Traditional Chinese medicine efficacy evaluation method and device based on node weighted network, equipment and storage medium

Abstract

The disclosure provides a traditional Chinese medicine efficacy evaluation method and device based on a node-weighted network, equipment and a storage medium. The method determines target disease protein targets and corresponding target weights based on the comprehensive scoring results of candidate disease protein targets by a public database and a large language model. When determining the protein targets of medicinal materials and the corresponding protein target weights of medicinal materials, the prescription ratio, chemical component information, and the interaction probability between the chemical components and the protein targets of each medicinal material in the traditional Chinese medicine prescription to be evaluated are comprehensively considered. The target disease protein targets and the corresponding disease protein target weights, as well as the protein targets of medicinal materials and the corresponding protein target weights of medicinal materials, are added to a pre-constructed protein interaction network. The obtained node-weighted network focuses on the real pharmacological basis, effectively improves the network biological interpretation, and the multi-dimensional network index determined accordingly is used to evaluate the regulation effect of the traditional Chinese medicine prescription to be evaluated on the target disease, which is more accurate.

Description

Method, apparatus, equipment and storage medium for evaluating the efficacy of traditional Chinese medicine based on node weighted networks Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method, apparatus, device, and storage medium for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network. Background Technology

[0002] Traditional Chinese Medicine (TCM) formulas, with their synergistic effects among multiple components and targets, possess unique value in the prevention and treatment of complex diseases. In recent years, the development of technologies such as network pharmacology and artificial intelligence has gradually provided new research pathways for analyzing the complex systems of TCM.

[0003] In related technologies, Large Language Models (LLMs) are typically used to perform semantic mining, relation extraction, and knowledge fusion from massive amounts of text related to diseases and traditional Chinese medicine (TCM) to establish a target network that reflects the association between diseases and medicinal materials. Furthermore, for each target in the network, its total frequency of occurrence in all related texts is used as its target weight. While the target network constructed in this way can assess the association between medicinal materials and diseases, simply using the frequency of target occurrences in text as weights is limited by factors such as the level of academic attention. A high frequency of a target may only represent high research interest and is unlikely to reflect the true biological mechanism of action between the disease and the medicinal material. Therefore, using this as the basis for evaluating the efficacy of medicinal materials in treating diseases results in low accuracy. Summary of the Invention

[0004] This disclosure provides a method, apparatus, equipment, and storage medium for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network, in order to at least solve the above-mentioned technical problems existing in the prior art.

[0005] The first aspect of this disclosure provides a method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network, the method comprising:

[0006] Identify candidate disease protein targets corresponding to the target disease, and determine the target disease protein target and its corresponding weight among the candidate disease protein targets based on the comprehensive scoring results of the candidate disease protein targets using public databases and large language models.

[0007] The prescription ratio and chemical composition of each herb in the Chinese medicine prescription to be evaluated are determined, and the protein target of the herb and the corresponding weight of the herb are determined based on the prescription ratio, the component information of the chemical composition and the interaction probability between the chemical composition and the target of the herb.

[0008] The target disease protein target and its corresponding weight are mapped to the medicinal material protein target and its corresponding weight to a preset protein interaction network to obtain the corresponding node weighted network.

[0009] Based on the node weighted network, multidimensional network indicators are determined for evaluating the efficacy of traditional Chinese medicine prescriptions.

[0010] In one possible implementation, determining the target disease protein target and its corresponding weight among the candidate disease protein targets based on the comprehensive scoring results of the candidate disease protein targets using a public database and a large language model includes:

[0011] Determine the search keywords, which include the names of the corresponding Western medicine diseases and the names of the diseases treated by the Chinese medicine formula to be evaluated;

[0012] For each candidate disease protein target, a search is performed in at least one public database based on the search keywords to determine the search score corresponding to the target.

[0013] Using a large language model, semantic features are extracted from the target Chinese and English documents to determine the semantic relevance score for each candidate disease protein target.

[0014] The search score and the semantic relevance score are combined to determine the comprehensive score of each candidate disease protein target as the comprehensive weight of that target.

[0015] Targets are screened based on the comprehensive weight of each candidate disease protein target to obtain the target disease protein target and its corresponding weight.

[0016] In one possible implementation, determining the prescription ratio and chemical composition corresponding to each medicinal material in the traditional Chinese medicine prescription to be evaluated, and determining the medicinal material protein target and its corresponding weight based on the prescription ratio, the component information of the chemical components, and the interaction probability between the chemical components and the medicinal material target, includes:

[0017] Determine the prescription ratio, chemical composition, and component information of each medicinal material in the Chinese medicine prescription to be evaluated, including the relative abundance of the chemical components;

[0018] All target links corresponding to each medicinal material are identified from the target knowledge base. Each target link is used to characterize the association between the medicinal material and its corresponding chemical components and the corresponding medicinal material protein targets.

[0019] For each target link, the link weight is determined based on the product of the prescription ratio of the corresponding medicinal material, the relative abundance of chemical components, and the probability of interaction between the chemical components and the protein target of the medicinal material, and is used as the medicinal protein target weight of the link.

[0020] In one possible implementation, for each target link, determining the link weight based on the product of the prescription ratio of the corresponding medicinal material on that link, the relative abundance of chemical components, and the probability of interaction between the chemical components and the protein target of the medicinal material, and using this product as the medicinal protein target weight of the link, includes:

[0021] For each target link, determine the normalized prescription ratio of the corresponding medicinal materials and the normalized relative abundance of chemical components on that link;

[0022] The molecular structure of the chemical components and the binding pocket structure of the medicinal protein target on the link are determined. The molecular structure and the binding pocket structure are input into the deep learning prediction model to determine the interaction probability between the corresponding chemical components and the medicinal protein target.

[0023] The link weight is determined based on the product of the normalized prescription ratio, the normalized relative abundance of chemical components, and the corresponding interaction probability between the chemical components and the medicinal protein target, and is used as the medicinal protein target weight in the link.

[0024] In one possible implementation, the step of determining the link weight for each target link based on the product of the prescription ratio of the corresponding medicinal material on that link, the relative abundance of chemical components, and the probability of interaction between the chemical components and the protein target of the medicinal material, and using this product as the medicinal protein target weight for that link, further includes:

[0025] For each medicinal protein target, all relevant target links are traversed, and the sum of the link weights of all relevant target links is used as the medicinal protein target weight of that target.

[0026] In one possible implementation, the step of determining the link weight for each target link based on the product of the prescription ratio of the corresponding medicinal material on that link, the relative abundance of chemical components, and the probability of interaction between the chemical components and the protein target of the medicinal material, and using this product as the medicinal protein target weight for that link, further includes:

[0027] The weights of all medicinal protein targets were normalized.

[0028] In one embodiment, the multidimensional network metrics include overlap coverage, direct connection coverage, Jaccard similarity, weighted proximity, and proximity Z-score.

[0029] A second aspect of this disclosure provides a device for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network, the device comprising:

[0030] The disease target processing module is used to determine the candidate disease protein targets corresponding to the target disease, and to determine the target disease protein target and the weight of the corresponding disease protein target among the candidate disease protein targets based on the comprehensive scoring results of the candidate disease protein targets using a public database and a large language model.

[0031] The herbal target processing module is used to determine the prescription ratio and chemical composition of each herb in the Chinese medicine prescription to be evaluated, and to determine the herbal protein target and the corresponding herbal protein target weight based on the prescription ratio, the component information of the chemical composition and the interaction probability between the chemical composition and the herbal target.

[0032] The target network construction module is used to map the target disease protein target and the corresponding disease protein target weights and the medicinal material protein target and the corresponding medicinal material protein target weights to a preset protein interaction network to obtain the corresponding node weighted network.

[0033] The evaluation result generation module is used to determine multidimensional network indicators for evaluating the efficacy of traditional Chinese medicine prescriptions based on the node weighted network.

[0034] A third aspect of this disclosure provides an electronic device comprising:

[0035] At least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the traditional Chinese medicine efficacy evaluation method based on node-weighted networks as described in this disclosure.

[0036] A fourth aspect of this disclosure provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the traditional Chinese medicine efficacy evaluation method based on a node-weighted network as described in this disclosure.

[0037] This disclosure discloses a method for evaluating the efficacy of traditional Chinese medicine (TCM) based on a node-weighted network. It uses the comprehensive scoring results of candidate disease protein targets from public databases and large language models to screen all candidate disease protein targets for the target disease, determining the target disease protein target and its corresponding weight. Furthermore, when determining the weights of medicinal herb protein targets and their corresponding weights, it comprehensively considers the prescription ratio, chemical composition information, and interaction probability between each herb in the TCM formula to be evaluated and the medicinal herb targets, resulting in more biologically reasonable medicinal herb protein target weights. After adding the target disease protein target and its corresponding weights, as well as the medicinal herb protein target and its corresponding weights, to a pre-constructed protein interaction network, the resulting node-weighted network focuses on the true pharmacological basis, effectively improving the network's biological explanatory power. Compared to related technologies that simply use the frequency of target occurrences in relevant texts as weights, the node-weighted network disclosed in this paper comprehensively considers the differences in network contribution resulting from differences in biological functions. That is, the drug targets and their corresponding weights in the network can better reflect the actual regulatory capacity of the drug or drug components. Based on this, the multidimensional network indicators determined can be used to evaluate the regulatory effect of the Chinese medicine prescription on the target disease with higher accuracy and can provide more reliable quantitative standards.

