Personalized Drug Recommendation System for Rheumatoid Arthritis Based on Graph Embedding Hierarchical Structure

Abstract

The present invention discloses a personalized drug recommendation system for rheumatoid arthritis based on a graph embedding hierarchical structure. The system obtains the medical records of patients diagnosed with rheumatoid arthritis and represents patient information using the probability distribution of drug categories; taking drugs as nodes, constructs EHR and DDI matrices based on the occurrence times of drug pairs and the interactions between drug pairs, and processes the EHR and DDI graph features through a graph convolutional network, combines the adjacency matrix and the feature matrix to update the node representation; combines patient features and graph embedding information, enhances the features by fusing the drug category probability distribution information with the weighted information obtained from the DDI and EHR matrices, realizes attention-based multi-level weighted feature fusion, and obtains the output result. The present invention combines the design of a drug graph enhancement encoder module and a hierarchical structure decoder to recommend a personalized drug plan with an increased remission rate after treatment for patients with single diagnosis of rheumatoid arthritis.

Description

Technical Field

[0001] The present invention relates to the field of medical health information technology, and particularly to a personalized drug recommendation system for rheumatoid arthritis based on a graph embedding hierarchical structure. Background Art

[0002] Rheumatoid Arthritis (RA) is a systemic autoimmune disease, and the condition usually recurs and progresses continuously, ultimately resulting in disability or even death due to major complications. Among them, Disease Activity (DA), as one of the important indicators for RA disease assessment, decisively affects the doctor's treatment strategy. Currently, the treatment of rheumatoid arthritis (RA) still has a low remission and response rate. Part of the reason is the current insufficient understanding of the treatment target strategy for rheumatoid diseases, resulting in rheumatologists being unable to fully support treatment based on treatment guidelines. In addition, some treatment methods themselves have failed to effectively improve the remission rate. At the same time, most clinicians and patients have not fully accepted the concept of treatment to target, resulting in the principle of treatment to target not being fully implemented in clinical practice.

[0003] Therefore, it is particularly crucial to formulate personalized and effective clinical treatment plans according to the specific conditions of different patients to ensure the optimal treatment outcome and improve the quality of life of patients.

[0004] Early work on drug recommendation focused on learning a series of rules from EHRs. Subsequent work used recurrent neural networks to model the sequential dependence of a patient's past visits to improve the effectiveness of drug recommendation. However, these methods did not consider drug interactions, which are crucial for minimizing drug adverse reactions.

[0005] Current research on drug recommendation models mainly focuses on diseases with relatively mature multi-disease and pathogenesis studies. In common multi-disease drug recommendation models, LEAP combines an external drug-drug interaction (DDI) database and uses a recurrent decoder to model the drug-disease relationship in the current visit to recommend a set of drugs. G-Bert uses all data to pre-train the diagnosis and drug encoders, but requires historical drug records in the fine-tuning stage. COGNet copies past medication records into the current visit information. REFINE proposes to model the severity of drug interactions and fine-grained drug doses. But these several networks rely on historical medication records as input to improve performance. This historical data-based model lacks effective support for single-visit patients.

[0006] In summary, existing research emphasizes the feasibility and clinical utility of using machine learning in rheumatology for patient stratification and prediction of treatment response. Current drug recommendation models mainly focus on general practice and diseases with more mature research. For chronic diseases such as RA, achieving reasonable drug selection based on drug interactions remains a key unmet need. Specifically, the existing technologies have the following deficiencies:

[0007] 1) Insufficient accurate prediction of treatment response rate and remission, over-reliance on multiple diagnostic records

[0008] Existing treatment regimens fail to accurately predict the treatment response rate and remission effect of rheumatoid arthritis patients, rely on multiple diagnostic records, and ignore the recommendation needs of single-visit patients. In the real-world healthcare system, single-visit patients cannot be excluded, indicating that this is a key area that needs improvement.

