Traditional Chinese medicine pair incompatibility prediction method and system
By constructing a taboo prediction system for the combination of traditional Chinese medicine, using attention aggregation and multi-view learning technology, the problems of component changes and dynamic changes in the combination of traditional Chinese medicine are solved, and the accuracy and understanding ability of the prediction of the combination of traditional Chinese medicine are improved.
Patent Information
- Application Number
- CN202411086340.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-08
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-08
AI Technical Summary
The prior art lacks consideration of component changes and dynamic changes in drug compatibility in traditional Chinese medicine, resulting in insufficient accuracy of compatibility prediction.
By obtaining the properties and taste of traditional Chinese medicine, the properties of traditional Chinese medicine, the efficacy data and the component correlation matrix are constructed, and the similarity matrix and feature matrix are obtained, the characteristic vectors of traditional Chinese medicine ingredients are fused to generate the characteristic vectors of medicine pairs, and the compatibility taboos are predicted. sex.
The accuracy of traditional Chinese medicine prediction of compatibility taboos is improved, and the ability to analyze and model Chinese medicine ingredient data is enhanced by adaptively adjusting the component feature weight and dynamic representation.
Smart Images

Figure CN119008036A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of drug compatibility, and in particular to a method and system for predicting incompatibility of traditional Chinese medicine pairs. Background Art
[0002] The compatibility of Chinese medicine involves multiple factors and dynamic changes. First of all, a single medicinal material usually has multiple components, and each component may play a different role in the pharmacological action. Changes in the types and proportions of these components often significantly change the therapeutic efficacy and toxicity of the medicinal materials. For example, a certain medicinal material may contain a high concentration of a certain active ingredient in a specific growth environment, while the content of this ingredient may be lower in other environments, so that the same medicinal material may have different therapeutic effects and safety in different situations. In the related art, the description of traditional Chinese medicine mainly uses descriptive data such as nature, taste, and meridians, and lacks consideration of ingredients, which affects the accuracy of the prediction.
[0003] Secondly, the same herb may play different roles in different Chinese medicine combinations. For example, astragalus may mainly play a role in replenishing qi in some combinations, while in other combinations it may have anti-inflammatory or diuretic functions. This difference depends not only on the composition of the herb itself, but also on the interaction with other compatible herbs. The relevant technology lacks consideration of the dynamic changes in drug combinations, which affects the accuracy of the prediction. Summary of the invention
[0004] In order to solve at least one of the deficiencies of the prior art, the object of the present invention is to provide a method and system for predicting incompatibility of Chinese medicine pairs, so as to improve the accuracy of predicting incompatibility of Chinese medicine pairs.
[0005] In order to achieve the above object, according to some embodiments, the first aspect of the present invention provides a method for predicting incompatibility of Chinese medicine pairs, comprising:
[0006] Obtain the data on the properties and meridians of traditional Chinese medicine, the data on the efficacy of traditional Chinese medicine, the association matrix between traditional Chinese medicine and its ingredients, and the characteristic data on the ingredients of traditional Chinese medicine;
[0007] Based on the data of TCM properties and meridians and the data of TCM efficacy, similarity matrices of properties and meridians and efficacy attributes were constructed respectively.
[0008] Based on the association matrix between Chinese medicine and ingredients and the characteristic data of Chinese medicine ingredients, the characteristic matrix of Chinese medicine ingredients is obtained through attention aggregation;
[0009] The similarity matrix of properties and meridians, the similarity matrix of efficacy properties and the characteristic matrix of Chinese medicinal ingredients are integrated to obtain the characteristic vector of each Chinese medicinal material.
