Drug interaction prediction method and system based on multi-source data
Through the integration and network fusion algorithm of multi-source data, combined with network diffusion and classic machine learning algorithms, the Stacking integrated learning strategy is integrated with the LightGBM algorithm, and the problems of unused inter-feature correlation and limitations of a single algorithm in drug interaction prediction are solved, significantly improving prediction accuracy and performance.
Patent Information
- Application Number
- CN202411886342.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-20
- Publication Date
- 2025-04-11
AI Technical Summary
Existing machine learning-based drug interaction prediction methods fail to fully consider the correlation between drug characteristics, and the utilization of a single machine learning algorithm limits the utilization rate of drug characteristic information and affects the prediction ability.
Multi-source data is used to construct the drug feature matrix and interaction matrix, and a multi-source feature similarity network and multi-feature matrix are generated through network fusion algorithms and feature splicing methods. Preliminary prediction is performed by combining network diffusion algorithms and classic machine learning algorithms. Finally, the Stacking integrated learning strategy is integrated with the LightGBM algorithm to output the final prediction results of drug interactions.
By fully leveraging the information between different characteristics of the drug and the advantages of multiple machine learning algorithms, the accuracy and performance of drug interaction prediction are improved and the utilization rate of feature information is improved.
Smart Images
Figure CN119811695A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of drug interaction prediction, and in particular relates to a drug interaction prediction method and system based on multi-source data. Background Art
[0002] Using multiple drugs together to treat a disease is a common treatment method, but taking two or more drugs at the same time or over a period of time may cause drug interactions, leading to adverse drug reactions, which in turn endanger health and may even cause death in severe cases.
[0003] Early drug interactions were mainly verified through in vivo and in vitro experiments. This experimental method is not only very time-consuming but also very expensive. With the advancement of computer technology, more and more computational methods based on machine learning are being used to predict potential drug interactions. At present, drug interaction prediction methods based on machine learning methods have some limitations. Some drug interaction prediction methods based on multiple drug characteristics do not consider the correlation between different drug characteristics, so they cannot fully learn the information between different drug characteristics, which affects the ability to predict drug interactions. In addition, different machine learning algorithms use drug characteristics in different ways. Predicting drug interactions based on only one machine learning method limits the utilization of drug characteristic information to a certain extent, thereby affecting the ability to predict drug interactions. Summary of the invention
[0004] The present invention aims to solve the problems of insufficient information learning between different characteristics of drugs in the process of drug interaction prediction and the limitations of prediction based on a single machine learning algorithm.
[0005] In one aspect, the present invention provides a method for predicting drug interactions based on multi-source data, comprising the following steps:
[0006] According to the standard names of marketed drugs in DrugBank, various characteristic information of drugs is obtained from databases such as DrugBank, PubChem, and KEGG, and a drug characteristic matrix is constructed with rows as drugs and columns as drug characteristic information;
[0007] Drug interaction information was obtained from TWOSIDES, and the drug interaction matrix was constructed with rows and columns as drugs;
[0008] Calculate the similarity between each pair of drugs in each drug feature matrix, so that each feature matrix generates a drug multi-source feature similarity network;
[0009] The generated drug multi-source feature similarity network is fused using a network fusion algorithm to obtain a drug feature fusion network;
[0010] The generated drug multi-source feature similarity network is spliced using the feature splicing method to obtain the drug multi-feature matrix;
[0011] The drug feature fusion network and drug interaction matrix were input into the network diffusion algorithm to obtain the preliminary prediction results of drug interactions;
[0012] The drug multi-feature matrix is input into the classical machine learning algorithm to obtain the preliminary prediction results of drug interactions;
[0013] Based on the Stacking ensemble learning strategy, LightGBM is used as the final predictor to integrate the base learners to output the final prediction results of drug interactions. A drug interaction prediction model is constructed with the network diffusion algorithm and the classical machine learning algorithm as the base predictors and LightGBM as the final predictor.
[0014] Furthermore, the above-mentioned drug characteristic information includes six types of drug property information and two types of drug side effect information.
[0015] Furthermore, the similarities between drugs in each drug feature matrix are calculated separately, so that each feature matrix generates a drug multi-source feature similarity network, including:
[0016] The Jaccard similarity calculation method was used to calculate the similarity between drugs in each drug feature matrix, and a multi-source feature similarity network of 8 drugs was generated.