[0038] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0039] The above and other objects, features, and advantages of this disclosure will become readily apparent from the following detailed description of exemplary embodiments, taken in conjunction with the accompanying drawings. Several embodiments of this disclosure are illustrated in the drawings by way of example and not limitation, in which:

[0040] In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0041] Figure 1 shows a schematic diagram of the implementation process of a method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network according to an embodiment of this disclosure;

[0042] Figure 2 shows a schematic diagram of the implementation process of a method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network according to an embodiment of this disclosure;

[0043] Figure 3 shows a schematic diagram of the composition of four PD-related prescriptions and a schematic diagram of literature attention in the embodiments of this disclosure;

[0044] Figure 4 shows the Venn diagram and the unweighted proximity analysis diagram between the herbal targets and the PD targets used to characterize the four herbal targets in the embodiments of this disclosure.

[0045] Figure 5 shows a schematic diagram outlining the three stratification methods for determining herbal target weights in embodiments of this disclosure;

[0046] Figure 6 shows a robustness analysis diagram of the herbal target weight ranking of four traditional Chinese medicine prescriptions under the condition of gradually increasing weight perturbation in the embodiments of this disclosure;

[0047] Figure 7 shows the target coverage analysis results of four classic Chinese medicine compound priority targets (top-N ranked by weight) in the embodiments of this disclosure on high priority protein targets (top 8, top 100, top 167) of Parkinson's disease.

[0048] Figure 8 shows the comparison and evaluation results of the overlap coverage, direct connection coverage and Jaccard similarity of the four traditional Chinese medicine compound targets in different Parkinson's disease target subsets in the embodiments of this disclosure.

[0049] Figure 9 shows a comparison of weighted and unweighted neighbor Z-scores for four traditional Chinese medicine prescriptions in embodiments of this disclosure.

[0050] Figure 10 shows a schematic diagram of the proportions of flavonoids and isoflavones in all identified compounds and the top 10 most abundant compounds in ginkgo leaves in this embodiment of the present disclosure.

[0051] Figure 11 shows a schematic diagram of the network proximity Z-score between weighted and unweighted Ginkgo biloba full-component target sets and PD-related targets on different disease subsets with different numbers of rankings in this disclosure.

[0052] Figure 12 shows a schematic diagram of the Z-scores of the proximity between weighted and unweighted subsets of Ginkgo biloba flavonoid / isoflavonoid derivative targets and PD targets in embodiments of this disclosure.

[0053] Figure 13 shows the predicted binding probability of flavonoids / isoflavones (right) and other compounds (left) in Ginkgo biloba leaves to targets of Parkinson's disease-related proteins in the embodiments of this disclosure;

[0054] Figure 14 shows a hierarchical clustering analysis diagram of the predicted interaction probabilities between flavonoid isoflavone molecules and PD-related proteins in the embodiments of this disclosure.

[0055] Figure 15 shows a schematic diagram of a traditional Chinese medicine efficacy evaluation device based on a node-weighted network according to an embodiment of the present disclosure;

[0056] Figure 16 shows a schematic diagram of the composition structure of an electronic device according to an embodiment of the present disclosure. Detailed Implementation

[0057] To make the objectives, features, and advantages of this disclosure more apparent and understandable, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0058] This disclosure provides a method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network, as shown in Figure 1. The method includes:

[0059] S101. Determine the candidate disease protein targets corresponding to the target disease. Based on the comprehensive scoring results of the candidate disease protein targets from the public database and the large language model, determine the target disease protein target and the weight of the corresponding disease protein target among the candidate disease protein targets.

[0060] In this step, multiple public databases are used to mine several protein targets related to the target disease as candidate disease protein targets. These public databases are disease-related databases, including the Human Genome Database (GeneCards), the Disease-Gene Association Database (DisGeNET), the Comparative Toxicogenomics Database (CTD), and the Therapeutic Target Database (TTD).

[0061] It should be noted that after identifying candidate disease protein targets, the scoring results of these targets based on public databases are combined with the scoring results using a large language model. The resulting comprehensive score is used as the weight for each candidate disease protein target, which then serves as the basis for selecting target disease protein targets from all candidate targets. In other words, all candidate disease protein targets are sorted according to their corresponding weights, and the top N protein targets are selected, resulting in a set of target disease protein targets and their corresponding weights. This step combines the scoring results of candidate disease protein targets determined by public databases and the large language model, using this as a basis to select target disease protein targets with high biological importance from among the candidate targets.

[0062] S102. Determine the prescription ratio and chemical composition of each medicinal material in the Chinese medicine prescription to be evaluated, and determine the medicinal protein target and corresponding weight of the medicinal protein target based on the prescription ratio, the component information of the chemical components and the interaction probability between the chemical components and the medicinal material target.

[0063] In this step, after obtaining the TCM prescription to be evaluated, the chemical composition of each herb in the prescription, its corresponding chemical composition information, and the proportion of each herb in the prescription can be determined using a local knowledge base and a public TCM database. The local knowledge base in this step is a pre-built, manually verified TCM database, while the public TCM database is a publicly available TCM database, such as BATMAN-TCM2.0.

[0064] It should be noted that in this step, each herb in the TCM formula to be evaluated can have its corresponding target pathway obtained from the aforementioned TCM database. Accordingly, each pathway represents the relationship between the herb and its corresponding chemical components and protein targets. Each herb contains at least one chemical component, and each chemical component corresponds to at least one protein target; that is, at least one target pathway can be obtained for each herb. For each pathway, the pathway contribution is quantified based on the prescription ratio of the corresponding herb in the TCM formula to be evaluated, the information on the chemical components of the herb in the pathway, and the interaction probability between the corresponding chemical components and the protein targets. The interaction probability between the chemical components and the protein targets is determined based on a deep learning prediction model. The obtained pathway contribution is used to characterize the target weights of the protein targets on the target pathway, thereby determining the protein targets and their weights corresponding to the TCM formula to be evaluated.

[0065] S103. Map the target disease protein target and the corresponding disease protein target weights and the medicinal material protein target and the corresponding medicinal material protein target weights to a preset protein interaction network to obtain the corresponding node weighted network.

[0066] In this step, the pre-constructed protein-protein interaction network is a global protein-protein interaction (PPI) network, including real and validated protein-protein interaction data. By mapping the target disease protein targets and their corresponding weights obtained in step S101, and the medicinal material protein targets and their corresponding weights obtained in step S102, to the pre-constructed protein-protein interaction network, a corresponding node-weighted network is obtained. That is, this step incorporates multi-dimensional weight information corresponding to the target disease protein targets and medicinal material protein targets into the protein-protein interaction network. The resulting node-weighted network can significantly improve the accuracy and interpretability of drug efficacy prediction and mechanism of action tracing.

[0067] S104. Determine the multidimensional network indicators for evaluating the efficacy of traditional Chinese medicine prescriptions based on the node weighted network.

[0068] In this step, a multidimensional network index is determined based on the node-weighted network obtained in step S103 to measure the therapeutic effect of the traditional Chinese medicine formula on the target disease. It should be noted that the node-weighted network used to determine the multidimensional network index in this step maps the target disease protein target and its corresponding weight, as well as the medicinal material protein target and its corresponding weight. This integrates multidimensional weight information such as the prescription ratio of medicinal materials, information on the chemical composition of medicinal materials, and the probability of interaction between the chemical composition information and the medicinal material protein targets. Furthermore, the targets in the network can distinguish the biological importance of different disease proteins, effectively improving the biological interpretability of the node-weighted network. Therefore, the multidimensional network index determined based on this has higher accuracy in assessing the therapeutic effect.

[0069] This disclosure provides a method for evaluating the efficacy of traditional Chinese medicine (TCM) based on a node-weighted network. Aiming for biological rationality, it systematically introduces a node weighting mechanism, comprehensively considering the prescription ratios of each medicinal material, chemical composition information, and the interaction between chemical composition information and medicinal material targets to determine the protein targets and corresponding target weights of the medicinal materials. These medicinal material protein targets and corresponding target weights, along with the target disease protein targets and corresponding target weights determined based on a comprehensive scoring result from a public database and a large language model, are mapped to a protein-protein interaction network. The resulting node-weighted network can evaluate the relative contributions of different medicinal materials and components. Using this as a basis, multidimensional network indicators are extracted to quantify the intervention effect of the prescription on the disease, resulting in higher accuracy and biological interpretability.

[0070] In one possible implementation, the target disease protein target and its corresponding weight are determined based on the comprehensive scoring results of candidate disease protein targets using a public database and a large language model, including:

[0071] Determine the search keywords, which include the names of the corresponding Western medicine diseases and traditional Chinese medicine diseases for the Chinese medicine prescriptions to be evaluated;

[0072] For each candidate disease protein target, a search is conducted in at least one public database based on the search keywords to determine the search score corresponding to the target.

[0073] Using a large language model, semantic features are extracted from the target Chinese and English documents to determine the semantic relevance score for each candidate disease protein target.

[0074] By combining the retrieval score and the semantic relevance score, the comprehensive score of each candidate disease protein target is determined and used as the comprehensive weight of that target.

[0075] Targets are screened based on the comprehensive weight of each candidate disease protein target to obtain the target disease protein target and its corresponding weight.

[0076] In this embodiment, when determining the weights of candidate disease protein targets, the combination of the Western medicine disease name and the traditional Chinese medicine disease name corresponding to the herbal formula to be evaluated is used as keywords. The candidate disease protein targets are searched in multiple public databases, and a search score is determined based on the search results. This aims to comprehensively and deeply explore the target knowledge corresponding to diseases in both Western and traditional Chinese medicine. In this embodiment, the disease name corresponding to the herbal formula to be evaluated refers to the Western medicine disease name. Taking Parkinson's disease as an example, the determined search keywords are a combination of the Western medicine disease name and the traditional Chinese medicine disease name. The search is conducted in public databases using "Parkinson's disease" OR "tremor" to determine the corresponding search score. Here, Parkinson's disease is the Western medicine disease name, and tremor is the corresponding traditional Chinese medicine disease name.

