A multi-model integrated drug repositioning system and method

Abstract

This invention belongs to the fields of bioinformatics and computer science, specifically relating to a multi-model integrated drug relocation system and method. The method includes extracting a dataset from a drug-disease database module; processing the dataset uniformly using a data integration module; inputting the uniformly processed dataset into a model integration module, where each algorithm model obtains its corresponding drug-disease association prediction results and inputs them into a list integration module; extracting relevant drug-disease association prediction lists from the list integration module based on the target drug or target disease, and extracting the performance values ​​of the corresponding algorithm models for each drug-disease association prediction list from a performance calculation module; and using an improved TOPSIS method to correct and optimize the relevant drug-disease association prediction lists based on the performance values, resulting in an optimized drug-disease association prediction list for the target drug or target disease. This invention improves the stability and universality of the prediction results.

Description

Technical Field

[0001] This invention belongs to the fields of bioinformatics and computer science, and specifically relates to a multi-model integrated drug repositioning system and method. Background Technology

[0002] With advancements in genomics, proteomics, and life science technologies, drug research and development technologies have developed rapidly over the past few decades. However, the number of new drugs approved by the U.S. Food and Drug Administration (FDA) has stagnated. New drug development remains an expensive and lengthy process. Statistics show that bringing a new drug to market takes more than 10 years and costs over $1 billion. To address this issue, drug repositioning (also known as "drug repurposing") technology is becoming an important strategy in drug development. Drug repositioning is a method of finding new indications for existing drugs to treat diseases beyond their original purpose. Because this strategy can significantly shorten drug development time, reduce overall development costs, avoid risks, and because existing drugs have undergone extensive experiments and clinical trials, providing safety, efficacy, and tolerability data, drug repositioning has become a widely used strategy.

[0003] The sheer number of diseases and known drugs makes it prohibitively expensive to solely rely on experimental screening for novel uses of existing drugs. Advances in bioscience and improved assay techniques have enabled research to access molecular data at various levels, such as genomics, proteomics, transcriptomics, and metabolomics. This accumulation of omics assays has gradually formed high-quality databases from fields such as chemistry, medicine, pharmacology, and biology, including DrugBank, OMIM, CTD, Uniprot, MimMiner, SIDER, and DO. Rational design of drug repositioning strategies using computational prediction methods can provide clues for large-scale experimental screening, further reducing costs and ushering in a stage where rational design and experimental screening are combined in drug repositioning.

[0004] In recent years, numerous methods for drug relocation computation have been proposed. Luo et al. proposed the MBIRW algorithm (see patent document CN105653846A) to enhance drug and disease similarity using existing information on drug-disease associations. This algorithm is implemented on a heterogeneous drug-disease network and utilizes several comprehensive similarity metrics and a double random walk algorithm to predict potential drug-disease associations. Luo et al. also proposed the DRRS drug relocation recommendation system (see patent document CN107545151). This method constructs a heterogeneous network using drug and disease similarity, represented by a drug-disease adjacency matrix. Finally, it uses the singular value thresholding (SVT) algorithm to complete the matrix and achieve drug relocation. Zhang et al. proposed SCMFDD, which predicts drug-disease associations based on known drug-disease associations, drug features, and disease semantic information using a similarity-constrained matrix factorization method. Liang et al. proposed a Laplacian regularized sparse subspace learning method, LRSSL, to identify new indications. This method integrates drug chemistry information, drug target domain information, and target annotation information. The BNNR algorithm employs bounded norm regularization under the low-rank assumption to complete the drug-disease matrix for drug-disease association prediction. Yang et al. designed a multi-similarity bilinear matrix factorization (MSBMF) method, which dynamically optimizes the fusion process and the drug-disease association training process simultaneously to fully utilize multiple similarity measures between paired drugs and paired diseases for identifying promising drug-related indications. Zeng et al. developed a network-based deep learning method, deepDR. First, it calculates the Pointed Mutual Information (PPMI) matrix as features from 10 drug-related networks. Then, it fuses the PPMI matrix using a multimodal deep autoencoder. Finally, it uses the fused features to infer new applications of existing drugs through a collective variational autoencoder. Zhang et al. proposed a new method called DRIMC, which integrates multi-source drug and disease data and models the correlation probability through matrix completion to predict drug-disease associations. Yu et al. proposed LAGCN, which uses graph convolutional networks to capture information and introduces an attention mechanism to combine the embeddings of different convolutional layers to predict drug-disease associations. Meng et al. proposed a similarity-constrained probability matrix factorization (SCPMF) model based on drug-disease similarity information to identify potential indications for drugs with unknown disease characteristics. The computational methods mentioned above can be mainly divided into three categories:

