Multi-task hepatitis B drug screening method and system based on knowledge graph assistance

Abstract

The invention relates to the technical field of knowledge maps, in particular to a multi-task hepatitis B drug screening method and system based on knowledge map assistance, and the method comprises the following steps: based on hepatitis B virus genotype sequence data recorded along with time, patient drug use records and drug sensitivity. According to the method, virus genotype sequence data, patient drug use records, drug sensitivity and other multi-dimensional dynamic time sequence data are integrated, interaction events between entities are marked through timestamps, and virus variation tracks, drug curative effect changes and other dynamic characteristics are embedded into atlas node attributes; and the knowledge representation can reflect the time dependence in a real scene. A hyperedge connection multi-entity set is defined based on biological pathway annotation information and protein interaction data, the limitation that only a binary relation is supported in a traditional knowledge graph is expanded, a drug combination and multi-target synergistic effect mechanism is explicitly modeled, and misjudgment of the combination effect caused by simplification of the interaction relation is avoided.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph technology, and in particular to a multi-task hepatitis B drug screening method and system based on knowledge graph assistance. Background Art

[0002] Knowledge graph is a technology that models entities and their relationships with graph structures. It uses semantic networks to perform structured representation and reasoning on multi-source heterogeneous data. Its core areas include entity extraction, relationship mining, graph embedding, and dynamic time series modeling.

[0003] Existing technologies rely on binary edges to describe entity interactions, making it difficult to express synergistic or competitive interactions among multiple entities, such as the complex interaction mechanisms between drug combinations and multiple target proteins. Simplified modeling can easily miss key signals of combinatorial effects. Entity vector representations typically optimize features based on a single task objective and fail to incorporate dynamic contextual information such as temporal correlation strength and neighborhood interaction patterns. This limits the generalization of features across tasks, such as drug sensitivity prediction and combination effect assessment, which require repeated training of different models. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to address the shortcomings of the prior art and propose a multi-task hepatitis B drug screening method and system based on knowledge graph assistance.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a multi-task hepatitis B drug screening method based on knowledge graph assistance, comprising the following steps: Based on the hepatitis B virus genotype sequence data recorded over time, patient drug use records and drug sensitivity, a temporal connection relationship between entities is established to generate a temporal hepatitis B knowledge graph structure; Based on the temporal hepatitis B knowledge graph structure, combined with input biological pathway annotation information, protein interaction data, and known drug combination effect records, identify the set of multiple entities that exist in the temporal hepatitis B knowledge graph structure that act together, define hyperedges connecting the multiple entity sets, obtain the multi-entity interaction set and hyperedge definition, and based on the multi-entity interaction set and hyperedge definition, expand the connection mode of the nodes in the temporal hepatitis B knowledge graph structure to establish a hepatitis B knowledge base that integrates hyperedge interactions; Based on the hepatitis B knowledge base integrated with hyperedge interaction, the initial vector expression of each hepatitis B virus entity, drug entity, and gene entity is calculated to obtain an initial entity vector representation. Based on the initial entity vector representation, the vector expression is updated and adjusted by aggregating the neighborhood node information and the connection information of different types of edges in the graph to establish a multi-task hepatitis B entity feature vector; Based on the multi-task hepatitis B entity feature vector, the feature vector corresponding to the target hepatitis B virus variant and the feature vector corresponding to the candidate drug are selected, the drug sensitivity score between the two is estimated, and a virus strain drug sensitivity score list is obtained. Based on the virus strain drug sensitivity score list and the multi-task hepatitis B entity feature vector corresponding to the candidate drug combination, the antagonistic effect of the drug combination is evaluated to obtain the hepatitis B drug combination potential evaluation result.

[0006] Preferably, the steps for obtaining the time-series hepatitis B knowledge graph structure are: Integrate hepatitis B virus genotype sequence data, patient drug use records, and drug sensitivity data, extract the timestamp information of each data item, convert the timestamp information into a numerical time label in a unified time format, and generate a timestamp-related dataset; Based on the timestamp association data set, the interval duration between time tags is calculated, continuous time periods are divided according to the interval duration and a preset time window threshold, an association edge is established between the hepatitis B virus genotype entity and the drug entity in each time period, the weight value of the time association edge is calculated, and a weighted time association edge set is generated; Based on the weighted time-related edge set, the related edges with weight values greater than the preset edge weight threshold are connected to the corresponding entities to form a time-series hepatitis B knowledge graph structure.

[0007] Preferably, the steps for obtaining the multi-entity interaction set and hyperedge definition are: Integrate the entity nodes, biological pathway annotation information, protein interaction data and drug combination effect records in the time-series hepatitis B knowledge graph structure, extract the gene regulatory pathway identifiers in the biological pathway annotation information and the target gene set identifiers corresponding to the drugs in the drug combination effect records, and generate a multi-source data set; Based on the multi-source data set, calculating the co-action strength of the gene entity set and the drug target gene set in the same time window; Based on the co-action strength, a gene set and a drug target gene set whose co-action strength is greater than a preset threshold are selected, a hyperedge connecting the sets is defined, and a multi-entity interaction set and a hyperedge definition are generated.

[0008] Preferably, the steps for acquiring the hepatitis B knowledge base integrated with hyperedge interaction are: Extract the node connection relationship between the multi-entity interaction set and the hyperedge definition and the time-series hepatitis B knowledge graph structure, traverse each hyperedge in the hyperedge definition, establish a bidirectional connection relationship between the hyperedge and all nodes in the corresponding entity set, and generate a hyperedge connection relationship table; Based on the hyperedge connection relationship table, detecting whether there are isolated nodes or redundant connections in the hyperedge connection relationship, if there are isolated nodes, supplementing the missing connections according to the protein interaction data, if there are redundant connections, merging the repeated hyperedge connections according to a preset redundancy threshold, and generating a verified hyperedge connection relationship table; Based on the verified hyperedge connection relationship table, the hyperedge connection relationship is mapped and replaced with the original node connection mode in the time-series hepatitis B knowledge graph structure, and the connection type and attributes between nodes are updated to form a hepatitis B knowledge base that integrates hyperedge interactions.

[0009] Preferably, the steps for obtaining the initial entity vector representation are: Traversing the hepatitis B virus entity, drug entity, and gene entity nodes in the hepatitis B knowledge base integrated with the hyperedge interaction, extracting the interaction event timestamp recorded in the time association edge of each node, counting the total number of times each node is associated with a multi-entity interaction set in the hyperedge information, recording the interaction event timestamp sequence and the number of hyperedge associations, and generating a node time series interaction log; Based on the node time-series interaction log, taking the first interaction timestamp as the benchmark, calculate the time difference between each subsequent timestamp and the benchmark to generate a time difference sequence, count the combined frequency of each node appearing simultaneously with other entities in the hyperedge information, calculate the ratio of the combined frequency to the length of the time difference sequence, and generate a set of multi-entity co-occurrence intensity factors; Based on the multi-entity co-occurrence intensity factor set, the time difference sequence is divided into windows with a period of 30 days, the mean and variance of the time difference in each window are calculated, the three sets of values of mean, variance and co-occurrence intensity factor are spliced, and the initial entity vector representation is output.

[0010] Preferably, the steps of obtaining the multi-task hepatitis B entity feature vector are: Traversing each entity node in the initial entity vector representation, extracting a set of directly connected neighboring nodes in the fusion hyperedge-interacted hepatitis B knowledge base, recording the entity type, edge type, and number of connections of the neighboring nodes, and generating a neighboring node information set; Based on the neighborhood node information set, counting the proportion of connections of different edge types in the neighborhood of each entity node, and multiplying the proportion value by a preset priority coefficient of the neighborhood node entity type to generate an edge type weight factor set; Based on the edge type weight factor set, the initial entity vector representation and the entity vector in the neighborhood node information set are weighted and spliced according to the weight factor to generate a multi-task hepatitis B entity feature vector.

