymmetric interaction prediction probability value between the medicine pair nodes.
6. The method of claim 5, wherein the step of determining the interaction prediction result for the drug node pair based on the magnitude of the asymmetric interaction prediction probability value comprises:
Acquiring a first target drug node in the drug node pair as a relationship source role, and acquiring a first asymmetric interaction prediction probability value when a second target drug node is a relationship target role; acquiring a second asymmetric interaction prediction probability value obtained when a first target drug node in the drug pair node is used as a relationship target role and a second target drug node is used as a relationship source role;
and determining an interaction prediction result of the drug node pair according to the first asymmetric interaction prediction probability value and the second asymmetric interaction prediction probability value.
7. The method of claim 6, wherein determining the interaction prediction result for the drug node pair based on the first asymmetric interaction prediction probability value and the second asymmetric interaction prediction probability value comprises:
When the first asymmetric interaction prediction probability value is greater than or equal to a first preset value and the second asymmetric interaction prediction probability value is smaller than the first preset value, determining that asymmetric interaction exists between the drug node pairs, wherein the asymmetric interaction is represented as an attacker by the first target drug node and a victim by the second target drug node;
When the first asymmetric interaction prediction probability value is smaller than a first preset value and the second asymmetric interaction prediction probability value is larger than or equal to the first preset value, determining that asymmetric interaction exists between the drug node pairs, wherein the asymmetric interaction is represented as a victim to the first target drug node and an attacker to the second target drug node;
And when the first asymmetric interaction prediction probability value is smaller than a first preset value and the second asymmetric interaction prediction probability value is smaller than the first preset value, judging that no asymmetric interaction exists between the medicine node pairs.
8. An end system, the end system comprising: a memory, a processor and an asymmetric drug interaction prediction program stored on the memory and executable on the processor, which asymmetric drug interaction prediction program when executed by the processor implements the steps of the asymmetric drug interaction prediction method of any of claims 1 to 7.
9. A computer-readable storage medium, wherein an asymmetric drug interaction prediction program is stored on the computer-readable storage medium, which when executed by a processor, implements the steps of the asymmetric drug interaction prediction method of any one of claims 1 to 7.
CN202410391016.3A 2024-04-02 2024-04-02 Asymmetric drug interaction prediction method, system and storage medium Pending CN117976245A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410391016.3A CN117976245A (en) 2024-04-02 2024-04-02 Asymmetric drug interaction prediction method, system and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410391016.3A CN117976245A (en) 2024-04-02 2024-04-02 Asymmetric drug interaction prediction method, system and storage medium

Publications (1)

Publication Number Publication Date
CN117976245A true CN117976245A (en) 2024-05-03

Family

ID=90861689

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410391016.3A Pending CN117976245A (en) 2024-04-02 2024-04-02 Asymmetric drug interaction prediction method, system and storage medium

Country Status (1)

Country Link
CN (1) CN117976245A (en)

Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080215402A1 (en) * 2005-09-22 2008-09-04 Pearson Ronald K Methods and Systems for Evaluating Interaction of Medical Products and Dependence on Demographic Variables
JP2011107169A (en) * 2011-03-10 2011-06-02 Keddem Bio-Science Ltd Drug discovery method
CN110223786A (en) * 2019-06-13 2019-09-10 重庆亿创西北工业技术研究院有限公司 Drug-drug interactions prediction technique and system based on non-negative tensor resolution
CN110310703A (en) * 2019-06-25 2019-10-08 中国人民解放军军事科学院军事医学研究院 Prediction technique, device and the computer equipment of drug
CN112070277A (en) * 2020-08-17 2020-12-11 杭州电子科技大学 Hypergraph neural network-based drug-target interaction prediction method
US20210142173A1 (en) * 2019-11-12 2021-05-13 The Cleveland Clinic Foundation Network-based deep learning technology for target identification and drug repurposing
CN112863696A (en) * 2021-04-25 2021-05-28 浙江大学 Drug sensitivity prediction method and device based on transfer learning and graph neural network
EP3859745A1 (en) * 2020-02-03 2021-08-04 National Centre for Scientific Research "Demokritos" System and method for identifying drug-drug interactions
CN113327644A (en) * 2021-04-09 2021-08-31 中山大学 Medicine-target interaction prediction method based on deep embedding learning of graph and sequence
CN114882970A (en) * 2022-06-02 2022-08-09 西安电子科技大学 Drug interaction effect prediction method based on pre-training model and molecular graph
CN116072213A (en) * 2023-03-03 2023-05-05 河南大学 DDI prediction method integrating multi-source information and improved LightGCN
WO2023141345A1 (en) * 2022-01-24 2023-07-27 Kenneth Bean System and method for predictive candidate compound discovery
CN116564555A (en) * 2023-05-16 2023-08-08 郑州大学 Drug interaction prediction model construction method based on deep memory interaction
US20230290435A1 (en) * 2022-03-10 2023-09-14 Wipro Limited Method and system for selecting candidate drug compounds through artificial intelligence (ai)-based drug repurposing
CN117095740A (en) * 2023-08-10 2023-11-21 合肥千手医疗科技有限责任公司 End-to-end neural network model for DTI prediction

