Artificial intelligence-driven polycystic ovarian syndrome drug target screening method

Abstract

The invention discloses an artificial intelligence-driven polycystic ovarian syndrome drug target screening method, which comprises the following steps: S1, collecting genome data and hormone level data from the body of a patient, and constructing a patient individualized data set; s2, carrying out denoising processing, missing value filling and standardization on the acquired individual data set of the patient; s3, inputting the standardized individual data set of the patient into a multi-layer neural network model; s4, training the dynamic gene-hormone interaction network model by adopting the training set data; s5, screening out dual regulation and control drug targets highly related to the PCOS pathological mechanism; s6, performing optimization through historical treatment data of the patient and target treatment effect feedback; and S7, outputting the dual regulation and control drug target list in a report form. According to the invention, the efficiency and accuracy of drug target screening are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to an artificial intelligence-driven method for screening drug targets for polycystic ovary syndrome. Background Art

[0002] With the rapid development of artificial intelligence technology and bioinformatics, accurate diagnosis and personalized treatment of diseases have gradually become the key research directions of modern medicine. Polycystic ovary syndrome, as a common endocrine and metabolic disease, has complex pathological mechanisms and significant individual differences, which puts high demands on the discovery and screening of drug targets. The screening of drug targets is of great significance for the treatment of PCOS and can help find the most effective intervention approach. However, there are still many challenges and shortcomings in the screening of PCOS drug targets in the existing technology.

[0003] At present, the screening of most PCOS drug targets relies on traditional laboratory analysis and analysis methods based on a single data source. Researchers usually discover potential targets related to PCOS through manual analysis based on a single type of data such as genome, hormone levels and metabolic indicators. However, traditional methods have obvious defects: on the one hand, since the pathology of PCOS involves multi-level factors such as gene regulation, endocrine disorders and metabolic abnormalities, analysis that relies solely on a single data source often fails to fully reveal the pathological mechanism of the disease and may miss key drug targets; on the other hand, the traditional experimental process is complex and time-consuming. When faced with massive amounts of data, the efficiency and accuracy of data analysis are difficult to guarantee.

[0004] In recent years, some scientific research institutions have begun to try to use computer algorithms and high-throughput data analysis technologies to assist in the screening of drug targets. However, these methods usually rely on static data analysis models, which make it difficult to capture the dynamic changes of individualized patient data, and are unable to perform real-time optimization based on clinical feedback from patients. More importantly, existing technologies make it difficult to simultaneously consider the complex interactions between genes and hormones, and the lack of ability to identify key nodes of dual regulation results in limited applicability of target screening results in clinical applications.

[0005] In addition, existing target screening technologies lack the ability to be deeply integrated with personalized medicine. It is usually difficult to recommend the most suitable targets and treatment plans based on the pathological characteristics of different patients. This not only affects the accuracy of treatment, but may also increase patients' medication risks and treatment costs. In the absence of effective data integration and dynamic analysis models, clinicians need to rely on experience and judgment when formulating treatment plans, which can easily lead to subjectivity and individual differences in target selection.

[0006] In summary, the existing technology has the following problems in the process of PCOS drug target screening: lack of deep fusion analysis capabilities of multiple data sources, low screening efficiency and inaccurate results, lack of dynamic optimization mechanism, and difficulty in achieving personalized target recommendation. The above shortcomings seriously affect the precise treatment effect of PCOS and the efficiency of drug development. Therefore, there is an urgent need for an artificial intelligence-based drug target screening method that can improve the efficiency and accuracy of screening through multi-level data integration analysis and dynamic optimization models to provide strong support for the treatment of PCOS. Summary of the invention

[0007] One purpose of the present invention is to propose an artificial intelligence-driven method for screening drug targets for polycystic ovary syndrome, which significantly improves the efficiency and accuracy of drug target screening.

[0008] An artificial intelligence-driven polycystic ovary syndrome drug target screening method according to an embodiment of the present invention comprises the following steps:

[0009] S1. Collect genomic data and hormone level data from patients to build a patient-specific data set;

[0010] S2. De-noising, missing value filling and standardization of the collected individualized data sets of patients;

[0011] S3, inputting the standardized individualized data set of the patient into the multi-layer neural network model, and constructing a dynamic gene-hormone interaction network model based on the individual data of the patient;

[0012] S4. Use training set data to train the dynamic gene-hormone interaction network model, use back propagation algorithm to adjust model parameters, combine reinforcement learning mechanism, and dynamically optimize the dynamic gene-hormone interaction network model through clinical feedback data;

[0013] S5. In the dynamic gene-hormone interaction network model after training and optimization, the path analysis algorithm is used to identify key nodes that have both gene regulation and hormone regulation functions, and to screen out dual-regulatory drug targets that are highly related to the pathological mechanism of PCOS. The screening results are scored and prioritized according to the regulatory intensity, network centrality and clinical relevance of the targets;

[0014] S6. Generate a target list suitable for each patient's pathological characteristics in the screening results of dual-regulatory drug targets, and optimize it through the patient's historical treatment data and target efficacy feedback;

[0015] S7. Output the list of dual-regulatory drug targets in the form of a report. Clinicians evaluate the model based on the treatment effect. The evaluation information is used to optimize the dynamic gene-hormone interaction network model.

[0016] Optionally, the S1 specifically includes:

[0017] S11, collecting genome data from the patient, wherein the genome data includes the patient's gene sequence G s , gene mutation site set M i ={m1,m2,…,m n}, Gene expression matrix E ij ,in:

[0018] G s represents the complete genome sequence of the patient;

[0019] M i Represents a set of gene mutation sites associated with PCOS, including n mutation sites;

[0020] E ij is the expression value of gene i in sample j, forming an N×J matrix, where N is the number of genes and J is the number of samples;

[0021] S12. Collect hormone level data from patient serum samples, including testosterone concentration T s , luteinizing hormone concentration LH s , follicle stimulating hormone concentration FSH s and estrogen concentration E s ,in:

[0022] T s It represents the testosterone concentration in serum in nanomoles / liter;

[0023] LH s It indicates the concentration of luteinizing hormone in international units per liter;

[0024] FSH s represents the concentration of follicle-stimulating hormone in international units per liter;

[0025] E s It indicates the estrogen concentration in picomoles / liter;

[0026] S13. Convert the collected genomic data and hormone level data into different data formats to construct the patient's personalized data set D p :

[0027] D p ={(G s ,M i ,E ij ),(T s ,LH s ,FSH s ,E s )}.

