Plateau sickness drug screening method and device based on virtual mitochondria modeling and disturbance prediction, equipment and medium

Abstract

The application discloses a high altitude disease drug screening method and device based on virtual mitochondria modeling and disturbance prediction, equipment and medium, comprising: carrying out heterogeneous graph Transformer pre-training based on a mitochondria specific heterogeneous knowledge graph to generate a graph embedding; constructing a double-track neural ordinary differential equation model based on the graph embedding and disease control condition labels, and obtaining a track model through multi-objective joint optimization training; calculating a screening score of track back pull caused by candidate gene disturbance, and screening a candidate intervention target set according to the screening score; grading the candidate intervention target through mechanism reasoning and external evidence verification; and obtaining a candidate drug screening result based on the grading evaluation result of the candidate intervention target and the mitochondria specific heterogeneous knowledge graph. The application belongs to the field of high altitude disease drug screening. The application provides a more accurate screening method for mitochondrial damage.

Description

Technical Field

[0001] This invention relates to the field of altitude sickness drug screening, and in particular to altitude sickness drug screening methods, devices, equipment and media based on virtual mitochondrial modeling and perturbation prediction. Background Technology

[0002] Altitude sickness is a group of diseases caused by low-pressure, hypoxic environments, and its common pathological basis is closely related to mitochondrial dysfunction. Under hypoxic conditions, the mitochondrial electron transport chain is damaged, oxidative phosphorylation capacity decreases, and reactive oxygen species accumulate, mitochondrial DNA is damaged, calcium homeostasis is abnormal, and mitophagy is disordered, ultimately leading to an imbalance in cellular energy metabolism. Therefore, screening for effective intervention targets and candidate drugs targeting the mitochondrial damage process is of great significance for the prevention and treatment of altitude sickness.

[0003] Existing computationally assisted drug screening methods primarily rely on whole-cell or whole-transcriptome data, using virtual cell models to predict perturbation responses or trajectory inference methods to describe disease state changes. However, these methods typically lack specific modeling for mitochondria, a crucial organelle, leading to mitochondrial-related signals being easily diluted by numerous irrelevant genes. Furthermore, some methods only predict single perturbation responses or depend on pseudo-time to describe state changes, making it difficult to characterize disease evolution and determine the optimal intervention window over the actual exposure timeline. In addition, existing continuous dynamic models often fail to distinguish between disease and normal evolutionary trajectories, making them susceptible to non-specific expression changes. Moreover, biological priors are often introduced only during the results analysis phase, lacking mechanistic constraints throughout the model training process, resulting in insufficient reliability of screening results. Therefore, providing a more accurate screening method for mitochondrial damage is an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method, device, equipment, and medium for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction. This solves the technical problem of insufficient reliability in screening for mitochondrial damage in the prior art and achieves the technical effect of improving the accuracy of screening for mitochondrial damage.

[0005] In a first aspect, this invention provides a method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction, comprising: A mitochondrial-specific heterogeneous knowledge graph is constructed, and a heterogeneous graph Transformer is pre-trained based on the mitochondrial-specific heterogeneous knowledge graph to generate graph embeddings. Based on map embedding and disease control condition labels, a dual-trajectory neural frequent differential equation model was constructed, and the trajectory model was obtained through multi-objective joint optimization training. The trajectory model was used to predict the dynamic evolution of mitochondrial genes in disease and normal states. Based on the trajectory model, perturbation analysis is performed on candidate genes, the screening score of trajectory pullback caused by candidate gene perturbation is calculated, and a set of candidate intervention targets is selected based on the screening score. The perturbation response characteristics of each candidate intervention target in the candidate intervention target set are fused with the mitochondrial-specific heterogeneous knowledge graph, and the candidate intervention targets are graded and evaluated through mechanism reasoning and external evidence verification. Based on the graded evaluation results of candidate intervention targets and the mitochondrial-specific isomorphism knowledge graph, candidate drug screening results were obtained.

[0006] Furthermore, a mitochondrial-specific heterogeneous knowledge graph is constructed, and a heterogeneous graph Transformer is pre-trained based on the mitochondrial-specific heterogeneous knowledge graph to generate graph embeddings, including: Obtain correlation data of genes, proteins, pathways, complexes and drugs from multiple mitochondrial-related biological knowledge databases; Entity identification alignment, duplicate relationship removal, and conflict relationship filtering are performed on the associated data to obtain entities and relationships related to mitochondrial function, and a mitochondrial-specific heterogeneous knowledge graph is constructed. Based on mitochondrial-specific heterogeneous knowledge graphs, a self-supervised pre-trained model is constructed using the heterogeneous graph Transformer, and graph embeddings corresponding to each candidate gene are generated through graph structure relationships.

