A drug repositioning method using neighborhood information and weighted fusion network

Abstract

The application discloses a drug repositioning method using neighborhood information and a weighted fusion network, and relates to a drug repositioning method. The application is used to solve the problem that the existing drug repositioning method is a shallow prediction model, cannot deeply understand the characteristics of drug and disease nodes, and the repositioning result is biased towards general infectious diseases. The steps of the application include: extracting drug-disease interaction data samples; in a drug-disease interaction graph, using a weighted graph convolution network, learning the interaction characteristics between the nearest neighbors of drugs and diseases; aggregating the learned drug and disease characteristics; obtaining hidden high-order nonlinear characteristics through a multi-layer fusion network; inputting the drug-disease interaction in the training set into the above model to train the model, and after multiple iterations, obtaining a final prediction model; inputting the drug in the test set into the prediction model to obtain a prediction result. The application belongs to the technical field of drug repositioning.

Description

Technical Field

[0001] This invention relates to a drug repositioning method, belonging to the field of drug repositioning technology. Background Technology

[0002] Drug development is an expensive process. Due to inaccurate identification of drug targets and related responses, only 10% of drugs that enter Phase I clinical trials are approved; the rest are unusable due to high toxicity or ineffectiveness. The high failure rate of clinical trials and the complexity of drug development lead to increasingly longer development cycles. The average development cycle for a new drug is 8.1 years, with most drugs averaging 4.6-11.9 years. This long development cycle increases costs. Studies indicate that the average total cost of developing a new drug is $2-3 billion. Despite researchers investing significant time, money, and resources, the number of successfully developed new drugs has not increased effectively. According to the FDA, its Center for Drug Evaluation and Research (CDER) approved 50 new therapies in 2021, including 36 new molecular entities and 14 biologics, fewer than in 2020. Over the past decade, CDER has approved an average of only 24 drugs per year. This demonstrates that traditional drug development is expensive, time-consuming, and difficult.

[0003] Drug repositioning, also known as drug repurposing, is a strategy used to discover existing or investigational drugs that extend beyond their original approved indications, expanding their scope and uses. Drug repositioning technology can effectively shorten drug development cycles, reduce development costs, and play a crucial role in advancing drug discovery. Currently, deep learning-based drug transfer methods can be broadly categorized into four types: transfer methods based on known similarity associations, transfer methods based on drug structure, and transfer methods based on heterogeneous information.

[0004] Migration methods based on known similarity associations rely on the "similarity principle," which states that similar compounds tend to have similar biological activities. Therefore, similar drugs are more likely to be associated with the same disease. This first type of method depends on calculating the similarity between drugs and diseases. However, these methods are shallow predictive models that fail to delve into the characteristics of drug and disease nodes, and over-reliance on similarity can lead to relocation results that are overly biased towards common infectious diseases.

[0005] Drug structure-based transfer methods attempt to predict drug function by learning the drug's molecular structure. These methods can model or represent drug structures well, but they cannot fully utilize heterogeneous multi-source data, thus limiting their effectiveness in drug relocation.

[0006] Based on heterogeneous information transfer methods, we can effectively calculate the similarity between drugs and diseases in knowledge graphs to identify effective treatments for specific diseases. Previous studies have shown that drug-target networks, drug-drug networks, drug-disease networks, and protein-protein interaction networks play important roles in determining the relationship between drugs and diseases, providing new opportunities for drug discovery or drug retargeting. Summary of the Invention

[0007] This invention addresses the problem that existing drug relocation methods are shallow prediction models that cannot deeply understand the characteristics of drugs and disease nodes, and whose relocation results are biased towards common infectious diseases. Therefore, this invention proposes a drug relocation method that utilizes neighborhood information and a weighted fusion network.

[0008] The technical solution adopted by the present invention to solve the above problems is as follows: The steps of the present invention include:

[0009] Step 1: Extract data samples on the relationship between drugs and diseases, generate a dataset, divide it into training and testing sets, and perform network training and testing;

[0010] Step 2: In the drug-disease relationship graph, a weighted graph convolutional network is used to learn the interaction characteristics between the nearest neighbors of the drug and the disease.