[0009] 2) Ignoring the deep integration of the complexity of the relationship between patient characteristics and drugs, insufficient drug relationship modeling ability

[0010] The differences in drug relationships in a single-disease scenario are not fully considered, and the combination of patients' individualized clinical information and drug characteristics is not deeply explored. Traditional drug recommendation methods often process patient information and drug characteristics separately, simply relying on basic information (such as age, gender, etc.), while ignoring the complex co-occurrence and interaction relationships between patients' clinical manifestations and drugs. Some methods attempt to combine patient information and drug characteristics through a graph embedding encoder, but there are still deficiencies in the deep modeling of drug relationships, especially unable to handle the complex dependence relationships between drugs.

[0011] 3) Lack of hierarchical selection of drug relationships, confusion and loss in the information aggregation process

[0012] In the processing of drug chemical relationships, traditional models fail to distinguish the hierarchical structure between drug categories and their subcategories, resulting in confusion and loss in the information aggregation process. In the scenario of multi-label drug recommendation, the model fails to fully consider the relationship between drug categories and subcategories, which makes the model unable to make accurate hierarchical decisions in complex drug selection. Therefore, when faced with multiple drug choices, drug recommendation systems often cannot provide personalized and refined suggestions. Summary of the Invention

[0013] The purpose of the present invention is to propose a personalized drug recommendation system for rheumatoid arthritis based on a graph embedding hierarchical structure in view of the deficiencies of the existing technology. By combining a drug graph enhancement encoder module and a hierarchical structure decoder design, a personalized drug regimen is recommended to increase the remission rate after treatment for single-diagnosis patients with rheumatoid arthritis.

[0014] The object of the present invention is achieved by the following technical solutions: A personalized drug recommendation system for rheumatoid arthritis based on a graph embedding hierarchical structure, the system comprising:

[0015] A feature processing and patient information representation module, configured to obtain the medical records of patients diagnosed with rheumatoid arthritis, perform data feature processing, and represent patient information based on the probability distribution of drug categories used by patients obtained from the electronic medical record information of the first diagnosis.

[0016] An EHR and DDI matrix module, configured to use drugs as nodes, and use the occurrence times of drug pairs in the prescription drug set and the interaction between drug pairs as weighted edges to construct a top-level EHR matrix, a bottom-level EHR matrix, and a bottom-level DDI matrix.

[0017] A drug graph embedding encoder module, configured to process the EHR and DDI graph features through a graph convolutional network, combine the adjacency matrix and the feature matrix, and update the node representation.

[0018] A hierarchical drug recommendation decoder, first fusing the graph embedding representation of the patient's electronic health record EHR matrix with its current drug category probability distribution to generate a preliminary medication classification result; subsequently, at the bottom layer, fusing the drug probability distribution information output from the top layer with the graph embedding representation of the bottom-level EHR data and the drug interaction DDI network for secondary fusion, and finally outputting an optimized drug recommendation result through dual knowledge integration.

[0019] Further, obtaining the medical records of patients diagnosed with rheumatoid arthritis specifically includes: selecting patients with two consecutive clinical records starting from the first admission, and comparing the treatment effects before and after medication through the condition assessment, and selecting the qualified patient samples.

[0020] Further, the data feature processing specifically includes: using variance filtering, recursive feature elimination RFE in the wrapper method, recursive feature elimination based on weight ranking, and feature selection methods such as L2 regularization in the embedding method to screen out the union.

[0021] Further, the patient information representation specifically includes: using a random forest model, taking the electronic medical record information of the patient's first diagnosis as the input feature data of the patient, and obtaining the output of the drug category probability distribution corresponding to the patient.

[0022] Further, in the EHR matrix, the co-occurrence frequency is calculated by calculating the occurrence times of drug pairs in the prescription drug set as the weighted edge, and in the DDI matrix, the interaction between drug pairs is used as the weighted edge in the drug interaction graph.

[0023] Further, the weight in the drug interaction graph represents the severity of the interaction between drugs, and the severity information is divided into major, moderate, minor, or unknown / no response.

[0024] Further, in the drug graph embedding encoder module, a two-layer graph convolutional network is used to update the node representation, enabling the features of each node to incorporate the information of neighboring nodes, thereby generating node embedding vectors that fuse graph structure information; and the node embedding vectors will be fused with the original input features.