[0007] Furthermore, based on atlas embedding and disease control condition labels, a dual-trajectory neural frequent differential equation model is constructed, and the trajectory model is obtained through multi-objective joint optimization training, including: Obtain the atlas embedding and disease control condition labels, and input the atlas embedding, time information, and disease control condition labels into the dual-track neural frequent differential equation model. By employing temporal conditional embedding, knowledge graph semantic fusion, latent variable encoding, and conditional neuron frequent differential equation evolution, we can learn the dynamic evolution of mitochondrial genes under disease and normal conditions. Multi-objective joint optimization training is performed based on reconstruction loss, latent variable constraint loss, temporal consistency loss, distribution alignment loss, and future state prediction loss to optimize the dual-trajectory neural ordinary differential equation model and obtain the trajectory model.

[0008] Furthermore, based on the trajectory model, perturbation analysis is performed on candidate genes to calculate the screening score of trajectory pullback caused by candidate gene perturbation. Based on the screening score, a set of candidate intervention targets is selected, including: The candidate gene is input into the trajectory model. By simulating the dynamic evolution of gene expression after the candidate gene is perturbed, the degree of change in the disease trajectory pulling back to the normal trajectory before and after the perturbation is calculated, and the screening score of the candidate gene is obtained. The screening scores of several candidate genes under multiple time windows are sorted and filtered to obtain a set of candidate intervention targets.

[0009] Furthermore, the perturbation response characteristics corresponding to each candidate intervention target in the candidate intervention target set are fused with the mitochondrial-specific isomorphic knowledge graph, and the candidate intervention targets are graded and evaluated through mechanism reasoning and external evidence verification, including: The perturbation response features corresponding to each candidate intervention target are obtained, and the perturbation response features are fused with the mitochondrial specific isomorphism knowledge graph to generate a structured evaluation of the candidate intervention target. Mechanism reasoning is performed based on structured evaluation to obtain the mechanism information corresponding to the candidate intervention target, and the information is verified by external evidence to obtain the comprehensive evaluation result of the candidate intervention target. Based on the comprehensive evaluation results, the candidate intervention targets are graded and evaluated to obtain the graded evaluation results of the candidate intervention targets.

[0010] Furthermore, based on the graded evaluation results of candidate intervention targets and the mitochondrial-specific isomorphism knowledge graph, candidate drug screening results were obtained, including: Based on the mitochondrial-specific isomer knowledge graph, the target information of candidate drugs was determined. Based on the target information of candidate drugs and the graded evaluation results of candidate intervention targets, candidate drugs are evaluated and ranked to obtain candidate drug screening results.

[0011] Furthermore, the mitochondrial-specific isomer knowledge graph includes: Gene nodes, protein nodes, pathway nodes, complex nodes, and drug nodes.

[0012] Secondly, this invention provides a high-altitude sickness drug screening device based on virtual mitochondrial modeling and perturbation prediction, comprising: The graph construction module is used to construct a mitochondrial-specific heterogeneous knowledge graph and perform heterogeneous graph Transformer pre-training based on the mitochondrial-specific heterogeneous knowledge graph to generate graph embeddings. The trajectory model training module is used to construct a dual-trajectory neural frequent differential equation model based on map embedding and disease control condition labels, and obtain the trajectory model through multi-objective joint optimization training. The trajectory model is used to predict the dynamic evolution of mitochondrial genes in disease and normal states. The intervention target set module is used to perform perturbation analysis on candidate genes based on trajectory models, calculate the screening score of trajectory pullback caused by candidate gene perturbation, and select the candidate intervention target set based on the screening score. The hierarchical evaluation module is used to fuse the perturbation response characteristics of each candidate intervention target in the candidate intervention target set with the mitochondrial specific isomorphic knowledge graph, and to perform hierarchical evaluation of the candidate intervention targets through mechanism reasoning and external evidence verification. The drug screening module is used to obtain candidate drug screening results based on the graded evaluation results of candidate intervention targets and the mitochondrial-specific isomorphism knowledge graph.

[0013] Thirdly, the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute a method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction, as provided in the first aspect.

[0014] Fourthly, the present invention provides a non-transitory computer-readable storage medium that, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to perform the altitude sickness drug screening method based on virtual mitochondrial modeling and perturbation prediction as provided in the first aspect.

[0015] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention proposes a virtual mitochondrial modeling and drug screening method targeting the mitochondrial damage mechanism of altitude sickness. By refining the modeling granularity from the traditional whole-cell and whole-transcriptome levels to the mitochondrial organelle level, a differentiable virtual mitochondrial model is constructed that reflects the dynamic changes of mitochondria during altitude exposure, improving the targeting and accuracy of disease mechanism characterization. Based on a disease-control dual-trajectory neural common differential equation dynamic model with a real timeline, the disease state and the normal state share the main dynamic evolutionary framework and achieve conditional bifurcation, effectively distinguishing between disease-specific changes and general expression fluctuations. Furthermore, by calculating the trajectory pullback amplitude through latent spatial gradient perturbation, quantitative screening of candidate genes and drugs is achieved, and the optimal intervention window can be output across multiple exposure time windows, providing a temporal dimension basis for precise intervention of altitude sickness.