[0077] When determining the retrieval score for each candidate disease protein target, the corresponding steps include: identifying any candidate disease protein target as the current target to be scored, and identifying any one of multiple public databases as the current public database; determining whether the current target to be scored exists in the retrieval results for the search keywords in the current public database; if the current target to be scored does not exist in the retrieval results, setting the score of the current target to be scored in the current public database to equal the preset minimum weight; if the current target to be scored exists in the retrieval results, determining whether the retrieval results contain a score for the current target to be scored; if the current target to be scored exists in the retrieval results, obtaining that score as the data basis for subsequently determining the comprehensive score result; and if the current target to be scored does not exist in the retrieval results, setting the score of the current target to be scored in the current public database to equal the preset maximum weight.

[0078] Similarly, for any candidate disease protein target, the scores of each public database for the target are determined according to the above steps. That is, the scores of the target are determined from four databases: the Human Genome Database, the Disease-Gene Association Database, the Comparative Toxicology-Genome Database, and the Therapeutic Target Database. The scores of each database are combined to obtain the public database-based score of the target.

[0079] Furthermore, the scoring results of the large language model for candidate disease protein targets are determined. In this embodiment, the large language model is preferably TCMChat, a large language model specifically designed for the field of traditional Chinese medicine based on the Transformer architecture. The TCMChat model automatically analyzes Chinese and English literature, extracting semantic features from the literature, identifying target disease-related protein targets, and generating semantic relevance scores.

[0080] It should be noted that the Chinese and English literature in this embodiment is literature related to diseases and corresponding treatment prescriptions obtained from the CNKI and SciFinder databases. In order to automatically extract and quantify disease-related entity information from unstructured literature, this embodiment utilizes a pre-constructed, manually verified corpus and database of traditional Chinese medicine, combined with the full text of Chinese and English literature obtained from the CNKI and SciFinder databases, to perform domain fine-tuning and instruction fine-tuning on TCMChat to obtain a quantifiable semantic relevance score, i.e., an LLM score, which is used as a weighting factor in subsequent weighted calculations.

[0081] Accordingly, the training process for the TCMChat model includes: integrating formulary literature, local chronicles and ancient texts, modern textbooks, a knowledge table of medicinal materials, components and targets, and disease and TCM syndrome terminology (including synonyms / synonyms), and establishing a mapping relationship between terminology ontology and synonyms; for easily confused TCM terms that require context for determination, i.e., the same word or phrase may correspond to multiple different concepts in the TCM context, such as phlegm may refer to tangible phlegm (coughing up phlegm). It may also refer to intangible phlegm (an abstract pathogenesis concept of phlegm and blood stasis), or different words that indicate the same concept, such as insomnia and sleeplessness. Therefore, positive and negative control samples are constructed, and contrastive loss is introduced to improve the model's ability to semantic alignment and disambiguation. Supervised fine-tuning, such as SFT, and parameter-efficient fine-tuning, such as LoRA / QLoRA, are used to train the model, with a fine-tuning sample size of about 40,000. During the inference stage, the model output is standardized, including unifying it to a standardized glossary of medicinal materials, chemical components, proteins / genes, diseases and syndromes. The original semantic relevance scores are monotonically calibrated to the (0,1] interval to obtain the LLM score.

[0082] Furthermore, for each candidate disease protein target, the scores from the four public databases and the semantic relevance score from the large language model are multiplied and weighted to obtain a comprehensive score, which serves as the overall weight for that target. The scoring formula is as follows:

[0083]

[0084] In the formula, The overall weight of candidate disease protein target i; CTD score for candidate disease protein target i; DisGeNET score for candidate disease protein target i; GeneCards score for candidate disease protein target i; The TTD score for candidate disease protein target i; A large language model score is given to candidate disease protein target i.

[0085] After obtaining the weighted fusion comprehensive score based on the above formula, monotonic calibration is performed to obtain the final comprehensive target weight of the candidate disease protein targets. Subsequently, all candidate disease protein targets are sorted in descending order according to the comprehensive target weight, and the top N targets after sorting are obtained as the target disease protein targets.

[0086] Therefore, this embodiment, in determining the target disease protein targets and their corresponding weights, is based on the combined results of scores from multiple public disease databases and a large language model in the field of traditional Chinese medicine (TCM). That is, the determination of target disease protein targets integrates scores from multiple databases and semantic scores from a monotonically calibrated large language model. This involves not only automated knowledge mining from the TCM large language model TCMChat but also the aggregation of multiple public databases such as DisGeNET, GeneCards, CTD, and TTD. In other words, in the process of assessing the importance of disease targets, multi-source information is integrated to comprehensively cover relevant targets in both TCM and Western medicine. After incremental domain library updates and parameter fine-tuning, the large language model can output quantifiable semantic evidence scores. Simultaneously, the large language model automatically crawls, mines knowledge, extracts entities, and performs semantic weighting from unstructured literature, fully utilizing the automation capabilities driven by big data and artificial intelligence to significantly improve knowledge integration and disease-related target identification, making it more efficient and objective than manual screening.

[0087] In one possible implementation, the prescription ratio and chemical composition of each medicinal material in the traditional Chinese medicine formula to be evaluated are determined. Based on the prescription ratio, the component information of the chemical components, and the interaction probability between the chemical components and the medicinal material targets, the medicinal material protein targets and their corresponding weights are determined, including:

[0088] Determine the prescription ratio, chemical composition, and component information of each medicinal material in the Chinese medicine prescription to be evaluated. The component information includes the relative abundance of chemical components.

[0089] All target links corresponding to each medicinal material are identified from the target knowledge base. Each target link is used to characterize the association between the medicinal material and its corresponding chemical components and the corresponding medicinal material protein targets.

[0090] For each target link, the link weight is determined based on the product of the prescription ratio of the corresponding medicinal material, the relative abundance of chemical components, and the probability of interaction between the chemical components and the protein target of the medicinal material, and is used as the medicinal protein target weight of the link.

[0091] In this embodiment, for each herb in the traditional Chinese medicine prescription to be evaluated, the connection relationships between the herb, its components, and its target are obtained based on a traditional Chinese medicine database. These connections, known as target links, serve as a priori basis for determining whether there is a correlation between chemical components and herb targets. Specifically, for each herb, the herb targets involved in all its target links are the herb protein targets corresponding to that herb.

[0092] Furthermore, in this embodiment, the chemical component information is preferably the relative abundance of chemical components. Accordingly, for each target link of the medicinal material, the weight of the link is comprehensively determined based on the prescription ratio of the corresponding medicinal material on the link, i.e., the dosage ratio of the medicinal material in the traditional Chinese medicine prescription to be evaluated, the relative abundance of the chemical component represented by the link, and the interaction probability between the chemical component represented by the link and the protein target of the medicinal material. The weight of the link is then used as the target weight of the protein target of the medicinal material on the link. The component list of the traditional Chinese medicine prescription to be evaluated is obtained from the traditional Chinese medicine database and the local knowledge base. The relative abundance of chemical components can be obtained from the local database, which is a pre-constructed database containing chemical component information of various medicinal materials that has been manually verified and calibrated. It includes standardized data of LC-MS small molecule components and abundance of nearly 1,000 kinds of traditional Chinese medicinal materials.

[0093] In other words, this embodiment, when determining the weights of medicinal protein targets, comprehensively considers the actual amount of the target medicinal material in the evaluated traditional Chinese medicine prescription, the true abundance of its chemical components, and the intensity of its molecular-level effects. This maximizes the reflection of the medicinal material's true biological intervention ability and functional specificity on the disease network. The medicinal protein targets and their corresponding weights obtained in this way, after being mapped to a protein-protein interaction network, can effectively improve the sensitivity and discriminative ability of the resulting node-weighted network to the actual intervention effects of the medicinal material. Furthermore, the weights of the medicinal protein targets are determined based on the product of the actual amount of the medicinal material, the true abundance of its chemical components, and the intensity of its analytical-level effects. This product operation is more sensitive to low values ​​of any factor, suppressing pseudo-strong terms with high proportions but low interactions / low abundance.

[0094] In one possible implementation, for each target link, the link weight is determined based on the product of the prescription ratio of the corresponding medicinal material on the link, the relative abundance of chemical components, and the probability of interaction between the chemical components and the protein target of the medicinal material. This weight is used as the medicinal protein target weight of the medicinal material in the link, including:

[0095] For each target link, determine the normalized prescription ratio of the corresponding medicinal materials and the normalized relative abundance of chemical components on that link;

[0096] The molecular structure of the chemical components and the binding pocket structure of the medicinal protein target in this link are determined. The molecular structure and binding pocket structure are then input into a deep learning prediction model to determine the interaction probability between the corresponding chemical components and the medicinal protein target.

[0097] The link weight is determined based on the product of the normalized prescription ratio, the normalized relative abundance of chemical components, and the corresponding interaction probability between the chemical components and the medicinal protein target, and is used as the medicinal protein target weight in the link.

[0098] In this embodiment, when determining the link weight of each target link, after obtaining the prescription ratio of the corresponding medicinal material in the Chinese medicine prescription to be evaluated, the prescription ratio is normalized, and the abundance information of the corresponding chemical components in the link is also normalized.

