Prediction and generation method for curative effect of traditional Chinese medicine prescription

Abstract

The invention provides a traditional Chinese medicine prescription curative effect prediction and generation method, and the method comprises the steps: constructing a network score model of a single herbal medicine, and the network score model is used for evaluating the effect of the input single herbal medicine on a target disease; constructing an empirical score model of the herbal medicine pair, wherein the empirical score model is used for evaluating the effect of the input herbal medicine on the target disease; constructing a prescription score model based on the network score model and the experience score model; compared with the prior art, the method for generating the traditional Chinese medicine prescription has the advantages that the curative effect of the target prescription for the target disease is predicted on the basis of the prescription score model, the prediction accuracy can be greatly improved, and in addition, the traditional Chinese medicine prescription generation method integrates the traditional experience of multi-target interaction of herbal medicines and the modern pharmacological mechanism, so that the traditional Chinese medicine prescription generation method is suitable for large-scale popularization and application. The effectiveness of generating the herbal medicine formula can be improved.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to a method for predicting and generating the efficacy of a traditional Chinese medicine prescription. Background Art

[0002] Traditional Chinese medicine has accumulated rich experience in prescription treatment in thousands of years of clinical practice; however, it also has the characteristics of complex ingredients and unclear mechanism of action. According to research, the experience involved in the Chinese medicine formula can be extracted through computational methods to reduce development costs and analyze the comprehensive effects of Chinese medicine, but the optimization of each algorithm has not yet been determined.

[0003] Disease modules can be screened using network structures, even from incomplete interactions, and drug functions can be selected by computer to obtain better therapeutic effects. A drug group prediction method based on complex networks is proposed to analyze the correlation between drug combinations. Among them, DrugCombDB is a comprehensive database for discovering the therapeutic effects of drug combinations, and biomarker-driven networks and network modeling can be used to analyze the driving signal pathways predicted by drug combinations. For example, network perturbation analysis is performed to clarify the mechanism of action of compounds, and a new method is proposed to reduce side effects and enhance efficacy. In addition, drug toxicity is reduced through drug combinations by modeling the target interference effect in the protein-protein interaction (PPI) network and the side effects of multi-drug combination using convolutional neural networks.

[0004] Existing disease gene and target predictions include network-based methods such as the CIPHER algorithm, which is based on the "phenotype network-molecular network" method, and the DrugCIPHER algorithm, which predicts disease genes in gene regulatory networks. Other methods include using multi-omics data to combine genomic and network features to predict the scalability of synergistic anticancer drugs, and the CE-Basic Local AlignmentSearch Tool, which calculates the distance between different antigens to design antiviral vaccines.

[0005] In the study of traditional Chinese medicine, systems pharmacology and network perturbation signal systems are used to describe the effects of traditional Chinese medicine on complex diseases. Integrative pharmacology was proposed based on the TCMIP database, aiming to transform the paradigm of traditional Chinese medicine compound research. A general template for traditional Chinese medicine research has been developed using the gene expression profiles of 102 traditional Chinese medicine ingredients. At the same time, the synergistic mechanism of each component of the traditional Chinese medicine compound was revealed from the perspective of network modularization. Bioinformatics-based methods have also made great progress in the field of new drug development of traditional Chinese medicine. Combining bioinformatics technology, multi-omics, biological networks and other pharmacological mechanism analysis methods has provided a new approach for the development of new drugs including traditional Chinese medicine compounds. In addition, corresponding progress has been made in the development of drugs derived from traditional Chinese medicine. Several studies on the prediction of traditional Chinese medicine prescriptions have been carried out based on methods such as data mining, machine learning, convolutional neural networks, deep learning, and graph neural networks. Moreover, topic models, FangNet, and recommendation systems based on multi-graph neural networks have been used for the prediction of traditional Chinese medicine prescriptions.

[0006] Although the combined drug research in the above methods has made some progress in improving efficacy and reducing toxic side effects, there are still many shortcomings. First, how to design the scores of Chinese herbal medicine prescriptions in training samples has not been determined; second, the classical prescription theory and modern pharmacological mechanism have not yet been intelligently combined; third, how to screen the best Chinese herbal compound for complex diseases has not been determined. Summary of the invention

[0007] The present invention provides a method for predicting and generating the efficacy of a Chinese medicine prescription, which is used to solve the technical problem that the existing Chinese medicine prescription efficacy prediction method is poor and cannot screen the optimal Chinese medicine prescription. In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0008] A method for predicting the efficacy of a traditional Chinese medicine prescription, comprising:

[0009] Constructing a network score model of a single herb, wherein the network score model is used to evaluate the effect of the input single herb on the target disease;

[0010] constructing an empirical score model for herbal medicine pairs, wherein the empirical score model is used to evaluate the effect of the input herbal medicine on the target disease;

[0011] Constructing a prescription score model based on the network score model and the experience score model;

[0012] The efficacy of the target prescription for a target disease is predicted based on the prescription score model.