[0011] Preferably, the steps for obtaining the virus strain drug sensitivity score list are: Traversing the target hepatitis B virus variants and candidate drug entities in the multi-task hepatitis B entity feature vector, extracting the virus variant feature vector, which includes the expression intensity and time correlation intensity of the gene mutation site, and the drug feature vector, which includes the target effect intensity and metabolic half-life, to generate a set of virus-drug feature vector pairs; Based on the set of virus-drug feature vector pairs, the sensitivity scores of virus and drug feature vectors are calculated; Based on the sensitivity scores, the strains are sorted from high to low according to the sensitivity scores to generate a list of drug sensitivity scores of the virus strains.

[0012] Preferably, the steps for obtaining the hepatitis B drug combination potential evaluation results are: Traverse each combination in the candidate drug combination list, extract the identifiers of all drug entities in the combination, obtain the sensitivity score of each drug to the target virus variant from the virus strain drug sensitivity score list, and simultaneously extract the feature vector of the drug entity from the multi-task hepatitis B entity feature vector to generate a drug combination feature-score dataset; Based on the drug combination feature-score dataset, the cosine similarity between the feature vectors of each pair of drugs in the combination is calculated. If the cosine similarity of any two drugs in the combination is higher than the preset mechanism overlap threshold and the sensitivity scores of both drugs are higher than the preset single-drug effectiveness threshold, the drug pair is determined to have a target competitive antagonistic effect, and an antagonistic effect marker set is generated; Based on the antagonistic effect marker set, combinations containing at least one antagonistic drug pair are eliminated, and the remaining combinations are sorted to generate a hepatitis B drug combination potential evaluation result.

[0013] The present invention provides a hepatitis B drug screening system, comprising: The time series graph construction module establishes the temporal connection relationship between entities based on the hepatitis B virus genotype sequence data recorded over time, the patient's drug use records and drug sensitivity, and generates a time series hepatitis B knowledge graph structure; A hyperedge fusion module, based on the temporal hepatitis B knowledge graph structure, combines input biological pathway annotation information, protein interaction data, and known drug combination effect records to identify a set of entities that act together in the temporal hepatitis B knowledge graph structure, define hyperedges connecting multiple entity sets, obtain multi-entity interaction sets and hyperedge definitions, and based on the multi-entity interaction sets and hyperedge definitions, expand the connection mode of nodes in the temporal hepatitis B knowledge graph structure to establish a hepatitis B knowledge base that integrates hyperedge interactions; A representation learning module calculates the initial vector expression of each hepatitis B virus entity, drug entity, and gene entity based on the hepatitis B knowledge base integrated with hyperedge interactions, obtains an initial entity vector representation, and updates and adjusts the vector expression based on the initial entity vector representation by aggregating neighborhood node information and connection information of different types of edges in the graph to establish a multi-task hepatitis B entity feature vector; The effect prediction module selects the feature vector corresponding to the target hepatitis B virus variant and the feature vector corresponding to the candidate drug based on the multi-task hepatitis B entity feature vector, estimates the drug sensitivity score between the two, obtains a list of virus strain drug sensitivity scores, and evaluates the antagonistic effect of the drug combination based on the virus strain drug sensitivity score list and the multi-task hepatitis B entity feature vector corresponding to the candidate drug combination to obtain the hepatitis B drug combination potential evaluation result.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In this invention, multidimensional dynamic time series data such as viral genotype sequence data, patient drug use records, and drug sensitivity are integrated. Interaction events between entities are marked by timestamps, and dynamic features such as viral mutation trajectories and changes in drug efficacy are embedded in graph node attributes, enabling knowledge representation to reflect the time dependency in real scenarios. Based on biological pathway annotation information and protein interaction data, a hyperedge-connected multi-entity set is defined, expanding the limitations of traditional knowledge graphs that only support binary relationships. The synergistic mechanism of drug combinations and multi-targets is explicitly modeled to avoid misjudgment of combination effects due to simplified interaction relationships. The initial entity vector represents temporal correlation features such as the relative time interval within the fusion time window and the hyperedge co-occurrence intensity factor. The vector expression is dynamically updated through neighborhood node information aggregation and edge type weight adjustment to enhance the feature differentiation of entities in different task scenarios. The virus strain drug sensitivity score combines cosine similarity and time decay factor to quantify the matching degree between drugs and virus variants on key targets and the timeliness of data, reducing the evaluation bias caused by outdated historical data. The antagonistic effect evaluation identifies competitive drug pairs with high similarity and high scores through the dual constraints of mechanism overlap threshold and single drug effectiveness threshold. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] See also Figure 1The present invention provides a technical solution, a multi-task hepatitis B drug screening method based on knowledge graph assistance, comprising the following steps: Based on the hepatitis B virus genotype sequence data recorded over time, patient drug use records and drug sensitivity, a temporal connection relationship between entities is established to generate a temporal hepatitis B knowledge graph structure; Based on the temporal hepatitis B knowledge graph structure, combined with input biological pathway annotation information, protein interaction data, and known drug combination effect records, we identify the sets of multiple entities that interact with each other in the temporal hepatitis B knowledge graph structure, define hyperedges connecting multiple entity sets, obtain the multi-entity interaction set and hyperedge definition, and based on the multi-entity interaction set and hyperedge definition, expand the connection method of the nodes in the temporal hepatitis B knowledge graph structure to establish a hepatitis B knowledge base that integrates hyperedge interactions. Based on the HBV knowledge base integrated with hyperedge interactions, the initial vector expression of each HBV virus entity, drug entity, and gene entity is calculated to obtain the initial entity vector representation. Based on the initial entity vector representation, the vector expression is updated and adjusted by aggregating the neighborhood node information and the connection information of different types of edges in the graph to establish a multi-task HBV entity feature vector. Based on the multi-task hepatitis B entity feature vector, the feature vector corresponding to the target hepatitis B virus variant and the feature vector corresponding to the candidate drug are selected, and the drug sensitivity score between the two is estimated to obtain a list of virus strain drug sensitivity scores. Based on the virus strain drug sensitivity score list and the multi-task hepatitis B entity feature vector corresponding to the candidate drug combination, the antagonistic effect of the drug combination is evaluated to obtain the potential evaluation results of hepatitis B drug combinations.

[0018] The steps for obtaining the time series hepatitis B knowledge graph structure are: Integrate hepatitis B virus genotype sequence data, patient drug use records, and drug sensitivity data, extract the timestamp information of each data item, convert the timestamp information into a numerical time label in a unified time format, and generate a timestamp-related dataset; Based on the timestamp association dataset, the interval length between time tags is calculated, and continuous time periods are divided according to the interval length and the preset time window threshold. In each time period, the association edge between the hepatitis B virus genotype entity and the drug entity is established using the formula: ; Calculate the weight value of the time-related edge ,in represents the time label of entity r, represents the time label of entity s, is the time interval reference value, For the time deviation tolerance coefficient, generate a weighted time-related edge set; Based on the set of weighted time-related edges, the related edges with weight values greater than the preset edge weight threshold are connected to the corresponding entities to form a time-series hepatitis B knowledge graph structure.

[0019] Specifically, the system integrates hepatitis B virus (HBV) genotype sequence data (such as nucleotide sequences in FASTA format) from clinical electronic medical record systems, gene sequencing platforms, and drug sensitivity testing laboratories, detailed patient drug use records (including drug name, dosage, start and end dates), and in vitro or clinical drug sensitivity test results (such as phenotypic test results). For analysis of drug resistance loci (e.g., genotype resistance loci), we first accurately extract timestamp information associated with each recorded event from various data sources, including the date of gene sequence determination, the first and last drug use dates, and the date of susceptibility testing. These timestamps in various formats (e.g., 'YYYY-MM-DD' and 'MM / DD / YYYYHH:MM') are then uniformly converted into standard numeric time tags. Specifically, the Unix timestamp format of Coordinated Universal Time (UTC) is used, which is the total number of seconds elapsed from 00:00:00 on January 1, 1970, to the time of the event. For example, the date '2023-10-26 14:30:00' is converted to the numeric value 1698301800 to ensure the consistency and computability of the time data. The original data is then associated with the converted numeric time tags to construct a structured dataset, in which each record contains the original entity information (e.g., virus sequence ID, drug ID, patient ID) and its corresponding standardized numeric time tag, generating a timestamp-associated dataset.