[0011] Step 3: Aggregate the learned features of drugs and diseases;

[0012] Step 4: Obtain hidden high-order nonlinear features through a multi-layer fusion network, and then combine them with the original features as features of new drugs or diseases for drug relocation prediction.

[0013] Step 5: Input the drug-disease relationship data from the training set into the above model, train the model, and obtain the final prediction model after multiple iterations.

[0014] Step 6: Input the drug pairs in the test set into the prediction model, obtain the prediction results, and perform result analysis.

[0015] Further, step 1 specifically includes: randomly dividing the dataset into a training set and a test set in a 4:1 ratio, with 80% being the training set and 20% being the test set; both the training set and the test set contain each action type; labeling the action types according to 0-(n-1), where n represents the total number of categories; and performing 5-fold cross-validation on the dataset.

[0016] Furthermore, step 2 specifically involves learning the interactions between the drug's nearest neighbors using a weighted linear aggregator, thus extending the drug feature representation. The formula is shown below;

[0017]

[0018] In formula (1), r d Let ω be the feature representation of drug d, i and j be the nearest neighbors of drug d, ω be the weight matrix, and W be the feature representation of drug d. ij The interaction weights between drug i and drug j are used; the weighted graph convolution operation aggregates pairwise interactions between neighbors.

[0019] The characteristics of acquiring the disease are represented as follows:

[0020]

[0021] Furthermore, step 3 specifically includes: for a given drug-disease pair [d, v], using generalized matrix factorization to learn the association between the drug-disease pairs, as shown in the following formula:

[0022] r d，v =F d ⊙F v (3),

[0023] In formula (3), r d，v Representing the association between drug-disease pairs, let d represent the embedding vectors of drug and disease v, respectively, and aggregate drug and disease features:

[0024]

[0025]

[0026]

[0027] In formulas (4), (5) and (6), and It is a representation of drugs and diseases obtained through the interaction between neighbors, and σ(·) is a nonlinear activation function.

[0028] Furthermore, step 4 specifically includes:

[0029] Step 401, Multi-layer fusion network;

[0030] Design a multi-layer fusion network to capture hidden high-order nonlinear features and fuse them with the original embedded features to predict drug-disease associations. The formula is as follows:

[0031] x 1+1 =αx0⊙(wx1+b)+(1-α)x1) (7),

[0032] h l+1 =f(w l h l +b l(8),

[0033] In equations (7) and (8), the initial input embeddings obtained from the encoder are x0, x1, and x 1+1 Let w and b represent the input and output of the l+1 layer fusion network, where w and b are the learnable weight matrices and biases; h l and h l+1 These are the inputs and outputs of the l-th layer, w l Weight matrix, b l is the bias vector, and f(·) is the ReLU activation function;

[0034] Step 402: Deepen Network Integration;

[0035] Deep neural networks and multi-layer fusion networks are combined using a parallel structure, as shown in the following formula:

[0036] f fusion +f deep (9).

[0037] The beneficial effects of this invention are as follows: This invention proposes a transfer method based on heterogeneous information, which can effectively calculate the similarity between drugs and diseases in a knowledge graph to identify effective treatment drugs for specific diseases; This invention provides new opportunities for drug discovery or drug repositioning; This invention expands the representation of drug features by learning the interactions between the nearest neighbors of drugs through a weighted linear aggregator. Attached Figure Description

[0038] Figure 1 This is a flowchart of a drug relocation method that utilizes neighborhood information and a weighted fusion network;

[0039] Figure 2 This is a framework diagram of a drug relocation method that utilizes neighborhood information and a weighted fusion network. Detailed Implementation

[0040] Specific implementation method one: as follows Figure 1 and Figure 2 As shown, a drug relocation method utilizing neighborhood information and a weighted fusion network includes the following steps:

[0041] Step 1: Extract data samples of drug-disease relationship, generate a dataset, divide it into training and test sets, and perform network training and testing; randomly divide the dataset into training and test sets in a 4:1 ratio, with 80% as the training set and 20% as the test set; both the training and test sets contain each type of effect; label the types of effects as 0-(n-1), where n represents the total number of categories, and perform 5-fold cross-validation on the dataset;