[0013] Preferably, the network score model is implemented by the following steps:

[0014] Construction of the adjacency matrix of the Alzheimer's disease target-ingredient-herb network

[0015]

[0016] in, and The calculations are as follows:

[0017]

[0018]

[0019] Among them, M IT is the component-target association matrix, M IT (i,j) is component I i and target T j The relationship between M IH is the ingredient-herb association matrix, M IH (i,j) is component I i and herbal H j The relationship between i represents the i-th component, T j represents the jth target, H j represents the jth herb;

[0020] Perform network propagation to calculate the network score Nscore for each herb:

[0021]

[0022] in, Representing the transition probability matrix of propagation k times, W is calculated as follows:

[0023]

[0024] Preferably, constructing an empirical score model for herbal medicine pairs comprises:

[0025] Collect classic prescriptions and extract herbal pairs from them;

[0026] The experience score of the herb pair was calculated, and the experience score of an herb pair was equal to its frequency of occurrence in the classical prescription.

[0027] Preferably, the prescription score model satisfies:

[0028]

[0029] Among them, HPscore is the target prescription score, Nscore i is the network score of the i-th herb in the target prescription, Nscore j is the network score of the jth herb in the target prescription; Escore ijis the empirical score of the i-th herb and the j-th herb pair in the target prescription, and n is the number of herbs in the target prescription.

[0030] A method for generating a traditional Chinese medicine prescription, comprising:

[0031] A prescription generation model is constructed based on reinforcement learning, wherein the prescription generation model takes the prescription as a state and the addition of herbs as an action, and uses the Chinese medicine prescription efficacy prediction method according to any one of claims 1 to 5 to predict the prescription score of the prescription after the newly added herbs are added, and the prescription score is used as a reward value of the prescription generation model;

[0032] A formula generation model is trained, and the formula generation model is used to generate an optimal Chinese medicine formula.

[0033] Preferably, using the formula generation model to generate an optimal Chinese medicine formula includes:

[0034] S1, Initialization: Initialize states, actions, and define the policy network;

[0035] S2, action selection and state update: At time step t, the current state S t Input to the policy network, get the probability of herb selection, and select action a based on this probability t , perform action a t , update the state to s t+1 ,s t+1 Contains newly selected herbs;

[0036] S3. Calculate immediate rewards and cumulative returns: Calculate action a t The reward of the policy network is calculated based on the reward.

[0037] S4, policy optimization: Use policy gradient ascent to update policy network parameters;

[0038] S5. Determine whether the current iteration has reached the termination condition. If not, repeat steps S2-S5. If the termination condition is reached, output the current optimal solution.

[0039] Preferably, the policy network is defined as π θ (a|s)=P(a|s,θ)

[0040] Among them, θ represents the parameters of the policy network, s represents the state, that is, the current herbal combination of the recipe, a represents the action, selecting a certain herb, π represents the policy network; P is the initial policy;

[0041] and / or

[0042] The cumulative return is satisfied if:

[0043]

[0044] Among them, G t represents the cumulative reward at time step t, T represents the final time step, γ is the discount factor, which is used to adjust the importance of future rewards in the current decision, and the value range is 0≤γ≤1;

[0045] and / or

[0046] Updating the policy network parameters is achieved through the following formula:

[0047]

[0048]

[0049] Among them, θ t+1 is the strategy parameter of the t+1th step, θ t is the strategy parameter of the tth step, ΔJ(θ t ) represents the expected cumulative return for the policy parameter θ at step t t The gradient of , η represents the learning rate, where b is the baseline and N represents the strategy according to the tth step The number of sampled trajectories, each trajectory represents a state-action-reward sequence, T n is the number of time steps in the nth trajectory, n is the nth trajectory, is the action selected at time step t in the nth trajectory, is the state observed at time step t in the nth trajectory.

[0050] Preferably, the disease is Alzheimer's disease, and the optimal Chinese medicine prescription is ginseng, cinnamon twig, Poria cocos, angelica, Atractylodes macrocephala, jujube, dried ginger, ginger, Vaccaria segetalis, and licorice.