[0099] It should be noted that the interaction probabilities between the corresponding chemical components and the medicinal protein targets along the link are determined using a deep learning prediction model. Accordingly, the molecular structures corresponding to the chemical components and the binding pocket structures of the medicinal protein targets are input into the deep learning prediction model to obtain the interaction probabilities, which are monotonically calibrated and mapped to the [0,1] interval. Among them, the probability of interaction between chemical components and the protein targets of medicinal materials. This model is used to characterize the interaction strength between chemical components and medicinal targets (protein targets) at the molecular level. It should also be noted that the prediction model in this embodiment encodes the molecular structure diagram corresponding to the chemical component and the three-dimensional conformation of the protein pocket corresponding to the medicinal protein target, employing a geometric isomorphic graph network and distillation training to improve the interaction probability. Cross-target comparability and robustness.

[0100] It is particularly important to emphasize that, in order to overcome the systematic errors caused by missing experimental data and lack of standardization, this embodiment employs a teacher-student structure domain deep learning framework for inference. Accordingly, in the input and encoding process, chemical components are represented by molecular diagrams. Representation; a 3D point cloud map of the protein within 5 angstroms of its small molecule binding sites. To be consistent with two-dimensional chemical diagrams Indicates. Regarding. and Extracting chemical embeddings using a message-passing neural network (MPNN) ;right Spatial embedding is extracted using equivariant graph convolution. Protein atom characterization is represented using Hadamard product fusion:

[0101]

[0102] Furthermore, a teacher-student distillation model is used for training and prediction. During the training period, a bipartite graph of the ligand-pocket correspondence complex is constructed. The teacher network explicitly learns the atom pair interaction weights; the student network only uses... and As input, the atomic-level interaction representation of the teacher is aligned through knowledge distillation loss, and the student regression value is output. Then, through monotonic calibration, such as Platt / iso-regression, the interaction probabilities are mapped to the [0,1] interval:

[0103]

[0104] α and β are fitted using a validation set to ensure comparability across targets and batches. This framework provides interpretable and calibrable data. This effectively overcomes the systematic bias caused by non-standardized and missing experiments.

[0105] Finally, the normalized proportions of medicinal materials, the normalized abundance information of chemical components, and the interaction probabilities between chemical components and medicinal protein targets are multiplied to quantify the contribution of each target link, i.e., to determine the corresponding link weight, and this link weight is used as the weight of the medicinal protein target on that link. Accordingly, for any medicinal material h, the target links corresponding to its chemical component i and medicinal protein target j are... The link contribution, i.e., the link weight. The formula is as follows:

[0106]

[0107] In the formula, The normalized proportions of medicinal materials in the prescription. The relative abundance of normalized chemical components within a single medicinal herb. This represents the probability of interaction between chemical components and the protein targets in medicinal materials.

[0108] Therefore, this embodiment, in determining the interaction probability between chemical components and medicinal protein targets, does not rely on scattered and dimensionlessly variable in vitro experimental values. Instead, it directly learns atomic pair interactions at the structural level based on a deep learning framework and outputs the results to [0,1] after probability calibration. This overcomes the limitations of missing experimental data and unstandardized systematic biases, providing interpretable and calibrable interaction strengths. Furthermore, based on deep learning... The prediction is not limited by the coverage of existing database edges. It can extrapolate the edge probability and estimate the strength of any given combination of proteins and small molecules beyond the existing prior edges of medicinal materials-components-targets. This significantly expands the evaluable space, improves the accuracy and completeness of network weighting, and reduces systematic bias caused by missing or missing data in the database.

[0109] In one possible implementation, for each target link, the link weight is determined based on the product of the prescription ratio of the corresponding medicinal material on the link, the relative abundance of chemical components, and the probability of interaction between the chemical components and the protein target of the medicinal material, and is used as the medicinal protein target weight of the medicinal material in the link, further comprising:

[0110] For each medicinal protein target, all relevant target links are traversed, and the sum of the link weights of all relevant target links is used as the medicinal protein target weight of that target.

[0111] In this embodiment, since each medicinal herb involves at least one chemical component, and each chemical component corresponds to at least one medicinal herb target, this embodiment determines the target weight of each medicinal herb protein target by traversing all relevant target links of the medicinal herb protein target and summing the link weights of all relevant target links. The resulting weight is then used as the medicinal herb protein target weight of the target.

[0112] Corresponding cumulative weight The formula is:

[0113] .

[0114] For example, a traditional Chinese medicine formula to be evaluated includes 30% herb A, 30% herb B, and 40% herb C. Herb A involves chemical components X and Y. Correspondingly, chemical component X corresponds to herb target H1 and herb target H2, and chemical component Y corresponds to herb target H2 and herb target H3. That is, the target pathways corresponding to herb A include herb A-chemical component X-herb target H1, herb A-chemical component X-herb target H2, herb A-chemical component Y-herb target H2, and herb A-chemical component Y-herb target H3. For medicinal material A, it acts on the same target point, namely medicinal target point H2, through two links: medicinal material A-chemical component X-medicinal material target H2 and medicinal material A-chemical component Y-medicinal material target H2. Therefore, the weight of medicinal target point H2 needs to be accumulated by the link weights of the above two links, and the accumulated link weight is used as the total weight of medicinal target point H2.

[0115] In one possible implementation, for each target link, the link weight is determined based on the product of the prescription ratio of the corresponding medicinal material on the link, the relative abundance of chemical components, and the probability of interaction between the chemical components and the protein target of the medicinal material, and is used as the medicinal protein target weight of the medicinal material in the link, further comprising:

[0116] The weights of all medicinal protein targets were normalized.

[0117] In this embodiment, after determining the target weights of all medicinal protein targets, the total weights of all medicinal targets are normalized and monotonically calibrated to obtain the probability values ​​in the interval [0,1], thereby ensuring the comparability of results between different medicinal materials and prescriptions.

[0118] Therefore, it can be seen that this embodiment will use the prescription ratio True abundance of chemical components Interaction strength at the molecular level The link-level product coupling and aggregation are applied to the target level, and the link weights are normalized to facilitate comparisons across medicinal materials and prescriptions. This fully reflects the real biological intervention capabilities and functional specificities of medicinal materials on the disease network, providing a dimensionally consistent and calibrable weighting basis for determining multidimensional network indicators in the subsequent node-weighted network.

[0119] In one implementation, the multidimensional network metrics include overlap coverage, direct connection coverage, Jaccard similarity, weighted proximity, and proximity Z-score.

[0120] In this embodiment, as shown in Figure 2, the node-weighted network used to determine the multidimensional network index is generated by mapping the weights of the target disease protein target and its corresponding disease protein target, as well as the weights of the medicinal material protein target and its corresponding medicinal material protein target, to a pre-constructed protein-protein interaction network. The global protein-protein interaction network (PPI) includes real data with actual protein-protein interactions. In one scenario, the PPI network comprises 16,677 proteins and 243,603 interactions. The data are sourced from the molecular interaction databases IntAct (https: / / www.ebi.ac.uk / intact / home) and MINT (https: / / mint.bio.uniroma2.it / ), the innate immune interaction knowledge base InnateDB (https: / / www.innatedb.com / ), the protein interaction network analysis PINA (https: / / omics.bjcancer.org / pina / ), the biological interaction database BioGRID (https: / / thebiogrid.org / ), the kinase interaction network Kinome NetworkX (https: / / bioinfo.uth.edu / kinomenetworkX / download.html), the kinase substrate interaction database PhosphositePlus (https: / / www.phosphosite.org / homeAction.action), and the Global Neurodegenerative Protein Consortium GNPC (https: / / www.neuroproteome.org / v1-harmonized-data-set-hds). All interactions are treated as undirected edges.

[0121] To comprehensively evaluate the intervention capability of traditional Chinese medicine (TCM) target clusters on disease targets and assess the efficacy of medicinal materials, this embodiment determines multidimensional network indicators from the aforementioned node-weighted network, including overlap coverage, direct connection coverage, Jaccard similarity, weighted proximity, and proximity Z-score, to comprehensively evaluate the targeting capability of TCM targets in the node-weighted network for disease-related targets. Among these, overlap coverage... This measure assesses the proportion of disease targets that are also targeted by herbal ingredients, reflecting the direct overlap between disease targets and medicinal targets. The corresponding formula is:

[0122]

[0123] In the formula Indicates the set of herbal target points. This represents a set of disease targets.

[0124] Direct connection coverage is used to measure how many nodes in the herbal target set belong to the disease target set or are directly connected to the disease target through a direct edge in a node weighted network, in order to capture potential indirect regulatory relationships.

[0125] In addition, this embodiment also calculates Jaccard similarity, i.e., the Jaccard similarity index, to quantify the degree of overlap between the herbal target set and the disease target set. The corresponding formula is:

[0126]

[0127] This embodiment also employs proximity metric to evaluate the ability of herbal targets to modulate disease-related targets in a node-weighted network, a set of herbal targets. Disease Targets Unweighted proximity is defined as the average shortest path length between all pairs of nodes:

[0128]

[0129] In the formula, This represents the length of the unweighted shortest path between nodes h and d in a node-weighted network, calculated using the breadth-first search (BFS) algorithm in the NetworkX package.

[0130] To reflect the biological importance of specific nodes, this embodiment determines that the multidimensional network index also includes weighted proximity, and the corresponding formula is:

[0131]

[0132] In the formula, the weight of the herbal target is labeled as follows: Disease protein target weight labeling is , This represents the shortest path between medicinal targets and disease targets in a node-weighted network.