[0025] Further, using the probability distribution information of the drug category features obtained for the patient, the graph embedding information processed by the top-level EHR matrix through the graph convolutional network is used for feature enhancement. Through feature weighted fusion based on the attention mechanism, the top-level classification result that combines patient features and graph embedding information is obtained; the top-level classification result is input into the bottom-level classifier, and the probability distribution information of the drug category features obtained at the bottom level is enhanced with the weighted information obtained from the bottom-level EHR matrix and the bottom-level DDI matrix, realizing attention-based multi-level weighted feature fusion to obtain the final output result.

[0026] Further, the predicted value of each drug category label is converted into a probability value and the loss is calculated. The Sigmoid activation function is used to map the value of each drug category label to the interval [0, 1], thereby obtaining the predicted probability of each label.

[0027] Further, by maximizing the difference between the predicted score of the correct label and other labels, the discrimination of the model for each label is improved.

[0028] Advantages of the present invention:

[0029] 1) Improve the prediction accuracy of the treatment remission rate for patients with a single diagnosis record

[0030] Existing drug recommendation models mainly rely on multiple medical records. However, for patients with only a single visit record, it is difficult for existing technologies to provide accurate personalized drug recommendations. The treatment of rheumatoid arthritis patients usually needs to be adjusted according to multiple treatment responses. For patients with a single visit, existing models cannot effectively capture the treatment responses of patients, resulting in drug regimens recommended that are insufficient to improve the remission rate of patients. The present invention aims to solve the problem of personalized drug recommendation for improving the remission rate after treatment of patients with a single visit, proposes a representation method based on data of patients who have achieved remission after treatment, and uses a variety of feature selection methods to screen effective features to improve the accuracy of recommendations for patients with a single visit.

[0031] 2) Achieve a deep fusion of patient personalized features and drug relationships, and enhance drug relationship modeling

[0032] The co-occurrence matrix of drugs is a matrix constructed using the frequency of a single drug in the treatment regimens of all patients in the training set within drug combinations. The Drug-Drug Interaction (DDI) matrix is used to represent the interactions between different drugs, enhance the safety of recommendations while avoiding adverse drug reactions, and enhance the therapeutic effect.

[0033] In this task, a two-layer graph convolutional network is used to process the graph features of the input EHR and DDI. The representations of the nodes are updated through the adjacency matrix and the feature matrix, enabling the features of each node to incorporate the features of its neighboring nodes, thereby generating a node embedding vector that fuses the graph structure information. This is then fused with the feature representation based on the model prediction probability to obtain the graph-enhanced information.

[0034] 3) Construct a hierarchical selection mechanism for drug relationships

[0035] Through a hierarchical drug recommendation decoding method, the drug recommendation problem is transformed into a hierarchical classification task. By leveraging the top-down hierarchical structure relationship of the category hierarchy, the classification performance of the model is improved. The major categories and their corresponding minor categories among the common first-line RA treatment drugs are used as the final recommended labels. The model effectively handles complex multi-label recommendation tasks by refining the classification granularity, solving the problem of ineffective aggregation and decision-making of the hierarchical structure and its interdependent information between the major and minor drug categories in drug recommendation. Description of the Drawings

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0037] Figure 1 It is a schematic diagram of a personalized drug recommendation system for rheumatoid arthritis based on a graph embedding hierarchical structure provided by the present invention.

[0038] Figure 2 It is a schematic diagram of the model of the present invention. Detailed Embodiments

[0039] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described below in conjunction with the drawings and implementation examples. It should be understood that the specific implementation examples described herein are only used to explain the present invention and are not used to limit the present invention.

[0040] Such as Figure 1As shown, the present invention provides a personalized drug recommendation system for rheumatoid arthritis based on a graph embedding hierarchical structure, which designs modules such as data preprocessing and feature extraction, patient information representation, graph embedding enhancement (including EHR and DDI matrix modules and drug graph embedding encoder modules), a hierarchical drug recommendation decoder, and loss function design to complete the personalized drug recommendation process; specifically, it includes the following modules:

[0041] Feature processing and patient information representation module:

