A knowledge graph-based terpenoid compound and disease correlation prediction method and system

Abstract

The present application relates to the technical field of artificial intelligence and pharmacology, and proposes a terpenoid compound and disease correlation prediction method and system based on a knowledge graph, comprising the following steps: collecting terpenoid compound and biological activity data, including terpenoid compound molecular structure, protein target, gene target, cell line and disease correlation information, and constructing a terpenoid compound biological activity knowledge graph; inputting the molecular descriptor of the terpenoid compound to be predicted into a convolutional neural network for feature extraction, to generate a first feature vector; performing feature extraction on the molecular structure of the terpenoid compound to be predicted through a graph convolution network, to generate a second feature vector; inputting the molecular embedding vector obtained by splicing the first feature vector and the second feature vector into a knowledge graph embedding model, to generate a terpenoid compound-disease prediction score matrix; and performing descending order sorting based on the terpenoid compound-disease prediction score, to generate a disease recommendation list related to the terpenoid compound.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence and pharmacology, and more particularly, to a terpenoid compound and disease correlation prediction method and system based on a knowledge graph. BACKGROUND

[0002] As a natural compound widely existing in plants and other organisms, terpenoids have attracted extensive attention in the field of drug research and development due to their diverse structures and extensive biological activities. Terpenoids exhibit potential in various pharmacological activities such as anti-inflammatory, anti-tumor, and anti-microbial. However, the confirmation of the pharmacological activity of terpenoids mainly relies on traditional experimental methods, including in vitro cell experiments, animal experiments, and clinical trials. Although these experimental methods are widely recognized and technically feasible, they are often time-consuming and accompanied by high resource and time investment, limiting the efficiency of new drug discovery and development.

[0003] With the increasing maturity of artificial intelligence in drug research and development, machine learning and deep learning-based algorithms can quickly process and analyze large amounts of data, significantly accelerating the process of drug activity screening, drug target discovery, and drug-disease correlation prediction. Currently, there are proposals to construct a knowledge graph dataset centered on drugs and targets, learn the feature information of drugs, targets, and their neighborhoods through artificial intelligence models, and quickly predict the potential therapeutic effect of a terpenoid compound on a specific disease, thereby reducing the time and cost of experimental verification. This type of model can extract the commonalities and regularities of existing terpenoid compounds and diseases based on existing association data, and apply it to the drug development of new terpenoid compounds, providing direction guidance for future drug research and development. However, there is currently a lack of comprehensive multi-dimensional relationship network data covering the activity profile of terpenoids and potential disease prediction models, resulting in limitations of disease correlation prediction results in specific scenarios. SUMMARY

[0004] To overcome the above-mentioned defects of single data source in the prior art, which leads to limitations of disease correlation prediction results in specific scenarios, the present application provides a terpenoid compound and disease correlation prediction method and system based on a knowledge graph.

[0005] To solve the above technical problems, the technical solutions of the present application are as follows:

[0006] A terpenoid compound and disease correlation prediction method based on a knowledge graph, comprising the following steps:

[0007] Collecting terpenoid compound and biological activity data, including terpenoid compound molecular structure, protein target, gene target, cell line, and disease correlation information, and constructing a terpenoid compound biological activity knowledge graph;

[0008] The molecular descriptor of the terpenoid compound to be predicted is input into a convolutional neural network for feature extraction, to generate a first feature vector related to the physical and chemical properties of the molecule;

[0009] The molecular structure of the terpenoid compound to be predicted is subjected to feature extraction by a graph convolution network, to generate a second feature vector related to the terpenoid molecule graph feature;

[0010] The molecular embedding vector obtained by splicing the first feature vector and the second feature vector is subjected to a knowledge graph embedding model, to generate a terpenoid compound-disease prediction score matrix;

[0011] Based on the terpenoid compound-disease prediction score, a descending order sorting is performed to generate a disease recommendation list related to the terpenoid compound.

[0012] Further, the present application also proposes a knowledge graph-based terpenoid compound and disease correlation prediction system, which applies the terpenoid compound and disease correlation prediction method proposed by the present application. The system comprises:

[0013] A knowledge graph construction module is configured to collect terpenoid compound and biological activity data, including terpenoid compound molecular structure, protein target, gene target, cell line and disease correlation information, and construct a terpenoid compound biological activity knowledge graph.