[0010] The Chinese medicine feature vectors of the Chinese medicine pair to be predicted are integrated to obtain the medicine pair feature vector;
[0011] According to the obtained drug pair feature vector, predict the possibility of incompatibility of Chinese medicine pairs;
[0012] Among them, the Chinese medicine ingredient feature matrix is obtained through attention aggregation, including: obtaining the ingredient feature vector of each ingredient based on the Chinese medicine ingredient feature data; calculating the attention weight corresponding to the ingredient according to the ingredient feature vector; aggregating the ingredient feature vectors according to the attention weight to obtain the Chinese medicine ingredient feature vector and the Chinese medicine ingredient feature matrix.
[0013] The second aspect of the present invention provides a Chinese medicine pair incompatibility prediction system, comprising:
[0014] A data acquisition module is configured to acquire the property data of Chinese medicine nature and flavor, the meridians of Chinese medicine, the efficacy data of Chinese medicine, the association matrix between Chinese medicine and ingredients, and the characteristic data of Chinese medicine ingredients;
[0015] A similarity matrix construction module is configured to construct a similarity matrix of properties and meridians and a similarity matrix of efficacy properties based on the properties and meridians attribute data of traditional Chinese medicine and the efficacy data of traditional Chinese medicine respectively;
[0016] The feature aggregation module is configured to obtain a Chinese medicine component feature matrix through attention aggregation based on the Chinese medicine and component association matrix and the Chinese medicine component feature data; wherein obtaining the Chinese medicine component feature matrix through attention aggregation includes: obtaining a component feature vector of each component based on the Chinese medicine component feature data; calculating an attention weight corresponding to the component according to the component feature vector; and aggregating each component feature vector according to the attention weight to obtain a Chinese medicine component feature vector and a Chinese medicine component feature matrix;
[0017] The first feature fusion module is configured to fuse the similarity matrix of the properties and meridians, the similarity matrix of the efficacy properties and the feature matrix of the Chinese medicinal ingredients to obtain the Chinese medicinal feature vector of each Chinese medicinal material;
[0018] The second feature fusion module is configured to fuse the Chinese medicine feature vectors of the Chinese medicine pair to be predicted to obtain a medicine pair feature vector;
[0019] The prediction module is configured to predict the possibility of incompatibility of Chinese medicine pairs according to the obtained drug pair feature vectors.
[0020] Compared with the prior art, the present invention has the following beneficial effects:
[0021] The present invention provides a method and system for predicting the incompatibility of Chinese medicine pairs. Based on the attention aggregation of the characteristics of each component of Chinese medicine, different attention weights can be adaptively assigned to each component according to the type and proportion of the components of the Chinese medicine, so as to obtain the characteristic vector of the Chinese medicine components through aggregation for predicting the incompatibility of Chinese medicine pairs. The method and system realize the effective analysis and modeling of complex Chinese medicine component data, and can significantly improve the recognition and understanding of the components in single medicinal materials and compatible medicinal materials, thereby improving the accuracy of predicting the incompatibility of Chinese medicine pairs.
[0022] The present invention introduces the idea of multi-view learning, regards data from different sources as a view, and designs a multi-view learning loss to enhance the information representation between data from different sources and reduce the information redundancy between data from different sources.
[0023] When fusing the traditional Chinese medicine feature vectors, the present invention uses a shared layer to fuse the traditional Chinese medicine feature vectors to obtain a medicine pair vector, and then uses the medicine pair vector to dynamically adjust the traditional Chinese medicine feature vector to obtain the adaptively adjusted feature vector representation of the same traditional Chinese medicine in different traditional Chinese medicine pairs, thereby realizing the dynamic adjustment of the information flow of the traditional Chinese medicine pair and improving the accuracy of the subsequent prediction of the incompatibility of the traditional Chinese medicine pair.