[0042] First, all medical records of patients with confirmed rheumatoid arthritis were extracted, and all outpatient records and hospitalization records with missing admission and discharge diagnoses were excluded. Patients with two consecutive clinical records since the first admission were selected, and the treatment effects before and after medication were compared by condition assessment. The existing results of four indicators such as DAS28CRP, DAS28ESR, CDAI, and SDAI were used as references. According to the evaluation criteria proposed by the European League Against Rheumatism, when the DAS28CRP score is less than 2.6, the DAS28ESR score is less than 2.6, the CDAI score is less than 2.8, and the SDAI score is less than 2.6, it is regarded as disease remission under the definition of each indicator. In order to reflect the average situation under different indicators, the present invention defines any one of the four indicators detected during the second visit as disease remission (target). When the DAS28CRP or DAS28ESR score is less than 5.1, and the difference between the two DAS28CRP or DAS28ESR scores is greater than 0.6, the treatment response is considered. If the difference between any of the two indicators detected during the second visit reaches the standard, it is defined as a response. In order to propose a medication recommendation model with a better prognosis, thereby improving the disease treatment target rate, the present invention selects patients who meet the target as research subjects.

[0043] Taking into account the difficulty of data collection, feature selection methods such as variance filtering, recursive feature elimination (RFE) in the wrapping method, recursive feature elimination based on weight sorting, and L2 regularization in the embedding method are used to screen out the union. The specific process is as follows:

[0044] Suppose there is a feature matrix , where n is the number of samples and m is the number of features, represents the corresponding j-th feature, represents the specific value of the i-th sample on the j-th feature, represents the mean of the jth feature. The process of variance filtering can be expressed as the following formula. The feature selection rule of variance filtering is to remove features with variance less than a certain threshold. Features, It is the feature set after variance retention.

[0045]

[0046]

[0047] Recursive feature elimination constructs a feature representation learning method based on the predicted probability of the model. The probability information obtained through ensemble learning methods such as random forest is used as new features, thereby enhancing the expression and prediction capabilities of downstream models when dealing with complex medical data. Each time of recursion, the selection of features can be expressed by the following formula. represents the finally selected k features, S is the set of features, and L is the loss function. represents using the parameter the model trained, referring to ensemble models such as random forest. The parameter represents, for example, the number of trees, depth, etc. in the random forest.

[0048]

[0049] Weight-ranking-based RFE is similar to standard RFE, but it further optimizes by considering the weight ranking of features. Assume that the model used can assign weights .

[0050]

[0051] In the embedding method, L2 regularization (also known as ridge regression) is a feature selection method that optimizes the model by adding a penalty term. The goal of L2 regularization is to minimize the objective function while penalizing the weights of features, making the weights of unimportant features close to zero. Assume that a linear regression model is used for training, n and m represent the total number of samples and the number of features respectively, is the true label of the i-th sample, is the feature vector of the i-th sample, is the weight of the j-th feature. And add the L2 regularization penalty term, is the regularization strength hyperparameter, and the expression of the objective function is as follows. Select the features with non-zero weights for the final model.

[0052]

[0053]

[0054] Comprehensive features The selection process is as follows:

[0055]

[0056] First, the present invention predicts the prediction probability of a given sample X by training a random forest model. The random forest consists of N trees, and each tree t gives a probability output regarding the drug category indicating the prediction probability of the drug category k of the nth patient on the tree t:

[0057]

[0058] The present invention uses the electronic medical record information of the first diagnosis of the disease as the input as the input feature data of the nth patient, is the drug probability distribution output of the nth patient, and the probability of each drug category , where k is the number of drug categories. Among them represents the parameters of the tree t.

[0059]

[0060] EHR and DDI matrix module:

[0061] The drug EHR and DDI matrices are respectively represented as Ge = { }e and Gd = { }, and each drug node . represents the co-occurrence relationship of drugs in the drug usage plan, and the weighted edge calculates the co-occurrence frequency by calculating the number of occurrences of drug pairs in the prescribed drug set in the EHR, where the weight represents the frequency of co-occurrence between drugs. represents the drug interaction graph, where the weighted edge represents the interaction between any pair of drugs i and j. Among them, the weight represents the severity of the interaction between drugs. In the DDInter database, the severity information is divided into major, moderate, minor, or unknown / no reaction. The present invention uses 3, 2, 1, and 0 to represent these severities, and the severity is normalized to the range between 0 and 1.