[0133] As shown in the above formula, weighted proximity scales the distance contribution between node pairs by multiplying the corresponding weights, thus allowing for the preferential consideration of node pairs with higher biological relevance. Simultaneously, a zero-distribution Z-score is generated through degree-matched random sampling. In this embodiment, to assess the statistical significance of the observed proximity values, weighted and unweighted proximity indices are compared with proximity scores obtained from a random set of 500 equally sized nodes. The random sampling process considers the degree of the nodes to preserve the topological characteristics of the protein-protein interaction network. Statistical significance is assessed using the Z-score (…). The specific calculation formula is as follows:

[0134]

[0135] In the formula This represents the observed proximity score. and represents the mean and standard deviation of the proximity score calculated from the random node set, respectively. The smaller the score, the lower the actual proximity of the medicinal target group and the disease target group in the network is compared to the expected proximity under random conditions. This indicates that the medicinal material's intervention ability on the disease network is significantly stronger than that of randomly selected node groups. This demonstrates that the medicinal material components studied have a stronger targeting and regulatory effect on disease targets at the biological network level, thus reflecting their superior therapeutic potential and efficacy.

[0136] Therefore, this embodiment utilizes knowledge mining assisted by a large language model to systematically integrate disease-related protein target information. Specifically, disease targets not only originate from automated knowledge mining using the TCM large language model TCMChat, but also aggregate data from databases such as DisGeNET (https: / / disgenet.com / ), GeneCards (https: / / www.genecards.org / ), CTD (https: / / ctdbase.org / ), and TTD (https: / / idrblab.net / ttd / ). By fusing multi-source information, it comprehensively covers disease-related targets in both TCM and Western medicine. The scores of the above five items are normalized and used as the comprehensive weight of the disease targets. Furthermore, multiple information such as the proportion of medicinal materials, the abundance of components, and the probability of compound-target interactions are integrated to obtain medicinal material targets and their weights. All targets are then mapped to a large-scale protein interaction network containing 16,677 proteins and 243,603 interactions, completing the construction and weighting of the node weighted network. Based on a node-weighted network, this study evaluates the disease intervention capabilities and potential of various traditional Chinese medicine (TCM) formulas through a multidimensional network index system. This not only quantifies the intervention effects of medicinal materials and components on disease networks but also supports the priority identification of core active ingredients and formula optimization. The node-weighted network, by introducing node weights, enables the quantitative differentiation of the biological importance of disease targets and medicinal material targets.

[0137] The following section will use Parkinson's disease as an example to explain in detail the implementation process and effects of the method disclosed herein.

[0138] Parkinson's disease (PD) is a complex and multifactorial neurodegenerative disease characterized by the progressive degeneration of dopaminergic neurons in the substantia nigra. Since there is currently no effective cure, multi-target treatment strategies, represented by traditional Chinese medicine, have gained increasing attention in recent years. This example uses Parkinson's disease as a typical complex disease model to systematically reveal the multi-target intervention mechanism of traditional Chinese medicine compound prescriptions.

[0139] S1. Determine the target protein targets and corresponding target weights for Parkinson's disease;

[0140] Multiple candidate disease protein targets related to Parkinson's disease were mined using multiple public databases. In order to systematically identify and prioritize protein targets related to Parkinson's disease (PD), knowledge from disease databases and Chinese and English literature was integrated.

[0141] Protein targets were sourced from the GeneCards, DisGeNET, CTD, and TTD databases. Each database provides different scoring criteria reflecting the association between targets and diseases. Since the TTD database does not provide explicit numerical scores, this embodiment assigns a value of one to targets listed in the TTD and a minimum value (10) to targets not listed. -20 Correspondingly, the same minimum value is assigned to targets that are missing in other databases.

[0142] To further incorporate semantic relevance, this embodiment adds a disease target weighting term, LLMscore, derived from TCMChat. Multiplying the scores of the five terms and normalizing them to the (0,1) interval yields the comprehensive weight, targetscore, for each target.

[0143] After standardization and post-processing, the PD-related targets are sorted in descending order based on the comprehensive score. To evaluate the impact of different target sets on subsequent analysis, this embodiment selects the top 100, 200, 300, 400, 500, 600, 700, 800, 900, and 1000 targets as subsets for downstream priority ranking and comparative evaluation.

[0144] S2. Determination of herbal targets and weight calculation;

[0145] This embodiment selects four commonly used traditional Chinese medicine formulas for treating Parkinson's disease, including Qianzheng San (QZS), Dabu Yin Wan (DBYW), Liuwei Dihuang Wan (LWDH), and Tianma Gouteng Tang (TGD). The corresponding association relationships (existence of edges) between medicinal materials, components, and targets, as well as the list of components, are obtained from public traditional Chinese medicine databases, such as BATMAN-TCM2.0 and local knowledge bases. The quantitative abundance of components is provided by the local database MedTCM.

[0146] To comprehensively quantify the target weights of medicinal materials, this embodiment compares three different methods for calculating the protein target weights of medicinal materials.

[0147] Method 1: Calculate target weights based solely on the proportion of medicinal materials;

[0148] Method 2 combines the proportion of medicinal materials and chemical composition data with Method 1, but does not consider the probability of interaction between compounds and targets.

[0149] Method 3, building upon Method 2, further integrates the proportions of medicinal materials, chemical components, and the probability of interactions between compounds and targets. All weight results are normalized to the 0–1 range for comparability across different prescriptions. The specific process includes:

[0150] Using each medicinal herb as a unit (h), it is first normalized according to its proportion in the prescription. ;

[0151] Chemical components in medicinal materials The relative abundance of compounds in each medicinal herb was obtained by normalizing the semi-quantitative abundance data from the local medicinal herb composition quantitative database MedTCM. ;

[0152] Considering the fragmented, in vitro assays from different sources ( / / Due to system differences (buffer, temperature, substrate, time window) and significant missing active tags, directly substituting raw experimental values ​​for interaction strengths would introduce uncontrollable bias. This embodiment employs a teacher-student domain deep learning framework for inference. This is to overcome the systematic bias caused by missing and unstandardized experimental data. The input and encoding process of the model, as well as the training and prediction process, are performed in accordance with the process for the teacher-student prediction model in the above embodiments; repeated details will not be elaborated upon.

[0153] The weight of each herb-component-target link is determined according to the link weight formula and used as the weight of the target on that link. For the same herb acting on the same target through multiple component links, the weights contributed by all components and targets are summed to obtain the total weight of that target. The weights of all herb and target are normalized to ensure the comparability of results between different herbs and prescriptions.

[0154] Therefore, it can be seen that Method 3, by multiplying and coupling the prescription ratio, true abundance, and molecular-level interaction strength at the link level and aggregating them to the target level, more fully reflects the true biological intervention ability and functional specificity of medicinal materials on the disease network. This provides a dimensionally consistent and calibrable weighted basis for subsequent network efficacy measurement of indicators such as weighted proximity Z-score and coverage / Jaccard on node-weighted networks.

[0155] S3, Construction of a node-weighted network;

[0156] The target disease protein and its corresponding weights for Parkinson's disease obtained in step S1, and the target herbal medicine and its corresponding weights obtained in step 2, are mapped to a pre-constructed global protein-protein interaction network containing 16,677 proteins and 243,603 interactions to obtain the corresponding node weighted network.

[0157] S4. Determination of multidimensional network evaluation indicators;

[0158] The corresponding overlap coverage, direct connection coverage, Jaccard similarity, weighted proximity, and proximity Z-score are determined from the constructed node weighted network to comprehensively evaluate the intervention capability of the traditional Chinese medicine target group on the disease target group.

[0159] It should be noted that, in order to evaluate the robustness and stability of the calculated protein node weights to random fluctuations or noise, this embodiment also performed a weight perturbation sensitivity analysis to evaluate the robustness of the ranking order of protein targets when the original weights change.

[0160] Each node will be affected by random noise to varying degrees. Let... This represents the original normalized weight of protein node i. For a given perturbation coefficient... The perturbated weights are composed of the original weights and random values ​​drawn from a uniform distribution. The convex combination constitutes:

[0161]

[0162] when When , it means there is no disturbance (i.e., the original weights). This indicates that the weights are completely randomized. For each perturbation level This embodiment calculates the Spearman rank correlation coefficient between the original weighted ranking and the perturbed ranking. This is used to quantify the sensitivity of the weighting scheme in this embodiment to disturbances. The weight disturbance sensitivity analysis can prove that each node is robust to random noise. Under different weight disturbance conditions, the ranking correlation always remains at a high level (the Spearman correlation coefficient Rs is high). This shows that the method in this embodiment is not only scientific and reasonable, but also has strong practical stability and is suitable for application in real complex data environments.

[0163] This embodiment uses Parkinson's disease (PD) as a representative disease model to systematically evaluate the ability of four traditional Chinese medicine formulas—Tianma Gouteng Decoction (TGD), Liuwei Dihuang Pill (LWDH), Qianzheng Powder (QZS), and Dabuyin Pill (DBYW)—as well as the neuroprotective single herb Ginkgo biloba leaf to intervene in the disease network. It demonstrates the advantages of networks with node weights compared to unweighted networks, and also reflects the ability to evaluate formulas and identify active ingredients in traditional Chinese medicine.