[0016] This invention integrates a mitochondrial-specific isomorphic knowledge graph throughout the entire process of model training, perturbation analysis, and result evaluation. By leveraging the prior knowledge graph, the model's biological semantic understanding capabilities are enhanced, enabling screening results to be mapped to specific genes, pathways, complexes, and drug-drug interactions. Furthermore, by combining knowledge graph-anchored mechanism reasoning with cross-validation using external independent databases, automatic credibility grading of candidate targets and drugs is achieved, reducing the risk of false positives while preserving potential innovative intervention signals. In addition, this method exhibits excellent transferability and scalability; simply replacing the time-series data and disease association information under different high-altitude exposure or hypoxia scenarios allows it to be used for target discovery and drug screening for various hypoxia-related diseases. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the altitude sickness drug screening method based on virtual mitochondrial modeling and perturbation prediction provided by the present invention. Figure 2 The flowchart and core module structure of the screening method provided by this invention are shown in the figure. Detailed Implementation

[0019] This invention provides a method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction, which solves the technical problem of insufficient reliability in screening for mitochondrial damage in existing technologies.

[0020] The technical solution of this invention is to solve the above-mentioned technical problems, and the overall idea is as follows: A method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction includes: constructing a mitochondrial-specific isomorphic knowledge graph and performing isomorphic graph Transformer pre-training based on the mitochondrial-specific isomorphic knowledge graph to generate graph embeddings; constructing a dual-trajectory neural frequent differential equation model based on the graph embeddings and disease control condition labels, and obtaining a trajectory model through multi-objective joint optimization training, wherein the trajectory model is used to predict the dynamic evolution of mitochondrial genes in disease and normal states; performing perturbation analysis on candidate genes based on the trajectory model, calculating the screening score of trajectory pullback caused by perturbation of candidate genes, and selecting a set of candidate intervention targets based on the screening score; fusing the perturbation response features corresponding to each candidate intervention target in the set of candidate intervention targets with the mitochondrial-specific isomorphic knowledge graph, and classifying and evaluating the candidate intervention targets through mechanism reasoning and external evidence verification; and obtaining candidate drug screening results based on the classification and evaluation results of candidate intervention targets and the mitochondrial-specific isomorphic knowledge graph.

[0021] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0022] First, it should be clarified that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0023] All hyperparameters or preset thresholds that may be involved in this invention can be determined based on experiments or historical experience, and this invention does not impose any limitations on them.

[0024] This invention provides, for example Figure 1 The altitude sickness drug screening method based on virtual mitochondrial modeling and perturbation prediction shown includes steps S11-S15: Step S11: Construct a mitochondrial-specific heterogeneous knowledge graph, and perform heterogeneous graph Transformer pre-training based on the mitochondrial-specific heterogeneous knowledge graph to generate graph embeddings.

[0025] A mitochondrial-specific heterogeneous knowledge graph was constructed, and a heterogeneous graph Transformer was pre-trained based on the mitochondrial-specific heterogeneous knowledge graph to generate graph embeddings. This process included: acquiring association data of genes, proteins, pathways, complexes, and drugs from multiple mitochondrial-related biological knowledge databases; performing entity label alignment, duplicate relationship removal, and conflict relationship filtering on the association data to screen out entities and associations related to mitochondrial function, and constructing the mitochondrial-specific heterogeneous knowledge graph; and using the heterogeneous graph Transformer to construct a self-supervised pre-trained model based on the mitochondrial-specific heterogeneous knowledge graph, and generating graph embeddings corresponding to each candidate gene through graph structure relationship learning.

[0026] The mitochondrial-specific isomorphism knowledge graph includes: gene nodes, protein nodes, pathway nodes, complex nodes, and drug nodes.

[0027] Specifically, it includes: The study retrieved correlation data of genes, proteins, pathways, complexes and drugs from multiple mitochondrial-related biological knowledge databases, including MitoCarta, Reactome, CORUM, DrugBank and STRING databases. Based on entity naming rules in different databases, the obtained gene symbols, protein identifiers, drug identifiers, and pathway complex identifiers are uniformly mapped and aligned with entity identifiers to eliminate the problem of multiple representations of the same entity caused by naming differences between different databases. Further, duplicate relationships are removed and conflicting relationships are filtered from the associated data. The same associated edges of duplicate records in multiple databases are removed, and the associated relationships that have contradictory directions, inconsistent functions, or cannot be mapped to the mitochondrial functional context are filtered out, thereby obtaining effective entities and associated relationships related to mitochondrial functional regulation.