[0062] The present invention involves a total of one top-level EHR matrix , one bottom-level EHR matrix and one bottom-level DDI matrix .

[0063] Drug graph embedding encoder module:

[0064] Graph Convolutional Network (GCN) is an effective method for node representation learning, which is widely used in graph-structured data and can capture local structural information between nodes (such as feature dependencies in clinical data). GCN combines the features of each node with the features of its neighbor nodes through convolutional operations on the adjacency matrix to generate deep node representations. GCN has been successfully applied to biomedical networks, especially drug-drug interaction (DDI) graphs, where drugs are modeled as nodes and DDI as the relationships between nodes, thus enhancing the performance of drug recommendation systems.

[0065] The calculation of GCN usually involves the following steps. First is the normalization of the adjacency matrix. To avoid the problem of dimension in feature propagation, GCN usually normalizes the adjacency matrix, aiming to eliminate the influence of node degree differences and make the graph convolutional operation more balanced. As shown in the following formula, is the original adjacency matrix, representing the connection relationships between nodes. is the identity matrix, is the degree matrix.

[0066]

[0067] In the task of the present invention, EHR and DDI graph features are processed by a two-layer GCN. GCN combines the adjacency matrix and the feature matrix to update the node representations, enabling the features of each node to incorporate the information of neighbor nodes, thus generating node embedding vectors that fuse graph structure information. The node embedding vectors processed by GCN will be fused with the original input features. The fusion method can adopt methods such as concatenation or weighted summation to combine the structural information extracted by the graph convolutional network with the original features and enhance the expressive power of the model.

[0068] First, graph convolution propagates information about the features of nodes through the normalized adjacency matrix is the trainable parameter matrix in the GCN layer, which is used for feature transformation and dimension mapping. Second, the node features are updated through matrix multiplication Then, after introducing non-linear interactions through the ReLU activation function, a second matrix multiplication is performed in combination with the weight and finally the final embedding representation of each node is obtained .

[0069]

[0070] The core objective of the present invention is to solve the problem of hierarchical modeling of drug relationships. By introducing GCN and a multi-level classification structure, the performance of the drug recommendation system in complex multi-label scenarios is improved, especially being able to make accurate decisions when facing multiple drug choices and effectively avoiding information loss. ​

[0071] Hierarchical Structure Drug Recommendation Decoder

[0072] Directly classifying all text categories may face the problem of uneven distribution of category samples and it is difficult to learn the relationships between different types of drugs. Therefore, the present invention transforms the problem into a hierarchical classification problem, and improves the classification performance of the model by utilizing the hierarchical structure relationship between categories, thus proposing a RA double-layer drug recommendation model.

[0073] Directly classifying or encoding patient information will face problems such as uneven distribution of category samples and difficulty in learning the relationships between drug types. Therefore, the research takes drug categories as the benchmark, transforms it into a classification problem, and improves the classification performance of the model by utilizing the hierarchical structure relationship between categories from top to bottom.

[0074] The specific steps are to utilize the patient to obtain the feature probability distribution information , and use the final embedding representation of the nodes after double-layer processing by GCN as the graph embedding enhancement module to enhance the features . Through feature weighted fusion based on the attention mechanism, combine patient features and graph embedding information. Feature enhancement is obtained by adding the graph matrices, is a custom scaling parameter used to adjust the weight ratio of the matrices.

[0075]

[0076]

[0077] Calculate the similarity matrix of patient features and graph embedding through matrix multiplication . Apply Softmax to normalize the similarity, control the aggregation intensity through the hyperparameter , and add it to the original features to retain the individual feature details and obtain the top-level classification result . The top-level output is used as the input, and features are further extracted through a bottom-level classifier (such as a linear layer or a shallow GCN).

[0078] Enhance the feature probability distribution obtained at the bottom layer with the weighted information obtained from the DDI and EHR matrices to achieve attention-based multi-level weighted feature fusion, and finally obtain the output result .