[0005] 1) Machine learning-based methods: deepDR, LAGCN, etc.

[0006] (ii) Network-based propagation methods: MBiRW

[0007] (iii) Methods based on matrix completion or matrix decomposition: DRRS, SCMFDD, BNNR, MSBMF, DRIMC, SCPMF.

[0008] Machine learning-based methods heavily rely on known sample labels in the dataset, but in practical applications, obtaining stable negative sample data is difficult, limiting the predictive power of these methods. Network propagation-based methods tend to select edges with larger weights during propagation, so nodes lacking relevant information may not be allocated resources for propagation for a considerable period, also affecting the accuracy of predictions. Matrix completion or matrix factorization methods offer better performance and are less dependent on input data than other methods; however, as the dataset grows, the computational resources and time consumption of these methods increase exponentially.

[0009] In summary, while many computational methods have been effectively applied in the field of drug repositioning, the following problems still exist:

[0010] 1) The prediction results are not stable enough. Different methods perform inconsistently on different datasets, making it difficult to obtain reliable drug-disease association prediction results.

[0011] 2) Many methods use relatively simple drug and disease representations or data sources, which cannot fully reflect the complex attributes of drugs and diseases, making the calculated drug and disease similarity less accurate and reliable.

[0012] 3) Using a single algorithm or network cannot integrate the effects of different advantageous algorithms. Directly sorting the results based on the results of a single algorithm can easily introduce biases and thus affect the accuracy of the results. Summary of the Invention

[0013] To address the aforementioned issues, this invention proposes a multi-model integrated drug relocation system and method, which integrates multi-source bioinformatics data and multiple drug relocation algorithms, overcoming the limitations of individual methods and improving the stability and accuracy of drug relocation prediction.

[0014] In a first aspect, the present invention proposes a multi-model integrated drug relocation system, comprising:

[0015] The drug repositioning literature statistics module is used to synchronize the PubMed database and retrieve drug repositioning literature from the synchronized PubMed database based on keywords.

[0016] The Drug Disease Database module is used to synchronize the DrugBank database, SIDER database, OMIM database, MimMiner database, Mesh database, and Disease Ontology database.

[0017] The data integration module is used to process the dataset in a unified manner to make it conform to the specifications and requirements of different algorithm models, and then provide the processed dataset to the model integration module.

[0018] The model integration module is used to collect the algorithm model used in each drug repositioning literature retrieved by the drug repositioning literature statistics module, receive the output results of the data integration module and perform drug repositioning prediction, and provide the drug-disease association prediction list of each algorithm model to the list integration module; the drug-disease association prediction results include multiple drug-disease association prediction lists.

[0019] The performance calculation module is used to calculate the performance of the data integration module output results of different algorithm models and provide them to the list integration module.

[0020] The list integration module is used to receive and store the drug-disease association prediction results of each algorithm model output by the model integration module;

[0021] The revised prediction module is used to extract a list of drug-disease association predictions based on the target drug or target disease, and then use an improved TOPSIS method to revise and optimize the list, ultimately obtaining an optimized drug-disease association prediction list.

[0022] Furthermore, the data integration module also includes:

[0023] The drug-disease information extraction module is used to extract datasets from the drug-disease database module; the datasets include information on various drugs, various diseases, and various drug-disease association information.