[0051] Preferably, the method further comprises the following steps:

[0052] The biological functions of the generated prescriptions on the diseases were evaluated by GO enrichment analysis and KEGG enrichment analysis.

[0053] A computer system comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the computer program.

[0054] The present invention has the following beneficial effects:

[0055] 1. The method for predicting the efficacy of a Chinese medicine prescription in the present invention comprises the following steps: constructing a network score model of a single herb, wherein the network score model is used to evaluate the effect of a single input herb on a target disease; constructing an empirical score model of a pair of herbs, wherein the empirical score model is used to evaluate the effect of an input herb on a target disease; constructing a prescription score model based on the network score model and the empirical score model; and predicting the efficacy of the target prescription for a target disease based on the prescription score model. Compared with the prior art, the accuracy of the prediction can be greatly improved. In addition, based on this, the present invention also provides a method for generating a Chinese medicine prescription, which can improve the effectiveness of generating a herbal formula by integrating traditional experience and modern pharmacological mechanisms of multi-target interactions of herbs.

[0056] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0058] Figure 1 A schematic diagram of a process for generating a Chinese medicine prescription for Alzheimer's disease based on network communication and reinforcement learning provided by the present invention;

[0059] Figure 2 A schematic diagram of the structure of the device for recommending and analyzing Chinese medicine prescriptions for Alzheimer's disease based on network communication and reinforcement learning provided by the present invention;

[0060] Figure 3 A schematic diagram of the structure of the device for recommending and analyzing Chinese medicine prescriptions for Alzheimer's disease based on network communication and reinforcement learning provided by the present invention.

[0061] Figure 4 The top 10 bar charts of biological processes (BP), cellular components (CC) and molecular functions (MF) of the traditional Chinese medicine prescription for Alzheimer's disease based on network propagation and genetic algorithm provided by the present invention, wherein BP is the biological process entry, CC is the cellular component entry, MF is the molecular function entry, Counts is the number of genes, -log 10 (pvalue) indicates the P value for whether the enrichment observed in the crossbar is significant.

[0062] Figure 5The top 10 bubble charts of the Kyoto Encyclopedia of Genomes (KEGG) ranking of the traditional Chinese medicine prescriptions for Alzheimer's disease based on network propagation and genetic algorithm provided by the present invention, wherein KEGG represents the Kyoto Encyclopedia of Genomes pathway enrichment entry. DETAILED DESCRIPTION

[0063] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0064] Please refer to Figure 1 , a method for generating a Chinese medicine prescription for Alzheimer's disease based on network communication and reinforcement learning in this embodiment, the method comprises the following steps:

[0065] S1, constructing the Alzheimer's disease target-ingredient-herb network and calculating the network score of individual herbs based on network propagation;

[0066] S2, collect herbal pairs in classic prescriptions and calculate the experience scores of herbal pairs based on TCM compatibility experience;

[0067] S3, formula scores were calculated based on the network scores of individual herbs and the empirical scores of herb pairs;

[0068] S4, generating effective prescriptions based on prescription scores and reinforcement learning;

[0069] S5, GO enrichment analysis and KEGG enrichment analysis were used to evaluate the biological functions of the generated prescriptions on the disease.

[0070] Specifically, in step S1, 29 targets, 2114 ingredients, and 446 herbs related to Alzheimer's disease were collected from the TCMSP database to construct an Alzheimer's disease target-ingredient-herb network; then network propagation was performed on this network to calculate the network score of a single herb.

[0071] Network propagation was performed on the Alzheimer's disease target-ingredient-herb network to calculate the network score of each herb, including the following steps:

[0072] Specifically, network propagation is performed on the Alzheimer's disease target-ingredient-herb network to calculate the network score of a single herb, including the following steps:

[0073] Step S11a, constructing the adjacency matrix of the Alzheimer's disease target-ingredient-herb network

[0074]

[0075] in, and The calculations are as follows:

[0076]

[0077]

[0078] Among them, I i represents the i-th component, T j represents the jth target, H j represents the jth herb.

[0079] Step S12a, executing network propagation to calculate the network score Nscore of each herb:

[0080]

[0081] in, Representing the transition probability matrix of propagation k times, W is calculated as follows:

[0082]

[0083] Specifically, in step S2, classic prescriptions are collected from the Chinese patent medicine value evaluation information database, and herbal pairs are extracted therefrom; then, the experience score Escore of the herbal pair is calculated, and the experience score of an herbal pair is equal to the frequency of its appearance in the classic prescription.