[0020] formula: , the formula is beneficial in that it uses the Gaussian kernel function to quantify the two event time points ( and ). It does not only consider the absolute value of the time interval, but also evaluates the relationship between this interval and an expected typical time interval ( ), while taking into account the acceptable range of variation in the time interval ( This approach gives higher weights to event pairs that are more consistent with expected patterns in time (for example, the common time span for drug sensitivity testing after viral mutations), while lower weights are given to event pairs with too short or too long time intervals. This allows for more accurate capture of the dynamic and time-dependent relationship between HBV mutations, drug use, and sensitivity changes, providing a quantitative basis for constructing a temporal knowledge graph that reflects real biomedical processes.

[0021] The time tag representing entity r is the time point of the event associated with entity r (for example, the time of determination of a HBV genotype sequence). This parameter is a numeric value in seconds and is directly extracted from the timestamp-related dataset in the previous step. For example, the timestamp associated with the HBV genotype G1896A sequencing event is '2023-05-10 09:00:00' from the timestamp-related dataset. Converting it to a Unix timestamp yields .

[0022] The time tag representing entity s is the time point at which the event associated with entity s (for example, the time when the lamivudine drug sensitivity test for the above-mentioned viral genotype was performed) occurred. This parameter is also a numeric value in seconds and is directly extracted from the timestamp association dataset in the previous step. For example, the timestamp associated with the lamivudine drug sensitivity test event found in the timestamp association dataset is '2023-08-15 11:30:00'. Converting it to a Unix timestamp yields .

[0023] It is the time interval benchmark value, representing the expected or most typical time interval between entity r and entity s. This parameter is a numeric value in seconds, and its setting should be based on the understanding of clinical diagnosis and treatment of hepatitis B and virological research, reflecting the standard cycle or average interval of related events. Acquisition method: Statistically analyze the time intervals of a large number of related event pairs (such as virus sequencing and subsequent drug sensitivity testing) in the timestamp associated data set, and calculate the average or median of these intervals. For example, analyze the time interval data of 1,000 pairs of "virus genotype sequencing" and "the first related drug sensitivity test immediately thereafter" and calculate the average interval length. If the calculated average interval is 95 days, then Second.

[0024] is the time deviation tolerance coefficient, which represents the deviation of the time interval between entity r and entity s from the baseline value The acceptable degree or common fluctuation range of the event. This parameter is a numerical value in seconds, reflecting the discrete degree of the time interval between events. At the same time, calculate the standard deviation of the time interval data used. For example, based on the time interval data of the above 1000 pairs of events, the calculated standard deviation is 25 days, then Second.

[0025] Preset time window threshold: This threshold is used to preliminarily filter possible related event pairs, and only calculates between event pairs whose time interval is less than this threshold. For example, set it to 180 days. Basis for setting: Based on clinical experience or data analysis, determine a period long enough to include the vast majority of meaningful related events while avoiding the calculation of unnecessary long-term event pairs. For example, analyze the time interval distribution of all event pairs in the data set and select the value that covers 95% of the intervals. If 95% of the intervals are less than 170 days, the threshold can be set to 180 days ( Second).

[0026] Substitute the parameters into the weight value calculated by the formula Represents the typical time interval considered Days and deviation tolerance In the case of days, The viral genotype sequencing events at the time point were compared with those occurring in Strength of temporal association between drug sensitivity testing events at time points.

[0027] The weighted time-related edge set calculated in the previous step contains all entity pairs (such as hepatitis B virus genotype entity and drug entity) within the preset time window (for example, 180 days) and their corresponding temporal association weights. , perform the edge screening process, specifically, it is necessary to set a preset edge weight threshold, which is designed to distinguish strong and weak associations, retain connections with significant temporal significance, and filter out connections with weak associations or those that may be generated by noise. This threshold can be determined by analyzing all calculated The distribution characteristics of the values are completed, such as drawing a histogram or cumulative distribution function of the weights, observing the inflection point of the weight distribution, or setting a value that can retain, for example, the top 70% of the strongly associated edges based on domain knowledge. For example, if the calculated weight values are mainly distributed between 0 and 0.016, through analysis, it is found that the time intervals corresponding to the edges with weights less than 0.005 are different from the benchmark intervals. The larger the deviation, the less meaningful the association, while a weight above 0.01 indicates that the time intervals are very close. , the correlation is strong, and a compromise value can be selected. For example, the preset edge weight threshold is set to 0.008. The calculation method can be to take the 30th percentile of all calculated weights (retain the top 70%) or set according to expert experience. Then, traverse each edge record in the weighted time-related edge set and compare its weight value. With the preset edge weight threshold of 0.008, if , then the associated edge is considered significant, and the edge and the two entities it connects (such as the viral genotype entity G1896A and the drug entity lamivudine) are included in the final graph structure. , then discard this edge and do not connect it. All associated edges that pass the threshold screening and their corresponding entity nodes are aggregated to form a collection of nodes (representing hepatitis B virus genotypes, drugs, and possibly patients, etc.) and weighted edges (representing event relationships that meet the temporal association strength threshold). Finally, a networked temporal hepatitis B knowledge graph structure is constructed.

[0028] The steps to obtain the multi-entity interaction set and hyperedge definition are: Integrate entity nodes, biological pathway annotation information, protein interaction data, and drug combination effect records in the time-series hepatitis B knowledge graph structure, extract gene regulatory pathway identifiers in the biological pathway annotation information and target gene set identifiers corresponding to drugs in the drug combination effect records, and generate a multi-source dataset; Based on multi-source data sets, the co-action strength of the gene entity set and the drug target gene set in the same time window is calculated using the following formula: ; in, is the joint action strength, is the gene set in the annotation of the i-th biological pathway, is the target gene set of the kth drug in the drug combination effect record, is the interaction frequency between genes m and n in the protein interaction data, is the frequency normalization constant, take The maximum value of Based on the co-action strength, the gene set and drug target gene set whose co-action strength is greater than the preset threshold are selected, the hyperedges connecting the sets are defined, and the multi-entity interaction set and hyperedge definition are generated.

[0029] Specifically, various entity nodes (such as hepatitis B virus genotypes, drugs, genes, etc.) contained in the previously constructed temporal hepatitis B knowledge graph structure are integrated and combined with biological pathway annotation information obtained from external public databases (for example, pathway definitions related to liver diseases and viral infections are downloaded from the KEGG database, such as hsa05161: Hepatitis B pathway, which contains a series of gene identifiers), as well as protein-protein interaction (PPI) data and drug combination effect records. On this basis, key identification information is systematically extracted. Specifically, the unique identifier of each pathway and its associated gene set (represented by standard gene symbols or EntrezID lists) are extracted from the biological pathway annotation information. At the same time, from the drug combination effect records, for each drug or drug combination, the target gene set identifier with clear effects is extracted. These data from different sources (temporal knowledge graph entities, pathway gene sets, PPI data, drug target gene sets) are normalized and mapped, and a unified identifier system is used (for example, EntrezGeneID is used uniformly) to construct a multidimensional, heterogeneous dataset containing entity information, pathway information, PPI information, and drug target information, thus generating a multi-source dataset.

[0030] formula: The formula is useful in that it evaluates a biological pathway (consisting of gene sets) by combining two dimensions. Representative) and a drug (or drug combination) target (by gene set The first part is the Jacquard index. , directly quantifies the degree of overlap between pathway genes and drug target genes. The higher the overlap, the more direct the potential functional association. Part II, , which introduces protein interaction information. Specifically, it considers genes in the overlapping regions of pathways and drug targets ( ) (e.g., mean interaction frequency), and logarithmic transformation and smoothing, as well as maximum frequency Normalized to measure the activity of this internal interaction. Multiplying the two together makes the joint action strength It not only reflects the direct gene member overlap, but also considers the functional collaboration within these overlapping genes, thereby more comprehensively evaluating the potential biological interaction strength between pathways and drug target sets.

[0031] Representative The set of genes included in the annotation information of a biological pathway. This is a set of gene identifiers. It can be obtained by querying the standard biological pathway database. For example, query the KEGG database to obtain the gene list related to the "Hepatitis B" pathway (hsa05161). Example: , containing 5 genes.