[0028] Optionally, the S2 specifically includes:

[0029] S21. Use threshold filtering method to filter the patient individualized data set D p Perform denoising to remove the gene expression matrix E ij Noisy data in:

[0030]

[0031] Among them, E′ ij is the gene expression value after denoising, ∈ is the threshold parameter of noise filtering;

[0032] S22. De-noised patient individualized dataset D′ p ={(G s ,M i ,E′ ij ),(T s ,LH s ,FSH s ,E s )} to fill in missing values, the gene expression matrix E′ ij Missing values ​​in the hormone level data are handled using the mean filling method:

[0033]

[0034] Among them, E″ ij is the gene expression value after missing value filling, X s ∈{T s ,LH s ,FSH s ,E s} is the hormone level data, P is the number of hormone samples;

[0035] S23, using Min-Max normalization to fill the patient individualized dataset D″ p Standardize the process;

[0036] S24. The dataset after denoising, missing value filling and standardization is represented as the final standardized patient individualized dataset:

[0037]

[0038] Optionally, the S3 specifically includes:

[0039] S31. Standardized patient individualized data set Input multi-layer neural network model NN m , gene expression data and hormone levels Combine to form the input feature vector x p :

[0040]

[0041] in, is the input feature vector of patient p, containing N gene expression values ​​and 4 hormone level values;

[0042] S32. Construction of a dynamic gene-hormone interaction network model GHIN based on individual patient data p , dynamic gene-hormone interaction network model based on multi-layer neural network NN m The structure and parameters of represent the relationship between gene regulation and hormone levels. The hidden layer of the multi-layer neural network adopts a nonlinear activation function. The forward propagation process of the dynamic gene-hormone interaction network model is:

[0043] h (1) =σ(W (1) x p +b (1) );

[0044] h (l) =σ(W (l) h (l-1) +b (l) ),l=2,3,…,L;

[0045] y p =σ(W (L+1) h (L) +b (L+1) );

[0046] Among them, h (1) is the neuron output vector of the lth hidden layer, W (1) is the weight matrix of the lth layer, indicating the connection strength from the previous layer to this layer, b (l) is the bias vector of the lth layer, L is the total number of hidden layers, y p The output of neurons in the output layer reflects the interaction between genes and hormones;

[0047] S33. In the dynamic gene-hormone interaction network model GHIN p The weights of the neuron connections Represents the regulatory relationship between gene i and hormone j at level l, by calculating the comprehensive influence I between gene and hormone ij and the importance of nodes in the network topology to identify key nodes in gene regulatory pathways and hormone feedback loops, with a comprehensive impact I ij The calculation is:

[0048]

[0049] in, is the connection weight from node i to node j in layer l, reflecting the direct influence of genes and hormones in this layer. is the activation value of node i in the l-1th layer, indicating the activation state of gene i in the network, is the activation value of node j in layer l, indicating the activation state of hormone j in the network;

[0050] Combined with betweenness centrality C b (i) Degree centrality C d (i) and eigenvector centrality C e (i) Topological index, computing node n i Importance score S in the network i :

[0051] S i =γ1C b (i)+γ2C d (i)+γ3C e (i);

[0052] in, σ st is the number of shortest paths from node s to node t, σ st (i) is the number of shortest paths passing through node i, reflecting the importance of node i in network information transmission. d (i) = deg(i), which is the degree of node i, indicating the number of nodes it is directly connected to, C e (i) is the eigenvector centrality of node i, which indicates the influence of the node in the overall network structure. γ1, γ2, and γ3 are weight coefficients, satisfying γ1+γ2+γ3=1, which are used to balance the importance of each indicator.

[0053] S34. Based on comprehensive influence I ij and node importance score S i , by calculating the score K of gene i as a key node i :

[0054] K i =β1I ij +β2S i ;

[0055] Among them, K i Score the key nodes of gene i. The higher the value, the more critical it is in gene regulation and hormone feedback. β1 and β2 are weight parameters, satisfying β1+β2=1, which are used to adjust the proportion of comprehensive influence and node importance in the score.

[0056] Sort the gene nodes in descending order according to the scoring results to generate a list of key nodes:

[0057] K p ={k1,k2,…,k m};

[0058] Among them, k m Indicates the mth key gene node after sorting. The key node is used for dual regulation in the pathological mechanism of polycystic ovary syndrome and serves as a potential drug target.

[0059] Optionally, the S4 specifically includes:

[0060] S41. Standardized individualized datasets obtained from multiple patients Composition of training set , and the sample sequence Indicates that x p is the input feature vector of patient p, to label the drug targets associated therewith;

[0061] S42. Using the back propagation algorithm to adjust the dynamic gene-hormone interaction network model GHIN p Parameters, including the weight matrix W (l) and the bias vector b (l) , to minimize the loss function L(θ), where θ={W (l) ,b (l)} represents the model parameter set, and the loss function is defined as:

[0062]

[0063] Among them, M is the number of training samples, N is the number of genes, λ1,λ2 are the weight coefficients of the loss function, y pi is the predicted value and true value of gene expression of the pth patient, H ps are the predicted and true values ​​of the hormone levels of the pth patient;

[0064] S43, in each iteration, updating the model parameters according to the gradient descent method;

[0065] S44. Combined with reinforcement learning mechanism, the dynamic gene-hormone interaction network model GHIN is trained through clinical feedback data p Dynamic optimization is performed, the dynamic gene-hormone interaction network model is regarded as an intelligent agent, the clinical feedback data is used as the environment, and the reward function R is defined to measure the effectiveness of the prediction results of the dynamic gene-hormone interaction network model:

[0066]

[0067] Among them, R(s t ,a t ) is in state s t Take action a t The reward value is, M is the number of patients, K is the number of key nodes, I pk ,S pk is the comprehensive influence and importance score of node k of the pth patient, is the indicator function, which takes 1 when the prediction is correct and 0 otherwise, and γ is the discount coefficient;

[0068] S45. In the reinforcement learning process, the dynamic gene-hormone interaction network model selects action a according to the strategy function π(a|s), the state s represents the current model parameters, the action a is the parameter update direction, and the strategy function is updated using the strategy gradient method to maximize the cumulative reward;

[0069] S46. Through multiple rounds of iterative training, the dynamic gene-hormone interaction network model adjusts the model parameters θ under the joint action of the training set data and clinical feedback data, so that the dynamic gene-hormone interaction network model can capture the individual differences of different patients and optimize the ability to understand the interaction between genes and hormones.