[0084] Specifically, in step S3, the Chinese medicine prescription score HPscore is calculated based on the network score of the single herb and the empirical score of the herb pair. If a Chinese medicine prescription contains herbs that are more effective for Alzheimer's disease, and the effect of the herb combination is stronger, then the Chinese medicine prescription is more effective; this is because the Chinese medicine prescription is not a simple superposition of herbs, but includes the relationship between the herb combination. In order to prevent the number of ingredients in the herbal formula from being too large, the calculation design of HPscore is as follows:

[0085]

[0086] Where n is the number of herbs contained in a TCM prescription, and HPscore is negatively correlated with the number of herbs; the more herbs there are, the lower the HPscore is, and vice versa.

[0087] Specifically, in step S4, generating an effective prescription based on the prescription score and reinforcement learning includes the following steps:

[0088] S11b, strategy formulation. The initial strategy P is defined as a neural network, called the strategy network π, expressed as follows:

[0089] π θ (a|s)=P(a|s,θ)

[0090] Where θ represents the parameters of the policy network, s represents the state (i.e. the current combination of herbs), and a represents the action (i.e. selecting a certain herb); the policy network takes the current state as input and outputs the probability distribution of all possible next herbs to be added to the prescription. The network architecture includes an input layer corresponding to the number of optional herbs, a hidden layer containing 128 neurons, and an output layer with a softmax activation function.

[0091] S12b, action selection and state update. At time step t, the current state is first input into the policy network, and the probability distribution of herb selection is obtained as follows:

[0092] a t ~π θ (a t |s t )

[0093] Then, based on this probability, an action a is chosen t , perform action a t Status updated to s t+1 ,s t+1 Contains newly selected herbs.

[0094] S13b, immediate reward and cumulative return. Immediate reward refers to the reward after an action is completed, which mainly reflects the short-term benefit of an action. Here, the immediate reward r t is the HPscore of the prescription with the newly selected herb added. Cumulative return is the discounted sum of the immediate rewards from the current state to the final prescription, which is used to evaluate the current strategy and guide strategy optimization. A higher return indicates that the strategy is moving in the right direction, while a lower return indicates that the current strategy is underperforming. The cumulative return is calculated as follows:

[0095]

[0096] Among them G t represents the cumulative reward at time step t, T represents the final time step, and γ is the discount factor. γ is used to adjust the importance of future rewards in the current decision, and its value range is 0≤γ≤1.

[0097] S14b, Policy Optimization. To find the optimal policy that maximizes the expected cumulative return, we use policy gradient ascent to update the policy network parameters as follows:

[0098]

[0099] Where η represents the learning rate, represents the gradient of the expected cumulative return with respect to the policy parameter θ. The gradient of the updated parameter is calculated as follows:

[0100]

[0101] Where b is the baseline, which is used to reduce G t Variance, thus making the gradient estimation more stable and efficient. N represents the variance according to the strategy π θ The number of trajectories sampled, each trajectory represents a state-action-reward sequence.

[0102] S15b, prescription generation. When the policy network is fully trained, it can generate prescriptions that are expected to be very effective for Alzheimer's disease. Here, the prescriptions with the highest HPscore are: ginseng, cinnamon twig, poria, angelica, atractylodes, jujube, dried ginger, ginger, fenugreek, and licorice.

[0103] Specifically, in step S5, the Metascape platform (https: / / metascape.org) was used to perform gene ontology (GO) enrichment analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis on drug-disease intersection targets. The following parameters were used for G0 enrichment analysis: species = H. sapiens, min·overlap = 3, P-value < 0.05, min enrichment = 1.5, and the following parameters were used for KEGG enrichment analysis: species = H. sapiens, min·overlap = 3, P-value < 0.05, min enrichment = 1.5. The data were then visualized on the Weishengxin website.

[0104] Embodiment 2

[0105] Please refer to Figure 2 , is an Alzheimer's disease Chinese medicine prescription recommendation and analysis device based on network communication and reinforcement learning in an embodiment, the Chinese medicine prescription recommendation and analysis device 100 includes a network score calculation module 11, an experience score calculation module 12, a prescription score calculation module 13, a Chinese medicine prescription generation module 14, and a Chinese medicine prescription analysis module 15.

[0106] The network score calculation module 11 constructs an Alzheimer's disease target-ingredient-herb network and calculates the network score of a single herb based on network propagation. The construction method adopted by the network score calculation module 11 can refer to step S1 in Example 1.