[0032] Representative The target gene set of a drug or drug combination that is clearly affected by the drug combination effect record. This is also a set of gene identifiers. Acquisition methods include querying drug databases to obtain known drug targets, or extracting target gene information related to the efficacy of specific combinations from drug combination effect research literature and databases. For example, for the drug combination entecavir + lamivudine, the host or viral gene set that it mainly affects is determined based on literature and database information. Example: , containing 5 genes.

[0033] Represents the interaction frequency of genes m and n in protein interaction data. First, determine the intersection , and then query all different gene pairs in the intersection from the protein interaction database Frequency of interaction (e.g., based on the amount of experimental evidence), calculate the average of these frequencies Example: For and , the intersection is The BioGRID database is queried and the interaction frequency between 'GeneC' and 'GeneX' is recorded as 8 times. Since the intersection has only two genes, there is only one pair ('GeneC', 'GeneX'). .

[0034] is the frequency normalization constant, defined as the number of observations in the entire data set (Here refers to the maximum value of the average interaction frequency of gene pairs in the intersection). Acquisition method: Calculate all considered pathway-drug target gene pairs The average interaction frequency of the intersection gene pairs , find the maximum value. This requires all relevant and Combined execution of the above Example: After analyzing all pathway and drug combinations, it is found that the average interaction frequency of the largest intersection gene pair is 150. Then set .

[0035] Calculation process: Using the above example parameters and : Calculate the intersection : ; Calculate the size of the intersection : ; Compute union : ; Calculate the size of the union : ; Calculate the Jacquard index part: ; Get the average interaction frequency of gene pairs in the intersection (According to the example acquisition process): ; Get the frequency normalization constant : ; Calculate the logarithmic part: ; ; ; Calculate the final interaction strength : ; The results show that: the calculated interaction strength It is a biological pathway (e.g., Hepatitis B pathway) and drug combinations A quantitative assessment of the functional association between target gene sets. This value combines the degree of overlap between the pathway and the target gene set (Jaccard index of 0.25) and the average interaction strength within the overlapping genes (the logarithm contributes a factor of approximately 0.075). The higher the value, the stronger the association between the pathway and the drug target set, and the greater the potential functional synergy or impact. The value is compared with a preset threshold to determine whether the pathway-drug target combination constitutes a multi-entity interaction set that requires attention. For example, if If the value is greater than the threshold, it is considered that the two sets have significant common effects, and a hyperedge can be defined to connect them.

[0036] Based on a series of joint action strengths calculated in the previous step Value, each The value corresponds to a set of gene entities (from biological pathways ) and drug target gene sets (from drug ) pairing, perform screening and definition steps, first set a preset threshold for judging whether the joint effect strength is significant, the determination of the threshold needs to be based on the Statistical analysis of the distribution of values or consideration of biological significance, for example, all The distribution of values, select the value in the top 10% (i.e., the 90th percentile) of the distribution as the threshold, or set an empirical threshold based on known strong interaction cases. The specific calculation process is: collect all the calculated values, sort them, find the value corresponding to the 90th percentile, if the value is 0.045, then the preset threshold can be set to 0.05, then traverse all the calculated Gene set pairs with values , each The value is compared with the preset threshold 0.05. If , then the gene entity set is considered Drug target gene panel There is a strong enough interaction between the two sets As the basis for the multi-entity interaction, for each selected set pair, a hyperedge is defined. This hyperedge is intended to explicitly represent the interaction between the two sets as a whole. The objects connected by the hyperedge are all the gene entity nodes that constitute the two sets (i.e. All gene nodes in ), or hyperedges can be connected to represent pathways Nodes and representative drugs nodes (if these are defined as independent entities in the knowledge graph), all set pairs that pass the threshold screening and their defined hyperedge information are summarized and recorded to form a clear list of multi-entity interaction sets and corresponding hyperedge definition descriptions, generating multi-entity interaction sets and hyperedge definitions.

[0037] The steps for obtaining the hepatitis B knowledge base integrated with hyperedge interaction are as follows: Extract the node connection relationship between the multi-entity interaction set and the hyperedge definition and the temporal hepatitis B knowledge graph structure, traverse each hyperedge in the hyperedge definition, establish a bidirectional connection relationship between the hyperedge and all nodes in the corresponding entity set, and generate a hyperedge connection relationship table; Based on the hyperedge connection relationship table, whether there are isolated nodes or redundant connections in the hyperedge connection relationship is detected. If there are isolated nodes, the missing connections are supplemented according to the protein interaction data. If there are redundant connections, the repeated hyperedge connections are merged according to the preset redundancy threshold to generate a verified hyperedge connection relationship table; Based on the verified hyperedge connection relationship table, the hyperedge connection relationship is mapped and replaced with the original node connection method in the time-series hepatitis B knowledge graph structure, and the connection type and attributes between nodes are updated to form a hepatitis B knowledge base that integrates hyperedge interactions.

[0038] Specifically, extract the information of existing nodes and edges in the "multi-entity interaction set and hyperedge definition" and the "time-series hepatitis B knowledge graph structure" generated in the previous step. Specifically, read each defined hyperedge and its associated entity set (the set contains entities from a specific biological pathway) from the "multi-entity interaction set and hyperedge definition" Gene entities and specific drug target sets Gene entity), and access the "Time Series Hepatitis B Knowledge Graph Structure" to obtain the identifiers of all entity nodes (including genes, drugs, virus variants, etc.) and their existing connection relationships. Then, for each hyperedge definition record in the "Multi-Entity Interaction Set and Hyperedge Definition", first create a new node representing the hyperedge itself in the knowledge graph data structure (for example, give it a unique hyperedge ID and mark its type as 'Hyperedge'), then find the entity set associated with the hyperedge (for example, the set ) is the node identifier of each entity member in the set For each entity node in the model, a connection is established from the entity node to the hyperedge node, and a connection is established from the hyperedge node back to the entity node, forming a bidirectional connection. These two connections can be given specific type labels, such as 'participatesInHyperedge' and 'hasParticipant'. The connection relationship between these newly established hyperedges and entities is recorded, including information such as hyperedge ID, entity ID, and connection type. The connection information of all hyperedges is summarized to generate a structured hyperedge connection relationship table.

[0039] Based on the "hyperedge connection relationship table" generated in the previous step, the connection relationship is verified and refined. First, isolated node detection is performed. Each hyperedge in the table and the set of entity nodes connected to it are traversed to check whether there is an entity node that is only connected to other nodes through the hyperedge, but lacks the expected direct connection relationship within the entity set connected by the hyperedge (that is, between other entities participating in the hyperedge). For example, a gene node belongs to a hyperedge set, but at the protein interaction level, it is expected to have a direct association with other genes belonging to the set but has no corresponding connection. If such a potential isolated node is detected, protein interaction (PPI) data (for example, querying the STRING database version 11.5 or the latest version of the BioGRID database) is used to supplement the possible missing Connection, specifically, query whether there is a known protein interaction between the isolated node and other nodes in the same hyperedge set, and set an interaction confidence threshold. This threshold is based on the score or evidence strength provided by the PPI database. For example, the comprehensive score of the STRING database is selected, and the threshold is set to 0.7. This threshold is determined by analyzing the distribution of known highly correlated hepatitis B-related protein interaction scores. The value that can filter out low-confidence connections (for example, scores below 0.7) while retaining medium and high-confidence connections is selected. Only when the queried PPI score is greater than 0.7, a standard binary edge representing the PPI is added to the knowledge graph to connect the two entity nodes. Then, redundant connection detection is performed, focusing on identifying repeated hyperedges connecting similar or identical entity sets, and calculating any two hyperedges. and The set of connected entities and The similarity between them is measured using the Jacquard index The measurement is performed and the preset redundancy threshold is set to 0.9. The basis for setting this threshold is that when the overlap of the node sets connected by two hyperedges reaches more than 90%, it is considered that the macro-interaction patterns they represent are highly similar and can be merged. This threshold is analyzed by the Jacquard index distribution of all hyperedge pairs in the data, and a critical point that can distinguish between high overlap and general overlap is selected (for example, the 95th percentile in the distribution is selected, if it is 0.88, it is rounded up to 0.9). If the calculated If it is greater than 0.9, the edge Mark as redundant and merge its connections into (that is, the original connection to The entity node is now connected to , then remove ), after the isolated nodes are supplemented with connections and redundant hyperedges are merged, the final connection relationship set is obtained to generate a verified hyperedge connection relationship table.