[0070] Optionally, the S5 specifically includes:

[0071] S51. In the dynamic gene-hormone interaction network model after training optimization, the path analysis algorithm is used to identify the key nodes that have both gene regulation and hormone regulation functions. The path analysis calculates the path strength of the influence path P(i, j) between the gene and hormone nodes based on the network structure of the dynamic gene-hormone interaction network model:

[0072]

[0073] in, represents the path strength between gene i and hormone j, is the connection weight between the gene and hormone nodes in the lth layer, and L is the number of hidden layers;

[0074] S52. Calculate the control intensity R of each node according to the network topology i and centrality indexes to evaluate the importance of nodes in the gene-hormone interaction network, and the regulatory strength R i Represents the comprehensive contribution of node i in multiple paths:

[0075]

[0076] in, is the set of nodes connected to node i, C b (i) is the betweenness centrality of node i, Cd (i) is its degree centrality, max(C d ) is the maximum degree centrality value in the network;

[0077] S53, the control intensity R of the combined node i and eigenvector centrality C e (i) Calculate the dual control score S of the node i :

[0078] S i =α1R i +α2C e (i);

[0079] Among them, α1 and α2 are weight parameters, satisfying α1+α2=1, which are used to balance the importance of regulation intensity and network centrality;

[0080] S54. Screen out key nodes k with high dual regulation scores i Generate a list of candidate targets:

[0081]

[0082] Among them, m is the number of key nodes screened;

[0083] S55. Prioritize candidate targets and generate a final target priority list

[0084]

[0085] in, The target k i β1, β2, and β3 are weight parameters. The first term calculates the weight and betweenness centrality contribution of gene i and other nodes j to evaluate its regulatory role in the network. The second term calculates the prediction accuracy of patient p and key node k according to the prediction accuracy of patient p and key node k. i Rating S i Calculate the average importance of the node, I(·) is the indicator function, which takes the value of 1 when the prediction is correct and 0 otherwise. The third term regulates the intensity R by standardization i Reflects the network importance of the target;

[0086] S56. Score based on comprehensive priority The candidate targets are sorted in descending order based on the calculation results to generate a final target priority list for subsequent targeted drug development and treatment plan design:

[0087]

[0088] Optionally, the S7 specifically includes:

[0089] S71. Generate a priority report of dual-regulatory drug targets, which includes the position of each target in the gene-hormone interaction network, the main regulatory pathway, the priority ranking, and the clinical relevance;

[0090] S72. Describe the regulatory pathway for each target. The regulatory pathway priority report includes all genes and hormone nodes connected to the target, as well as the regulatory relationships between the nodes. The key nodes and influential regulatory links in each pathway are highlighted in the priority report.

[0091] S73. Analyze the central position of each target in the gene-hormone interaction network. Use the number of connections and importance of the target in the gene-hormone interaction network to mark the regulatory ability of the node. The priority report indicates whether the target is in the core position of the gene-hormone interaction network and its involvement in different pathological pathways.

[0092] S74. Based on the patient's historical treatment data and the performance of different targets in individualized treatment of patients, the priority report provides clinical relevance analysis for each target, including the target's response rate, effectiveness feedback, and the target's adaptability evaluation in different patient groups;

[0093] S75. Clinicians evaluate the targets generated by the dynamic gene-hormone interaction network model based on the target information and treatment effects in the priority report, and record the evaluation information for future optimization of the dynamic gene-hormone interaction network model. The doctors' feedback will become a reference for target priority sorting.

[0094] The beneficial effects of the present invention are:

[0095] (1) The present invention uses a multi-layer neural network model to deeply integrate the patient's genomic data, hormone level data and metabolic index multi-source data. Different from the traditional single data source analysis, the present invention realizes the modeling of the complex relationship between gene regulation and hormone level by constructing a dynamic gene-hormone interaction network model. The dynamic gene-hormone interaction network model analyzes the path strength between gene and hormone nodes in the network topology structure to automatically identify the regulatory path and key nodes. The algorithm utilizes the combination of path analysis algorithm and centrality calculation to ensure that the screening results cover the multidimensional pathological mechanism of PCOS and avoid missing key targets. Compared with traditional methods, it reveals the pathological process of PCOS more comprehensively and significantly improves the integrity of target screening.

[0096] (2) The present invention introduces a reinforcement learning mechanism during the training process of the dynamic gene-hormone interaction network model, and continuously optimizes the model parameters through clinical feedback data. The dynamic gene-hormone interaction network model adopts a policy gradient algorithm during the training process so that the target screening results can dynamically adapt to the individualized data of different patients. The reward function is designed based on prediction accuracy and treatment feedback to ensure that the model captures the individual differences of patients during the iteration process. This not only improves the accuracy of the screening results, but also supports personalized target recommendations, providing patients with more individualized treatment plans. Experimental results show that after multiple iterations, the consistency between the screened targets and clinical treatment feedback is improved by more than 13%.

[0097] (3) The present invention provides a detailed network analysis report for each screened target through dual regulation score calculation and multi-layer priority sorting. The dual regulation score combines the regulation strength and centrality indicators with the response rate of patient treatment data and clinical feedback to generate a comprehensive priority sorting. The present method specifically designs an indicator function for calculating the response of the target and ensures the objectivity and stability of the score through standardized network indicators. The generated priority report provides clear decision support for clinicians, helping them to be more accurate and effective in target selection and treatment plan formulation. Compared with traditional target screening technology, the effectiveness and applicability of the method of the present invention in clinical applications are significantly improved, and the success rate of the screened targets in PCOS treatment is increased by 15%. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0099] Figure 1 A flow chart of an artificial intelligence-driven polycystic ovary syndrome drug target screening method proposed in the present invention;

[0100] Figure 2 This is a schematic diagram of the structure of a dynamic gene-hormone interaction network model in an artificial intelligence-driven polycystic ovary syndrome drug target screening method proposed in the present invention. DETAILED DESCRIPTION