[0024] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0026] Figure 1 The present invention provides a flowchart of a method for predicting incompatibility of traditional Chinese medicine pairs. DETAILED DESCRIPTION
[0027] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0028] Embodiment 1
[0029] Embodiment 1 of the present invention provides a method for predicting incompatibility of Chinese medicine pairs. Figure 1 As shown, including:
[0030] S1. Obtaining the data of properties and meridians of traditional Chinese medicine, the data of efficacy of traditional Chinese medicine, the association matrix between traditional Chinese medicine and ingredients, and the characteristic data of traditional Chinese medicine ingredients;
[0031] S2. Based on the data of properties and meridians of traditional Chinese medicine and the data of efficacy of traditional Chinese medicine, a similarity matrix of properties and meridians and a similarity matrix of efficacy properties are constructed respectively;
[0032] S3. Based on the Chinese medicine and ingredient association matrix and the Chinese medicine ingredient characteristic data, a Chinese medicine ingredient characteristic matrix is obtained by attention aggregation; including: based on the Chinese medicine ingredient characteristic data, a component characteristic vector of each component is obtained; according to the component characteristic vector, an attention weight corresponding to the component is calculated; according to the attention weight, each component characteristic vector is aggregated to obtain a Chinese medicine ingredient characteristic vector and a Chinese medicine ingredient characteristic matrix;
[0033] S4, integrating the similarity matrix of properties and meridians, the similarity matrix of efficacy properties and the characteristic matrix of Chinese medicinal ingredients to obtain the characteristic vector of each Chinese medicinal material;
[0034] S5, fusing the Chinese medicine feature vectors of the Chinese medicine pair to be predicted to obtain a medicine pair feature vector;
[0035] S6. Predict the possibility of incompatibility of Chinese medicine pairs based on the obtained drug pair feature vectors.
[0036] In step S1, firstly, the property data of Chinese medicine nature and flavor, Chinese medicine efficacy data, Chinese medicine and ingredient association matrix and Chinese medicine ingredient characteristic data are obtained.
[0037] The property and flavor attribute data refers to the properties and taste of Chinese medicine, which are obtained from the theory of Chinese medicine properties; the meridian attribute data refers to the selective effect of Chinese medicine on a certain part of the body. Each Chinese medicine has a property and flavor meridian attribute vector with a vector length of 44. When the Chinese medicine has a certain property and flavor / meridian, the corresponding element of the property and flavor meridian attribute vector is 1, otherwise it is 0.
[0038] The efficacy data of traditional Chinese medicine refers to the properties and functions related to the therapeutic effects of traditional Chinese medicine. Each traditional Chinese medicine has an efficacy vector, and the length of the vector is the number of the efficacy of the traditional Chinese medicine. When the traditional Chinese medicine has a certain efficacy, the corresponding element of the efficacy vector is 1, otherwise it is 0.
[0039] The association matrix between Chinese medicine and ingredients refers to the matrix formed by the association between Chinese medicine and the ingredients it contains. The number of rows in the matrix is the number of all Chinese medicines, and the number of columns in the matrix is the number of all ingredients. When a Chinese medicine contains a certain ingredient, the corresponding row and column elements of the association matrix between Chinese medicine and ingredients are 1, otherwise they are 0.
[0040] Component feature data refers to the Morgan fingerprint feature of the component; specifically, this feature is the molecular substructure feature obtained after processing the SMILES data of the component through the RDKit toolkit, that is, the Morgan fingerprint feature. Each component feature can be represented as a 512-dimensional component feature vector, and the elements of the vector are the number of corresponding molecular substructures in the component.
[0041] In step S2, a similarity matrix is constructed based on the Chinese medicine property, flavor and meridian attribute data and the Chinese medicine efficacy data obtained in step S1. For multiple Chinese medicines to be studied, a property, flavor and meridian attribute matrix is obtained according to the property, flavor and meridian attribute vectors of each Chinese medicine, represented by P; similarly, an efficacy matrix is obtained according to the efficacy vectors of each Chinese medicine, represented by E. Using Jaccard similarity, the property, flavor and meridian attribute matrix and the efficacy matrix are processed separately to obtain a property, flavor and meridian attribute similarity matrix and an efficacy attribute similarity matrix.