[0079]

[0080]

[0081] Loss function design module:

[0082] The loss function is divided into two parts: The first part uses binary cross-entropy loss, a commonly used standard loss function in multi-label tasks. In this task, the output of the model is usually unactivated logits, which represent the predicted values for each label. To convert these logits into probability values and calculate the loss, the Sigmoid activation function is used to map each logit value to the interval [0, 1], thereby obtaining the predicted probability for each label.

[0083] The second part of the loss function uses multi-label marginal loss. Different from traditional multi-class cross-entropy loss, multi-label marginal loss does not require mutual exclusivity between labels. Instead, it optimizes the performance of the model by focusing on the marginal differences between labels. The design concept of marginal loss is to improve the discrimination of the model for each label by maximizing the difference between the predicted scores of the correct label and other labels.

[0084] To comprehensively consider the advantages of binary cross-entropy loss and multi-label marginal loss, the weighted sum of the two loss functions is used as the final total loss function. The weighting coefficients are controlled by hyperparameters, and the values of these hyperparameters need to be tuned according to the actual task. The specific loss function is as follows. is the true label of sample i, represents the predicted output for sample i, which is converted into a probability through the Sigmoid function

[0085]

[0086] Data processing and model result comparison: The present invention collected data of rheumatoid arthritis patients from 352 clinical hospitals in multiple provinces between 2000 and 2023. After excluding patients without drug treatment plans, patients without records of re-evaluation of the condition after drug treatment, and patients whose treatment did not reach the standard after the first treatment, 4975 patients who reached the standard after single treatment were obtained.

[0087] Considering the ease of data collection, a total of 44 clinical features were selected from the basic information, diagnosis confirmation, medical history, joint assessment, organ involvement, condition assessment, laboratory tests, etc. of the patients in the enrollment cohort. Using feature selection methods such as variance filtering, recursive feature elimination (RFE) in the wrapper method, recursive feature elimination based on weight ranking, and L2 regularization in the embedding method, after screening and taking the union, 37 features were left.

[0088] Select four categories of drugs, namely glucocorticoids, immunosuppressants, special immunosuppressants, and nonsteroidal anti-inflammatory drugs (NSAIDs) among the common first-line RA treatment drugs, and select two specific drug chemical names with the most frequent medication times under each drug category, a total of 8 subcategories, as the final recommended labels.

[0089] Table 1 Comparison of indicators of the drug recommendation model

[0090]

[0091] Compare the model proposed in the present invention with the classical multi-label model CC chain classifier, common machine learning algorithms such as SVM, LightGBM, RF, classical neural networks MLP and CNN, and the multi-disease drug recommendation model LEAP that can be based on a single visit. At the same time, the present invention also includes the non-hierarchical model with a graph embedding enhancement module (Our model (-) in Table 1) in the comparison. The comparison indicators include three basic indicators, namely Micro F1, AUC, and Micro Jaccard, to evaluate the performance of the classification model, as well as the number of drug recommendations. The model of the present invention (Our model in Table 1) has the best comprehensive performance. The evaluation adopts five-fold cross-validation. The t-test is used to evaluate the significant statistical difference with p < 0.05 in AUC between LEAP and the model proposed in the present invention.

[0092] To more intuitively and comprehensively display the results of drug recommendations, show the average number of recommended drugs, the average number of additional drugs, and the average number of missed drugs in the present invention and the comparative experiment. It can be seen that although the chain classifier recommends more drugs, it additionally recommends incorrect drugs, resulting in missing more drugs. The model of the present invention misses the fewest drugs, recommends more correct drugs, and achieves better results.

[0093] The above embodiments are used to explain the present invention, rather than limiting the present invention. Any modifications and changes made within the spirit and scope of the protection of the present invention fall within the protection scope of the present invention.