[0017] Furthermore, the above-mentioned network fusion algorithm is used to fuse the generated drug multi-source feature similarity network and the feature splicing method is used to splice the generated drug multi-source feature similarity network to obtain a drug feature fusion network and a drug multi-feature matrix, respectively, including:
[0018] The fusion algorithm based on the Pearson coefficient is used to fuse the 8 drug multi-source feature similarity networks to generate a drug feature fusion network; the 8 drug multi-source feature similarity networks are converted into column matrices of 8 drug features, and the matrices are merged using a column splicing method to generate a drug multi-feature matrix.
[0019] Furthermore, the above network diffusion algorithm is a label propagation algorithm and a neighbor recommendation algorithm, and the above classic machine learning algorithm is a random forest algorithm.
[0020] Furthermore, the label propagation algorithm and the neighbor recommendation algorithm were used to obtain preliminary prediction results of two drug interactions based on the drug feature fusion network and the drug interaction matrix, and the random forest algorithm was used to obtain preliminary prediction results of the third drug interaction based on the drug multi-feature matrix.
[0021] Furthermore, based on the Stacking ensemble learning strategy, LightGBM is used as the final predictor to integrate the base learners to output the final results of drug interactions, including:
[0022] The Stacking ensemble learning strategy was used to convert the preliminary prediction results of the above three drug interactions into column matrices, which were concatenated and input into the LightGBM predictor to obtain the final prediction results of drug interactions.
[0023] The beneficial effects of the present invention are as follows:
[0024] Based on the fused feature information of multiple drugs, a drug interaction prediction framework based on ensemble learning is constructed to improve the performance of drug interaction prediction. Compared with directly using feature information of multiple drugs, the prediction method described in the present invention adds a network fusion method to the feature processing part to make more effective use of the information between different drug features. Compared with the prediction algorithm using a machine learning algorithm, the prediction method described in the present invention integrates different machine learning algorithms through Stacking ensemble learning in the prediction stage, thereby improving the utilization rate of drug features and thus improving the accuracy of drug interaction prediction results.
[0025] On the other hand, based on the same inventive concept, the present application also provides a drug interaction prediction system based on multi-source data, comprising:
[0026] A data acquisition module, used to acquire multi-source drug feature data;
[0027] A data processing module is used to construct and fuse the drug multi-source feature similarity network to obtain a drug multi-feature fusion network and a drug multi-feature matrix;
[0028] The drug interaction prediction module is used to input the drug feature fusion network and the drug multi-feature matrix into the base learners based on the network diffusion algorithm and the classical machine learning algorithm respectively to make a preliminary prediction of drug interaction, and integrate the preliminary prediction results of drug interaction based on ensemble learning to obtain the final prediction results of drug interaction. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] The accompanying drawings are only used to illustrate specific embodiments and are not to be considered as limiting the present invention. In the accompanying drawings:
[0030] Figure 1 A flowchart of a method for predicting drug interactions based on multi-source data according to an embodiment of the present invention;
[0031] Figure 2 The block diagram of a drug interaction prediction system based on multi-source data described in an embodiment of the present invention. DETAILED DESCRIPTION
[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention, wherein the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0033] Reference Figure 1 Specifically describing this embodiment, a drug interaction prediction method based on multi-source data described in this embodiment includes:
[0034] Step S1: obtaining various characteristic information of the drug, the specific steps are as follows:
[0035] The dataset constructed in this example contains 548 drugs. The standard names of the drugs are obtained from DrugBank. Drug interaction information is obtained from the TWOSIDES dataset based on the standard names. Drug attribute information is obtained from DrugBank, PubChem, and KEGG. Drug side effect information is obtained from the SIDER side effect dataset and the OFFSIDER side effect dataset.
[0036] Among them, drug interaction information of 548 drugs was obtained from the TWOSIDES dataset, and a matrix with drugs as rows and drugs as columns was constructed to obtain a drug interaction matrix with a dimension of 548×548. Chemical structure information of 881 drugs was obtained from PubChem, 780 drug target information, 78 drug transporter information, 129 drug enzyme information, 4897 drug indication information were obtained from DrugBank, and 253 drug pathway information were obtained from KEGG. Side effect information of 4897 drugs was obtained from the SIDER side effect dataset, and Off side effect information of 9496 drugs was obtained from the OFFSIDES side effect dataset.