[0042] Based on Jaccard similarity, construct the similarity matrix X of nature, flavor and meridian attributes p :
[0043]
[0044] Among them, Sim jac (·) indicates the calculation of Jaccard similarity, P indicates the property matrix of nature, flavor and meridians, P i and P j They respectively represent the property, taste and meridian attribute vector of the i-th Chinese medicinal material and the property, taste and meridian attribute vector of the j-th Chinese medicinal material.
[0045] Construct the efficacy attribute similarity matrix X e :
[0046]
[0047] Where E represents the efficacy matrix, E i and E j They respectively represent the efficacy vector of the i-th Chinese medicinal material and the efficacy vector of the j-th Chinese medicinal material.
[0048] In step S3, for Chinese medicinal materials, a single medicinal material is composed of multiple components, and the changes in the types and proportions of the components cause considerable complexity. Such changes can cause the Chinese medicine formula to exhibit different pharmacological properties. Taking into account the different meanings of different components of Chinese medicine and the collective interaction of multiple components, the embodiment of the present invention uses an attention-based aggregation method to aggregate multiple component features in a Chinese medicinal material into a Chinese medicine component feature vector when aggregating multiple component feature vectors of Chinese medicine.
[0049] Specifically, for M i For the i-th Chinese herbal medicine with a component, we first search the i-th row according to the association matrix between Chinese medicine and components to obtain the index of the corresponding component of the i-th Chinese herbal medicine, and then use the attention-based aggregation method to aggregate M i The component feature vectors are aggregated to obtain the Chinese medicinal component feature vector of the i-th Chinese medicinal material
[0050]
[0051] Among them, t m,i is the component feature vector of the mth component of the i-th Chinese medicinal material, and the attention weight a m,i is the weight of the mth component of the ith Chinese medicinal material, expressed as follows:
[0052]
[0053] Among them, tanh(·) is the tanh activation function, sigm(·) is the sigmoid activation function, w, W v , W u is a learnable parameter matrix. The Chinese medicinal ingredient features of each Chinese medicinal material constitute the Chinese medicinal ingredient feature matrix X t .
[0054] In step S4, based on multi-view data enhancement, the fusion of data from different sources is regarded as a multi-view learning task, and a multi-view learning loss function is designed to reduce the information redundancy between multiple view data and enhance the feature representation of Chinese medicinal materials in different views. For a Chinese medicinal material, its nature, flavor, meridian attributes, efficacy and composition are regarded as data from different data sources. In the multi-view learning task, data from different data sources can be regarded as a view, and different views contain different information because they capture different perspectives. This embodiment introduces the idea of multi-view learning, thereby enhancing the information representation between data from different sources and reducing the information redundancy between data from different sources.
[0055] Specifically, the similarity matrix X of the properties and meridians obtained in step S2 p , efficacy attribute similarity matrix X e and the Chinese medicine ingredient feature matrix X obtained in step S3 t , each row of the matrix represents a feature vector of the corresponding data source of traditional Chinese medicine. A set of independent autoencoders are used to project the data into the same latent space and reduce the dimension to the same dimension to complete the data alignment.
[0056] Each set of autoencoders consists of an encoder and a decoder They are used to obtain the potential feature vector of the i-th Chinese medicinal material and the resulting reconstruction vector
[0057]
[0058] Among them, s∈S={e,p,t} represents the data source, Represents the original feature vector of the sth data source of the ith Chinese medicinal material. For example, when s=t, it represents the Chinese medicinal ingredient feature vector of the ith Chinese medicinal material. is the learnable weight matrix.
[0059] The loss function L of the autoencoder rec Reconstruction loss is used to enhance the representation ability of feature vectors. The loss function is the mean square error between the original data vector and the reconstructed vector:
[0060]
[0061] Among them, MSE(·) is the mean square error, N is the total number of Chinese medicinal materials, is the original feature vector of the sth data source of the ith Chinese medicinal material, is the reconstructed vector of the sth data source of the ith Chinese medicinal material.