Patent Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080215402A1 (en) * 2005-09-22 2008-09-04 Pearson Ronald K Methods and Systems for Evaluating Interaction of Medical Products and Dependence on Demographic Variables
JP2011107169A (en) * 2011-03-10 2011-06-02 Keddem Bio-Science Ltd Drug discovery method
CN110223786A (en) * 2019-06-13 2019-09-10 重庆亿创西北工业技术研究院有限公司 Drug-drug interactions prediction technique and system based on non-negative tensor resolution
CN110310703A (en) * 2019-06-25 2019-10-08 中国人民解放军军事科学院军事医学研究院 Prediction technique, device and the computer equipment of drug
US20210142173A1 (en) * 2019-11-12 2021-05-13 The Cleveland Clinic Foundation Network-based deep learning technology for target identification and drug repurposing
EP3859745A1 (en) * 2020-02-03 2021-08-04 National Centre for Scientific Research "Demokritos" System and method for identifying drug-drug interactions
CN112070277A (en) * 2020-08-17 2020-12-11 杭州电子科技大学 Hypergraph neural network-based drug-target interaction prediction method
CN113327644A (en) * 2021-04-09 2021-08-31 中山大学 Medicine-target interaction prediction method based on deep embedding learning of graph and sequence
CN112863696A (en) * 2021-04-25 2021-05-28 浙江大学 Drug sensitivity prediction method and device based on transfer learning and graph neural network
WO2023141345A1 (en) * 2022-01-24 2023-07-27 Kenneth Bean System and method for predictive candidate compound discovery
US20230290435A1 (en) * 2022-03-10 2023-09-14 Wipro Limited Method and system for selecting candidate drug compounds through artificial intelligence (ai)-based drug repurposing
CN114882970A (en) * 2022-06-02 2022-08-09 西安电子科技大学 Drug interaction effect prediction method based on pre-training model and molecular graph
CN116072213A (en) * 2023-03-03 2023-05-05 河南大学 DDI prediction method integrating multi-source information and improved LightGCN
CN116564555A (en) * 2023-05-16 2023-08-08 郑州大学 Drug interaction prediction model construction method based on deep memory interaction
CN117095740A (en) * 2023-08-10 2023-11-21 合肥千手医疗科技有限责任公司 End-to-end neural network model for DTI prediction

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
FENG YY, ET AL: "Directed graph attention networks for predicting asymmetric drug-drug interactions", BRIEF BIOINFORM, vol. 23, no. 03, 13 May 2022 (2022-05-13), pages 3 - 4 *
何亚峰，等: "基于亲和探针的药物靶点鉴定技术研究进展", 药学进展, vol. 14, no. 01, 25 January 2017 (2017-01-25), pages 178 - 182 *
华阳，等: "注意力特征融合的蛋白质-药物相互作用预测", 计算机研究与发展, vol. 01, no. 09, 31 December 2022 (2022-12-31), pages 1 - 5 *
展鹏，等: ""精准医疗"背景下的分子靶向药物研究――精准药物设计策略浅析", 化学进展, vol. 28, no. 09, 31 December 2016 (2016-12-31), pages 1363 - 1386 *
李淑怡，等: "基于符号图卷积网络的药物互作用关系预测", 现代计算机, vol. 145, no. 16, 5 June 2020 (2020-06-05), pages 1 - 5 *
饶晓洁，等: "基于多层注意力和消息传递网络的药物相互作用预测方法", 自动化学报, vol. 49, no. 12, 31 December 2023 (2023-12-31), pages 2507 - 2519 *
马洁，等: "基于代谢酶靶标介导的他汀类药物相互作用预测查询数据库的设计与开发", 中南药学, vol. 12, no. 05, 20 May 2020 (2020-05-20), pages 12 - 17 *

Similar Documents

Publication Publication Date Title
Vinayagam et al. Applying support vector machines for gene ontology based gene function prediction
WO2023029506A1 (en) Illness state analysis method and apparatus, electronic device, and storage medium
CN113628059B (en) Associated user identification method and device based on multi-layer diagram attention network
Yu Three principles of data science: predictability, computability, and stability (PCS)
CN112201359B (en) Method and device for identifying severe inquiry data based on artificial intelligence
WO2023178971A1 (en) Internet registration method, apparatus and device for seeking medical advice, and storage medium
CN111666477A (en) Data processing method and device, intelligent equipment and medium
Hou et al. Remote homolog detection using local sequence–structure correlations
Melvin et al. SVM-Fold: a tool for discriminative multi-class protein fold and superfamily recognition
CN113223735B (en) Diagnosis method, device, equipment and storage medium based on dialogue characterization
CN116386899A (en) Graph learning-based medicine disease association relation prediction method and related equipment
CN112035611A (en) Target user recommendation method and device, computer equipment and storage medium
Zheng et al. CAMU: Cycle-consistent adversarial mapping model for user alignment across social networks
Dyrka et al. A stochastic context free grammar based framework for analysis of protein sequences
Mahapatra et al. Boosting predictions of Host-Pathogen protein interactions using Deep neural networks
WO2021139220A1 (en) Epidemic monitoring and controlling method and apparatus, computer device, storage medium
CN117457064A (en) Graph structure self-adaption based medicine-medicine interaction prediction method and device
CN110175516B (en) Biological characteristic model generation method, device, server and storage medium
CN117976245A (en) Asymmetric drug interaction prediction method, system and storage medium
CN114724630B (en) Deep learning method for predicting post-translational modification site of protein
CN114121181B (en) Heterogeneous graph neural network traditional Chinese medicine target prediction method based on attention mechanism
Yin et al. Stroke risk prediction: Comparing different sampling algorithms
CN113764035A (en) Knowledge graph-based traditional Chinese medicine compound target prediction method
CN117095741B (en) Graph self-attention-based microorganism-drug association prediction method
WO2020190359A1 (en) System and method for data curation

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