[0014] A first feature extraction module is configured to input the molecular descriptor of the terpenoid compound to be predicted into a convolutional neural network for feature extraction, to generate a first feature vector related to the physical and chemical properties of the molecule.

[0015] A second feature extraction module is configured to input the molecular structure of the terpenoid compound to be predicted into a graph convolution network for feature extraction, to generate a second feature vector related to the terpenoid molecule graph feature.

[0016] A knowledge graph embedding module is configured to input the molecular embedding vector obtained by splicing the first feature vector and the second feature vector into a knowledge graph embedding model, to generate a terpenoid compound-disease prediction score matrix.

[0017] An output module is configured to perform descending order sorting based on the terpenoid compound-disease prediction score, to generate a disease recommendation list related to the terpenoid compound.

[0018] Further, the present application also proposes a device comprising a memory and a processor, wherein the memory stores computer readable instructions, and when the computer readable instructions are executed by the processor, the processor executes all or part of the steps of the knowledge graph-based terpenoid compound and disease correlation prediction method according to the present application.

[0019] Further, the present application also proposes a storage medium having computer readable instructions stored thereon, wherein the computer readable instructions, when executed by a processor, implement all or part of the steps of the knowledge graph-based terpenoid and disease correlation prediction method according to the present application.

[0020] Compared with the prior art, the beneficial effects of the technical scheme of the present application are:

[0021] The present application collects terpenoid and biological activity related data from multiple public databases and constructs a biological activity knowledge graph of terpenoids, which comprehensively integrates multi-dimensional information such as terpenoids, target proteins, genes, cell lines and diseases, solves the problem of single data source and incomplete coverage in the prior art, and further combines a knowledge graph embedding model to predict the related diseases of target terpenoids, thereby providing a reliable data basis for the research of terpenoids.

[0022] The present application combines molecular physical and chemical characteristics, two-dimensional graph structure characteristics and knowledge graph embedding technology, constructs molecular embedding vectors embedded with rich molecular structure and biological information, and further fills in the relationship information in the incomplete knowledge graph through a prediction model based on knowledge graph embedding technology, thereby effectively improving the accuracy of correlation prediction.

[0023] The present application can recommend candidate diseases related to terpenoids, and predict diseases with higher scores as potential target diseases, thereby providing important candidate diseases for subsequent experimental or clinical verification. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 A flowchart of the knowledge graph-based terpenoid and disease correlation prediction method according to an embodiment of the present application is shown.

[0025] Figure 2 A flowchart of the terpenoid and disease correlation prediction method according to an embodiment of the present application is shown.

[0026] Figure 3 An architecture diagram of the biological activity knowledge graph of terpenoids according to an embodiment of the present application is shown.

[0027] Figure 4 An experimental result diagram of testing the anti-melanoma effect of terpenoid TKC014820 on A375 cells is shown.

[0028] Figure 5 An experimental result diagram of testing the anti-melanoma effect of terpenoid TKC117556 on A375 cells is shown.

[0029] Figure 6A schematic diagram of a terpenoid compound and disease correlation prediction system based on a knowledge graph according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0030] The exemplary embodiments will be described in detail herein below with reference to the accompanying drawings. In the following description, the same or similar components are designated by the same or similar reference numerals, and a repeated description thereof will be omitted. The embodiments described in the following exemplary embodiments do not represent all the technical ideas falling within the scope of the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application, as recited in the appended claims.

[0031] The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0032] It is to be understood that the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. It is to be further understood that the terms "comprise", "comprising", "comprises", "including", "comprising", "comprises", "include" and / or "comprising", when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0033] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] Embodiment 1

[0035] The present embodiment proposes a terpenoid compound and disease correlation prediction method based on a knowledge graph, as shown in Figure 1 , 2 The present embodiment proposes a terpenoid compound and disease correlation prediction method based on a knowledge graph, as shown in

[0036] The present embodiment proposes a terpenoid compound and disease correlation prediction method based on a knowledge graph, as shown in

[0037] S1, collecting terpenoid compound and biological activity data, including terpenoid compound molecular structure, protein target, gene target, cell line and disease correlation information, and constructing a terpenoid compound biological activity knowledge graph;