[0101] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0102] refer to Figure 1-Figure 2 , an artificial intelligence-driven method for screening drug targets for polycystic ovary syndrome, comprising the following steps:

[0103] S1. Collect genomic data and hormone level data from patients to build a patient-specific data set;

[0104] S2. De-noising, missing value filling and standardization of the collected individualized data sets of patients;

[0105] S3, inputting the standardized individualized data set of the patient into the multi-layer neural network model, and constructing a dynamic gene-hormone interaction network model based on the individual data of the patient;

[0106] S4. Use training set data to train the dynamic gene-hormone interaction network model, use back propagation algorithm to adjust model parameters, combine reinforcement learning mechanism, and dynamically optimize the dynamic gene-hormone interaction network model through clinical feedback data;

[0107] S5. In the dynamic gene-hormone interaction network model after training and optimization, the path analysis algorithm is used to identify key nodes that have both gene regulation and hormone regulation functions, and to screen out dual-regulatory drug targets that are highly related to the pathological mechanism of PCOS. The screening results are scored and prioritized according to the regulatory intensity, network centrality and clinical relevance of the targets;

[0108] S6. Generate a target list suitable for each patient's pathological characteristics in the screening results of dual-regulatory drug targets, and optimize it through the patient's historical treatment data and target efficacy feedback;

[0109] S7. Output the list of dual-regulatory drug targets in the form of a report. Clinicians evaluate the model based on the treatment effect. The evaluation information is used to optimize the dynamic gene-hormone interaction network model.

[0110] In this implementation, S1 specifically includes:

[0111] S11. Collect genomic data from the patient, including the patient's gene sequence G s , gene mutation site set M i ={m1,m2,…,m n}, Gene expression matrix E ij ,in:

[0112] G s represents the complete genome sequence of the patient;

[0113] M i Represents a set of gene mutation sites associated with PCOS, including n mutation sites;

[0114] E ij is the expression value of gene i in sample j, forming an N×J matrix, where N is the number of genes and J is the number of samples;

[0115] S12. Collect hormone level data from patient serum samples, including testosterone concentration T s , luteinizing hormone concentration LH s , follicle stimulating hormone concentration FSH s and estrogen concentration E s ,in:

[0116] T s It represents the testosterone concentration in serum in nanomoles / liter;

[0117] LH s It indicates the concentration of luteinizing hormone in international units per liter;

[0118] FSH s represents the concentration of follicle-stimulating hormone in international units per liter;

[0119] E s It indicates the estrogen concentration in picomoles / liter;

[0120] S13. Convert the collected genomic data and hormone level data into different data formats to construct the patient's personalized data set D p :

[0121] D p ={(G s ,M i ,E ij ),(T s ,LH s ,FSH s ,E s )}.

[0122] In this implementation, S2 specifically includes:

[0123] S21. Use threshold filtering method to filter the patient individualized data set D p Perform denoising to remove the gene expression matrix E ij Noisy data in:

[0124]

[0125] Among them, E′ ij is the gene expression value after denoising, ∈ is the threshold parameter of noise filtering;

[0126] S22. De-noised patient individualized dataset D′ p ={(G s ,M i ,E′ ij ),(T s ,LH s ,FSHs ,E s )} to fill in missing values, the gene expression matrix E′ ij Missing values ​​in the hormone level data are handled using the mean filling method:

[0127]

[0128] Among them, E″ ij is the gene expression value after missing value filling, X s ∈{T s ,LH s ,FSH s ,E s} is the hormone level data, P is the number of hormone samples;

[0129] S23, using Min-Max normalization to fill the patient individualized data set D″ p Standardize the process;

[0130] S24. The dataset after denoising, missing value filling and standardization is represented as the final standardized patient individualized dataset:

[0131]

[0132] In this implementation, S3 specifically includes:

[0133] S31. Standardized patient individualized data set Input multi-layer neural network model NN m , gene expression data and hormone levels Combine to form the input feature vector x p :

[0134]

[0135] in, is the input feature vector of patient p, containing N gene expression values ​​and 4 hormone level values;

[0136] S32. Construction of a dynamic gene-hormone interaction network model GHIN based on individual patient data p , dynamic gene-hormone interaction network model based on multi-layer neural network NN m The structure and parameters of represent the relationship between gene regulation and hormone levels. The hidden layer of the multi-layer neural network adopts a nonlinear activation function. The forward propagation process of the dynamic gene-hormone interaction network model is:

[0137] h (1) =σ(W (1) xp +b (1) );

[0138] h (l) =σ(W (l) h (l-1) +b (l) ),l=2,3,…,L;

[0139] y p =σ(W (L+1) h (L) +b (L+1) );

[0140] Among them, h 91) is the neuron output vector of the lth hidden layer, W (1) is the weight matrix of the lth layer, indicating the connection strength from the previous layer to this layer, b (l) is the bias vector of the lth layer, L is the total number of hidden layers, y p The output of neurons in the output layer reflects the interaction between genes and hormones;

[0141] S33. In the dynamic gene-hormone interaction network model GHIN p The weights of the neuron connections Represents the regulatory relationship between gene i and hormone j at level l, by calculating the comprehensive influence I between gene and hormone ij and the importance of nodes in the network topology to identify key nodes in gene regulatory pathways and hormone feedback loops, with a comprehensive impact I ij The calculation is:

[0142]

[0143] in, is the connection weight from node i to node j in layer l, reflecting the direct influence of genes and hormones in this layer. is the activation value of node i in the l-1th layer, indicating the activation state of gene i in the network, is the activation value of node j in layer l, indicating the activation state of hormone j in the network;

[0144] Combined with betweenness centrality C b (i) Degree centrality C d (i) and eigenvector centrality C e (i) Topological index, computing node n i Importance score S in the network i :

[0145] S i =γ1C b (i)+γ2Cd (i)+γ3C e (i);

[0146] in, σ st is the number of shortest paths from node s to node t, σ st (i) is the number of shortest paths passing through node i, reflecting the importance of node i in network information transmission. d (i) = deg(i), which is the degree of node i, indicating the number of nodes it is directly connected to, C e (i) is the eigenvector centrality of node i, which indicates the influence of the node in the overall network structure. γ1, γ2, and γ3 are weight coefficients, satisfying γ1+γ2+γ3=1, which are used to balance the importance of each indicator.