[0024] The information matrix conversion module is used to convert various types of information obtained by the drug disease information extraction module into a matrix.

[0025] The similarity calculation module is used to calculate various similarity matrices based on multiple matrices from the information matrix transformation module.

[0026] Secondly, a multi-model integrated drug relocation method for realizing the above system is proposed, including the following steps:

[0027] S1. Extract the dataset from the drug disease database module, and use the data integration module to process the dataset in a unified manner to make it conform to the specifications and requirements of different algorithm models.

[0028] S2. Input the uniformly processed dataset into the model integration module. Each algorithm model obtains the corresponding drug-disease association prediction results according to its own operation and inputs them into the list integration module.

[0029] S3. Extract relevant drug-disease association prediction lists from the list integration module based on the target drug or target disease, and extract the performance values ​​of the algorithm model corresponding to each drug-disease association prediction list from the performance calculation module;

[0030] S4. Based on the performance values, the improved TOPSIS method is used to revise and optimize the relevant drug-disease association prediction list to obtain the optimized drug-disease association prediction list for the target drug or target disease.

[0031] Furthermore, the processing in step S4 includes:

[0032] S11. Obtain the performance value of each algorithm model on the dataset, and use the linear scaling transformation method to obtain the performance weight of each algorithm model based on the performance value.

[0033] S12. For the target drug or target disease, obtain the relevant drug-disease association prediction list calculated by each algorithm model, extract the maximum length from all drug-disease association prediction lists, and complete the drug-disease association prediction lists whose length is not equal to the maximum length to obtain a complete list.

[0034] S13. Convert the complete list into a complete matrix, and standardize the matrix to obtain a standardized score matrix; where the number of columns in the standardized score matrix is ​​the number of algorithm models, and the number of rows is the number of candidate entries;

[0035] S14. Based on the performance weight, calculate the optimal solution distance and the worst solution distance for each candidate item, obtain the fit degree of each candidate item based on the optimal solution distance and the worst solution distance, and sort them in descending order according to the fit degree to finally obtain the optimized drug-disease association prediction list.

[0036] Furthermore, the complete matrix is ​​standardized and expressed as:

[0037]

[0038] Where, x ij Z represents the element in the i-th row and j-th column of the complete matrix; ij Let n be the element in the i-th row and j-th column of the standardized rating matrix, where n is the number of rows in the complete matrix.

[0039] Furthermore, the performance weight of each algorithm model is obtained using a linear scaling transformation method based on the performance values, expressed as:

[0040]

[0041] Among them, w j Auc represents the performance weight of the j-th algorithm model. jAuc represents the performance value of the j-th algorithm model. min Auc represents the minimum performance value. max This represents the maximum performance value.

[0042] Furthermore, the ranking is aggregated based on the performance weighted calculation results to obtain the final drug-disease association ranking list, represented as follows:

[0043]

[0044]

[0045]

[0046]

[0047]

[0048] Where Z represents the standardized scoring matrix, Let represent the element in the i-th row and j-th column of the standardized rating matrix, where n is the number of rows in the standardized rating matrix. This represents the largest element in the j-th column of the standardized scoring matrix. w represents the smallest element in the j-th column of the standardized rating matrix. j Let m represent the performance weight of the j-th algorithm model, and m represent the number of algorithm models. and Let C represent the distance to the best solution and the distance to the worst solution for the i-th candidate item, respectively. i This represents the fit of the i-th candidate entry.

[0049] The beneficial effects of this invention are:

[0050] This invention can fully utilize drug and disease information from multiple data sources, integrate the advantages of existing drug relocation algorithms, fully extract drug attribute features and disease attribute features, weight the integrated algorithms according to their performance, and further perform weighted TOPSIS ranking aggregation on the prediction results to improve the stability, universality and accuracy of the prediction results, and more effectively and accurately discover new indications for known drugs. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the principle framework of the multi-model integrated drug relocation system of the present invention;

[0052] Figure 2 This is a flowchart of the multi-model integrated drug relocation method of the present invention;

[0053] Figure 3This invention describes a process for revising and optimizing a list of drug-disease association predictions using an improved TOPSIS method. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] This invention proposes a multi-model integrated drug relocation system, such as... Figure 1 As shown, it includes:

[0056] The drug repositioning literature statistics module is used to synchronize the PubMed database and retrieve drug repositioning literature from the synchronized PubMed database based on keywords.