[0062] Based on multi-view learning, the latent feature vector of Chinese medicinal materials obtained by the encoder is enhanced. The loss function of dual view Designed for:
[0063]
[0064] Among them, N represents the total number of Chinese medicinal materials. Indicates the sth of the i-th Chinese medicinal material 1 The potential feature vector of the data source and the sth 2 The joint distribution of latent feature vectors from the data sources, Indicates the sth of the i-th Chinese medicinal material 1 The potential feature vector of the data source and the sth feature vector of the jth Chinese medicinal material 2 Whether the potential feature vectors of the data sources come from the same Chinese medicinal material:
[0065]
[0066] Among them, exp is an exponential function with a natural constant as the base, Respectively represent the sth 1 The potential feature vector of the data source and the sth 2 The potential feature vectors of the data sources. Since the number of data sources is greater than 2, that is, the total number of views is greater than 2, the multi-view learning loss function L c It is expressed as the sum of the two-view learning losses between any two combinations of views:
[0067]
[0068] Here, |S| represents the number of data sources.
[0069] The potential feature vectors of the three data sources after fusion enhancement are combined into a TCM feature vector by splicing:
[0070]
[0071] Among them, d i represents the Chinese medicinal feature vector of the i-th Chinese medicinal material, is the potential feature vector of the efficacy attribute of the i-th Chinese medicinal material, is the potential feature vector of the nature, flavor and meridian attributes of the i-th Chinese medicinal material, is the potential feature vector of the components of the i-th Chinese medicinal material.
[0072] In step S5, since the same Chinese medicine may have different effects in different Chinese medicine pairs, it is necessary to dynamically adjust the Chinese medicine feature vectors of various Chinese medicine combinations. This embodiment uses a dynamic representation layer, which includes a gated attention unit GAU, which is used to adjust the information flow of each Chinese medicine feature vector in the Chinese medicine pair.
[0073] Specifically, suppose the Chinese medicine pair a and b to be predicted include Chinese medicine a and Chinese medicine b, and the Chinese medicine feature vectors of Chinese medicine a and Chinese medicine b are d a and d b First, use a shared layer Sharedlayer to fuse d a and d b , get the initial fusion vector d of the drug pair p :
[0074] d p =SharedLayer(d a ,d b )=W k ((W l d a )||(W l d b ));
[0075] Among them, W k and w l is the learnable weight matrix of the shared layer, and || is the vector concatenation operation.
[0076] Then, the initial fusion vector of the drug pair is used to dynamically optimize the TCM feature vectors of drug pair a and b.
[0077] Adjustment. Use the gated attention unit GAU to initially fuse the drug pair vector d pThe Chinese medicine feature vector of Chinese medicine a is input into the first gated attention unit GAU 1 , the drug pair is initially fused into vector d p The Chinese medicine feature vector of Chinese medicine b is input into the second gated attention unit GAU 2 , use the drug pair to initially fuse the vector d p For the Chinese medicine feature vector d a and d b Dynamic adjustments are made respectively to obtain the adaptively adjusted feature vector representation of the same Chinese medicine in different Chinese medicine pairs:
[0078] z a =GAU 1 (d a ,d p )=(W a d a )⊙(m a ⊙(w p d p ));
[0079] z b =GAU 2 (d b ,d p )=(W b d b )⊙(m b ⊙(w p d p ));
[0080] Among them, W a , W b and w p is the learnable weight matrix, m a 、m b They are the first gated attention unit GAU 1 and the second gated attention unit GAU 2 The attention weight vector, z a is the adaptively adjusted feature vector of Chinese medicinal material a, z b is the adaptively adjusted feature vector of Chinese medicinal material b.
[0081] The attention weight vector m of the first gated attention unit and the second gated attention unit a 、m b They are respectively expressed as follows:
[0082]
[0083] Among them, d k Yes p d p The dimension, W t,a, W s,a and W t,b , W s,b They are the learning weight matrices of the attention matrices of Chinese medicinal materials a and b respectively.