[0147] S34. Based on comprehensive influence I ij and node importance score S i , by calculating the score K of gene i as a key node i :

[0148] K i =β1I ij +β2S i ;

[0149] Among them, K i Score the key nodes of gene i. The higher the value, the more critical it is in gene regulation and hormone feedback. β1 and β2 are weight parameters, satisfying β1+β2=1, which are used to adjust the proportion of comprehensive influence and node importance in the score.

[0150] Sort the gene nodes in descending order according to the scoring results to generate a list of key nodes:

[0151] K p ={k1,k2,…,k m};

[0152] Among them, k m Indicates the mth key gene node after sorting. The key node is used for dual regulation in the pathological mechanism of polycystic ovary syndrome and serves as a potential drug target.

[0153] In this implementation, S4 specifically includes:

[0154] S41. Standardized individualized datasets obtained from multiple patients Composition of training set , and the sample sequence Indicates that x p is the input feature vector of patient p, to label the drug targets associated therewith;

[0155] S42. Using the back propagation algorithm to adjust the dynamic gene-hormone interaction network model GHIN p Parameters, including the weight matrix W (l) and the bias vector b (l) , to minimize the loss function L(θ), where θ={W (l) ,b (l)} represents the model parameter set, and the loss function is defined as:

[0156]

[0157] Among them, M is the number of training samples, N is the number of genes, λ1,λ2 are the weight coefficients of the loss function, y pi is the predicted value and true value of gene expression of the pth patient, H ps are the predicted and true values ​​of the hormone levels of the pth patient;

[0158] S43, in each iteration, updating the model parameters according to the gradient descent method;

[0159] S44. Combined with reinforcement learning mechanism, the dynamic gene-hormone interaction network model GHIN is trained through clinical feedback data p Dynamic optimization is performed, the dynamic gene-hormone interaction network model is regarded as an intelligent agent, the clinical feedback data is used as the environment, and the reward function R is defined to measure the effectiveness of the prediction results of the dynamic gene-hormone interaction network model:

[0160]

[0161] Among them, R(s t ,a t ) is in state s t Take action a t The reward value is, M is the number of patients, K is the number of key nodes, I pk ,S pk is the comprehensive influence and importance score of node k for the pth patient, I(·) is the indicator function, which takes 1 when the prediction is correct and 0 otherwise, and γ is the discount coefficient;

[0162] S45. In the reinforcement learning process, the dynamic gene-hormone interaction network model selects action a according to the strategy function π(a|s), the state s represents the current model parameters, the action a is the parameter update direction, and the strategy function is updated using the strategy gradient method to maximize the cumulative reward;

[0163] S46. Through multiple rounds of iterative training, the dynamic gene-hormone interaction network model adjusts the model parameters θ under the joint action of the training set data and clinical feedback data, so that the dynamic gene-hormone interaction network model can capture the individual differences of different patients and optimize the ability to understand the interaction between genes and hormones.

[0164] In this implementation, S5 specifically includes:

[0165] S51. In the dynamic gene-hormone interaction network model after training and optimization, the path analysis algorithm is used to identify the key nodes that have both gene regulation and hormone regulation functions. The path analysis calculates the influence path between gene and hormone nodes based on the network structure of the dynamic gene-hormone interaction network model. Path strength:

[0166]

[0167] in, represents the path strength between gene i and hormone j, is the connection weight between the gene and hormone nodes in the lth layer, and L is the number of hidden layers;

[0168] S52. Calculate the control intensity R of each node according to the network topology i and centrality indexes to evaluate the importance of nodes in the gene-hormone interaction network, and the regulatory strength R i Represents the comprehensive contribution of node i in multiple paths:

[0169]

[0170] in, is the set of nodes connected to node i, C b (i) is the betweenness centrality of node i, C d (i) is its degree centrality, max(C d ) is the maximum degree centrality value in the network;

[0171] S53, the control intensity R of the combined node i and eigenvector centrality C e (i) Calculate the dual control score S of the node i :

[0172] S i =α1R i +α2C e (i);

[0173] Among them, α1 and α2 are weight parameters, satisfying α1+α2=1, which are used to balance the importance of regulation intensity and network centrality;

[0174] S54. Screen out key nodes k with higher dual regulation scores i Generate a list of candidate targets:

[0175]

[0176] Among them, m is the number of key nodes screened;

[0177] S55. Prioritize candidate targets and generate a final target priority list

[0178]

[0179] in, The target k i β1, β2, and β3 are weight parameters. The first term calculates the weight and betweenness centrality contribution of gene i and other nodes j to evaluate its regulatory role in the network. The second term calculates the prediction accuracy of patient p and key node k according to the prediction accuracy of patient p and key node k. i Rating S i Calculate the average importance of the node, I(·) is the indicator function, which takes the value of 1 when the prediction is correct and 0 otherwise. The third term regulates the intensity R by standardization i Reflects the network importance of the target;

[0180] S56. Score based on comprehensive priority The candidate targets are sorted in descending order based on the calculation results to generate a final target priority list for subsequent targeted drug development and treatment plan design:

[0181]

[0182] In this implementation, S7 specifically includes:

[0183] S71. Generate a priority report of dual-regulatory drug targets, which includes the position of each target in the gene-hormone interaction network, the main regulatory pathway, the priority ranking, and the clinical relevance;

[0184] S72. Describe the regulatory pathway for each target. The regulatory pathway priority report includes all genes and hormone nodes connected to the target, as well as the regulatory relationships between the nodes. The key nodes and influential regulatory links in each pathway are highlighted in the priority report.

[0185] S73. Analyze the central position of each target in the gene-hormone interaction network. Use the number of connections and importance of the target in the gene-hormone interaction network to mark the regulatory ability of the node. The priority report indicates whether the target is in the core position of the gene-hormone interaction network and its involvement in different pathological pathways.

[0186] S74. Based on the patient's historical treatment data and the performance of different targets in individualized treatment of patients, the priority report provides clinical relevance analysis for each target, including the target's response rate, effectiveness feedback, and the target's adaptability evaluation in different patient groups;

[0187] S75. Clinicians evaluate the targets generated by the dynamic gene-hormone interaction network model based on the target information and treatment effects in the priority report, and record the evaluation information for future optimization of the dynamic gene-hormone interaction network model. The doctors' feedback will become a reference for target priority sorting.