Claims

1. A personalized drug recommendation system for rheumatoid arthritis based on a graph embedding hierarchical structure, characterized in that, The system includes: A feature processing and patient information representation module, which is used to obtain the medical records of patients diagnosed with rheumatoid arthritis, perform data feature processing, and represent patient information based on the probability distribution of drug categories used by patients obtained from the electronic medical record information of the first diagnosis disease; An EHR and DDI matrix module, which is used to construct a top-level EHR matrix, a bottom-level EHR matrix, and a bottom-level DDI matrix with drugs as nodes and the occurrence times of drug pairs in the prescription drug set and the interactions between drug pairs as weighted edges; A drug graph embedding encoder module, which is used to process the top-level EHR matrix, the bottom-level EHR matrix, and the bottom-level DDI matrix through a graph convolutional network, and update the node representation by combining the adjacency matrix and the feature matrix; In the top layer of the hierarchical drug recommendation decoder, first, the graph embedding representation of the top-level EHR matrix is feature-fused with its current drug category probability distribution to generate a preliminary medication classification result; subsequently, in the bottom layer, the drug probability distribution information output from the top layer is secondarily fused with the graph embedding representations of the bottom-level EHR matrix and the bottom-level DDI matrix, and the optimized drug recommendation result is finally output through double knowledge integration.

2. The personalized drug recommendation system for rheumatoid arthritis based on the graph embedding hierarchical structure according to claim 1, wherein Obtaining the medical records of patients diagnosed with rheumatoid arthritis specifically means: selecting patients with two consecutive clinical records starting from the first admission, and comparing the treatment effects before and after medication through the condition assessment, and selecting the qualified patient samples.

3. The personalized drug recommendation system for rheumatoid arthritis based on the graph embedding hierarchical structure according to claim 1, wherein The data feature processing specifically means: using the feature selection methods of variance filtering, recursive feature elimination RFE in the wrapper method, recursive feature elimination based on weight ranking, and L2 regularization in the embedding method to screen out the union.

4. The personalized drug recommendation system for rheumatoid arthritis based on the graph embedding hierarchical structure according to claim 1, characterized in that, The patient information representation specifically means: using a random forest model, taking the electronic medical record information of the patient's first diagnosis disease as the input feature data of the patient, and obtaining the output of the probability distribution of drug categories corresponding to the patient.

5. The personalized drug recommendation system for rheumatoid arthritis based on the graph embedding hierarchical structure according to claim 1, wherein In the top-level EHR matrix and the bottom-level EHR matrix, the co-occurrence frequency is calculated by calculating the occurrence times of drug pairs in the prescription drug set as the weighted edge, and in the bottom-level DDI matrix, the interaction between drug pairs in the drug interaction graph is used as the weighted edge.

6. The personalized drug recommendation system for rheumatoid arthritis based on the graph embedding hierarchical structure according to claim 5, wherein The weight in the drug interaction graph represents the severity of the interaction between drugs, and the severity information is divided into major, moderate, minor, or unknown / no response.

7. A personalized drug recommendation system for rheumatoid arthritis based on a graph embedding hierarchical structure according to claim 1, characterized in that, In the drug graph embedding encoder module, a two-layer graph convolutional network is used to update the node representation, so that the features of each node can combine the information of neighboring nodes, thereby generating a node embedding vector that fuses the graph structure information; And the node embedding vector will be fused with the original input features.

8. The personalized drug recommendation system for rheumatoid arthritis based on the graph embedding hierarchical structure according to claim 1, characterized in that, Using the probability distribution information of the drug category features obtained by the patient, the graph embedding information after the top-level EHR matrix is processed by the graph convolutional network is used for feature enhancement, and through the feature weighted fusion based on the attention mechanism, the top-level classification result combining the patient features and the graph embedding information is obtained; the top-level classification result is input into the bottom-level classifier, and the probability distribution information of the drug category features obtained at the bottom level is feature-enhanced with the weighted information obtained from the bottom-level EHR matrix and the bottom-level DDI matrix, realizing the feature fusion of multi-level weighted based on attention, and obtaining the final output result.

9. The personalized drug recommendation system for rheumatoid arthritis based on the graph embedding hierarchical structure according to claim 1, wherein, Convert the predicted values of each drug class label into probability values and calculate the loss. Use the Sigmoid activation function to map the value of each drug class label to the interval [0, 1], so as to obtain the predicted probability of each label.

10. The personalized drug recommendation system for rheumatoid arthritis based on the graph embedding hierarchical structure according to claim 1, wherein, By maximizing the difference between the predicted scores of the correct label and other labels, improve the discrimination of the model for each label.

Images