[0028] Based on the screened entities and their relationships, a mitochondrial-specific isomorphic knowledge graph is constructed. The mitochondrial-specific isomorphic knowledge graph includes gene nodes, protein nodes, pathway nodes, complex nodes, and drug nodes. Five types of heteromorphic relationship edges are established: gene-gene, gene-pathway, gene-complex, drug-target, and pathway-pathway. This allows the knowledge graph to simultaneously express mitochondrial gene regulation relationships, functional pathway relationships, protein complex composition relationships, and drug intervention relationships. In this invention, the constructed mitochondrial-specific heterogeneous knowledge graph includes approximately 2,549 nodes and approximately 23,019 relational edges. Nodes are used to represent different types of mitochondrial-related biological entities, and relational edges are used to represent the biological relationships between different entities.

[0029] The mitochondrial-specific isomorphic knowledge graph is input into the isomorphic graph Transformer model. Through the feature aggregation mechanism designed for different node types and different relationship types, the association information between gene / protein nodes and their neighborhood heteromorphic nodes such as pathways, complexes, and drugs is encoded. High-confidence gene association edges in the STRING database are used as positive samples, and non-edges are randomly sampled from node pairs that do not have direct connections as negative samples. Link prediction training data is constructed according to a preset ratio. The model parameters are optimized through a self-supervised link prediction task, enabling the model to learn the potential structural relationships in the mitochondrial-specific heterogeneous knowledge graph. After pre-training, a corresponding (64)-dimensional map embedding is generated for each candidate gene to characterize the structural location, functional association and potential regulatory relationship of the gene in the mitochondrial functional network.

[0030] Step S12: Based on the map embedding and disease control condition labels, a dual-trajectory neural frequent differential equation model is constructed, and the trajectory model is obtained through multi-objective joint optimization training. The trajectory model is used to predict the dynamic evolution of mitochondrial genes in disease and normal states. This includes: acquiring atlas embeddings and disease control condition labels, and inputting the atlas embeddings, temporal information, and disease control condition labels into a dual-trajectory neural frequent differential equation model. Through temporal condition embedding, knowledge graph semantic fusion, latent variable encoding, and conditional neural frequent differential equation evolution, the dynamic evolution of mitochondrial genes in disease and normal states is learned. Based on reconstruction loss, latent variable constraint loss, temporal consistency loss, distribution alignment loss, and future state prediction loss, multi-objective joint optimization training is performed to optimize the dual-trajectory neural frequent differential equation model and obtain the trajectory model.

[0031] The specific form of the loss function can be the sum of the losses of each term, with the weights of each term appended before each loss term.

[0032] Specifically, it includes: Obtain the 64-dimensional map embedding corresponding to the candidate gene and the disease control condition label. The disease control condition label is used to characterize the current state of the sample, with a value of 0 or 1, corresponding to the normal state and the disease state, respectively. Simultaneously, the expression vectors of mitochondrial-related genes at different sampling times and the corresponding normalized timestamps (i.e., time information) are obtained, and the map embedding, gene expression vectors, time information, and disease control condition labels are jointly input into the dual-track neural ordinary differential equation model. (a) Temporally conditional gene embedding: The temporal conditional embedding module is used to dynamically enhance the map embedding. The temporal correction term generated by the sinusoidal position encoding is superimposed onto the original map embedding in a residual manner, so that the same gene can form a time-discriminative gene representation under different time windows. (b) Knowledge graph semantic fusion gating: Furthermore, continuous gating weights are generated based on temporal conditional embedding, and gene expression features are dynamically reweighted channel by channel through a knowledge graph semantic fusion gating mechanism. This enables the model to combine global semantic information in the mitochondrial-specific heterogeneous knowledge graph and adaptively screen mitochondrial-related gene information that plays a key regulatory role in different disease evolution stages.

[0033] (c) Variational encoder: Based on gated gene expression features and disease control condition variables, the input features are mapped to the latent variable space through a variational encoder to obtain the mean and variance of the latent variables, and a continuous latent state representation is generated based on the reparameterized sampling method. (d) Conditional constant differential equation: Furthermore, the hidden state is input into the conditional neuron ordinary differential equation module, with real time as the continuous evolution variable, and the global context information provided by the mitochondrial-specific heterogeneous knowledge graph and the disease control condition labels are used to parameterize the dynamic function, so that the disease state trajectory and the normal state trajectory share the same dynamic evolution backbone, and only the condition-related direction produces state bifurcation. (e) Decoder: During the continuous evolution of the hidden state, the fourth-order Runge-Kutta integral method is used to numerically solve the neural ordinary differential equation to obtain the evolution results of the hidden state at different time points. The hidden state is then combined with the temporal conditional embedding, the global context of the graph, and the conditional embedding through the decoder to reconstruct the mitochondrial gene expression state corresponding to the target time point, thereby learning the dynamic evolution law of mitochondrial genes under disease and normal states.