[0188] Embodiment 1:

[0189] In order to verify the feasibility and effectiveness of the artificial intelligence-driven polycystic ovary syndrome drug target screening method of the present invention, the experimenters conducted a detailed experimental simulation. The experimental scenario comprehensively demonstrated the operation flow, screening process and comparative analysis of the method of the present invention with traditional screening methods based on real clinical data and public medical data sets.

[0190] The experiment was conducted from March 2024 to September 2024. 150 PCOS patients from three tertiary hospitals in City A, City B and City C were selected, and their genome, hormone levels and treatment feedback data were collected. At the same time, in order to further verify the generalization ability of the model, the experimenters selected the GSE34526 dataset from the Gene Expression Omnibus (GEO) database as an external test dataset, which contains gene expression information of PCOS patients and healthy women. The experimental design aims to screen out the drug targets with the most dual regulatory functions from the data of these patients, and optimize the treatment plan through clinical feedback.

[0191] 1. Acquisition of genome and hormone level data:

[0192] Genetic data: Gene sequences were extracted from the blood samples of each patient, and the gene mutations of INSR and FSHR, which are highly correlated with PCOS, were analyzed. Taking patient A as an example, genomic analysis showed that the FSHR gene had specific mutation sites and its gene expression was significantly abnormal, indicating a disorder in its follicle-stimulating hormone levels.

[0193] Hormone level data: The patient's hormone levels were measured from serum, including testosterone concentration (8.5nmol / L, higher than normal), luteinizing hormone (14IU / L) and estrogen level (68pmol / L). Hormone abnormalities indicated that the patient had typical PCOS symptoms.

[0194] 2. Data standardization and model input:

[0195] All collected data were standardized after noise filtering and missing value filling, and constructed into an individualized data set. The data set was input into the dynamic gene-hormone interaction network model. Through the forward propagation algorithm of the multi-layer neural network, the model captured the complex interaction relationship between genes and hormones.

[0196] The model was trained using the individualized data of 150 PCOS patients, and the model parameters were updated through the back propagation algorithm and the Adam optimizer. After 200 epochs of training, the model accuracy reached 93%. At the same time, through the reinforcement learning mechanism, the dynamic gene-hormone interaction network model uses clinical feedback as a reward function to optimize the target screening results in real time.

[0197] In the dynamic gene-hormone interaction network model, INSR and FSHR were identified as dual-regulated drug targets. Abnormal expression of the INSR gene is closely related to insulin resistance, while mutations in the FSHR gene affect the patient's luteinizing hormone level and lead to ovulation disorders. The dual regulation scores of the targets were calculated and a priority report was generated for clinicians' reference. The report content included the target's regulatory pathway, network centrality, and association with the pathological mechanism of PCOS.

[0198] Based on the priority report, the clinician developed a personalized treatment plan for patient A, using metformin (to improve insulin sensitivity) and the ovulation-inducing drug clomiphene combined for treatment. Three months later, patient A's testosterone level dropped to 3.5nmol / L, luteinizing hormone returned to normal levels, and ovulation occurred successfully.

[0199] During the treatment process, doctors record the patient's treatment response and input the feedback information into the dynamic gene-hormone interaction network model as the reward function of reinforcement learning. The dynamic gene-hormone interaction network model continuously adjusts the weight parameters based on the feedback, so that it can more accurately recommend targets suitable for different patients.

[0200] In order to verify the effect of the method of the present invention in practical application, the experimenters compared it with the traditional laboratory screening method in detail, and the data are shown in Table 1:

[0201] Table 1 Comparison between the method of the present invention and the traditional screening method

[0202] index Method of the present invention Traditional methods Data processing time 1 hour 5 hours Target screening time 2 days 14 days Screening accuracy 93％ 76％ Personalized treatment success rate 85％ 64％ Side effect incidence 9.2％ 15.8％ Target Prioritization Report Support support Not supported

[0203] The present invention realizes accurate screening of PCOS drug targets and recommendation of personalized treatment plans through multi-source data fusion, dynamic gene-hormone interaction network model and reinforcement learning mechanism. Experimental results show that compared with traditional screening methods, the present invention has significant advantages in data processing efficiency, screening accuracy and personalized treatment success rate. In addition, the target priority report provides strong decision-making support for clinicians, reduces the occurrence of side effects, and significantly improves the treatment effect of patients.

[0204] The innovation of the present invention lies not only in its in-depth analysis of multi-level data, but also in the continuous optimization of the dynamic gene-hormone interaction network model through reinforcement learning so that it can adapt to the individual needs of different patients. Through the verification of real data and public data sets, the method of the present invention has shown excellent applicability and reliability in clinical applications, providing a new solution for the precise treatment of PCOS.

[0205] The present invention uses a multi-layer neural network model to deeply integrate the patient's genomic data, hormone level data and metabolic index multi-source data. Different from the traditional single data source analysis, the present invention realizes the modeling of the complex relationship between gene regulation and hormone levels by constructing a dynamic gene-hormone interaction network model. The dynamic gene-hormone interaction network model analyzes the path strength between gene and hormone nodes in the network topology structure to automatically identify the regulatory path and key nodes. The algorithm utilizes the combination of path analysis algorithm and centrality calculation to ensure that the screening results cover the multidimensional pathological mechanism of PCOS and avoid missing key targets. Compared with traditional methods, it reveals the pathological process of PCOS more comprehensively and significantly improves the integrity of target screening.

[0206] The present invention introduces a reinforcement learning mechanism in the training process of the dynamic gene-hormone interaction network model, and continuously optimizes the model parameters through clinical feedback data. The dynamic gene-hormone interaction network model adopts a policy gradient algorithm in the training process so that the target screening results can dynamically adapt to the individualized data of different patients. The reward function is designed based on prediction accuracy and treatment feedback to ensure that the model captures the individual differences of patients during the iteration process, which not only improves the accuracy of the screening results but also supports personalized target recommendations, providing patients with more individualized treatment plans. Experimental results show that after multiple iterations, the consistency between the screened targets and clinical treatment feedback is improved by more than 13%.