[0040] Based on the "verified hyperedge connection relationship table" generated in the previous step, these verified and refined hyperedge connection relationships are integrated into the original "temporal hepatitis B knowledge graph structure" to perform mapping replacement and update operations. The specific process is to traverse each record in the "verified hyperedge connection relationship table", which describes the connection between a hyperedge node and its participating entity nodes, or the binary PPI connection between supplementary entities, and add these connection information to the graph representation of the "temporal hepatitis B knowledge graph structure" (for example, updating the adjacency matrix, adjacency list or edge list data structure). For hyperedge connections, it does not directly replace the original binary edges based on time association or other relationships, but is added as a new type of connection, that is, the connection method between nodes is expanded, and the original binary edges (such as edges representing temporal associations) and the newly added hyperedge connections (indicating participation in multi-entity interactions) can be used together. This change is reflected by updating the connection types and attributes between nodes. For example, for a gene node, its connection list may now contain temporal edges pointing to other genes, temporal edges pointing to drugs, and 'participatesInHyperedge' type edges pointing to one or more hyperedge nodes. At the same time, the attributes of the edges may also need to be updated. For example, an attribute can be added to the nodes participating in the hyperedge to record the hyperedge ID list to which they belong, or an attribute can be added to the hyperedge node itself to describe the biological significance represented by the hyperedge (such as associated pathways and drug combinations). In this way, the multi-entity collective behavior information represented by the hyperedge is seamlessly integrated into the knowledge graph, rather than simply replacing the original connection. The final network structure is a hepatitis B knowledge base that integrates hyperedge interaction information, has more diverse connection methods, and can better reflect the interactive characteristics of complex biological systems.

[0041] The steps to obtain the initial entity vector representation are: Traverse the hepatitis B virus entity, drug entity, and gene entity nodes in the hepatitis B knowledge base integrated with hyperedge interactions, extract the interaction event timestamp recorded in the time-related edge of each node, count the total number of times each node is associated with a multi-entity interaction set in the hyperedge information, record the interaction event timestamp sequence and the number of hyperedge associations, and generate a node time series interaction log; Based on the node time series interaction log, taking the first interaction timestamp as the benchmark, the time difference between each subsequent timestamp and the benchmark is calculated to generate a time difference sequence. The combination frequency of each node appearing simultaneously with other entities in the hyperedge information is counted, and the ratio of the combination frequency to the length of the time difference sequence is calculated to generate a set of multi-entity co-occurrence intensity factors. Based on the set of multi-entity co-occurrence intensity factors, the time difference series is divided into windows with a period of 30 days. The mean and variance of the time difference in each window are calculated. The three sets of values of mean, variance and co-occurrence intensity factor are spliced together to output the initial entity vector representation.

[0042] Specifically, all target entity nodes stored in the "Hepatitis B Knowledge Base Integrated with Hyperedge Interaction" are systematically accessed, including hepatitis B virus entities (such as specific genotypes or variant identifiers), drug entities (such as entecavir, lamivudine) and gene entities (such as human or viral gene identifiers). For each entity node, its associated edge information is carefully examined, and the time-related edges that record the timestamps of the interaction events are selected (these edges are usually established in the early stage of knowledge graph construction based on the time information in the original data). The timestamp values attached to these edges (such as Unix timestamp format) are accurately extracted. For the timestamps associated with the same node, Multiple timestamps are arranged in chronological order to form a timestamp list. At the same time, the node's participation in hyperedge interactions is checked. By finding the 'hasParticipant' type connection pointing to the node or the 'participatesInHyperedge' type connection starting from the node, the total number of different hyperedges associated with the node is counted. The ordered interaction event timestamp sequence of each node and the total number of hyperedges it participates in are structured and recorded. For each hepatitis B virus entity, drug entity, and gene entity node in the knowledge base, a corresponding node time series interaction log is generated.

[0043] Based on the "node time series interaction log" generated for each entity in the previous step, the time series features and co-occurrence features are calculated. First, for each node's interaction event timestamp sequence , select the first timestamp in the sequence As the time reference point for the first recorded interaction of the entity, and then calculate all subsequent timestamps (in From 1 to ) relative to the reference point The time difference is calculated as , these differences (in seconds or converted to days) form an ordered sequence of time differences , which reflects the relative time distribution of entity interaction events. Then, the total number of hyperedges in which the node participates is extracted from the corresponding “node time series interaction log” (denoted as ), this number represents the frequency of the node forming a multi-entity interaction set with other entities in the knowledge base. Then, the "multi-entity co-occurrence intensity factor" of the node is calculated by adding the total number of hyperedge associations Divide by the length of the time difference series (i.e. the total number of subsequent interaction events, if The factor is 0 or set as needed), that is, factor = ,This ratio quantifies the average intensity of a node’s participation in a complex ,interaction in a unit interaction event.,The “multi-entity co-occurrence intensity factors” calculated for all HBV ,entity, drug entity, and gene entity nodes in the knowledge base are ,collected together to generate a multi-entity co-occurrence intensity factor set.

[0044] Based on the previously calculated "multi-entity co-occurrence intensity factor" of each node (from the "multi-entity co-occurrence intensity factor set") and the corresponding "time difference sequence", the initial vector representation of each entity is constructed. First, the "time difference sequence" of each node is Divide the time window and use a fixed 30-day time window period (for example, the first window contains Value seconds of interaction, the second window contains For each divided time window, calculate all the time differences that fall within the window. If no timestamp falls within a window, its mean and variance can be set to 0 or processed by interpolation. In this way, each node will get a sequence consisting of the means of all windows (for example ) and a sequence consisting of all window variances (e.g. ), then, concatenate these three sets of numerical information: concatenate the mean sequence of all windows, the variance sequence of all windows, and the "multi-entity co-occurrence intensity factor" of the node (as a scalar feature) to form a long vector. For example, if the time span covers windows, the generated vector structure is , the vector dimension is , summarize the vectors generated by this method for all hepatitis B virus entities, drug entities and gene entities, and output the initial entity vector representation.

[0045] The steps for obtaining the multi-task hepatitis B entity feature vector are: Traverse each entity node in the initial entity vector representation, extract the set of directly connected neighboring nodes in the hepatitis B knowledge base integrated with hyperedge interaction, record the entity type, edge type and number of connections of the neighboring nodes, and generate a neighboring node information set; Based on the neighborhood node information set, the proportion of connections of different edge types in the neighborhood of each entity node is counted, and the proportion value is multiplied by the preset priority coefficient of the neighborhood node entity type to generate a set of edge type weight factors; Based on the edge type weight factor set, the initial entity vector representation and the entity vector in the neighborhood node information set are weighted and spliced according to the weight factor to generate a multi-task hepatitis B entity feature vector.

[0046] Specifically, for each entity node (including hepatitis B virus entity, drug entity, gene entity) with "initial entity vector representation" in the "hepatitis B knowledge base integrated with hyperedge interaction", a neighborhood information extraction operation is performed. For the central entity node currently being processed, the graph structure data of the knowledge base is queried to obtain the set of all one-hop neighboring nodes directly connected to the central node. For each neighboring node in the set, its key information is recorded in detail, including the unique identifier of the neighboring node, the entity type of the node (for example, distinguishing between 'Gene', 'Drug', 'VirusVariant', 'Hyperedge', etc.), and the central node is connected to the neighboring node. The edge type (for example, 'temporal_association' represents temporal association, 'ppi_interaction' represents protein interaction, 'participatesInHyperedge' represents entity participation in hyperedge, 'hasParticipant' represents hyperedge containing entity, etc.), and the number of connections of a specific type of edge between the pair of nodes (if parallel edges are allowed, then count; if not, the number is 1). The identifiers, entity types, edge types and connection counts of all collected neighboring nodes are sorted and stored for each central entity node to generate a neighboring node information set for the node.