[0107] The experience score calculation module 12 collects the herbal medicine pairs in the classic prescriptions and calculates the experience scores of the herbal medicine pairs based on the TCM compatibility experience. The calculation method of the experience score calculation module 12 can refer to step S2 in Example 1.

[0108] The prescription score calculation module 13 calculates the prescription score based on the network score of the single herb and the experience score of the herb pair. The prediction method of the prescription score calculation module 13 can refer to step S3 in Example 1.

[0109] The Chinese medicine prescription generation module 14 generates effective prescriptions based on the prescription score and reinforcement learning. The prediction method of the prescription score calculation module 14 can refer to step S4 in Example 1.

[0110] The Chinese medicine formula analysis module 15 evaluates the biological function of the generated formula on the disease through GO enrichment analysis and KEGG enrichment analysis. The prediction method of the Chinese medicine formula analysis module 15 can refer to step S5 in Example 1.

[0111] Embodiment 3

[0112] Please refer to Figure 3 , this embodiment is a device for recommending and analyzing Chinese medicine prescriptions for Alzheimer's disease based on network communication and reinforcement learning. The Chinese medicine prescription recommendation and analysis device 200 includes a memory 21, a processor 22, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for recommending and analyzing Chinese medicine prescriptions for Alzheimer's disease based on network communication and reinforcement learning are executed.

[0113] It is understandable that the technical solution of the present invention, in essence or in other words, the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, electronic device, or network device, etc.) to perform all or part of the steps of the method of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0114] Embodiment 4

[0115] Please refer to Figure 4, GO enrichment analysis was performed on the 142 predicted targets. In the analysis results, 498 entries were obtained under the condition of P < 0.05, including 270 biological process (BP) entries, 87 cellular components (CC) entries, and 141 molecular functions (MF) entries. According to the P value (the P value is used to evaluate the statistical significance of the research results, which measures the probability of the results observed in the hypothesis test under the condition that the null hypothesis (i.e., the invalid hypothesis) is established), the top 10 entries were selected from small to large to make a bar chart.

[0116] Most of the biological processes in the results are related to the pathological process of AD, which can be roughly divided into responses to reduced oxygen content, cellular responses to organic nitrogen compounds, blood circulation, and responses to oxygen content. The cellular components in the enriched results mainly involve membrane rafts, membrane microdomains, synaptic membranes, postsynapses, receptor complexes, and dendritic trees, indicating that most of the predicted targets and their products are distributed in the membrane, and a small part is distributed in synapses and cell bodies. The molecular functions in the enriched results include G protein-coupled amine receptor activity, neurotransmitter receptor activity, postsynaptic neurotransmitter receptor activity, adrenergic receptor activity, α-adrenergic receptor activity, and protein homodimerization activity. The richness of binding indicates that most genes or gene products have selective interactions with ligands and have binding affinity. In general, the targets of these active compounds are mostly located in membranes, synapses and cell bodies. They are mainly involved in biological processes such as response to reduced oxygen content, response to organic nitrogen compounds, blood circulation and response to oxygen content. They have molecular functions such as G protein-coupled amine receptor activity, neurotransmitter receptor activity, postsynaptic neurotransmitter receptor activity, adrenergic receptor activity, α-adrenergic receptor activity and protein homodimerization activity.

[0117] Please refer to Figure 5 , KEGG enrichment analysis was performed on the 142 predicted targets. In the analysis results, 157 signaling pathways were screened under the condition of P < 0.05, and the top 10 items were selected from small to large according to their biological significance according to the P value (P value is used to evaluate the statistical significance of the research results, which measures the probability of the results observed in the hypothesis test under the condition that the null hypothesis (i.e., the invalid hypothesis) is established) to draw a bubble chart. The main cancer pathogenesis pathways include lipid and atherosclerosis signaling pathway, calcium signaling pathway, chemical carcinogenesis-receptor activation signaling pathway, hepatitis C signaling pathway, neuroactive ligand and receptor interaction signaling pathway, fluid shear stress and atherosclerosis signaling pathway, and human cytomegalovirus infection signaling pathway.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the efficacy of a traditional Chinese medicine prescription, characterized in that: include: Constructing a network score model of a single herb, wherein the network score model is used to evaluate the effect of the input single herb on the target disease; constructing an empirical score model for herbal medicine pairs, wherein the empirical score model is used to evaluate the effect of the input herbal medicine on the target disease; Constructing a prescription score model based on the network score model and the experience score model; The efficacy of the target prescription for a target disease is predicted based on the prescription score model.