[0207] The present invention provides a detailed network analysis report for each screened target through dual regulation score calculation and multi-layer priority sorting. The dual regulation score combines the regulation strength and centrality indicators with the response rate of patient treatment data and clinical feedback to generate a comprehensive priority sorting. The method specially designs an indicator function for calculating the response of the target and ensures the objectivity and stability of the score through standardized network indicators. The generated priority report provides clear decision support for clinicians, helping them to be more accurate and effective in target selection and treatment plan formulation. Compared with traditional target screening technology, the effectiveness and applicability of the method of the present invention in clinical applications are significantly improved, and the success rate of the screened targets in the treatment of PCOS is increased by 15%.

[0208] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An artificial intelligence-driven method for screening drug targets for polycystic ovary syndrome, characterized in that: The steps include: S1. Collect genomic data and hormone level data from patients to build a patient-specific data set; S2. De-noising, missing value filling and standardization of the collected individualized data sets of patients; S3, inputting the standardized individualized data set of the patient into the multi-layer neural network model, and constructing a dynamic gene-hormone interaction network model based on the individual data of the patient; S4. Use training set data to train the dynamic gene-hormone interaction network model, use back propagation algorithm to adjust model parameters, combine reinforcement learning mechanism, and dynamically optimize the dynamic gene-hormone interaction network model through clinical feedback data; S5. In the dynamic gene-hormone interaction network model after training and optimization, the path analysis algorithm is used to identify key nodes that have both gene regulation and hormone regulation functions, and to screen out dual-regulatory drug targets that are highly related to the pathological mechanism of PCOS. The screening results are scored and prioritized according to the regulatory intensity, network centrality and clinical relevance of the targets; S6. Generate a target list suitable for each patient's pathological characteristics in the screening results of dual-regulatory drug targets, and optimize it through the patient's historical treatment data and target efficacy feedback; S7. Output the list of dual-regulatory drug targets in the form of a report. Clinicians evaluate the model based on the treatment effect. The evaluation information is used to optimize the dynamic gene-hormone interaction network model.

2. The artificial intelligence-driven polycystic ovary syndrome drug target screening method according to claim 1, characterized in that: The S1 specifically includes: S11, collecting genome data from the patient, wherein the genome data includes the patient's gene sequence G s , gene mutation site set M i ={m1,m2,…,m n }, Gene expression matrix E ij ,in: G s represents the complete genome sequence of the patient; M i Represents a set of gene mutation sites associated with PCOS, including n mutation sites; E ij is the expression value of gene i in sample j, forming an N×J matrix, where N is the number of genes and J is the number of samples; S12. Collect hormone level data from patient serum samples, including testosterone concentration T s , luteinizing hormone concentration LH s , follicle stimulating hormone concentration FSH s and estrogen concentration E s ,in: T s It represents the testosterone concentration in serum in nanomoles / liter; LH s It indicates the concentration of luteinizing hormone in international units per liter; FSH s represents the concentration of follicle-stimulating hormone in international units per liter; E s It indicates the estrogen concentration in picomoles / liter; S13. Convert the collected genomic data and hormone level data into different data formats to construct the patient's personalized data set D p : D p ={(G s ,M i ,E ij ),(T s ,LH s ,FSH s ,E s )}。 3. The artificial intelligence-driven polycystic ovary syndrome drug target screening method according to claim 1, characterized in that: The S2 specifically includes: S21. Use threshold filtering method to filter the patient individualized data set D p Perform denoising to remove the gene expression matrix E ij Noisy data in: Among them, E′ ij is the gene expression value after denoising, ∈ is the threshold parameter of noise filtering; S22. De-noised patient individualized dataset D′ p ={(G s ,M i ,E′ ij ),(T s ,LH s ,FSH s ,E s )} to fill in missing values, the gene expression matrix E′ ij Missing values ​​in the hormone level data are handled using the mean filling method: Among them, E″ ij is the gene expression value after missing value filling, X s ∈{T s ,LH s ,FSH s ,E s } is the hormone level data, P is the number of hormone samples; S23, using Min-Max normalization to fill the patient individualized data set D″ p Standardize the process; S24. The dataset after denoising, missing value filling and standardization is represented as the final standardized patient individualized dataset:

4. The artificial intelligence-driven polycystic ovary syndrome drug target screening method according to claim 1, characterized in that: The S3 specifically includes: S31. Standardized patient individualized data set Input multi-layer neural network model NN m , gene expression data and hormone levels Combine to form the input feature vector x p : in, is the input feature vector of patient p, containing N gene expression values ​​and 4 hormone level values; S32. Construction of a dynamic gene-hormone interaction network model GHIN based on individual patient data p , dynamic gene-hormone interaction network model based on multi-layer neural network NN m The structure and parameters of represent the relationship between gene regulation and hormone levels. The hidden layer of the multi-layer neural network adopts a nonlinear activation function. The forward propagation process of the dynamic gene-hormone interaction network model is: h (1) =σ(W (1) x p +b (1) ); h (l) =σ(W (l) h (l-1) +b (l) ),l=2,3,…,L; y p =σ(W (L+1) h (L) +b (L+1) ); Among them, h (1) is the neuron output vector of the lth hidden layer, W (1) is the weight matrix of the lth layer, indicating the connection strength from the previous layer to this layer, b (l) is the bias vector of the lth layer, L is the total number of hidden layers, y p The output of neurons in the output layer reflects the interaction between genes and hormones; S33. In the dynamic gene-hormone interaction network model GHIN p The weights of the neuron connections Represents the regulatory relationship between gene i and hormone j at level l, by calculating the comprehensive influence I between gene and hormone ij and the importance of nodes in the network topology to identify key nodes in gene regulatory pathways and hormone feedback loops, with a comprehensive impact I ij The calculation is: in, is the connection weight from node i to node j in layer l, reflecting the direct influence of genes and hormones in this layer. is the activation value of node i in the l-1th layer, indicating the activation state of gene i in the network, is the activation value of node j in layer l, indicating the activation state of hormone j in the network; Combined with betweenness centrality C b (i) Degree centrality C d (i) and eigenvector centrality C e (i) Topological index, computing node n i Importance score S in the network i : S i =γ1C b (i)+γ2C d (i)+γ3C e (i); in, σ st is the number of shortest paths from node s to node t, σ st (i) is the number of shortest paths passing through node i, reflecting the importance of node i in network information transmission. d (i) = deg(i), which is the degree of node i, indicating the number of nodes it is directly connected to, C e (i) is the eigenvector centrality of node i, which indicates the influence of the node in the overall network structure. γ1, γ2, and γ3 are weight coefficients, satisfying γ1+γ2+γ3=1, which are used to balance the importance of each indicator. S34. Based on comprehensive influence I ij and node importance score S i , by calculating the score K of gene i as a key node i : K i =β1I ij +β2S i ; Among them, K i Score the key nodes of gene i. The higher the value, the more critical it is in gene regulation and hormone feedback. β1 and β2 are weight parameters, satisfying β1+β2=1, which are used to adjust the proportion of comprehensive influence and node importance in the score. Sort the gene nodes in descending order according to the scoring results to generate a list of key nodes: K p ={k1,k2,…,k m }; Among them, k m Indicates the mth key gene node after sorting. The key node is used for dual regulation in the pathological mechanism of polycystic ovary syndrome and serves as a potential drug target.