[0037] For drug attribute information, taking the drug chemical structure characteristics as an example, the chemical structure characteristics have 881 chemical substructures, and a matrix with drugs as rows and chemical substructures as columns is constructed to obtain a chemical structure feature matrix with a dimension of 548 × 881. Repeat the above operation to obtain the remaining 5 drug attribute matrices, namely: drug target feature matrix with a dimension of 548 × 780, drug transporter feature matrix with a dimension of 548 × 78, drug enzyme feature matrix with a dimension of 548 × 129, drug indication feature matrix with a dimension of 548 × 4897, and drug pathway feature matrix with a dimension of 548 × 253.
[0038] For drug side effect information, take SIDER side effect features as an example. There are 4897 side effects in the SIDER side effect dataset. A matrix with drugs as rows and side effects as columns is constructed to obtain a SIDER side effect feature matrix with a dimension of 548 × 4897. Repeat the above operation to obtain the OFFSIDES side effect feature matrix with a dimension of 548 × 9496.
[0039] Step S2: Calculate the drug feature similarity based on the similarity algorithm and construct a drug multi-source feature similarity network. The specific steps are as follows:
[0040] The Jaccard similarity calculation method was used to calculate the drug similarity of 8 drug feature matrices, and 8 548×548-dimensional drug multi-source feature similarity networks were generated. The 8 drug multi-source feature similarity networks were: drug chemical structure similarity network, drug target similarity network, drug transporter similarity network, drug enzyme similarity network, drug indication similarity network, drug pathway similarity network, SIDER side effect similarity network, and OFFSIDES side effect similarity network. The Jaccard calculation formula is:
[0041]
[0042] Among them, P and Q are the feature sets of two drugs in the same drug feature matrix.
[0043] The similarity calculation method is defined as: the ratio of the size of the intersection of two set samples P and Q to the size of the union of two set samples P and Q, which is used to describe the similarity between sets, and its value range is between 0 and 1.
[0044] Taking drugs as nodes and node connections as the similarities between drugs, 8 drug multi-source feature similarity networks were generated.
[0045] Step S3: Based on the fusion method, the drug multi-source feature similarity network is fused to obtain the drug feature fusion network and the drug multi-feature matrix. The specific steps are as follows:
[0046] The network fusion algorithm is used to fuse the above 8 drug multi-source feature similarity networks to obtain a 548×548-dimensional drug feature fusion network. The network fusion algorithm formula is as follows:
[0047]
[0048] Where P is the drug feature fusion network, in this example, k = 7, a m is the weight corresponding to feature m, M m It is the drug multi-source feature similarity network corresponding to feature m.
[0049] Weight am The calculation formula is:
[0050]
[0051] In this example, k=7, c mn is the Pearson correlation coefficient between feature m and the remaining features n.
[0052] Pearson correlation coefficient c mn The calculation formula is:
[0053]
[0054] Among them, M and N represent two different drug multi-source feature similarity networks, Cov(M,N) is the covariance of M and N, Var(M) and Var(N) are the variances of M and N. The value range of Pearson correlation coefficient is [-1,1].
[0055] The 8 drug multi-source feature similarity networks are respectively converted into column feature vectors, and the 8 column feature vectors are spliced column by column using a feature splicing method to obtain a 300304×8-dimensional drug multi-feature matrix.
[0056] Step S4: construct a base learner based on the network diffusion algorithm and the classical machine learning algorithm, and then integrate the base learners based on the ensemble learning algorithm to construct a drug interaction prediction model to obtain drug interaction prediction results. The specific steps are as follows:
[0057] In this example, the network diffusion algorithm is the label propagation algorithm and the neighbor recommendation algorithm, and the classic machine learning algorithm is the random forest algorithm. The drug feature fusion network and the drug interaction matrix are input into the label propagation algorithm for drug interaction prediction to obtain the first preliminary prediction result; the drug feature fusion network and the drug interaction matrix are input into the neighbor recommendation algorithm for drug interaction prediction to obtain the second preliminary prediction result; the drug multi-feature matrix is input into the random forest algorithm for drug interaction prediction to obtain the third preliminary prediction result.
[0058] Based on the Stacking ensemble learning strategy, the three preliminary prediction results are converted into column matrices, and the three column matrices are concatenated and input into the LightGBM algorithm for integration to obtain the final prediction results of drug interactions.