[0034] The joint optimization objectives include gene expression remodeling loss, latent variable constraint loss, temporal consistency loss, distribution alignment loss, and future state prediction loss; Gene expression reconstruction loss is used to constrain the model to restore the gene expression state at the real time point; latent variable constraint loss is used to restrict the distribution of latent variables generated by the variational encoder to conform to the preset probability distribution; temporal consistency loss is used to maintain the continuous evolution characteristics of gene expression trajectory in disease or normal states; distribution alignment loss is used to constrain the distribution difference of samples in different states in the latent space through maximum mean difference (MMD); and future state prediction loss is used to improve the model's ability to predict the trend of gene expression changes at subsequent time points.

[0035] By updating the parameters of the dual-trajectory neural frequent differential equation model through the above multi-objective joint optimization method, the trajectory model can predict the dynamic evolution of mitochondrial genes in both disease and normal states.

[0036] Step S13: Perform perturbation analysis on candidate genes based on trajectory model, calculate the screening score of trajectory pullback caused by candidate gene perturbation, and select a set of candidate intervention targets based on the screening score. This includes: inputting candidate genes into a trajectory model, simulating the dynamic evolution of gene expression after perturbation of candidate genes, calculating the degree of change in the disease trajectory pulling back to the normal trajectory before and after perturbation, and obtaining the screening score of the candidate gene; sorting and screening the screening scores of several candidate genes under multiple time windows to obtain a set of candidate intervention targets.

[0037] Specifically, it includes: The trained trajectory model is used as a differentiable dynamic simulator. The candidate gene to be evaluated is input into the trajectory model, and small gradient perturbations are applied along the expression feature channels corresponding to the candidate gene in multiple preset time windows or differentiation time windows to simulate the expression change process of the candidate gene after being subjected to external intervention. The perturbation signal is input into the trajectory model through a complete forward propagation process, which allows the perturbation information to pass through time-conditional embedding, knowledge graph semantic fusion, latent variable encoding, continuous evolution of conditional neural ordinary differential equations, and decoding and reconstruction processes in sequence, thereby obtaining the dynamic evolution trajectory of the disease state under the perturbation effect. By comparing the changes in trajectory correlation between the disease trajectory and the normal trajectory before and after the perturbation, the degree of change Δr caused by the perturbation to pull the disease trajectory back towards the normal state is calculated, the trajectory pullback amplitude corresponding to the candidate gene is obtained, and the trajectory pullback amplitude is used as the screening score of the candidate gene.

[0038] Furthermore, perturbation analysis was performed on the same candidate gene under multiple time windows, and the screening score corresponding to each time window was calculated. The optimal intervention time window for the candidate gene was determined based on the changing trend of the screening score under different time windows. In this model, the conditional neural frequent differential equations are parameterized based on disease control condition labels, allowing disease state trajectories and normal state trajectories to share the same dynamic evolution backbone and branch along condition-related directions. Therefore, perturbations of the same candidate gene can act on both disease and normal trajectories for dynamic tracking. By comparing the differences in perturbation responses under the two conditions, the regulatory role of candidate genes in disease-specific dynamic processes can be identified, thereby reducing the impact of non-specific expression fluctuations and general expression noise on screening results.

[0039] Based on the screening scores obtained by multiple candidate genes at different time windows, all candidate genes are sorted and screened, and a predetermined number of candidate genes with the highest screening scores are selected to form a set of candidate intervention targets. Among them, the higher the screening score, the greater the recovery of the dynamic evolution trajectory of mitochondrial genes in the disease state to the normal state after the corresponding candidate gene is disturbed, indicating that the candidate gene has a higher potential intervention value in the process of disease-related mitochondrial dysfunction.

[0040] Step S14: The perturbation response features corresponding to each candidate intervention target in the candidate intervention target set are fused with the mitochondrial specific isomorphic knowledge graph, and the candidate intervention targets are graded and evaluated through mechanism reasoning and external evidence verification. The process includes: acquiring the perturbation response features corresponding to each candidate intervention target, fusing the perturbation response features with the mitochondrial-specific isomorphic knowledge graph to generate a structured evaluation of the candidate intervention target; performing mechanism reasoning based on the structured evaluation to obtain the mechanism of action information corresponding to the candidate intervention target, verifying it with external evidence to obtain a comprehensive evaluation result of the candidate intervention target; and conducting a graded evaluation of the candidate intervention target based on the comprehensive evaluation result to obtain a graded evaluation result of the candidate intervention target.

[0041] Specifically, it includes: Obtain the perturbation response characteristics corresponding to each candidate intervention target. The perturbation response characteristics include the perturbation sensitivity of the candidate intervention target, the response difference between the disease state and the normal state, the gating weight in the trajectory model, and the trajectory pullback amplitude change curve under multiple time windows. The perturbation response features are fused with the mitochondrial-specific isomorphic knowledge graph to extract local knowledge subgraph information corresponding to each candidate intervention target. The local knowledge subgraph information includes the functional pathway to which the candidate intervention target belongs, the protein complex involved, the confidence interaction relationship with other genes or proteins, and the association with known targeted drugs. By jointly representing the perturbation response features with structural association information in the knowledge graph, a structured evaluation for evaluating candidate intervention targets is generated.