5. The artificial intelligence-driven polycystic ovary syndrome drug target screening method according to claim 1, characterized in that: The S4 specifically includes: S41. Standardized individualized datasets obtained from multiple patients Composition of training set And the sample sequence Indicates that x p is the input feature vector of patient p, to label the drug targets associated therewith; S42. Using the back propagation algorithm to adjust the dynamic gene-hormone interaction network model GHIN p Parameters, including the weight matrix W (l) and the bias vector b (l) , to minimize the loss function L(θ), where θ={W (l) ,b (l) } represents the model parameter set, and the loss function is defined as: Among them, M is the number of training samples, N is the number of genes, λ1,λ2 are the weight coefficients of the loss function, y pi is the predicted value and true value of gene expression of the pth patient, H ps are the predicted and true values ​​of the hormone levels of the pth patient; S43, in each iteration, updating the model parameters according to the gradient descent method; S44. Combined with reinforcement learning mechanism, the dynamic gene-hormone interaction network model GHIN is trained through clinical feedback data p Dynamic optimization is performed, the dynamic gene-hormone interaction network model is regarded as an intelligent agent, the clinical feedback data is used as the environment, and the reward function R is defined to measure the effectiveness of the prediction results of the dynamic gene-hormone interaction network model: Among them, R(s t ,a t ) is in state s t Take action a t The reward value is, M is the number of patients, K is the number of key nodes, I pk ,S pk is the comprehensive influence and importance score of node k of the pth patient, Ⅱ(·) is the indicator function, which takes 1 when the prediction is correct and 0 otherwise, and γ is the discount coefficient; S45. In the reinforcement learning process, the dynamic gene-hormone interaction network model selects action a according to the strategy function π(a|s), the state s represents the current model parameters, the action a is the parameter update direction, and the strategy function is updated using the strategy gradient method to maximize the cumulative reward; S46. Through multiple rounds of iterative training, the dynamic gene-hormone interaction network model adjusts the model parameters θ under the joint action of the training set data and clinical feedback data, so that the dynamic gene-hormone interaction network model can capture the individual differences of different patients and optimize the ability to understand the interaction between genes and hormones.

6. The artificial intelligence-driven polycystic ovary syndrome drug target screening method according to claim 1, characterized in that: The S5 specifically includes: S51. In the dynamic gene-hormone interaction network model after training and optimization, the path analysis algorithm is used to identify the key nodes that have both gene regulation and hormone regulation functions. The path analysis calculates the influence path between gene and hormone nodes based on the network structure of the dynamic gene-hormone interaction network model. Path strength: in, represents the path strength between gene i and hormone j, is the connection weight between the gene and hormone nodes in the lth layer, and L is the number of hidden layers; S52. Calculate the control intensity R of each node according to the network topology i and centrality indexes to evaluate the importance of nodes in the gene-hormone interaction network, and the regulatory strength R i Represents the comprehensive contribution of node i in multiple paths: in, is the set of nodes connected to node i, C b (i) is the betweenness centrality of node i, C d (i) is its degree centrality, max(C d ) is the maximum degree centrality value in the network; S53, the control intensity R of the combined node i and eigenvector centrality C e (i) Calculate the dual control score S of the node i : S i =α1R i +α2C e (i); Among them, α1 and α2 are weight parameters, satisfying α1+α2=1, which are used to balance the importance of regulation intensity and network centrality; S54. Screen out key nodes k with higher dual regulation scores i Generate a list of candidate targets: Among them, m is the number of key nodes screened; S55. Prioritize candidate targets and generate a final target priority list in, The target k i β1, β2, and β3 are weight parameters. The first term calculates the weight and betweenness centrality contribution of gene i and other nodes j to evaluate its regulatory role in the network. The second term calculates the prediction accuracy of patient p and key node k according to the prediction accuracy of patient p and key node k. i Rating S i Calculate the average importance of the node, II(·) is the indicator function, which takes the value of 1 when the prediction is correct and 0 otherwise. The third term regulates the intensity R by standardization i Reflects the network importance of the target; S56. Score based on comprehensive priority The candidate targets are sorted in descending order based on the calculation results to generate a final target priority list for subsequent targeted drug development and treatment plan design:

7. The artificial intelligence-driven polycystic ovary syndrome drug target screening method according to claim 1, characterized in that: The S7 specifically includes: S71. Generate a priority report of dual-regulatory drug targets, which includes the position of each target in the gene-hormone interaction network, the main regulatory pathway, the priority ranking, and the clinical relevance; S72. Describe the regulatory pathway for each target. The regulatory pathway priority report includes all genes and hormone nodes connected to the target, as well as the regulatory relationships between the nodes. The key nodes and influential regulatory links in each pathway are highlighted in the priority report. S73. Analyze the central position of each target in the gene-hormone interaction network. Use the number of connections and importance of the target in the gene-hormone interaction network to mark the regulatory ability of the node. The priority report indicates whether the target is in the core position of the gene-hormone interaction network and its involvement in different pathological pathways. S74. Based on the patient's historical treatment data and the performance of different targets in individualized treatment of patients, the priority report provides clinical relevance analysis for each target, including the target's response rate, effectiveness feedback, and the target's adaptability evaluation in different patient groups; S75. Clinicians evaluate the targets generated by the dynamic gene-hormone interaction network model based on the target information and treatment effects in the priority report, and record the evaluation information for future optimization of the dynamic gene-hormone interaction network model. The doctors' feedback will become a reference for target priority sorting.

Images