[0057] The Drug Disease Database module is used to synchronize the DrugBank database, SIDER database, OMIM database, MimMiner database, Mesh database, and Disease Ontology database.

[0058] The data integration module is used to process the dataset in a unified manner to make it conform to the specifications and requirements of different algorithm models, and then provide the processed dataset to the model integration module.

[0059] The model integration module is used to collect the algorithm model used in each drug repositioning literature retrieved by the drug repositioning literature statistics module, receive the output results of the data integration module and perform drug repositioning prediction, and provide the drug-disease association prediction results of each algorithm model to the list integration module; the drug-disease association prediction results include multiple drug-disease association prediction lists.

[0060] The performance calculation module is used to calculate the performance of the data integration module output results of different algorithm models and provide them to the list integration module.

[0061] The list integration module is used to receive and store the drug-disease association prediction results of each algorithm model output by the model integration module;

[0062] The revised prediction module is used to extract a list of drug-disease association predictions based on the target drug or target disease, and then use an improved TOPSIS method to revise and optimize the list, ultimately obtaining an optimized drug-disease association prediction list.

[0063] Specifically, the drug repositioning literature statistics module synchronizes with the PubMed database, collecting published drug repositioning literature by entering the keywords "drug repositioning" or "drug repurposing" into the PubMed database.

[0064] Specifically, the data integration module also includes:

[0065] The drug-disease information extraction module is used to extract relevant datasets from the drug-disease database module based on all drug and disease data involved in drug relocation literature; the datasets include multiple drug information, multiple disease information, and multiple drug-disease association information.

[0066] The information matrix conversion module is used to convert various types of information obtained by the drug disease information extraction module into a matrix.

[0067] The similarity calculation module is used to calculate various similarity matrices based on multiple matrices from the information matrix transformation module.

[0068] Specifically, in the drug-disease information extraction module, all drug and disease data involved in the drug relocation literature are organized. Based on the organized data, the association data between different drugs and different diseases, the keyword IDs of different drugs (such as DrugBankID), and the keyword IDs of different diseases (such as OMIMID, MESHID) are obtained. Then, in the drug-disease database module, drugs obtain their own SMILES molecular structure information, protein association data, side effect information, and interaction information with other drugs based on the keyword ID; diseases obtain MESH description information, ontology description information, etc. based on the keyword ID.

[0069] Specifically, in the information matrix transformation module, a multidimensional vector is constructed using a 0 / 1 vector pattern for feature representation. Then, based on the acquired information, different drug-disease association matrices, drug chemical structure adjacency matrices, drug target adjacency matrices, drug-drug interaction connection matrices, and drug side effect adjacency matrices are established.

[0070] Specifically, in the similarity calculation module, based on the adjacency matrix of drug chemical structures, the molecular fingerprint is calculated using the R language open-source toolkit RCDK and processed using the Tanimot formula to obtain the drug chemical structure similarity matrix. Simultaneously, the similarity between pairs of drugs is calculated using the Jaccard formula to obtain the drug target similarity matrix, drug-drug interaction similarity matrix, and drug side effect similarity matrix. For diseases, the disease phenotype similarity matrix is ​​downloaded from the MimMiner database; the disease ontology similarity matrix is ​​calculated using the R package DOSE from the Disease Ontology database. The Jaccard formula is expressed as:

[0071]

[0072] Here, X1 and X2 represent the feature vectors of the two drugs, respectively.