[0042] The perturbation response features and local knowledge subgraph structures corresponding to candidate intervention targets are input into the mechanism reasoning model to generate a mechanism reasoning chain consisting of candidate intervention targets, related regulatory nodes, functional pathways and disease phenotypes. Each reasoning step in the mechanism reasoning chain is constrained to correspond to the real nodes and relationships in the mitochondrial specific isomorphic knowledge graph, thereby ensuring that the reasoning process is based on the existing biological knowledge structure. Further consistency checks were performed on the mechanism reasoning chain. The entities involved in the reasoning chain, such as genes, proteins, pathways, complexes, and drugs, were matched with the node information in the mitochondrial specific isomorphism knowledge graph. Reasoning content that could not be mapped to knowledge graph nodes or lacked relevant evidence was eliminated, and the mechanism of action information was obtained after knowledge constraints.

[0043] Furthermore, external evidence was used to verify the candidate intervention targets after mechanism verification. By introducing external database information that is independent of the source of the mitochondrial specific isomorphism knowledge graph, the reliability of the candidate intervention targets was evaluated. The study utilized the OpenTargets database to obtain association scores between candidate intervention targets and diseases, the PubMed database to obtain co-occurrence density of relevant research literature on candidate intervention targets, and the Ensembl database to obtain genotype and annotation quality information of candidate intervention targets. These multi-source external evidences were then weighted and fused to obtain a comprehensive external evaluation score for the candidate intervention targets. The comprehensive evaluation results were then graded according to preset grading thresholds to obtain graded evaluation results for candidate intervention targets, including HIGH, MEDIUM, and LOW levels. Candidate intervention targets with complete mechanistic reasoning and sufficient external evidence were marked as HIGH; those with reasonable mechanisms but insufficient external evidence were marked as MEDIUM and given priority in the experimental validation process; while LOW targets had their subsequent screening priority reduced.

[0044] Step S15: Based on the graded evaluation results of candidate intervention targets and the mitochondrial-specific isomorphism knowledge graph, obtain the candidate drug screening results.

[0045] This includes: determining the target information of candidate drugs based on mitochondrial-specific isomorphism knowledge graphs; evaluating and ranking candidate drugs based on the target information and the graded evaluation results of candidate intervention targets, and obtaining candidate drug screening results.

[0046] Specifically, it includes: Based on the mitochondrial specific isomorphism knowledge graph, the drug node information corresponding to the candidate compounds to be screened or known drugs is obtained, and the target information of each candidate drug is determined by the drug-target association in the knowledge graph. Among them, the target information is used to characterize the mitochondrial-related gene or protein nodes that the candidate drug can directly act on or regulate; Furthermore, the target information is matched with the set of candidate intervention targets to determine the mitochondrial intervention targets corresponding to the candidate drugs, and the perturbation response characteristics and graded evaluation results corresponding to the target are obtained, thus extending the screening process from candidate intervention targets to candidate drugs.

[0047] The trajectory pullback amplitude obtained by perturbation analysis of the target of the candidate drug is used as the basis for evaluating the drug intervention effect. By simulating the dynamic evolution process of mitochondrial genes after the target of the candidate drug is regulated, the ability of the candidate drug to promote the recovery of the trajectory from hypoxic mitochondrial damage to normal state is evaluated. Simultaneously, by combining the integrity of the mechanism of action obtained from mechanism reasoning in step S14 with the HIGH, MEDIUM, and LOW grading results obtained from external evidence verification, the reliability of the candidate drug's action is evaluated using a weighted approach, thereby obtaining the comprehensive score corresponding to the candidate drug.

[0048] Among them, candidate drugs whose targets can significantly reverse the disease trajectory and whose corresponding intervention targets have high mechanistic credibility and external evidence support are given higher evaluation weights.

[0049] Furthermore, the candidate drugs are ranked according to the comprehensive evaluation results to obtain the candidate drug screening results; Among them, the candidate drug screening results are used to characterize the potential intervention ability and priority of different candidate drugs on hypoxic mitochondrial damage.

[0050] By utilizing the drug-target edges in the mitochondrial-specific isomorphic knowledge graph to map the association between drugs and candidate intervention targets, and combining trajectory pullback and graded evaluation results, the screening of gene targets can be extended to the screening of candidate drugs.

[0051] Figure 2 The flowchart and core module structure of the screening method provided by this invention are shown in the figure.