[0038] S2, input the molecular descriptor of the terpenoid compound to be predicted into the convolutional neural network for feature extraction, to generate a first feature vector related to the physical and chemical properties of the molecule;

[0039] S3, the molecular structure of the terpenoid compound to be predicted is extracted by a graph convolutional network to generate a second feature vector related to the terpenoid molecular graph feature;

[0040] S4, the molecular embedding vector obtained by splicing the first feature vector and the second feature vector is input into a knowledge graph embedding model to generate a terpenoid compound-disease prediction score matrix;

[0041] S5, based on the terpenoid compound-disease prediction score, descending order sorting is performed to generate a disease recommendation list related to the terpenoid compound.

[0042] The embodiment collects terpenoid compounds and their biological activity related data from multiple public databases and constructs a biological activity knowledge graph of terpenoid compounds, which comprehensively integrates multi-dimensional information such as terpenoid compounds, target proteins, genes, cell lines and diseases, solves the problem of single data source and incomplete coverage in the prior art, and further combines a knowledge graph embedding model to predict the related diseases of the target terpenoid compound, thereby providing a reliable data basis for the research of terpenoid compounds.

[0043] The embodiment further combines molecular physical and chemical features, two-dimensional graph structure features and knowledge graph embedding technology, constructs a global molecular embedding vector, which embeds rich molecular structure and biological information, so that the molecular feature description is more comprehensive, and the prediction model based on the knowledge graph embedding technology fills in the relationship information in the incomplete knowledge graph, thereby effectively improving the accuracy of the correlation prediction. In addition, the embodiment can recommend candidate diseases related to terpenoid compounds, and predict diseases with higher scores as potential target diseases, thereby providing important candidate diseases for subsequent experiments or clinical verification. This data-driven method can reduce research and development costs and improve research efficiency.

[0044] In an optional embodiment, when collecting terpenoid compounds and their biological activity data, data can be collected from public open source databases such as DrugBank, ChEMBL, PubChem, TTD, CTD, GeneCards, Cellosauru, Uniport, TCMBank, TeroKIT, etc. Further, the collected data is structured and standardized, and a terpenoid compound biological activity knowledge graph is constructed.

[0045] For example, the substructure information, protein targets, cell targets and the association data of terpenoid compounds with their molecules and targets are collected from the TeroKit platform, approved terpenoid small molecule drugs are extracted from the DrugBank database, herbal ingredients classified as terpenes are extracted from the TCMbank database, and cell targets and their association information with rare human diseases are collected from the Cellosaurus database.

[0046] In another example, by integrating the CTD, GeneCards, and TTD databases, using genes as a bridge, protein targets were classified by disease type, mapped with the TeroKit database, and data collection for data analysis related to the interaction between terpenoid compounds and protein targets was completed.

[0047] In an optional embodiment, when constructing the knowledge graph of terpenoid bioactivity in step S1, the following steps are included:

[0048] Preprocessing the collected terpenoid compounds and their biological activity data;

[0049] Integrate the activity associations among five biological entities: terpenoids, protein targets, genes, cell lines, and diseases;

[0050] After filtering the interaction data through a preset activity threshold, the data are stored in the form of triples to construct a knowledge graph of the bioactivity of terpenoids.

[0051] In this example, by integrating the association between terpenoid compounds and targets, and using genes as a bridge to classify targets and diseases, the relationship between terpenoid compound entities and disease entities is obtained, and a knowledge graph of the bioactivity of terpenoid compounds is constructed in the form of triples (head entity-relationship-tail entity).

[0052] like Figure 3 , which is the architecture diagram of the terpenoid compound bioactivity knowledge graph of this embodiment.

[0053] For example, the preset activity threshold is set to 10 μM to filter low-confidence interaction data.

[0054] Further optionally, when pre-processing the collected terpenoid compounds and their biological activity data, the following steps are included:

[0055] Standardize the data, including cleaning, identifier unification, and data deduplication.

[0056] This embodiment standardizes the collected data before constructing the knowledge graph, which can adjust the scale of the data so that data with different characteristics have the same distribution characteristics, thereby improving the efficiency and effectiveness of model training.