[0073] Specifically, in the performance calculation module, a pipeline program is written using the Python programming language and its modules `os` and `sklearn`. This pipeline program takes input data and feeds it into the drug repositioning calculation method (i.e., the method proposed in the drug repositioning literature), obtaining the AUC value and prediction results on the dataset being run. The pipeline program is scheduled by the Python `os` module; the FPR and TPR values ​​are calculated using the `roc_curve` function under the `sklearn.metrics` module; and the precision and recall values ​​are calculated using the `precision_recall_curve` function under the `sklearn.metrics` module, thus obtaining the AUC value. The formulas for TPR (Recall), FPR, precision, and AUC are as follows:

[0074]

[0075]

[0076]

[0077]

[0078] TP, FP, TN, and FN represent True Positive (predicted as positive, and is indeed positive), False Positive (predicted as positive, but is actually negative), True Negative (predicted as negative, and is indeed negative), and False Negative (predicted as negative, but is actually positive), respectively. TPR (True Positive Rate) represents the proportion of correctly predicted positive samples out of all positive samples. Recall, calculated using the same formula as TPR, represents the recall rate. FPR (False Positive Rate) represents the proportion of incorrectly predicted positive samples out of all positive samples. Precision represents the proportion of correctly predicted positive samples out of all predicted positive samples. This represents the ranking of the i-th positive sample after sorting the drug-disease association scores predicted by the method according to the score size. Ins represents the positive sample category. i The index represents the positive sample, that is, the i-th positive sample ranked by score. This represents the sum of the ranking indices of the positive samples, where M and N are the number of positive and negative samples, respectively. The AUC value is a performance metric for the predictive performance of the method. Positive samples indicate a correlation between the drug and the disease, while negative samples indicate no correlation.

[0079] Specifically, after retrieving drug relocation literature from the synchronized PubMed database based on keywords, the datasets used in each drug relocation literature need to be collected and organized. Based on the organized datasets, association data between different drugs and different diseases, keyword IDs for different drugs (e.g., DrugBankID), and keyword IDs for different diseases (e.g., OMIMID, MESHID) are obtained. Then, in the drug database module, drugs obtain their own SMILES molecular structure information, protein association data, side effect information, and interaction information with other drugs based on their keyword IDs; diseases obtain MESH description information and ontology description information based on their keyword IDs. Then, matrix transformation and similarity calculations are performed on all the collected information. Next, different algorithm models are used for drug relocation, and the performance of all algorithm models and drug-disease prediction results on all datasets are obtained. These results are then stored in the performance calculation module and the list integration module, respectively. When predicting a target drug or disease belonging to the extracted datasets, the prediction data can be directly extracted for subsequent correction and optimization.

[0080] In one embodiment, a multi-model integrated drug relocation method based on the above system is employed, such as... Figure 2As shown, it includes the following steps:

[0081] S1. Extract the dataset from the drug disease database module, and use the data integration module to process the dataset in a unified manner to make it conform to the specifications and requirements of different algorithm models.

[0082] S2. Input the uniformly processed dataset into the model integration module. Each algorithm model obtains the corresponding drug-disease association prediction results according to its own operation and inputs them into the list integration module.

[0083] S3. Extract relevant drug-disease association prediction lists from the list integration module based on the target drug or target disease, and extract the performance values ​​of the algorithm model corresponding to each drug-disease association prediction list from the performance calculation module;

[0084] S4. Based on the performance values, the improved TOPSIS method is used to revise and optimize the relevant drug-disease association prediction list to obtain the optimized drug-disease association prediction list for the target drug or target disease.

[0085] Specifically, the processing steps of the list integration module include:

[0086] S11. Obtain the performance value of each algorithm model on the drug dataset, and use the linear scaling transformation method to obtain the performance weight of each algorithm model based on the performance value.

[0087] Specifically, the performance weight of each algorithm model is obtained using a linear scaling transformation method based on its performance values, and is expressed as follows:

[0088]

[0089] Among them, w j Auc represents the performance weight of the j-th algorithm model. j Auc represents the performance value of the j-th algorithm model. min Auc represents the minimum performance value. max This represents the maximum performance value.