[0042] Step 2: In the drug-disease interaction graph, a weighted graph convolutional network is used to learn the interaction features between the nearest neighbors of the drug and the disease; a weighted linear aggregator is used to learn the interactions between the nearest neighbors of the drug, thus expanding the drug feature representation. The formula is shown below;

[0043]

[0044] In formula (1), r d Let ω be the feature representation of drug d, i and j be the nearest neighbors of drug d, ω be the weight matrix, and W be the feature representation of drug d. ij The interaction weights between drug i and drug j are used; the weighted graph convolution operation aggregates pairwise interactions between neighbors.

[0045] The characteristics of acquiring the disease are represented as follows:

[0046]

[0047] Step 3: Aggregate the learned drug and disease features; for a given drug-disease pair [d, v], use generalized matrix factorization to learn the association between drug-disease pairs, as shown in the following formula:

[0048] r d，v =F d ⊙F v (3),

[0049] In formula (3), r d，v Representing the association between drug-disease pairs, let d represent the embedding vectors of drug and disease v, respectively, and aggregate drug and disease features:

[0050]

[0051]

[0052]

[0053] In formulas (4), (5) and (6), and It is a representation of drugs and diseases obtained through the interactions between neighbors, and σ(·) is a nonlinear activation function;

[0054] Step 4: Obtain hidden high-order nonlinear features through a multi-layer fusion network, and then combine them with the original features as features of new drugs or diseases for drug relocation prediction.

[0055] Specifically, it includes:

[0056] Step 401, Multi-layer fusion network;

[0057] Design a multi-layer fusion network to capture hidden high-order nonlinear features and fuse them with the original embedded features to predict drug-disease associations. The formula is as follows:

[0058] x 1+1 =ax0⊙(wx1+b)+(1-α)x1) (7),

[0059] h l+1 =f(w l h l +b l (8),

[0060] In equations (7) and (8), the initial input embeddings obtained from the encoder are x0, x1, and x 1+1 Let w and b represent the input and output of the l+1 layer fusion network, where w and b are the learnable weight matrices and biases; h l and h l+1 These are the inputs and outputs of layer 1, w l Weight matrix, b l is the bias vector, and f(·) is the ReLU activation function;

[0061] Step 402: Deepen Network Integration;

[0062] Deep neural networks and multi-layer fusion networks are combined using a parallel structure, as shown in the following formula:

[0063] f fusion +f deep (9);

[0064] Step 5: Input the drug-disease relationship data from the training set into the above model, train the model, and obtain the final prediction model after multiple iterations.

[0065] Step 6: Input the drug pairs in the test set into the prediction model, obtain the prediction results, and perform result analysis.

[0066] Dataset

[0067] The model was evaluated using three real-world datasets.

[0068] Fdataset includes 1933 known drugs and 593 diseases associated with diseases from the 1933 Drug Library database, and 313 diseases from the OMIM database.

[0069] Cdataset contains 2,532 known drug-disease associations, including 663 drugs extracted from the Drug Library database and 409 diseases listed in the OMIM database.

[0070] LRSSL includes 3,051 validated drug-disease associations, primarily comprising 763 drugs extracted from drug library databases and 681 diseases obtained from the MeSH database.

[0071] Table 1

[0072]

[0073]

[0074] Evaluation indicators

[0075] AUC and AUPR scores are used as evaluation metrics, and the calculation formulas for these metrics are as follows:

[0076] True Positive Rate

[0077] TPR is used to measure the proportion of correctly classified positive samples out of the actual number of positive samples. The calculation formula is as follows:

[0078]

[0079] False Positive Rate

[0080] FPR is used to measure the proportion of misclassified positive samples out of the actual negative samples. The calculation formula is as follows:

[0081]

[0082] The definitions of TP, TN, FP, and FN are shown in Table 2.

[0083] Table 2

[0084]

[0085] AUC

[0086] AUC stands for Area Under Curve, which reflects the classification ability. Its horizontal axis represents the false positive rate (FPR), and its vertical axis represents the true positive rate (TPR). The closer the AUC is to 1, the better the model results.