[0084] Finally, the adaptively adjusted TCM feature vectors are fused to obtain the drug pair feature vectors. a 、z b , a new shared layer is used for concatenation and fusion, and the obtained drug pair feature vector is expressed as follows:
[0085] z p =SharedLayer 2 (z a ,z b )=W k2 ((W l2 z a )||(W l2 z b ));
[0086] Among them, z p is the drug pair feature vector of drug pair a and b, W k2 , W l2 For z a and z b Shared learnable weight matrix.
[0087] In step S6, the incompatibility of the Chinese medicine pair is evaluated. In step S5, the drug pair feature vector of drug pair a and b is obtained through adaptive adjustment. In order to identify the incompatibility of the drug pair, a two-layer multi-layer perceptron is used to predict the possibility of the incompatibility of the Chinese medicine pair:
[0088]
[0089] in, It indicates the probability that the drug pair predicted by the model is incompatible, and its value range is (0,1). The closer it is to 1, the more likely the drug pair is to become incompatible, and vice versa, the less likely it is to become incompatible. Relu is the activation function of the multi-layer perceptron, W d1 , W d2 are the learnable weight matrices of the first and second layers of the multilayer perceptron, b d1 , b d2 They are the learnable bias vectors of the first and second layers of the multilayer perceptron respectively.
[0090] Finally, we introduce a binary classification task loss function for the imbalance of positive and negative samples to constrain the prediction results not to deviate from the true label and balance the problem of excessive sample difficulty in the binary classification task, thus achieving efficient and accurate evaluation and optimization of the incompatibility of Chinese medicine pairs. The binary classification task loss function L cls It is expressed as follows:
[0091]
[0092] Among them, α and γ are hyperparameters, which are used to balance the classification difficulty of positive and negative samples and difficult and easy samples respectively, and y represents the true label of the drug incompatibility.
[0093] The overall loss function L total for:
[0094] L total =L cls +λ 1 L C +λ 2 L rec
[0095] where λ 1 , 2 It is the balance factor among the three loss functions and is used to adjust the weights of the three loss functions.
[0096] Embodiment 2
[0097] This embodiment provides a Chinese medicine pair incompatibility prediction system, including:
[0098] A data acquisition module is configured to acquire the property data of Chinese medicine nature and flavor, the meridians of Chinese medicine, the efficacy data of Chinese medicine, the association matrix between Chinese medicine and ingredients, and the characteristic data of Chinese medicine ingredients;
[0099] A similarity matrix construction module is configured to construct a similarity matrix of properties and meridians and a similarity matrix of efficacy properties based on the properties and meridians attribute data of traditional Chinese medicine and the efficacy data of traditional Chinese medicine respectively;
[0100] The feature aggregation module is configured to obtain a Chinese medicine component feature vector through attention aggregation based on the Chinese medicine and component association matrix and the Chinese medicine component feature data; wherein obtaining the Chinese medicine component feature vector through attention aggregation includes: obtaining a component feature vector of each component based on the Chinese medicine component feature data; calculating an attention weight corresponding to the component according to the component feature vector; and obtaining the Chinese medicine component feature vector by aggregating each component feature vector according to the attention weight;
[0101] The first feature fusion module is configured to fuse the similarity matrix of the property and meridian attributes, the similarity matrix of the efficacy attributes and the feature vector of the Chinese medicine components to obtain the feature vector of the Chinese medicine;
[0102] The second feature fusion module is configured to fuse the Chinese medicine feature vectors of the Chinese medicine pair to be predicted to obtain a medicine pair feature vector;
[0103] The prediction module is configured to predict the possibility of incompatibility of Chinese medicine pairs according to the obtained drug pair feature vectors.
[0104] It should be noted here that the various modules in this embodiment correspond one-to-one to the steps of the method in Example 1, and the specific implementation process is the same, which will not be repeated here.