[0047] Based on the "neighborhood node information set" generated for each central entity node in the previous step, the weight factor for subsequent neighborhood information aggregation is calculated. First, the "neighborhood node information set" of the current central node is statistically analyzed to calculate the total number of its neighborhood connections (that is, the sum of all neighborhood connections). Then, for each different edge type appearing in the neighborhood (such as 'temporal_association', 'ppi_interaction', 'participatesInHyperedge', etc.), the total number of connections of this type of edge is counted separately, and the proportion of the number of connections of each edge type to the total number of neighborhood connections (proportion value) is calculated. Then, the preset priority coefficient of the neighborhood node entity type is introduced. The coefficient is set according to the importance of different entity types in the hepatitis B drug screening task. For example, the 'Drug' type node directly related to the drug screening target has the highest priority and is assigned a coefficient of 1.2, followed by the 'VirusVariant' type node, which is assigned a coefficient of 1.1, and the 'Gene' type node is assigned a coefficient of 1.0. The abstract 'Hyperedge' type node has a relatively low priority and is assigned a coefficient of 0.9. These coefficient values are initially set based on domain knowledge and can be subsequently tuned through the validation set effect during model training. For example, these coefficient values can be adjusted through grid search or gradient-based optimization to maximize downstream task performance. Next, the edge type weight factor is calculated according to an explanatory method: for each edge type of the central node , find all the pass types The edges connected to the neighboring nodes of , calculate the average value of the priority coefficients of these neighboring nodes , and then the proportion of the edge type The corresponding average priority coefficient of the neighboring nodes Multiply them together to get the weight factor of the edge type, that is, , perform this calculation on all edge types in the neighborhood of the central node, summarize the obtained edge type weight factors, and generate the edge type weight factor set of the central node.

[0048] Using the "edge type weight factor set" of each central node generated in the previous step, combined with the "initial entity vector representation" of the central node and its neighboring nodes, weighted aggregation and splicing of neighborhood information are performed to form the final feature vector. The specific operation is as follows: For a central entity node , obtain its own "initial entity vector representation" , and access its "neighborhood node information collection" to obtain all neighboring nodes "Initial entity vector representation" , at the same time, for the connection and Edge, determine its edge type , and find the corresponding weight factor from the "edge type weight factor set" (Here It is calculated based on the edge type The associated weight factor) calculates the weighted aggregation representation of neighborhood information by adding each neighborhood node The initial entity vector Multiply by the weight factor corresponding to the connection edge type , then sum the weighted vectors of all neighboring nodes to get the aggregated neighborhood vector Finally, perform the splicing operation to connect the central node Its own initial entity vector and its aggregated neighborhood vector Concatenate across dimensions to form longer, more informative vectors , this concatenated vector It combines the node's own timing and hyperedge participation characteristics (from ) and the structure and type information of its neighborhood environment (from weighted aggregation ), as the multi-task learning optimized feature representation of the entity node, repeat this process for all target entity nodes in the knowledge base to generate a multi-task hepatitis B entity feature vector.

[0049] The steps to obtain the virus strain drug sensitivity score list are: Traverse the target hepatitis B virus variants and candidate drug entities in the multi-task hepatitis B entity feature vector, extract the virus variant feature vector, which includes the expression intensity and time correlation intensity of the gene mutation site, and the drug feature vector, which includes the target effect intensity and metabolic half-life, to generate a set of virus-drug feature vector pairs; Based on the virus-drug feature vector pair set, the sensitivity scores of the virus and drug feature vectors are calculated using the following formula: ; in, Score sensitivity. is the dot product of the virus and drug feature vectors, which measures the positive correlation between the two in the key feature dimensions. It is the modulus-length product of the vector, used to normalize the dot product value, and the output range is [-1, 1] cosine similarity. is the interval between the time of recording the virus variant genotype and the time of drug testing (unit: month), is the maximum time interval in the data set; Based on the sensitivity scores, the strains were sorted from high to low to generate a list of drug sensitivity scores for the virus strains.

[0050] Specifically, traverse the set of "multi-task hepatitis B entity feature vectors" generated in the previous step, and screen out specific target hepatitis B virus variant entities (for example, according to user input or preset list, select specific drug-resistant variants such as L180M+M204V) and a group of candidate drug entities (for example, including entecavir, tenofovir, lamivudine and other drugs currently commonly used in clinical practice or under research). For each selected target hepatitis B virus variant entity, extract its corresponding "multi-task hepatitis B entity feature vector". This vector is generated by the previous step, and its internal dimensions contain encoding information about the characteristics of the virus variant, such as the expression intensity of gene mutation sites and the time correlation intensity and other multi-dimensional features. Similarly, for each candidate drug entity, its corresponding "multi-task hepatitis B entity feature vector" is also extracted. This vector encodes the characteristics of the drug, such as the target effect intensity and metabolic half-life and other related information. The extracted target hepatitis B virus variant feature vector is paired with the feature vector of each candidate drug. For example, if there is a target virus variant V1 and three candidate drugs D1, D2, and D3, three pairs are formed: (Vector(V1), Vector(D1)), (Vector(V1), Vector(D2)), and (Vector(V1), Vector(D3)). All these pairs are collected to generate a set of virus-drug feature vector pairs.

[0051] formula: The benefit of the formula is that it innovatively combines the semantic similarity between entity feature vectors learned based on the knowledge graph and the time difference information of the event to predict the sensitivity of the virus to the drug. The first part is the cosine similarity calculation ( ), which uses the virus feature vector learned in the previous step and drug feature vector The directional consistency in high-dimensional space is used to evaluate the potential correlation between them. This correlation encodes complex biomedical information (such as the impact of viral mutations, drug action mechanisms, time series patterns, etc.). The closer the vector direction (the higher the cosine similarity), the higher the potential drug sensitivity. The second part is the time decay factor , which introduces the consideration of time dimension, It is the time interval between the viral genotype data recording time and the drug evaluation time (such as the drug sensitivity test time or the medication time point). This factor makes the calculated sensitivity score lower when the time interval is longer (the older the viral data), thereby simulating the reality that the virus may evolve over time and develop drug resistance, making the prediction results more timely and clinically relevant.

[0052] Representative target HBV variants This is a numerical vector calculated in the previous main step. Its dimension and value reflect the comprehensive characteristics of the virus variant. Acquisition method: From the set of "multi-task hepatitis B entity feature vectors" generated in the previous step, according to the virus variant The identifier is directly extracted. For example, for virus variant V1, its feature vector is , the vector dimension is determined by the aforementioned vector construction method.

[0053] Representative drug candidates This is also the numerical vector calculated in the previous main step, reflecting the drug Comprehensive features of. Acquisition method: From the "Multi-task Hepatitis B Entity Feature Vector" set, according to the drug The identifier is directly extracted. For example, for drug D1, its feature vector is , the vector dimension is same.

[0054] Representative virus variants Genotype recording time and drug The time interval between the test or evaluation time, in months. Acquisition method: It is necessary to search for virus variants from the knowledge base or metadata. The most relevant timestamp associated with the entity (e.g. sequencing date ) and with medications Reference timestamps relevant to the assessment (e.g., drug susceptibility testing date or clinical medication start date) ), calculate the time difference between the two and convert it into months. Calculation example: virus The sequencing date is January 15, 2024, and the drug The sensitivity test date is July 20, 2024. The time interval is about 6 months and a few days. Then divide by the average number of days per month (e.g. 30.44) to get the number of months. For example, if the interval is 185 days, then moon.

[0055] The maximum time interval between the virus record time and the drug evaluation time observed in the dataset, in months. Acquisition method: It is necessary to analyze all the historical data used to build the knowledge base or all the relevant virus-drug timestamp pairs in the current evaluation dataset to find For example, after analyzing a dataset containing thousands of records, the largest time interval is found to be 5 years and 2 months. moon.

[0056] Calculation process: Using example parameters: moon, moon; Calculating the dot product : ; ; Calculate the vector modulus : ; ; Calculate the vector modulus : ; ; Calculate the module length product : ; Calculate the cosine similarity part: ; Calculate the time decay factor part: ; ; Calculation of final sensitivity score : ; The results show that the calculated sensitivity score This is a quantitative prediction of the sensitivity of the target HBV variant V1 to the candidate drug D1. A score close to 1 indicates high sensitivity to the drug, based on the strong similarity between the feature vectors learned by the model (cosine similarity approximately 0.92) and the relatively short time interval (decay factor approximately 0.90). The score can theoretically range from -1 to 1 (the time decay factor can be 0 or negative, but this design ensures it is non-negative). A higher score indicates greater antiviral efficacy. In the next step, this score will be used to rank all candidate drugs.