[0057] In an optional embodiment, in the step S2, when inputting the molecular descriptor of the terpenoid to be predicted into the convolutional neural network for feature extraction, the following steps are included:

[0058] calculating the molecular descriptor of the terpenoid; the molecular descriptor includes Wildman-Crippen distribution coefficient, topological polar surface area, molecular weight, drug-likeness, molar refractivity, number of hydrogen bond acceptors, number of hydrogen bond donors, number of aromatic rings, number of rotatable bonds, and fraction of carbon atoms SP3 hybridization (FractionCSP3);

[0059] splicing all the molecular descriptors to obtain a molecular feature matrix; inputting the molecular feature matrix into the convolutional neural network for feature extraction to obtain a first feature vector representing the physical and chemical properties of the terpenoid.

[0060] In this embodiment, the molecular descriptor is used to represent the characteristics of the terpenoid, wherein the molecular descriptor of the terpenoid is calculated by the RDKit library.

[0061] The Wildman-Crippen distribution coefficient is used to measure the distribution of the compound in water and lipid, and is used to quantify the hydrophobicity of each atom in the molecule.

[0062] The topological polar surface area (TPSA) represents the surface area of polar atoms in the compound, which is commonly used to evaluate the solubility of the compound, the transportability of the drug in the cell, and the small intestine absorption of the drug in the human body, Caco-2 monolayer permeability and blood brain barrier penetration, i.e. to predict the ability of the molecule to penetrate the cell membrane.

[0063] The molecular weight represents the molecular mass of the compound, which has a certain influence on the pharmacokinetic characteristics of the drug.

[0064] Drug-likeness is a comprehensive score calculated according to the physical and chemical properties of the drug, which is used to evaluate the potential of the compound as a drug.

[0065] Molar refractivity represents the polarization of the molecule, which is related to the volume and density of the molecule.

[0066] The number of hydrogen bond acceptors is the number of atoms or groups in the compound that can form hydrogen bonds.

[0067] The number of hydrogen bond donors is the number of atoms or groups in the compound that provide hydrogen bonds.

[0068] The number of aromatic rings is the number of aromatic rings in the compound, which affects the stability and binding ability of the molecule.

[0069] The number of rotatable bonds is the number of single bonds in a molecule that can rotate freely, which affects the conformational freedom and flexibility of the molecule.

[0070] Fraction of carbon atoms sp 3 Fraction of sp 3 Hybridized carbon atoms, often used to evaluate the complexity of molecular structure.

[0071] In this embodiment, all molecular descriptors are spliced to generate a molecular feature matrix, forming a comprehensive feature representation combining molecular physical and chemical characteristics, and embedding biological information to make the molecular feature description more comprehensive. Further, the molecular feature matrix is input into a convolutional neural network, and then sequentially passes through a convolutional layer, a pooling layer, a normalization layer, and a Tanh activation layer to generate a feature vector representing the physical and chemical properties of the terpenoid compound.

[0072] In an optional embodiment, when the molecular structure of the terpenoid compound to be predicted is feature-extracted by a graph convolutional network in the S3 step, the following steps are included:

[0073] Generating a molecular graph based on the molecular structure of the terpenoid compound; wherein atoms are used as nodes and chemical bonds are used as edges to form a two-dimensional graph structure of the terpenoid compound molecule;

[0074] Extracting an atomic adjacency matrix from the molecular graph, and generating an atomic feature matrix based on the atomic type, then inputting the atomic adjacency matrix and the atomic feature matrix into a graph convolutional network to generate a second feature vector representing global molecular features.

[0075] In this embodiment, the chemical molecule is represented as a two-dimensional graph structure, which can capture the structural features of the chemical molecule using the structural relationship between nodes and edges, especially the representation of chemical bonds. Among them, the atomic feature matrix is generated based on the atomic types of C, N, O, S, etc., the atomic adjacency matrix representing the spatial interconnection of atoms within the molecule is extracted from the molecular graph, and then input into a graph convolutional network (GCN), and then sequentially pass through a Readout layer, a normalization layer and a Tanh activation layer to generate a global molecular feature vector.