[0090] S12. For the target drug or target disease, obtain the list of relevant drug-disease association predictions calculated by each algorithm model, such as... Figure 3 As shown in list L, the first column is the list of drug-disease association predictions by algorithm model P1 for target drug R1; the maximum length is extracted from all drug-disease association prediction lists, and the drug-disease association prediction lists whose length is not equal to the maximum length are completed to obtain a complete list.

[0091] S13. Obtain all candidate entries involved in the complete list, and convert the complete list into a complete matrix, such as... Figure 3 As shown in matrix T, each column of the complete matrix is ​​the prediction result of an algorithm model for all candidate items; the complete matrix is ​​standardized to obtain the standardized score matrix, where the number of columns in the standardized score matrix is ​​the number of algorithm models and the number of rows is the number of candidate items.

[0092] Specifically, when step S12 is performed on a target drug, the candidate entries are the diseases involved in the extracted relevant drug-disease association prediction list; when step S12 is performed on a target disease, the candidate entries are the drugs involved in the extracted relevant drug-disease association prediction list.

[0093] Specifically, the obtained complete matrix is ​​standardized and represented as follows:

[0094]

[0095] Where, x ij Z represents the element in the i-th row and j-th column of the complete matrix; ij Let n be the element in the i-th row and j-th column of the standardized rating matrix, where n is the number of rows in the complete matrix.

[0096] S14. Based on the performance weight, calculate the optimal solution distance and the worst solution distance for each candidate item, obtain the fit degree of each candidate item based on the optimal solution distance and the worst solution distance, and sort them in descending order according to the fit degree to finally obtain the optimized drug-disease association prediction list.

[0097] Specifically, the distance between each candidate item and the optimal and worst solutions is calculated using a weighted average based on performance weights, and then the items are sorted according to the calculated fit, as shown below:

[0098]

[0099]

[0100]

[0101]

[0102]

[0103] Where Z represents the standardized scoring matrix, Let represent the element in the i-th row and j-th column of the standardized rating matrix, where n is the number of rows in the standardized rating matrix. This represents the largest element in the j-th column of the standardized scoring matrix. w represents the smallest element in the j-th column of the standardized rating matrix. j Let m represent the performance weight of the j-th algorithm model, and m represent the number of algorithm models. and Let C represent the distance to the best solution and the distance to the worst solution for the i-th candidate item, respectively. i This represents the fit of the i-th candidate entry.

[0104] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "setting," "connection," "fixing," "rotation," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Unless otherwise explicitly limited, those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0105] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-model comprehensive integrated drug relocation system, characterized in that, Comprise: Drug repositioning literature statistics module for synchronizing pubmed database, retrieving drug repositioning literature from the synchronized pubmed database according to keywords; Drug disease database module for synchronizing DrugBank database, SIDER database, OMIM database, MimMiner database, Mesh database and Disease Ontology database; Data integration module for uniformly processing data sets to meet the specifications and requirements of different algorithm models, and providing the processed data sets to the model integration module; The model integration module is used for collecting the algorithm model used by each drug repositioning literature retrieved by the drug repositioning literature statistics module, receiving the output results of the data integration module and performing drug repositioning prediction, and providing the drug disease association prediction results of each algorithm model to the list integration module; The drug disease association prediction results include a plurality of drug disease association prediction lists; The data integration module further comprises: Drug disease information extraction module for extracting data sets from the drug disease database module; the data set comprises a plurality of drug information, a plurality of disease information and a plurality of drug disease association information; In the drug disease information extraction module, all drug and disease data involved in drug repositioning literature are collated, and the association data of different drugs and different diseases, the keyword ID of different drugs and the keyword ID of different diseases are obtained according to the collated data; then in the drug disease database module, the drug obtains its SMILES molecular structure information, the association data of protein, the side effect information and the interaction information with other drugs according to the keyword ID; the disease obtains MESH description information and ontology description information according to the keyword ID; Information matrix conversion module for converting various information obtained by the drug disease information extraction module into matrix; In the information matrix conversion module, a multi-dimensional vector is constructed for feature representation by using 0 / 1 vector mode, and then different drug disease association matrices, drug chemical structure adjacency matrices, drug target adjacency matrices, drug-drug interaction connection matrices and drug side effect adjacency matrices are established according to the obtained information; Similarity calculation module for calculating a plurality of similarity matrices according to the plurality of matrices of the information matrix conversion module; Performance calculation module for calculating the performance of different algorithm models processing the output results of the data integration module, and providing the performance to the list integration module; List integration module for receiving and storing the drug disease association prediction results of each algorithm model output by the model integration module; The correction prediction module is used for extracting the relevant drug disease association prediction list according to the target drug or target disease, and adopting the improved TOPSIS method for correction and optimization, and finally obtaining the optimized drug disease association prediction list.