2. The method for predicting the efficacy of a Chinese medicine prescription according to claim 1, characterized in that: The network score model is implemented by the following steps: Construction of the adjacency matrix of the Alzheimer's disease target-ingredient-herb network in, and The calculations are as follows: Among them, M IT is the component-target association matrix, M IT (i,j) is component I i and target T j The relationship between M IH is the ingredient-herb association matrix, M IH (i,j) is component I i and herbal H j The relationship between i represents the i-th component, T j represents the jth target, H j represents the jth herb; Perform network propagation to calculate the network score Nscore for each herb: in, Representing the transition probability matrix of propagation k times, W is calculated as follows:

3. The method for predicting the efficacy of a Chinese medicine prescription according to claim 1, characterized in that: Construct an empirical score model for herb pairs, including: Collect classic prescriptions and extract herbal pairs from them; The experience score of the herb pair was calculated, and the experience score of an herb pair was equal to its frequency of occurrence in the classical prescription.

4. The method for predicting the efficacy of a Chinese medicine prescription according to any one of claims 1 to 3, characterized in that: The prescription score model satisfies: Among them, HPscore is the target prescription score, Nscore i is the network score of the i-th herb in the target prescription, Nscore j is the network score of the jth herb in the target prescription; Escore ij is the empirical score of the i-th herb and the j-th herb pair in the target prescription, and n is the number of herbs in the target prescription.

5. A method for generating a Chinese medicine prescription, characterized in that: include: A prescription generation model is constructed based on reinforcement learning, wherein the prescription generation model takes the prescription as a state and the addition of herbs as an action, and uses the Chinese medicine prescription efficacy prediction method according to any one of claims 1 to 5 to predict the prescription score of the prescription after the newly added herbs are added, and the prescription score is used as a reward value of the prescription generation model; The formula generation model is trained and used to generate an optimal Chinese medicine formula.

6. The method for generating a Chinese medicine prescription according to claim 5, characterized in that: The optimal Chinese medicine formula is generated using the formula generation model, including: S1, Initialization: Initialize states, actions, and define the policy network; S2, action selection and state update: At time step t, the current state S t Input to the policy network, get the probability of herb selection, and select action a based on this probability t , perform action a t , update the state to s t+1 ,s t+1 Contains newly selected herbs; S3. Calculate immediate rewards and cumulative returns: Calculate action a t The reward of the policy network is calculated based on the reward. S4, policy optimization: Use policy gradient ascent to update policy network parameters; S5. Determine whether the current iteration has reached the termination condition. If not, repeat steps S2-S5. If the termination condition is reached, output the current optimal solution.

7. The method for generating a Chinese medicine prescription according to claim 6, characterized in that: The policy network is defined as p θ (a|s)=P(a|s,θ) Among them, θ represents the parameters of the policy network, s represents the state, that is, the current herbal combination of the recipe, a represents the action, selecting a certain herb, π represents the policy network; P is the initial policy; and / or The cumulative return is satisfied if: Among them, G t represents the cumulative reward at time step t, T represents the final time step, γ is the discount factor, which is used to adjust the importance of future rewards in the current decision, and the value range is 0≤γ≤1; and / or Updating the policy network parameters is achieved through the following formula: Among them, θ t+1 is the strategy parameter of the t+1th step, θ t is the strategy parameter of the tth step, ΔJ(θ t ) represents the expected cumulative return for the policy parameter θ at step t t The gradient of , η represents the learning rate, where b is the baseline and N represents the strategy according to the tth step The number of sampled trajectories, each trajectory represents a state-action-reward sequence, Tn is the number of time steps in the nth trajectory, n is the nth trajectory, is the action selected at time step t in the nth trajectory, is the state observed at time step t in the nth trajectory.

8. The method for generating a traditional Chinese medicine prescription according to any one of claims 5 to 7, characterized in that: The disease is Alzheimer's disease, and the optimal Chinese medicine prescription is ginseng, cinnamon twig, Poria cocos, angelica, Atractylodes macrocephala, jujube, dried ginger, ginger, fenugreek, and licorice.

9. The method for generating a traditional Chinese medicine prescription according to any one of claims 5 to 7, characterized in that: The following steps are also included: The biological functions of the generated prescriptions on the diseases were evaluated by GO enrichment analysis and KEGG enrichment analysis.

10. A computer system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images