[0105] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for predicting incompatibility of Chinese medicine pairs, characterized in that: include: Obtain the data on the properties and meridians of traditional Chinese medicine, the data on the efficacy of traditional Chinese medicine, the association matrix between traditional Chinese medicine and its ingredients, and the characteristic data on the ingredients of traditional Chinese medicine; Based on the data of TCM properties and meridians and the data of TCM efficacy, similarity matrices of properties and meridians and efficacy attributes were constructed respectively. Based on the association matrix between Chinese medicine and ingredients and the characteristic data of Chinese medicine ingredients, the characteristic matrix of Chinese medicine ingredients is obtained through attention aggregation; The similarity matrix of properties and meridians, the similarity matrix of efficacy properties and the characteristic matrix of Chinese medicinal ingredients are integrated to obtain the characteristic vector of each Chinese medicinal material. The Chinese medicine feature vectors of the Chinese medicine pair to be predicted are integrated to obtain the medicine pair feature vector; According to the obtained drug pair feature vector, predict the possibility of incompatibility of Chinese medicine pairs; Among them, the Chinese medicine ingredient feature matrix is obtained through attention aggregation, including: obtaining the ingredient feature vector of each ingredient based on the Chinese medicine ingredient feature data; calculating the attention weight corresponding to the ingredient according to the ingredient feature vector; aggregating the ingredient feature vectors according to the attention weight to obtain the Chinese medicine ingredient feature vector and the Chinese medicine ingredient feature matrix.
2. A method for predicting incompatibility of Chinese medicine pairs as claimed in claim 1, characterized in that: The calculation formula of the attention weight is: Among them, a m,i represents the attention weight of the mth component of the ith Chinese herbal medicine, tanh(·) is the tanh activation function, sigm(·) is the sigmoid activation function, w, W v , W u is the learnable parameter matrix, t m,i is the component feature vector of the mth component of the ith Chinese medicinal material.
3. A method for predicting incompatibility of Chinese medicine pairs as claimed in claim 1, characterized in that: The fusion of the property and flavor meridian attribute similarity matrix, the efficacy attribute similarity matrix and the Chinese medicine component feature vector to obtain the Chinese medicine feature vector includes: Obtain the original feature vectors of each data source of each Chinese medicinal material in the similarity matrix of properties and meridians, the similarity matrix of efficacy properties and the feature matrix of Chinese medicinal ingredients, and calculate the corresponding potential feature vectors; Enhance latent feature vectors based on multi-view learning; The enhanced latent feature vectors of each data source are fused to obtain the TCM feature vector.
4. A method for predicting incompatibility of Chinese medicine pairs as claimed in claim 3, characterized in that: Loss function L for multi-view learning c for: in, is the loss function of the dual view, represents the joint distribution of the latent feature vector of the s1th data source of the i-th Chinese medicinal material and the latent feature vector of the s2th data source of the j-th Chinese medicinal material, Indicates whether the latent feature vector of the s1th data source of the i-th Chinese medicinal material and the latent feature vector of the s2th data source of the j-th Chinese medicinal material come from the same Chinese medicinal material.
5. A method for predicting incompatibility of Chinese medicine pairs as claimed in claim 1, characterized in that: The step of fusing the Chinese medicine feature vectors of the Chinese medicine pair to be predicted to obtain the medicine pair feature vector includes: The Chinese medicine feature vectors of each Chinese medicine in the Chinese medicine pair to be predicted are integrated to obtain a preliminary integration vector of the medicine pair; Using the preliminary fusion vector of the medicine pair, dynamically optimizing and adjusting the Chinese medicine feature vector of each Chinese medicinal material in the medicine pair, and obtaining the Chinese medicine feature vector of each Chinese medicinal material after adaptive adjustment; The adaptively adjusted TCM feature vectors are fused to obtain the drug pair feature vectors.