[0057] Based on the sensitivity scores between all target HBV variants and each candidate drug calculated in the previous step , sorting and ranking these scoring results by collecting all the calculated scores, each of which is associated with a specific virus variant and a specific drug candidate , these (virus variant identifiers, drug identifiers, sensitivity scores ) triples as a list, and then use a standard sorting algorithm (such as quick sort or merge sort) to process the list, sorting it according to the sensitivity score The values are arranged in descending order, that is, the drug pairs with the highest scores are at the front of the list and the ones with the lowest scores are at the end. The sorted list clearly shows the predicted sensitivity order of each candidate drug to the target hepatitis B virus variant, and generates a virus strain drug sensitivity score list.

[0058] The steps for obtaining the results of the hepatitis B drug combination potential assessment are as follows: Traverse each combination in the candidate drug combination list, extract the identifiers of all drug entities in the combination, obtain the sensitivity score of each drug to the target virus variant from the virus strain drug sensitivity score list, and simultaneously extract the feature vector of the drug entity from the multi-task hepatitis B entity feature vector to generate a drug combination feature-score dataset; Based on the drug combination feature-score dataset, the cosine similarity between the feature vectors of each pair of drugs in the combination is calculated. If the cosine similarity of any two drugs in the combination is higher than the preset mechanism overlap threshold and the sensitivity scores of both drugs are higher than the preset single-drug effectiveness threshold, the drug pair is determined to have a target competitive antagonistic effect, and an antagonistic effect marker set is generated; Based on the antagonistic effect marker set, combinations containing at least one antagonistic drug pair are eliminated, and the remaining combinations are ranked to generate the potential evaluation results of hepatitis B drug combinations.

[0059] Specifically, the user-provided or pre-defined "candidate drug combination list" is iteratively processed, and the list contains drug combinations whose potential needs to be evaluated, such as [(entecavir, adefovir), (tenofovir, lamivudine), (entecavir, telbivudine)], etc. For each drug combination in the list (for example, when processing the combination (entecavir, adefovir)), first identify and extract the unique identifiers of all individual drug entities contained in the combination (i.e., entecavir ID and adefovir ID), then use these drug identifiers to query the "virus strain drug sensitivity score list" generated in the previous step, and find out the sensitivity score values of these individual drugs (entecavir, adefovir) corresponding to the current target hepatitis B virus variant. and At the same time, based on the same drug identifier, the complete feature vectors Vector(entecavir) and Vector(adefovir) corresponding to each drug (entecavir, adefovir) are retrieved and extracted from the "multi-task hepatitis B entity feature vector" set. The identifier of each candidate drug combination, the sensitivity score of each drug contained in it to the target virus, and the feature vector of each drug are integrated together to create a record for each combination. The records of all combinations are summarized to generate a drug combination feature-score dataset.

[0060] Based on the "drug combination feature-scoring dataset" generated in the previous step, the antagonistic effect of each drug combination is evaluated. The specific operations are as follows: For each drug combination record in the dataset, first determine all possible drug pairings in the combination (for binary combinations such as (A, B), there is only one pair (A, B); for ternary combinations (A, B, C), there are three pairs (A, B), (A, C), (B, C)), and then perform the following judgment process for each drug pair (for example, drug A and drug B): The first step is to calculate the cosine similarity between the feature vector Vector(A) of drug A and the feature vector Vector(B) of drug B. This calculation is achieved by dividing the vector dot product by the vector modulus product to obtain a value between -1 and 1. , indicating the degree of similarity between the two in the feature space; the second step is to compare the calculated cosine similarity with a "preset mechanism overlap threshold", which is used to determine whether there is a high mechanism or target similarity between the drugs. Its setting can be based on the feature vector similarity distribution analysis of known antagonistic (especially competitive antagonistic) and non-antagonistic drug pairs, and select a value that can better distinguish between these two types of situations, such as 0.8. This value is determined by analyzing the observation that the vector similarity of competitive antagonistic drug pairs annotated in literature reports or databases (such as DrugComb) is usually higher than 0.8, while the similarity of synergistic or unrelated drug pairs is lower; the third step is to check the sensitivity scores of the two drugs (A and B) in the drug pair to the target virus variant. and Whether both drugs are above a "pre-set single-drug efficacy threshold" is used to ensure that their antagonistic effects are only considered when both drugs have a certain degree of efficacy. The threshold setting can refer to the overall distribution of sensitivity scores or clinical judgment criteria. For example, if the score range is -1 to 1, the effectiveness threshold can be set to 0.5, indicating a prediction of medium sensitivity or above. This value is determined by analyzing the distribution of effective drug scores in historical data. For example, if the 60th percentile of the score distribution is 0.48, it is set to 0.5. The fourth step is to determine the antagonistic effect. If and only if a drug pair meets the above two conditions simultaneously, that is, "the cosine similarity is above the preset mechanism overlap threshold of 0.8" and "the sensitivity scores of both drugs are above the preset single-drug efficacy threshold of 0.5", the drug pair is determined to have potential target competitive antagonism. The determination result (for example, marking the drug pair as 'antagonistic') is recorded. After completing the determination for all drug pairs in a combination, the information of all drug pairs marked as 'antagonistic' is collected to generate an antagonistic effect marker set.

[0061] Based on the "antagonistic effect marker set" generated in the previous step, the original "candidate drug combination list" is screened and sorted. First, each drug combination in the "candidate drug combination list" is traversed to check whether there is any drug pair in the combination that is marked as having an antagonistic effect by the "antagonistic effect marker set". If a combination contains at least one drug pair marked as antagonistic, the combination is removed from the candidate list, and the remaining combination is the drug combination that is predicted to have no obvious target competitive antagonism risk. Then, the remaining drug combinations after screening are sorted. The basis for sorting can be the comprehensive performance of the sensitivity scores of each drug in the combination to the target virus variant, for example, calculating the sensitivity scores of all drugs in each remaining combination. The average value or sum of the comprehensive scores is sorted in descending order from high to low according to the comprehensive score. The higher the comprehensive score, the greater the overall predictive potential of the combination. The drug combination list after eliminating antagonism risks and sorting is used as the final evaluation output to generate the hepatitis B drug combination potential evaluation results.

[0062] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A multi-task hepatitis B drug screening method based on knowledge graph assistance, characterized in that: The following steps are involved: Based on the hepatitis B virus genotype sequence data recorded over time, patient drug use records and drug sensitivity, a temporal connection relationship between entities is established to generate a temporal hepatitis B knowledge graph structure; Based on the temporal hepatitis B knowledge graph structure, combined with input biological pathway annotation information, protein interaction data, and known drug combination effect records, identify the set of multiple entities that exist in the temporal hepatitis B knowledge graph structure that act together, define hyperedges connecting the multiple entity sets, obtain the multi-entity interaction set and hyperedge definition, and based on the multi-entity interaction set and hyperedge definition, expand the connection mode of the nodes in the temporal hepatitis B knowledge graph structure to establish a hepatitis B knowledge base that integrates hyperedge interactions; Based on the hepatitis B knowledge base integrated with hyperedge interaction, the initial vector expression of each hepatitis B virus entity, drug entity, and gene entity is calculated to obtain an initial entity vector representation. Based on the initial entity vector representation, the vector expression is updated and adjusted by aggregating the neighborhood node information and the connection information of different types of edges in the graph to establish a multi-task hepatitis B entity feature vector; Based on the multi-task hepatitis B entity feature vector, the feature vector corresponding to the target hepatitis B virus variant and the feature vector corresponding to the candidate drug are selected, the drug sensitivity score between the two is estimated, and a virus strain drug sensitivity score list is obtained. Based on the virus strain drug sensitivity score list and the multi-task hepatitis B entity feature vector corresponding to the candidate drug combination, the antagonistic effect of the drug combination is evaluated to obtain the hepatitis B drug combination potential evaluation result.