[0052] In summary, this invention proposes a virtual mitochondrial modeling and drug screening method for mitochondrial damage mechanisms in altitude sickness. By refining the modeling granularity from the traditional whole-cell and whole-transcriptome levels to the mitochondrial organelle level, a differentiable virtual mitochondrial model reflecting the dynamic changes of mitochondria during altitude exposure is constructed, improving the targeting and accuracy of disease mechanism characterization. A disease-control dual-trajectory neural common differential equation dynamic model based on a real timeline allows disease and normal states to share the main dynamic evolutionary framework and achieve conditional bifurcation, effectively distinguishing between disease-specific changes and general expression fluctuations. Furthermore, by calculating the trajectory pullback amplitude through latent spatial gradient perturbation, quantitative screening of candidate genes and drugs is achieved, and the optimal intervention window can be output across multiple exposure time windows, providing a temporal dimension basis for precise intervention in altitude sickness.

[0053] This invention integrates a mitochondrial-specific isomorphic knowledge graph throughout the entire process of model training, perturbation analysis, and result evaluation. By leveraging the prior knowledge graph, the model's biological semantic understanding capabilities are enhanced, enabling screening results to be mapped to specific genes, pathways, complexes, and drug-drug interactions. Furthermore, by combining knowledge graph-anchored mechanism reasoning with cross-validation using external independent databases, automatic credibility grading of candidate targets and drugs is achieved, reducing the risk of false positives while preserving potential innovative intervention signals. In addition, this method exhibits excellent transferability and scalability; simply replacing the time-series data and disease association information under different high-altitude exposure or hypoxia scenarios allows it to be used for target discovery and drug screening for various hypoxia-related diseases.

[0054] Based on the same inventive concept, a high-altitude sickness drug screening device based on virtual mitochondrial modeling and perturbation prediction includes: The graph construction module is used to construct a mitochondrial-specific heterogeneous knowledge graph and perform heterogeneous graph Transformer pre-training based on the mitochondrial-specific heterogeneous knowledge graph to generate graph embeddings. The trajectory model training module is used to construct a dual-trajectory neural frequent differential equation model based on map embedding and disease control condition labels, and obtain the trajectory model through multi-objective joint optimization training. The trajectory model is used to predict the dynamic evolution of mitochondrial genes in disease and normal states. The intervention target set module is used to perform perturbation analysis on candidate genes based on trajectory models, calculate the screening score of trajectory pullback caused by candidate gene perturbation, and select the candidate intervention target set based on the screening score. The hierarchical evaluation module is used to fuse the perturbation response characteristics of each candidate intervention target in the candidate intervention target set with the mitochondrial specific isomorphic knowledge graph, and to perform hierarchical evaluation of the candidate intervention targets through mechanism reasoning and external evidence verification. The drug screening module is used to obtain candidate drug screening results based on the graded evaluation results of candidate intervention targets and the mitochondrial-specific isomorphism knowledge graph.

[0055] Based on the same inventive concept, the present invention also provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute a method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction, as described above.

[0056] Based on the same inventive concept, the present invention also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to execute the altitude sickness drug screening method based on virtual mitochondrial modeling and perturbation prediction as described above.

[0057] Since the electronic device described in this embodiment is an electronic device used to implement the information processing method in the embodiments of the present invention, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the information processing method described in the embodiments of the present invention. Therefore, how the electronic device implements the method in the embodiments of the present invention will not be described in detail here. Any electronic device used by those skilled in the art to implement the information processing method in the embodiments of the present invention falls within the scope of protection of the present invention.

[0058] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0059] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0060] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0061] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0062] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0063] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction, characterized in that, include: A mitochondrial-specific heterogeneous knowledge graph is constructed, and a heterogeneous graph Transformer is pre-trained based on the mitochondrial-specific heterogeneous knowledge graph to generate graph embeddings. Based on map embedding and disease control condition labels, a dual-trajectory neural frequent differential equation model was constructed, and the trajectory model was obtained through multi-objective joint optimization training. The trajectory model is used to predict the dynamic evolution of mitochondrial genes in disease and normal states. Based on the trajectory model, perturbation analysis is performed on candidate genes, the screening score of trajectory pullback caused by candidate gene perturbation is calculated, and a set of candidate intervention targets is selected based on the screening score. The perturbation response characteristics of each candidate intervention target in the candidate intervention target set are fused with the mitochondrial-specific heterogeneous knowledge graph, and the candidate intervention targets are graded and evaluated through mechanism reasoning and external evidence verification. Based on the graded evaluation results of candidate intervention targets and the mitochondrial-specific isomorphism knowledge graph, candidate drug screening results were obtained.