6. A method for predicting incompatibility of Chinese medicine pairs as claimed in claim 5, characterized in that: The calculation formula for dynamically optimizing and adjusting the Chinese medicinal material characteristic vectors of each Chinese medicinal material in the medicine pair is: z a =GAU1(d a ,d b )=(W a d a )⊙(m a ⊙(w p d p )); z b =GAU2(d b ,d b )=(W b d b )⊙(m b ⊙(W p d p )); Among them, W a , W b and w p is the learnable weight matrix, m a 、m b are the attention weight vectors of the first gated attention unit GAU1 and the second gated attention unit GAU2, respectively, a is the adaptively adjusted feature vector of Chinese medicinal material a, z b is the adaptively adjusted feature vector of Chinese medicinal material b, d p It is the initial fusion vector of drug pair.
7. A method for predicting incompatibility of Chinese medicine pairs as claimed in claim 6, characterized in that: The attention weight vector m of the first gated attention unit a and the attention weight vector m of the second gated attention unit b Respectively expressed as: Among them, d k Yes p d p The dimension, W t,a , W s,a is the learning weight matrix of the attention matrix of Chinese medicinal material a; W t,b , W s,b It is the learning weight matrix of the attention matrix of Chinese medicinal material b.
8. A method for predicting incompatibility of Chinese medicine pairs as claimed in claim 1, characterized in that: A multi-layer perceptron is used to predict the possibility of incompatibility of traditional Chinese medicine pairs.
9. A method for predicting incompatibility of Chinese medicine pairs as claimed in claim 1, characterized in that: Based on Jaccard similarity, similarity matrices of properties and flavors and meridians and efficacy properties were constructed respectively.
10. A Chinese medicine pair incompatibility prediction system, characterized in that: include: A data acquisition module is configured to acquire the property data of Chinese medicine nature and flavor, the meridians of Chinese medicine, the efficacy data of Chinese medicine, the association matrix between Chinese medicine and ingredients, and the characteristic data of Chinese medicine ingredients; A similarity matrix construction module is configured to construct a similarity matrix of properties and meridians and a similarity matrix of efficacy properties based on the properties and meridians attribute data of traditional Chinese medicine and the efficacy data of traditional Chinese medicine respectively; The feature aggregation module is configured to obtain a Chinese medicine component feature matrix through attention aggregation based on the Chinese medicine and component association matrix and the Chinese medicine component feature data; wherein obtaining the Chinese medicine component feature matrix through attention aggregation includes: obtaining a component feature vector of each component based on the Chinese medicine component feature data; calculating an attention weight corresponding to the component according to the component feature vector; and aggregating each component feature vector according to the attention weight to obtain a Chinese medicine component feature vector and a Chinese medicine component feature matrix; The first feature fusion module is configured to fuse the similarity matrix of the properties and meridians, the similarity matrix of the efficacy properties and the feature matrix of the Chinese medicinal ingredients to obtain the Chinese medicinal feature vector of each Chinese medicinal material; The second feature fusion module is configured to fuse the Chinese medicine feature vectors of the Chinese medicine pair to be predicted to obtain a medicine pair feature vector; The prediction module is configured to predict the possibility of incompatibility of Chinese medicine pairs according to the obtained drug pair feature vectors.
Citation Information
Patent Citations
Traditional Chinese medicine toxicity qualitative evaluation method and system thereof
CN103049673A
Traditional Chinese medicine and western medicine combined medication contraindication early-warning system
CN104794341A
Chinese medicine incompatibility prediction method based on supervision learning framework
CN110619960A
Prediction method and device for drug interaction, equipment and storage medium
CN114694791A
Drug-drug interaction prediction method and system based on multiple views
CN117079835A
Cited By
Graph convolutional network-based traditional Chinese medicine prescription drug attribute classification method
CN120951099A
Traditional Chinese medicinal material attribute feature coding and retrieval system based on machine learning
CN121725997A
Traditional Chinese medicine property feature coding and retrieval system based on machine learning
CN121725997B