[0164] To comprehensively identify potential protein targets related to Parkinson's disease and clarify their relative importance, this embodiment utilizes four biomedical databases to obtain disease-related target information. Simultaneously, it introduces the large language model TCMChat for automated literature knowledge mining. TCMChat extracts semantic association information of Parkinson's disease-related protein targets through in-depth analysis of a large amount of classic and modern research literature in traditional Chinese medicine, and quantifies and generates corresponding LLM knowledge weights. The information provided by the four biomedical databases has a certain degree of overlap and can complement each other. Combining this with the LLM knowledge weights obtained from mining the full text of nearly 33,000 Chinese and English documents related to Parkinson's disease using the large language model further strengthens the semantic relevance of the disease protein target weights. That is, in this embodiment, the standardized scores of the four databases are multiplied by the LLM knowledge weights to obtain the comprehensive weight of each protein target. The results show, as shown in Table 1, that high-weight targets such as α-synuclein (SNCA), tyrosine hydroxylase (TH), leucine enriched repeat kinase 2 (LRRK2), dopa decarboxylase (DDC), and microtubule-associated protein tau (MAPT) have the highest scores, which is highly consistent with the current medical understanding of the pathophysiology of Parkinson's disease.

[0165] Table 1. Top 30 Parkinson's Disease-Related Protein Targets and Their Overall Weights

[0166]

[0167] Based on the aforementioned comprehensive weighting, this embodiment sorts the targets from highest to lowest importance, selecting a subset of targets (top 100 to 1000) of varying numbers as target disease protein targets for subsequent network analysis and systematic evaluation of traditional Chinese medicine formulas. Integrating semantic knowledge mining with a large language model can significantly improve the identification accuracy and biological interpretability of disease-related targets, providing a reliable data-driven basis for subsequent optimization of traditional Chinese medicine formulas and identification of active ingredients.

[0168] To evaluate the effectiveness of this implementation method in assessing traditional Chinese medicine (TCM) compound prescriptions, this embodiment analyzes Parkinson's disease and four TCM prescriptions related to Parkinson's disease to verify the effectiveness of the node-weighted network. Specifically, this includes:

[0169] In order to verify the effectiveness of the method disclosed herein, the weighting and evaluation of herbal targets were performed by analyzing four traditional Chinese medicine formulas (QZS, DBYW, LWDH, and TGD) related to PD.

[0170] The level of attention given to traditional Chinese medicine formulas was assessed through a systematic search of the CNKI and SciFinder databases using formula names combined with Parkinson's disease as keywords. As shown in Figure 3, literature analysis revealed that among the four formulas, Tianma Gouteng Decoction was the most effective for TGD. The presence of commonly used PD treatment formulas in the CNKI and SciFinder databases indicates that this field has received significant attention and possesses potential clinical value.

[0171] The connections and interaction probabilities between each herb and its target were determined using the BATMAN-TCM2.0 database and a deep learning model. As shown in Figure 4, TGD contains the most herbs and components, and its herb target coverage is the broadest (5980 proteins), followed by LWDH, QZS, and DBYW.

[0172] In related technologies, only the overlap of direct targets is usually considered. However, this disclosure adds consideration to the indirect regulatory role mediated by protein-protein interaction networks (PPI networks). Specifically, all herbal targets and high-priority drug targets, i.e., the top N drug targets after descending order, are mapped to the same global PPI network, and the therapeutic potential is evaluated through topology-based proximity analysis. As shown in Figure 4, TGD consistently exhibits the smallest proximity value in the expanded drug target set (N=100-1000), indicating a stronger ability to influence drug targets, consistent with the high occurrence rate reported in the literature.

[0173] In addition, this disclosure adds consideration for the weighting of herbal target data, employing three hierarchical weighting strategies to optimize the priority ranking of herbal targets. Method 1 only considers the herbal proportion without considering the cumulative compound effect. Method 2 further combines the herbal proportion and compound data based on Method 1, as shown in Figure 5. Method 3 comprehensively considers the herbal proportion and compound data, while also incorporating the compound-target interaction probability. Furthermore, the robustness of the above three weighting methods is evaluated through weight perturbation experiments, where the noise factor... The change from 0 to 1. With... As the weights increase, the performance of all methods gradually decreases, demonstrating their sensitivity to weight fluctuations. Notably, even with the addition of detailed biochemical information to the weights, the stability of the node weights remained unaffected; as shown in Figure 6, all three methods maintained considerable stability across the entire perturbation range. These findings indicate that despite increased complexity, weighted methods retain robustness and reliability in networks.

[0174] This disclosure also evaluated whether high-weighted herbal targets (Top-N) could effectively cover high-weighted PD targets. As shown in Figure 7, TGD achieved the highest coverage, especially when using fewer herbal targets (from Top 100 to Top 600), indicating more efficient prioritization. This pattern was consistent across different PD target sets (Top 8, Top 100, Top 167).

[0175] In robustness and target coverage analyses, the three weighting methods showed similar trends, indicating that the choice of weighting strategy has a relatively small impact on the overall conclusions. However, Method 3, which comprehensively integrates herbal proportions, chemical composition, and compound-target interaction probabilities, is the most biologically sound and robust strategy and is the best among the three methods.

[0176] In evaluating the weighted proximity of four traditional Chinese medicine (TCM) formulas, this disclosure employs three node-weighted network-based metrics to comprehensively assess how each formula targets disease targets. Specifically, overlap coverage measures direct overlap, i.e., the proportion of herbal targets and disease targets that directly share targets; direct connectivity coverage assesses indirect regulation, referring to the proportion of herbal targets that are disease targets or have direct connections to disease targets; and Jaccard similarity measures the overall similarity between the herbal target set and the disease target set. As shown in Figure 8, among all metrics, TGD exhibits the strongest target involvement, followed by QZS, LWDH, and DBYW. The Jaccard similarity scores between herbal targets and disease targets are generally low, indicating limited direct overlap. Notably, the direct connectivity coverage values ​​for all formulas are close to 1.0, reflecting dense connections in the protein-protein interaction network and highlighting the significant potential of TCM formulas in indirectly regulating disease targets.

[0177] To systematically evaluate the role of node weights in network analysis, this disclosure calculates the weighted network proximity Z-scores for both cases with and without weights, considering both disease protein target weights and herbal target weights. The performance of the four formulas is compared across different PD disease target subsets (from Top 100 to Top 1000). A null distribution is generated through 500 rounds of isometric random sampling, serving as a reference for significance testing. Each Z-score reflects the deviation of the proximity between the herbal target set and the disease target set from the random result. The results show that, regardless of whether it's the Top 100 or Top 1000 disease target subsets, the weighted proximity Z-scores of all formulas are significantly lower than the unweighted case, indicating that the weighted network model can more sensitively capture the biological relevance of herbal interventions in disease networks. As shown in Figure 9, statistical tests show that the weighted proximity Z-scores of all formulas are significantly lower than the unweighted Z-scores (all p-values ​​are less than 0.05), demonstrating statistical significance.

[0178] Figure 9 shows a comparison of weighted and unweighted neighbor Z-scores for four traditional Chinese medicine formulas, highlighting the significant improvement effect of node weights. The neighbor Z-score is the mean and standard deviation calculated based on 500 random samples for each formula. Between-group statistical significance was assessed using a paired-samples t-test, comparing the neighbor Z-scores of each of the top N disease target sets with and without node weights. Weighted and unweighted represent the neighbor Z-scores calculated with and without node weights, respectively. A smaller Z-score indicates that the actual proximity between the medicinal herb target group and the disease target group in the network is much lower than expected under random conditions, suggesting that the intervention ability of medicinal herbs on the disease network is significantly stronger than that of randomly selected node groups.

[0179] It should be noted that Gastrodia and Uncaria Decoction (TGD) consistently showed the lowest (most negative) weighted proximity Z-score among all tested formulas, indicating its significant therapeutic potential against Parkinson's disease-related proteins at the network level. This finding is highly consistent with the literature mining results published in this paper, further confirming its therapeutic advantages. The above results demonstrate that constructing a node-weighted network by introducing node weights can significantly improve the discriminative power and predictive reliability of TCM formula network analysis, providing a more biologically meaningful quantitative basis for evaluating and optimizing TCM formulas. This supports TGD as the optimal choice for evaluating TCM formulas used for Parkinson's disease intervention.

[0180] In addition to the four common Chinese medicine prescriptions mentioned above, this embodiment further verifies the applicability and advantages of node-weighted networks in the mining of active ingredients of single Chinese medicines. Taking Ginkgo biloba leaf, a single medicinal material with neuroprotective effects, as an example, the potential of it to intervene in PD based on node-weighted networks is systematically evaluated.

[0181] First, the types and abundance information of small molecules in Ginkgo biloba leaves were obtained based on a local database. Then, by combining the relative abundance of compounds with the predicted compound-protein interaction probabilities, all components were screened and weighted. This embodiment constructed a weighted herbal target spectrum containing 1585 protein targets regulated by Ginkgo biloba leaf components. Further structural classification of the constituent compounds revealed dozens of chemical types, including aromatic polyketones, sesquiterpenes, flavonoids, lignans, and coumarins. As shown in Figure 10, flavonoids and isoflavones (collectively referred to as flavonoids) accounted for 46% of the total. Among them, 40% of the top 10 most abundant compounds belonged to flavonoids, collectively targeting 650 unique protein targets.

[0182] Furthermore, the intervention capabilities of the entire Ginkgo biloba compound target set and the flavonoid compound subset on the PD disease target network were evaluated separately. Proximity and the corresponding Z-scores were calculated using 500 rounds of degree-matched random sampling. As shown in Figures 11 and 12, the Z-score of the flavonoid subset was consistently lower than that of the complete compound set, indicating a stronger topological association with PD-related proteins. In addition, after node weighting, the proximity scores of Ginkgo biloba targets to the Top-N PD target sets significantly decreased (paired t-test, p < 1 × 10⁻⁶). -6 This further demonstrates that weighted networks can more accurately and sensitively reveal the biological associations between key active ingredients and disease networks.