[0076] In an optional embodiment, in the S4 step, the molecular embedding vector obtained by splicing the first feature vector and the second feature vector is input into a knowledge graph embedding model to generate a terpenoid compound-disease prediction score matrix, including the following steps:

[0077] Splicing the first feature vector and the second feature vector through a fully connected layer to generate a molecular embedding vector;

[0078] Generating a disease entity embedding vector using a random adjacency matrix and a relationship embedding vector using a random initialization vector of the same dimension;

[0079] The molecular embedding vector, disease embedding vector and relationship embedding vector are input into the DistMult model, the score of each triple is calculated, and a terpenoid-disease prediction score matrix is obtained.

[0080] The embodiment combines CNN, GCN and knowledge graph embedding (KGE), constructs a global molecular feature vector, and embeds rich structural and biological information, so that the molecular feature description is more comprehensive. Further, the DistMult model is used as the knowledge graph embedding model, which represents the interaction between entities and relationships through a bilinear model, aiming to combine the context of knowledge graph and relationship data, represent and infer the relationship between entities, and help to improve the current incomplete terpenoid bioactivity knowledge graph by filling in the missing relationship information.

[0081] Specifically, the DistMult model uses a simple and effective scoring function based on inner product to capture the second-order correlation between entities, which is represented as:

[0082] f r (h,t)=h⊙(r⊙t)

[0083] Where h, r and t represent the vector representation of the head entity, the relationship and the tail entity respectively, and represents the element-wise product between vectors.

[0084] Further optionally, the scoring function f r configured in the DistMult model in the embodiment is represented as:

[0085] f r (t,d)=t T W r d

[0086] Where t represents the molecular embedding vector, d represents the disease entity embedding vector, W r is the weight matrix of the relationship r between the molecular embedding vector and the disease embedding vector.

[0087] The embodiment evaluates the credibility of the binary pair (t, d) through the scoring function f r , and obtains the terpenoid-disease prediction score.

[0088] Further, according to the output of the scoring function f r , the prediction scores of all potential terpenoid-disease associations are sorted in descending order, and a disease recommendation list related to the terpenoid is generated. The diseases ranked at the top in the recommendation list are the potential action diseases of the terpenoid.

[0089] The scoring function in this embodiment quantifies the association between each disease and the terpenoid compound according to the feature expression information in the terpenoid compound biological activity knowledge graph constructed in S1 step, can automatically recommend candidate diseases related to the terpenoid compound, and predict diseases with higher scores as potential target diseases, providing important candidate diseases for subsequent experimental or clinical verification.

[0090] Exemplarily, the structures of two terpenoid compounds TKC014820 and TKC117556 to be tested are input into the model in the form of SMILES sequence. The model includes the terpenoid compound biological activity knowledge graph proposed in this embodiment, an encoding layer composed of convolutional neural networks, graph convolutional networks, etc., and a decoding layer composed of knowledge graph embedding models, etc. The model can predict all association scores of the two terpenoid compounds with the 775 disease entities contained in the knowledge graph. According to the scores, a complete disease recommendation list is obtained by ranking from high to low. The top diseases in the list are inferred as the potential related diseases of the two terpenoid compounds.

[0091] The prediction result shows that the prediction scores of melanoma are higher, ranking 1 and 2 respectively, indicating the potential association between the two compounds and melanoma. Therefore, the anti-melanoma effect of the two compounds on human malignant melanoma A375 cells is tested.

[0092] The specific operation of the experiment is as follows: A375 cells are seeded in a 96-well plate at a density of 1x10^4 cells per well, fresh culture medium is added and incubated overnight. After incubation, the cells are treated with different concentrations of compounds for 24-48 hours. After removing the supernatant, the cells are labeled with a cell counting kit-8 (CCK8) solution according to the manufacturer's instructions. Then incubate at 37°C for 3 hours, and read the absorbance at 450 nm using a microplate reader.

[0093] As shown in Figure 4 , 5 , the experimental results of this embodiment are shown schematically. The IC50 values of the two compounds for anti-melanoma at the cellular level are 3.842 nM and 3.27 μM respectively, indicating that the two terpenoid compounds are indeed significantly related to melanoma disease as shown in the prediction results of the model.

[0094] Example 2

[0095] This embodiment applies the terpenoid compound and disease correlation prediction method proposed in Example 1 to propose a knowledge graph-based terpenoid compound and disease correlation prediction system, as shown in Figure 6 , the architecture diagram of the terpenoid compound and disease correlation prediction system of this embodiment.