[0087] AUPR

[0088] AUPR stands for Area under the Precision / Recall curve. The x-axis of the PR curve represents recall, and the y-axis represents precision. The PR curve is easily affected by the sample distribution (the ratio of positive to negative samples in the training dataset). Therefore, AUPR can be used to measure the predictive performance on imbalanced datasets. An AUPR value closer to 1 indicates better model performance.

[0089] The experimental results were evaluated by comparing this invention with DRIMC, SCMFDD, NIMCGCN, BNNR, and LAGCN. The results are shown in Table 3.

[0090] Table 3

[0091]

[0092] The results show that BNNR achieves a slightly higher AUC than the model proposed in this invention on both the F and C datasets. However, BNNR's AUPR is lower than that of the model proposed in this invention on all three datasets. Generally, the unknown drug-disease pairs are much larger than the known drug-disease pairs, so negative samples are much larger than positive samples. AUPR is also an indispensable evaluation metric for imbalanced datasets. Compared with other models, the model proposed in this invention performs better.

[0093] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A drug relocation method utilizing neighborhood information and a weighted fusion network, characterized in that, The specific steps include: Step 1: Extract data samples on the relationship between drugs and diseases, generate a dataset, divide it into training and testing sets, and perform network training and testing; Step 2: In the drug-disease interaction graph, a weighted graph convolutional network is used to learn the interaction features between the nearest neighbors of the drug and the disease. Specifically, this includes learning the interactions between the nearest neighbors of the drug through a weighted linear aggregator, which expands the drug feature representation, as shown in the following formula: （1）， In formula (1), It is a drug Feature representation, and It is a drug The nearest neighbor, It is a weight matrix. For drugs and medicine Interaction weights; weighted graph convolution operations aggregate pairwise interactions between neighbors; The characteristics of acquiring the disease are represented as follows: （2）； Step 3: Aggregate the learned features of drugs and diseases; Step 4: Obtain hidden high-order nonlinear features through a multi-layer fusion network, and then combine them with the original features as features of new drugs or diseases for drug relocation prediction. Step 5: Input the drug-disease relationship data from the training set into the above model, train the model, and obtain the final prediction model after multiple iterations. Step 6: Input the drug pairs in the test set into the prediction model, obtain the prediction results, and perform result analysis.

2. The drug relocation method utilizing neighborhood information and a weighted fusion network according to claim 1, characterized in that, Step 1 specifically includes: randomly dividing the dataset into a training set and a test set in a 4:1 ratio, with 80% being the training set and 20% being the test set; both the training set and the test set contain each type of action; labeling the action types with numbers from 0 to (n-1), where n represents the total number of categories; and performing 5-fold cross-validation on the dataset.

3. The drug relocation method utilizing neighborhood information and a weighted fusion network according to claim 1, characterized in that, Step 3 specifically includes: For a given drug-disease pair [d, v], using generalized matrix factorization to learn the association between the drug-disease pairs, as shown in the following formula: （3）， In formula (3), Indicating the association between drug-disease pairs, and They represent drugs and disease Embedding vectors; aggregating drug and disease features: （4）， （5）， （6）， In formulas (4), (5) and (6), and It is the representation of drugs and diseases obtained through interactions between neighbors, and It is a non-linear activation function.

4. The drug relocation method utilizing neighborhood information and a weighted fusion network according to claim 1, characterized in that, Step 4 specifically includes: Step 401, Multi-layer fusion network; Design a multi-layer fusion network to capture hidden high-order nonlinear features and fuse them with the original embedded features to predict drug-disease associations. The formula is as follows: （7）， （8）， In formulas (7) and (8), the initial input embedding obtained from the encoder is , and express The input and output of the layer fusion network. and It is a learnable weight matrix and bias; and It is the first Layer inputs and outputs, Represents the weight matrix. It is a bias vector. It is the activation function of ReLU; Step 402: Deepen Network Integration; Deep neural networks and multi-layer fusion networks are combined using a parallel structure, as shown in the following formula: （9）。

Images