2. The multi-task hepatitis B drug screening method based on knowledge graph assistance according to claim 1 is characterized in that: The steps for obtaining the time series hepatitis B knowledge graph structure are: Integrate hepatitis B virus genotype sequence data, patient drug use records, and drug sensitivity data, extract the timestamp information of each data item, convert the timestamp information into a numerical time label in a unified time format, and generate a timestamp-related dataset; Based on the timestamp association data set, the interval duration between time tags is calculated, continuous time periods are divided according to the interval duration and a preset time window threshold, an association edge is established between the hepatitis B virus genotype entity and the drug entity in each time period, the weight value of the time association edge is calculated, and a weighted time association edge set is generated; Based on the weighted time-related edge set, the related edges with weight values greater than the preset edge weight threshold are connected to the corresponding entities to form a time-series hepatitis B knowledge graph structure.

3. The multi-task hepatitis B drug screening method based on knowledge graph assistance according to claim 1 is characterized in that: The steps for obtaining the multi-entity interaction set and hyperedge definition are: Integrate the entity nodes, biological pathway annotation information, protein interaction data and drug combination effect records in the time-series hepatitis B knowledge graph structure, extract the gene regulatory pathway identifiers in the biological pathway annotation information and the target gene set identifiers corresponding to the drugs in the drug combination effect records, and generate a multi-source data set; Based on the multi-source data set, calculating the co-action strength of the gene entity set and the drug target gene set in the same time window; Based on the co-action strength, a gene set and a drug target gene set whose co-action strength is greater than a preset threshold are selected, a hyperedge connecting the sets is defined, and a multi-entity interaction set and a hyperedge definition are generated.

4. The multi-task hepatitis B drug screening method based on knowledge graph assistance according to claim 1 is characterized in that: The steps for obtaining the hepatitis B knowledge base integrated with hyperedge interaction are as follows: Extract the node connection relationship between the multi-entity interaction set and the hyperedge definition and the time-series hepatitis B knowledge graph structure, traverse each hyperedge in the hyperedge definition, establish a bidirectional connection relationship between the hyperedge and all nodes in the corresponding entity set, and generate a hyperedge connection relationship table; Based on the hyperedge connection relationship table, detecting whether there are isolated nodes or redundant connections in the hyperedge connection relationship, if there are isolated nodes, supplementing the missing connections according to the protein interaction data, if there are redundant connections, merging the repeated hyperedge connections according to a preset redundancy threshold, and generating a verified hyperedge connection relationship table; Based on the verified hyperedge connection relationship table, the hyperedge connection relationship is mapped and replaced with the original node connection mode in the time-series hepatitis B knowledge graph structure, and the connection type and attributes between nodes are updated to form a hepatitis B knowledge base that integrates hyperedge interactions.

5. The multi-task hepatitis B drug screening method based on knowledge graph assistance according to claim 1 is characterized in that: The steps for obtaining the initial entity vector representation are: Traversing the hepatitis B virus entity, drug entity, and gene entity nodes in the hepatitis B knowledge base integrated with the hyperedge interaction, extracting the interaction event timestamp recorded in the time association edge of each node, counting the total number of times each node is associated with a multi-entity interaction set in the hyperedge information, recording the interaction event timestamp sequence and the number of hyperedge associations, and generating a node time series interaction log; Based on the node time-series interaction log, taking the first interaction timestamp as the benchmark, calculate the time difference between each subsequent timestamp and the benchmark to generate a time difference sequence, count the combined frequency of each node appearing simultaneously with other entities in the hyperedge information, calculate the ratio of the combined frequency to the length of the time difference sequence, and generate a set of multi-entity co-occurrence intensity factors; Based on the multi-entity co-occurrence intensity factor set, the time difference sequence is divided into windows with a period of 30 days, the mean and variance of the time difference in each window are calculated, the three sets of values of mean, variance and co-occurrence intensity factor are spliced, and the initial entity vector representation is output.

6. The multi-task hepatitis B drug screening method based on knowledge graph assistance according to claim 1 is characterized in that: The steps for obtaining the multi-task hepatitis B entity feature vector are: Traversing each entity node in the initial entity vector representation, extracting a set of directly connected neighboring nodes in the fusion hyperedge-interacted hepatitis B knowledge base, recording the entity type, edge type, and number of connections of the neighboring nodes, and generating a neighboring node information set; Based on the neighborhood node information set, counting the proportion of connections of different edge types in the neighborhood of each entity node, and multiplying the proportion value by a preset priority coefficient of the neighborhood node entity type to generate an edge type weight factor set; Based on the edge type weight factor set, the initial entity vector representation and the entity vector in the neighborhood node information set are weighted and spliced according to the weight factor to generate a multi-task hepatitis B entity feature vector.

7. The multi-task hepatitis B drug screening method based on knowledge graph assistance according to claim 1 is characterized in that: The steps for obtaining the virus strain drug sensitivity score list are: Traversing the target hepatitis B virus variants and candidate drug entities in the multi-task hepatitis B entity feature vector, extracting the virus variant feature vector, which includes the expression intensity and time correlation intensity of the gene mutation site, and the drug feature vector, which includes the target effect intensity and metabolic half-life, to generate a set of virus-drug feature vector pairs; Based on the set of virus-drug feature vector pairs, the sensitivity scores of virus and drug feature vectors are calculated; Based on the sensitivity scores, the strains are sorted from high to low according to the sensitivity scores to generate a list of drug sensitivity scores of the virus strains.

8. The multi-task hepatitis B drug screening method based on knowledge graph assistance according to claim 1 is characterized in that: The steps for obtaining the hepatitis B drug combination potential evaluation results are: Traverse each combination in the candidate drug combination list, extract the identifiers of all drug entities in the combination, obtain the sensitivity score of each drug to the target virus variant from the virus strain drug sensitivity score list, and simultaneously extract the feature vector of the drug entity from the multi-task hepatitis B entity feature vector to generate a drug combination feature-score dataset; Based on the drug combination feature-score dataset, the cosine similarity between the feature vectors of each pair of drugs in the combination is calculated. If the cosine similarity of any two drugs in the combination is higher than the preset mechanism overlap threshold and the sensitivity scores of both drugs are higher than the preset single-drug effectiveness threshold, the drug pair is determined to have a target competitive antagonistic effect, and an antagonistic effect marker set is generated; Based on the antagonistic effect marker set, combinations containing at least one antagonistic drug pair are eliminated, and the remaining combinations are sorted to generate a hepatitis B drug combination potential evaluation result.

9. The hepatitis B drug screening system based on the knowledge graph-assisted multi-task hepatitis B drug screening method according to any one of claims 1 to 8, characterized in that: include: The time series graph construction module establishes the temporal connection relationship between entities based on the hepatitis B virus genotype sequence data recorded over time, the patient's drug use records and drug sensitivity, and generates a time series hepatitis B knowledge graph structure; A hyperedge fusion module, based on the temporal hepatitis B knowledge graph structure, combines input biological pathway annotation information, protein interaction data, and known drug combination effect records to identify a set of entities that act together in the temporal hepatitis B knowledge graph structure, define hyperedges connecting multiple entity sets, obtain multi-entity interaction sets and hyperedge definitions, and based on the multi-entity interaction sets and hyperedge definitions, expand the connection mode of nodes in the temporal hepatitis B knowledge graph structure to establish a hepatitis B knowledge base that integrates hyperedge interactions; A representation learning module calculates the initial vector expression of each hepatitis B virus entity, drug entity, and gene entity based on the hepatitis B knowledge base integrated with hyperedge interactions, obtains an initial entity vector representation, and updates and adjusts the vector expression based on the initial entity vector representation by aggregating neighborhood node information and connection information of different types of edges in the graph to establish a multi-task hepatitis B entity feature vector; The effect prediction module selects the feature vector corresponding to the target hepatitis B virus variant and the feature vector corresponding to the candidate drug based on the multi-task hepatitis B entity feature vector, estimates the drug sensitivity score between the two, obtains a list of virus strain drug sensitivity scores, and evaluates the antagonistic effect of the drug combination based on the virus strain drug sensitivity score list and the multi-task hepatitis B entity feature vector corresponding to the candidate drug combination to obtain the hepatitis B drug combination potential evaluation result.

Images