[0096] The terpenoid compound and disease related prediction system provided in the embodiment comprises:

[0097] A knowledge graph construction module is configured to collect terpenoid compound and biological activity data, including terpenoid compound molecular structure, protein target, gene target, cell line and disease correlation information, and construct a terpenoid compound biological activity knowledge graph.

[0098] A first feature extraction module is configured to input the molecular descriptor of the terpenoid compound to be predicted into a convolutional neural network for feature extraction, and generate a first feature vector related to the physical and chemical properties of the molecule.

[0099] A second feature extraction module is configured to extract the molecular structure of the terpenoid compound to be predicted through a graph convolution network to generate a second feature vector related to the terpenoid molecule graph feature.

[0100] A knowledge graph embedding module is configured to input the molecular embedding vector obtained by splicing the first feature vector and the second feature vector into a knowledge graph embedding model to generate a terpenoid compound-disease prediction score matrix.

[0101] An output module is configured to sort the terpenoid compound-disease prediction scores in descending order to generate a disease recommendation list related to the terpenoid compound.

[0102] It can be understood that the system of the embodiment corresponds to the method of embodiment 1 described above, and the optional items in embodiment 1 also apply to the embodiment, so they will not be described here.

[0103] Embodiment 3

[0104] The embodiment provides a computer device, which comprises a memory and a processor, and the memory stores computer readable instructions, and the computer readable instructions are executed by the processor to make the processor execute all or part of the steps of the terpenoid compound and disease related prediction method provided in embodiment 1.

[0105] Embodiment 4

[0106] The embodiment provides a storage medium, which stores computer readable instructions, wherein the computer readable instructions are executed by a processor to realize all or part of the steps of the terpenoid compound and disease related prediction method provided in embodiment 1.

[0107] Exemplarily, the storage medium includes but is not limited to a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0108] Exemplarily, the instructions, programs, code sets or instruction sets can be implemented in a conventional programming language.

[0109] Exemplarily, the processor includes but is not limited to a smart phone, a personal computer, a server, a network device, etc., and is used to execute all or part of the steps of the terpenoid compound and disease correlation prediction method described in Embodiment 1.

[0110] Each of the embodiments in the present application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, they are described more simply, and the relevant parts can be referred to the part of the method embodiments. The device embodiments described above are only exemplary, and the modules described as separate components can or can not be physically separated, and the functions of each module can be implemented in the same or multiple software and / or hardware when implementing the solutions of the present application. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment.

[0111] Obviously, the above embodiments of the present application are only examples for clearly illustrating the present application, and are not intended to limit the implementation manner of the present application. Any modification, equivalent replacement and improvement made on the basis of the above description for those skilled in the art should be included in the protection scope of the claims of the present application.

Claims

1. A method for predicting the correlation between terpenoid compounds and diseases based on knowledge graph, characterized in that: The following steps are involved: Collect data on terpenoid compounds and their bioactivity, including molecular structures, protein targets, gene targets, cell lines, and disease association information, to construct a knowledge graph of terpenoid bioactivity; The molecular descriptors of the terpenoid compounds to be predicted are input into a convolutional neural network for feature extraction to generate a first feature vector of the molecular physicochemical properties; The molecular structure of the terpenoid compound to be predicted is subjected to feature extraction through a graph convolutional network to generate a second feature vector of the terpenoid molecular graph features; The molecular embedding vector obtained by concatenating the first eigenvector and the second eigenvector is embedded in the knowledge graph model to generate a terpenoid compound-disease prediction score matrix; The terpenoid-disease prediction scores were sorted in descending order to generate a recommended list of diseases related to terpenoids.

2. The method for predicting the correlation between terpenoid compounds and diseases according to claim 1, wherein: The construction of the terpenoid bioactivity knowledge graph comprises the following steps: Preprocessing the collected terpenoid compounds and their biological activity data; Integrate the activity associations among five biological entities: terpenoids, protein targets, genes, cell lines, and diseases; After filtering the interaction data through a preset activity threshold, the data are stored in the form of triples to construct a knowledge graph of the bioactivity of terpenoid compounds.