2. The method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction as described in claim 1, characterized in that, A mitochondrial-specific heterogeneous knowledge graph is constructed, and a heterogeneous graph Transformer is pre-trained based on the mitochondrial-specific heterogeneous knowledge graph to generate graph embeddings, including: Obtain correlation data of genes, proteins, pathways, complexes and drugs from multiple mitochondrial-related biological knowledge databases; The associated data is aligned with entity identifiers, duplicate relationships are removed, and conflict relationships are filtered to obtain entities and relationships related to mitochondrial function, and a mitochondrial-specific heterogeneous knowledge graph is constructed. Based on the mitochondrial-specific heterogeneous knowledge graph, a self-supervised pre-trained model is constructed using the heterogeneous graph Transformer, and graph embeddings corresponding to each candidate gene are generated through graph structure relationship learning.

3. The method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction as described in claim 1, characterized in that, Based on atlas embedding and disease control condition labels, a dual-trajectory neural frequent differential equation model was constructed, and the trajectory model was obtained through multi-objective joint optimization training, including: Obtain the atlas embedding and disease control condition labels, and input the atlas embedding, time information, and disease control condition labels into the dual-track neural frequent differential equation model. By employing temporal conditional embedding, knowledge graph semantic fusion, latent variable encoding, and conditional neuron frequent differential equation evolution, we can learn the dynamic evolution of mitochondrial genes under disease and normal conditions. Multi-objective joint optimization training is performed based on reconstruction loss, latent variable constraint loss, temporal consistency loss, distribution alignment loss, and future state prediction loss to optimize the dual-trajectory neural ordinary differential equation model and obtain the trajectory model.

4. The method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction as described in claim 3, characterized in that, Based on the trajectory model, perturbation analysis is performed on candidate genes, and a screening score for trajectory pullback caused by candidate gene perturbation is calculated. Based on the screening score, a set of candidate intervention targets is selected, including: The candidate gene is input into the trajectory model. By simulating the dynamic evolution of gene expression after the candidate gene is perturbed, the degree of change in the disease trajectory pulling back to the normal trajectory before and after the perturbation is calculated, and the screening score of the candidate gene is obtained. The screening scores of several candidate genes under multiple time windows are sorted and filtered to obtain a set of candidate intervention targets.

5. The method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction as described in claim 4, characterized in that, The perturbation response characteristics of each candidate intervention target in the candidate intervention target set are fused with the mitochondrial-specific isomorphic knowledge graph, and the candidate intervention targets are graded and evaluated through mechanism reasoning and external evidence verification, including: The perturbation response features corresponding to each candidate intervention target are obtained, and the perturbation response features are fused with the mitochondrial specific isomorphism knowledge graph to generate a structured evaluation of the candidate intervention target. Mechanism reasoning is performed based on structured evaluation to obtain the mechanism information corresponding to the candidate intervention target, and the information is verified by external evidence to obtain the comprehensive evaluation result of the candidate intervention target. Based on the comprehensive evaluation results, the candidate intervention targets are graded and evaluated to obtain the graded evaluation results of the candidate intervention targets.

6. The method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction as described in claim 5, characterized in that, Based on the hierarchical evaluation results of candidate intervention targets and the mitochondrial-specific isomer knowledge graph, candidate drug screening results were obtained, including: Based on the mitochondrial-specific isomer knowledge graph, the target information of candidate drugs is determined. Based on the target information of candidate drugs and the graded evaluation results of candidate intervention targets, candidate drugs are evaluated and ranked to obtain candidate drug screening results.

7. The method for screening drugs for altitude sickness based on virtual mitochondrial modeling and perturbation prediction as described in claim 2, characterized in that, A knowledge graph of mitochondrial-specific isomers, including: Gene nodes, protein nodes, pathway nodes, complex nodes, and drug nodes.

8. A high-altitude sickness drug screening device based on virtual mitochondrial modeling and perturbation prediction, characterized in that, include: The graph construction module is used to construct a mitochondrial-specific heterogeneous knowledge graph and perform heterogeneous graph Transformer pre-training based on the mitochondrial-specific heterogeneous knowledge graph to generate graph embeddings. The trajectory model training module is used to construct a dual-trajectory neural frequent differential equation model based on map embedding and disease control condition labels, and obtain the trajectory model through multi-objective joint optimization training. The trajectory model is used to predict the dynamic evolution of mitochondrial genes in disease and normal states. The intervention target set module is used to perform perturbation analysis on candidate genes based on the trajectory model, calculate the screening score of trajectory pullback caused by candidate gene perturbation, and select the candidate intervention target set according to the screening score. The hierarchical evaluation module is used to fuse the perturbation response characteristics of each candidate intervention target in the candidate intervention target set with the mitochondrial-specific isomorphic knowledge graph, and to perform hierarchical evaluation of the candidate intervention targets through mechanism reasoning and external evidence verification. The drug screening module is used to obtain candidate drug screening results based on the graded evaluation results of candidate intervention targets and the mitochondrial-specific isomorphism knowledge graph.

9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the altitude sickness drug screening method based on virtual mitochondrial modeling and perturbation prediction as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform the altitude sickness drug screening method based on virtual mitochondrial modeling and perturbation prediction as described in any one of claims 1 to 7.

Images