[0183] Furthermore, to evaluate the ability of the method in this embodiment to identify key active ingredients and elucidate the molecular mechanism of Ginkgo biloba's anti-Parkinson's disease (PD) action, this embodiment uses the Boltz-2 model system to compare the predicted binding probabilities of flavonoids and non-flavonoids on 30 high-priority PD-related proteins. Boltz-2 is an open-source protein-small molecule binding affinity prediction tool developed by MIT, which can predict the strength of interactions between proteins and chemical ligands. Official tests show that it approaches the accuracy level of free energy perturbation (FEP) on standard benchmark sets, while maintaining an inference efficiency improvement of more than 1000 times. The input to Boltz-2 is the Simplified Molecular Input Line Entry System (SMILES) for small molecules and the amino acid sequence of the protein. The SMILES data of the compounds were obtained from the PubChem database, and the amino acid sequences of the proteins were obtained from the UniProt database. Because some proteins have a full length of more than 1,000 amino acids, which is beyond the range predicted by the model, the following three proteins were analyzed using truncated amino acid sequences: IGF1R (residues 961-1270), LRRK2 (residues 1879-2140), and ATP13A2 (residues 181-920); all other proteins were analyzed using their full-length sequences.

[0184] As shown in Figure 13, the y-axis of Figure 13 displays the probability of ligand-protein interaction calculated by the Boltz-2 model (range: 0 to 1). The higher the value, the greater the likelihood of binding. Statistical significance between groups was assessed using an independent two-sample Welch's t-test, with the significance level as follows: ns (p>0.05). (p≤0.05) (p≤0.01) (p≤0.001). In each violin plot, white dots represent the median, and thick black lines represent the quartile range (25th to 75th percentile). Differences in binding probabilities for each protein were assessed using an independent two-sample Welch's t-test. The results showed that among the 30 proteins, 24 proteins had significantly higher predicted binding probabilities for flavonoids than for non-flavonoids (p<0.001), further supporting the network pharmacological prediction that flavonoids tend to act on key PD targets.

[0185] Furthermore, as shown in Figure 14(A), this embodiment also utilizes a hierarchical clustering heatmap to visualize the predicted binding probability matrix between all flavonoid compounds and the aforementioned 24 high-confidence PD protein targets. The clustering analysis results show that, as shown in Figure 14(B), some flavonoids (such as isoflavones and their analogues) tend to act on protein clusters involved in processes such as neuronal apoptosis, dopamine metabolism, autophagy, and astrocyte development (as shown in Figure 14(C)). The heatmap in Figure 14(A) also displays the predicted binding probability matrix (range: 0-1, color-coded) between all flavonoid compounds from Ginkgo biloba leaves and the 24 high-priority Parkinson's disease (PD) protein targets. Clustering analysis is performed simultaneously on compounds (rows) and proteins (columns) according to similarity in the figure. Figure 14(B) shows the representative chemical structures of isoflavones and their analogues, which are shaded in Figure 14(A). These compounds, including GBA1, DDC, ATP13A2, PRKN, GCH1, IL1B, GFAP, SNCA, CASP9, and MAPT, were identified as candidate molecules with preferential binding ability to key PD proteins. They were selected based on their high average binding probability among 30 protein targets, indicating broad-spectrum binding potential. These compounds were selected due to their high average binding probability among docking with 30 disease-related proteins, suggesting their broad-spectrum binding potential. The 30 proteins involved in the docking include mitogen-activated protein kinase 1 (MAPK1), leucine-rich repeat serine / threonine protein kinase 2 (LRRK2), insulin-like growth factor 1 receptor (IGF1R), synaptic vesicle amine transporter (SLC18A2), lysosomal acid glucosylceramidinase (GBA1), aromatic L-amino acid decarboxylase (DDC), polyamine transport ATPase 13A2 (ATP13A2), E3 ubiquitin protein ligase Parkin (PRKN), GTP cyclohydrolase 1 (GCH1), interleukin-1β (IL1B), and glial fibrillary acidic protein (GFAP). The protein clusters involved in flavonoid activity, as shown in Figure 14 (C), include α-synuclein (SNCA), caspase-9 (CASP9), microtubule-associated protein tau (MAPT), MAPK-activated protein kinase 3 (MAPK3), RAC-α serine / threonine protein kinase (AKT1), β-nerve growth factor (NGF), tyrosine hydroxylase (TH), caspase-3 (CASP3), Parkinson's disease protein 7 (PARK7), superoxide dismutase [Cu-Zn] (SOD1), mitochondrial manganese superoxide dismutase (SOD2), catechol-O-methyltransferase (COMT), and heme oxygenase 1 (HMOX1). The functional enrichment analysis of the flavonoid clusters in (C) involves Gene Ontology terms including neuronal apoptosis, dopamine metabolism, autophagy, astrocyte development, and proteolysis (enrichment significance is expressed as log). 10(p) indicates this. This analysis highlights the functional clustering characteristics of PD targets most susceptible to flavonoid regulation.

[0186] Therefore, flavonoids and isoflavones are the main active chemical categories in Ginkgo biloba that exert anti-PD effects. This embodiment integrates node importance indicators into a proximity-based network model, significantly enhancing the predictive power of network pharmacology analysis. This not only enables quantitative assessment of target effects at the system level but also facilitates the identification of key active ingredients, providing a scientific basis for the simplification and optimization of traditional Chinese medicine prescriptions.

[0187] In summary, because related technologies typically focus only on the number of simple connections or overlaps between nodes, they cannot truly reflect the actual ability of medicinal materials or components to regulate diseases. Evaluation results are often interfered with by highly expressed but biologically limited redundant nodes. Furthermore, when performing tasks such as screening different formulations for superiority or inferiority, explaining pharmacodynamic mechanisms, and identifying core targets, there is a lack of objective and quantifiable ranking criteria, affecting the scientific validity and guiding value of the analysis. For needs such as optimizing medicinal materials and simplifying formulations, related technologies struggle to sensitively distinguish the essential differences between highly effective formulations and random combinations, easily leading to inflated or understated intervention evaluations. This disclosure, however, systematically compares the effects of introducing a node-weighted network (node-weighted network) and not introducing a weighted network on four typical traditional Chinese medicine formulas and the single herb Ginkgo biloba leaf. Specifically, it calculates the network proximity Z-score between the herbal target and the disease target set under weighted and unweighted protein interaction networks. To ensure the objectivity of the comparison, this application generates a zero distribution through 500 rounds of isometric random sampling, strictly controlling the network topology attributes, which serve as a reference for significance testing. Each Z-score reflects the degree of deviation between the distance between the medicinal target and the disease target network and the random result.

[0188] The results showed that, across different disease target subsets such as the Top 100 and Top 1000, the weighted proximity Z-scores of all classic Chinese medicine compound formulas were significantly lower than those of the unweighted cases, and this difference was statistically highly significant (t-test p-values ​​were all less than 0.05). This result indicates that node-weighted networks can more sensitively capture the biological relevance of medicinal interventions to disease networks, effectively overcoming the information loss and spurious intervention risks caused by the neglect of node functional differences in traditional unweighted models. Specifically, a lower Z-score indicates a more significant intervention effect of the medicinal material or component on the disease network, reflecting a stronger ability to reflect actual biological effects.

[0189] Figure 9 shows a comparison of weighted and unweighted neighborhood Z-scores for four TCM formulas, highlighting the significant effect of node weighting. Neighborhood Z-scores are calculated based on the mean and standard deviation of 500 random samples for each formula. Between-group statistical significance was assessed using a paired-samples t-test, comparing the neighborhood Z-scores of each of the top N disease target sets with and without node weights. Weighted and unweighted represent the neighborhood Z-scores calculated with and without node weights, respectively. A smaller Z-score indicates that the actual proximity between the herbal target group and the disease target group in the network is much lower than expected under random conditions, suggesting that the herbal intervention ability on the disease network is significantly stronger than that of randomly selected node groups.

[0190] Furthermore, to verify the applicability of this technology in the identification of single medicinal materials and their active ingredients, this embodiment uses Ginkgo biloba leaves and their flavonoids / isoflavones as examples. The same weighted and unweighted methods were used to calculate their respective intervention capabilities on the PD-related target network. The results show that the Z-score of the flavonoid / isoflavone subset is consistently lower than that of the complete compound set, indicating a stronger topological association with PD-related proteins. Particularly under the weighted network, the proximity score of Ginkgo biloba targets to the Top-N PD target set was significantly reduced (paired t-test, p < 1 × 10⁻⁶). -6 This further validates that the weighted scheme can more accurately and sensitively reveal the essential biological links between key active ingredients and disease networks.

[0191] Therefore, compared to unweighted node networks, this disclosure, based on a node-weighted network, can significantly improve the ability to resolve the true biological associations of formulations or ingredients within disease networks. It assigns quantitative importance weights to disease protein targets and medicinal protein targets, avoiding information dilution and honorific noise caused by equal node weights, and preventing inflated interventions or underestimation of efficacy. This provides a more objective, quantitative, and sensitive evaluation tool for practical applications such as medicinal material optimization, formulation simplification, and identification of core effective components. It possesses good generalization and adaptability, applicable to various scenarios involving the evaluation of diseases, medicinal materials, active ingredients, and formulations, enriching research on the modernization and precision formulation of traditional Chinese medicine.

[0192] To implement the above method, this application also provides an example of a traditional Chinese medicine efficacy evaluation device 1500 based on a node-weighted network, as shown in Figure 15. This device includes:

[0193] The disease target processing module 1501 is used to determine the candidate disease protein targets corresponding to the target disease, and to determine the target disease protein targets and their corresponding weights among the candidate disease protein targets based on the comprehensive scoring results of the candidate disease protein targets from the public database and the large language model.