3. The method for predicting the correlation between terpenoid compounds and diseases according to claim 2, wherein: The preprocessing of the collected terpenoid compounds and their biological activity data comprises the following steps: Standardize the data, including cleaning, identifier unification, and data deduplication.

4. The method for predicting the correlation between terpenoid compounds and diseases according to claim 1, wherein: The method of inputting the molecular descriptors of the terpenoid compounds to be predicted into a convolutional neural network for feature extraction comprises the following steps: Calculate molecular descriptors of terpenoids; the molecular descriptors include: Wildman-Crippen partition coefficient, topological polar surface area, molecular weight, drug-like properties, molar refractive index, number of hydrogen bond acceptors, number of hydrogen bond donors, number of aromatic rings, number of rotatable bonds, and carbon atom sp 3 hybrid fraction; All molecular descriptors are spliced ​​to obtain a molecular feature matrix; the molecular feature matrix is ​​input into a convolutional neural network for feature extraction to obtain a first eigenvector characterizing the physicochemical properties of terpenoid compounds.

5. The method for predicting the correlation between terpenoid compounds and diseases according to claim 1, wherein: The molecular structure of the terpenoid compound to be predicted is subjected to feature extraction through a graph convolutional network, comprising the following steps: Generate a corresponding molecular graph based on the molecular structure of terpenoid compounds; wherein, atoms are used as nodes and chemical bonds are used as edges to form a two-dimensional graph structure of terpenoid compound molecules; After extracting an atomic adjacency matrix from the molecular graph and generating an atomic feature matrix based on the atomic type, the atomic adjacency matrix and the atomic feature matrix are input into a graph convolutional network to generate a second eigenvector representing the global molecular features.

6. The method for predicting the correlation between terpenoid compounds and diseases according to any one of claims 1 to 5, characterized in that: The molecular embedding vector obtained by concatenating the first eigenvector and the second eigenvector is embedded into a knowledge graph model to generate a terpenoid compound-disease prediction score matrix, comprising the following steps: The first eigenvector and the second eigenvector are concatenated and passed through a fully connected layer to generate a molecule embedding vector; Generate disease entity embedding vectors using a random adjacency matrix and generate relationship embedding vectors using a random initialization vector of the same dimension; The molecule embedding vector, disease embedding vector, and relationship embedding vector are input into the DistMult model, and the score of each triple is calculated to obtain a terpenoid compound-disease prediction score matrix.

7. The method for predicting the correlation between terpenoid compounds and diseases according to claim 6, characterized in that: The DistMult model is configured with a scoring function f r ; Its expression is: f r (t,d)=t T W r d Where t represents the terpenoid molecule embedding vector, d represents the disease entity embedding vector; W r is the weight matrix of the relationship r between the molecule embedding vector and the disease embedding vector.

8. A terpenoid compound and disease correlation prediction system based on a knowledge graph, applying the terpenoid compound and disease correlation prediction method according to any one of claims 1 to 7, characterized in that: include: A knowledge graph construction module is used to collect data on terpenoid compounds and their bioactivities, including terpenoid molecular structures, protein targets, gene targets, cell lines, and disease association information, to construct a knowledge graph on terpenoid bioactivities; A first feature extraction module is used to input the molecular descriptors of the terpenoid compound to be predicted into a convolutional neural network for feature extraction, and generate a first feature vector of the molecular physicochemical properties; A second feature extraction module is used to extract features of the molecular structure of the terpenoid compound to be predicted through a graph convolutional network to generate a second feature vector of the terpenoid molecular graph features; A knowledge graph embedding module, configured to embed a molecular embedding vector obtained by concatenating the first eigenvector and the second eigenvector into a knowledge graph embedding model to generate a terpenoid compound-disease prediction score matrix; The output module is used to generate a recommended list of diseases related to terpenoids by sorting them in descending order based on the terpenoid-disease prediction scores.

9. A device comprising a memory and a processor, wherein the memory stores computer-readable instructions, characterized in that: When the computer-readable instructions are executed by the processor, the processor executes all steps of the method for predicting the association between terpenoid compounds and diseases according to any one of claims 1 to 7.

10. A storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed by a processor, all steps of the method for predicting the association between terpenoid compounds and diseases according to any one of claims 1 to 7 are implemented.

Images