2. A multi-model comprehensive integrated drug relocation method using the multi-model comprehensive integrated drug relocation system of claim 1, characterized in that, Comprise the following steps: S1. Extracting data sets in the drug disease database module, uniformly processing the data sets by the data integration module to meet the specifications and requirements of different algorithm models; S2. The unified processed data set is input into the model integration module, each algorithm model obtains the corresponding drug-disease association prediction result according to its own operation and inputs the list integration module; S3. The relevant drug-disease association prediction list is extracted from the list integration module according to the target drug or target disease, and the performance value of the algorithm model corresponding to each drug-disease association prediction list is extracted from the performance calculation module; S4. According to the performance value, the improved TOPSIS method is used to correct and optimize the relevant drug-disease association prediction list, and the optimized drug-disease association prediction list of the target drug or target disease is obtained.

3. The multi-model comprehensive integrated drug relocation method according to claim 2, characterized in that, The processing process of step S4 includes: S11. Obtain the performance value of each algorithm model on the data set, and obtain the performance weight of each algorithm model by using the linear proportional conversion method according to the performance value; S12. For the target drug or target disease, obtain the relevant drug-disease association prediction list calculated by each algorithm model, extract the maximum length from all drug-disease association prediction lists, and complete the drug-disease association prediction list with a length not equal to the maximum length to obtain a complete list; S13. Convert the complete list into a complete matrix, and standardize the matrix to obtain a standardized score matrix; wherein the column number of the standardized score matrix is the number of algorithm models, and the row number is the number of candidate entries; S14. According to the performance weight, calculate the optimal solution distance and the worst solution distance of each candidate entry, obtain the fitting degree of each candidate entry according to the optimal solution distance and the worst solution distance, and sort the fitting degrees in descending order according to the size, and finally obtain the optimized drug-disease association prediction list.

4. The multi-model comprehensive integrated drug relocation method according to claim 3, characterized in that, The complete matrix is standardized, which is represented as: where x ij represents the element in the i-th row and j-th column of the complete matrix; Z ij represents the element in the i-th row and j-th column of the normalized score matrix, and n represents the number of rows of the complete matrix.

5. The multi-model comprehensive integrated drug relocation method according to claim 3, characterized in that, According to the performance value, the performance weight of each algorithm model is obtained by using the linear proportional conversion method, which is represented as: wherein w j represents the performance weight of the jth algorithm model, auc j represents the performance value of the jth algorithm model, auc min represents the minimum performance value, auc max represents the maximum performance value.

6. The multi-model comprehensive integrated drug relocation method according to claim 3, characterized in that, Ranking aggregation is performed according to the weighted calculation result of the performance weight, which is represented as: wherein Z represents a normalized score matrix, represents an element of the i-th row and the j-th column of the normalized score matrix, and n represents the number of rows of the normalized score matrix, represents the maximum element in the j-th column of the normalized score matrix, represents the minimum element in the j-th column of the normalized score matrix, and w j represents the performance weight of the j-th algorithm model, and m represents the number of algorithm models, and respectively represent the optimal solution distance and the worst solution distance of the i-th candidate item, and C i represents the fitness of the i-th candidate item.

Images