[0059] In this embodiment, drug interaction prediction is performed based on the Stacking ensemble learning strategy, which can combine the advantages of multiple base predictors and minimize the shortcomings of each base predictor. In addition, in the selection of base learners, not only the label propagation algorithm and the neighbor recommendation algorithm are used to obtain the topological structure information of the drug feature fusion network, but also the random forest algorithm is used to obtain the feature information of the drug multi-feature matrix, so that the entire prediction model can fully obtain the useful information in the data, thereby improving the prediction ability of drug interactions.
[0060] Based on the same inventive concept, the present application also provides a drug interaction prediction system based on multi-source data, comprising:
[0061] A data acquisition module, used to acquire multi-source drug feature data;
[0062] A data processing module, used for constructing a drug multi-source feature similarity network and fusing the drug multi-source feature similarity network to obtain a drug feature fusion network and a drug multi-feature matrix;
[0063] The drug interaction prediction module is used to input the drug feature fusion network and the drug multi-feature matrix into the base learners based on the network diffusion algorithm and the classical machine learning algorithm respectively to make a preliminary prediction of drug interaction, and integrate the preliminary prediction results of drug interaction based on ensemble learning to obtain the final prediction results of drug interaction.
[0064] The above method embodiments and system embodiments are based on the same principle, and their related parts can be used for reference, and can achieve the same technical effect. The specific implementation process refers to the above embodiment, which will not be repeated here.
[0065] 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 changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.
Claims
1. A drug interaction prediction method based on multi-source data, characterized in that: include: Obtain various characteristic information of drugs, calculate the similarity of drug characteristics based on similarity algorithm, and construct a drug multi-source characteristic similarity network; Based on the fusion method, the drug multi-source feature similarity network is fused to obtain the drug feature fusion network and the drug multi-feature matrix; the base learner is constructed based on the network diffusion algorithm and the classical machine learning algorithm, and then the base learner is integrated based on the ensemble learning algorithm to construct a drug interaction prediction model and obtain the drug interaction prediction results.
2. The drug interaction prediction method based on multi-source data according to claim 1, characterized in that: The multiple characteristic information of the drug includes 6 types of drug attribute information, 2 types of drug side effect information and drug interaction information; The six types of drug attribute information are: drug chemical structure information, drug target information, drug transporter information, drug enzyme information, drug indication information, and drug pathway information; The two types of drug side effect information are: SIDER side effect information and OFFSIDES side effect information.
3. The drug interaction prediction method based on multi-source data according to claim 1, characterized in that: The drug multi-source feature similarity network comprises: using the Jaccard similarity calculation method to respectively calculate the similarity between drugs for multiple drug feature information, so that each feature generates a drug multi-source feature similarity network.
4. The method for predicting drug interactions based on multi-source data according to claim 1, characterized in that: The multi-source feature fusion includes: Based on the Pearson fusion algorithm, useful information is fully extracted, the correlation between drug features is obtained, and a drug feature fusion network is obtained; the drug multi-source feature similarity network is converted into multiple column matrices, and the multiple column matrices are spliced using a column splicing method to generate a drug multi-feature matrix.
5. The method for predicting drug interactions based on multi-source data according to claim 1, characterized in that: The base learner is constructed based on the network diffusion algorithm and the classic machine learning algorithm, including: Based on the drug feature fusion network, a network diffusion method is used to obtain topological structure information to obtain preliminary prediction results of drug interactions; Based on the drug multi-feature matrix, a classical machine learning algorithm is used to obtain useful information between features to obtain preliminary prediction results of drug interactions; The network diffusion method includes: label propagation algorithm and neighbor recommendation algorithm; The classic machine learning algorithm is: random forest algorithm.
6. The method for predicting drug interactions based on multi-source data according to claim 1, characterized in that: Based on the ensemble learning algorithm and the integrated base learner, a drug interaction prediction model is constructed to obtain drug interaction prediction results, including: Using an integrated learning strategy, the preliminary prediction of drug interaction is input into the LightGBM algorithm for integration to obtain the final prediction result of drug interaction.
7. A drug interaction prediction system based on multi-source data, characterized in that: include: A data acquisition module, used to acquire multi-source drug feature data; A data processing module, used for constructing a drug multi-source feature similarity network and fusing the drug multi-source feature similarity network to obtain a drug feature fusion network and a drug multi-feature matrix; The drug interaction prediction module is used to input the drug feature fusion network and the drug multi-feature matrix into the base learners based on the network diffusion algorithm and the classical machine learning algorithm respectively to make a preliminary prediction of drug interaction, and integrate the preliminary prediction results of drug interaction based on the ensemble learning algorithm to obtain the final prediction results of drug interaction.