[0194] The herbal target processing module 1502 is used to determine the prescription ratio and chemical composition of each herb in the Chinese medicine prescription to be evaluated, and to determine the herbal protein target and the corresponding herbal protein target weight based on the prescription ratio, the component information of the chemical composition and the interaction probability between the chemical composition and the herbal target.

[0195] The target network construction module 1503 is used to map the target disease protein target and the corresponding disease protein target weights and the medicinal material protein target and the corresponding medicinal material protein target weights to a pre-constructed protein interaction network to obtain the corresponding node weighted network.

[0196] The evaluation result generation module 1504 is used to determine multidimensional network indicators for evaluating the efficacy of traditional Chinese medicine prescriptions based on the node weighted network.

[0197] By way of example, this application also provides an electronic device, including:

[0198] processor;

[0199] Memory used to store processor-executable instructions;

[0200] The processor is used to read executable instructions from memory and execute the instructions to implement the above-mentioned method for evaluating the efficacy of traditional Chinese medicine based on node-weighted networks.

[0201] For example, this application also provides a computer-readable storage medium storing a computer program for executing the above-described method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network.

[0202] Figure 16 illustrates a schematic block diagram of an example electronic device that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0203] As shown in Figure 16, device 400 includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 402 or a computer program loaded from storage unit 408 into random access memory (RAM) 403. RAM 403 may also store various programs and data required for the operation of device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0204] Multiple components in device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of monitors, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0205] The computing unit 401 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as a method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network. For example, in some embodiments, a method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed on device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of a method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network described above can be performed. Alternatively, in other embodiments, computing unit 401 may be configured by any other suitable means (e.g., by means of firmware) to perform a method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network.

[0206] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0207] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0208] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0209] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this application can be achieved, and this is not limited herein.

[0210] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means two or more, unless otherwise explicitly specified.

[0211] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network, characterized in that, The method includes: identifying candidate disease protein targets corresponding to the target disease; determining the weights of the target disease protein targets and corresponding disease protein targets among the candidate disease protein targets based on the comprehensive scoring results of the candidate disease protein targets using public databases and large language models; wherein, determining the weights of the target disease protein targets and corresponding disease protein targets among the candidate disease protein targets based on the comprehensive scoring results of the candidate disease protein targets using public databases and large language models includes: determining search keywords, wherein the search keywords include the Western medicine disease name and the traditional Chinese medicine disease name corresponding to the traditional Chinese medicine prescription to be evaluated; for each candidate disease protein target, performing searches in at least one public database based on the search keywords. The process involves: retrieving and determining the retrieval score corresponding to the target; when using multiple public databases, combining the scores from each public database to obtain the target's public database-based score; extracting semantic features from the target English and Chinese literature using a large language model to determine the semantic relevance score for each candidate disease protein target; multiplying and weighting the retrieval score and the semantic relevance score to determine the comprehensive score for each candidate disease protein target as its comprehensive weight; screening targets based on their comprehensive weights to obtain the target disease protein targets and their corresponding weights; and determining the prescription ratio and chemical composition of each medicinal material in the traditional Chinese medicine formula to be evaluated. Based on the prescription ratio, the component information of the chemical components, and the interaction probability between the chemical components and the medicinal material targets, the medicinal material protein targets and their corresponding weights are determined. Specifically, determining the prescription ratio and chemical components corresponding to each medicinal material in the traditional Chinese medicine prescription to be evaluated, and determining the medicinal material protein targets and their corresponding weights based on the prescription ratio, the component information of the chemical components, and the interaction probability between the chemical components and the medicinal material targets, includes: determining the prescription ratio, chemical components, and component information corresponding to each chemical component in the traditional Chinese medicine prescription to be evaluated, wherein the component information includes the relative abundance of the chemical components; and determining the corresponding... All target links, each used to characterize the association between the medicinal material and its corresponding chemical components and corresponding medicinal material protein targets; for each target link, the link weight is determined based on the product of the prescription ratio of the corresponding medicinal material, the relative abundance of the chemical components, and the probability of interaction between the chemical components and the medicinal material protein targets, and is used as the medicinal material protein target weight in the target link; the target disease protein targets and their corresponding disease protein target weights are mapped to a preset protein interaction network with the medicinal material protein targets and their corresponding medicinal material protein target weights to obtain the corresponding node weighted network; and multidimensional network indicators for evaluating the efficacy of traditional Chinese medicine prescriptions are determined based on the node weighted network.

2. The method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network according to claim 1, characterized in that, For each target link, the link weight is determined based on the product of the prescription ratio of the corresponding medicinal material, the relative abundance of chemical components, and the probability of interaction between the chemical components and the protein target of the medicinal material, and is used as the protein target weight of the medicinal material in that link. This includes: for each target link, determining the normalized prescription ratio and the normalized relative abundance of chemical components of the corresponding medicinal material in that link; determining the molecular structure of the chemical components and the binding pocket structure of the protein target of the medicinal material in that link, inputting the molecular structure and the binding pocket structure into a deep learning prediction model to determine the interaction probability between the corresponding chemical components and the protein target of the medicinal material; and determining the link weight based on the product of the normalized prescription ratio, the normalized relative abundance of chemical components, and the probability of interaction between the corresponding chemical components and the protein target of the medicinal material, and using this weight as the protein target weight of the medicinal material in that link.

3. The method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network according to claim 2, characterized in that, For each target link, the link weight is determined based on the product of the prescription ratio of the corresponding medicinal material, the relative abundance of chemical components, and the probability of interaction between the chemical components and the protein target of the medicinal material, and is used as the protein target weight of the medicinal material in the link. The method further includes: for each protein target, traversing all relevant target links and summing the link weights of all relevant target links as the protein target weight of the target.

4. The method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network according to claim 3, characterized in that, For each target link, the link weight is determined based on the product of the prescription ratio of the corresponding medicinal material, the relative abundance of chemical components, and the probability of interaction between the chemical components and the protein target of the medicinal material, and is used as the protein target weight of the medicinal material in the link. The method also includes: normalizing the protein target weights of all medicinal material protein targets.

5. The method for evaluating the efficacy of traditional Chinese medicine based on node-weighted networks according to any one of claims 1-4, characterized in that, The multidimensional network metrics include overlap coverage, direct connection coverage, Jaccard similarity, weighted proximity, and proximity Z-score.

6. A device for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network, characterized in that, The device includes: a disease target processing module, used to determine candidate disease protein targets corresponding to a target disease, and to determine the target disease protein target and its corresponding weight among the candidate disease protein targets based on the comprehensive scoring results of the candidate disease protein targets using a public database and a large language model; wherein, the step of determining the target disease protein target and its corresponding weight among the candidate disease protein targets based on the comprehensive scoring results of the candidate disease protein targets using a public database and a large language model includes: determining search keywords, the search keywords including the Western medicine disease name and the traditional Chinese medicine disease name corresponding to the traditional Chinese medicine prescription to be evaluated; for each candidate disease protein target, performing [further analysis] in at least one public database based on the search keywords. The process involves several steps: First, a search is performed to determine the search score corresponding to the target. When using multiple public databases, the scores from each database are combined to obtain a database-based score for the target. Second, a large language model is used to extract semantic features from the target English and Chinese literature to determine the semantic relevance score for each candidate disease protein target. Third, the search score and the semantic relevance score are multiplied and weighted to determine the comprehensive score for each candidate disease protein target, which serves as the comprehensive weight for that target. Fourth, target selection is performed based on the comprehensive weight of each candidate disease protein target to obtain the target disease protein target and its corresponding weight. Finally, a medicinal herb target processing module is used to determine the prescription ratio and chemical composition of each medicinal herb in the traditional Chinese medicine formula to be evaluated. Based on the prescription ratio, the component information of the chemical components, and the interaction probability between the chemical components and the medicinal material targets, the medicinal material protein targets and their corresponding weights are determined. Specifically, determining the prescription ratio and chemical components corresponding to each medicinal material in the traditional Chinese medicine prescription to be evaluated, and determining the medicinal material protein targets and their corresponding weights based on the prescription ratio, the component information of the chemical components, and the interaction probability between the chemical components and the medicinal material targets, includes: determining the prescription ratio, chemical components, and component information corresponding to each chemical component in the traditional Chinese medicine prescription to be evaluated, where the component information includes the relative abundance of the chemical components; and determining all target links corresponding to each medicinal material from the target knowledge base, with each target... The link is used to characterize the association between the medicinal material and its corresponding chemical components and corresponding medicinal material protein targets. For each target link, the link weight is determined based on the product of the prescription ratio of the corresponding medicinal material, the relative abundance of the chemical components, and the probability of interaction between the chemical components and the medicinal material protein targets on the target link. This weight is then used as the medicinal material protein target weight in the target link. The target network construction module is used to map the target disease protein targets and their corresponding disease protein target weights to the medicinal material protein targets and their corresponding medicinal material protein target weights onto a preset protein interaction network to obtain the corresponding node weighted network. The evaluation result generation module is used to determine multidimensional network indicators for evaluating the efficacy of traditional Chinese medicine prescriptions based on the node weighted network.

7. An electronic device, characterized in that, include: At least one processor; The at least one processor is connected to a memory; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the traditional Chinese medicine efficacy evaluation method based on a node-weighted network as described in any one of claims 1-5.

8. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the method for evaluating the efficacy of traditional Chinese medicine based on a node-weighted network according to any one of claims 1-5.

Images