CN116153391B - Antiviral drug screening method, system and storage medium based on joint projection - Google Patents

Abstract

The invention provides an antiviral drug screening method, system and storage medium based on joint projection, which belongs to the interdisciplinary technical field of bioinformatics, computational biology and artificial intelligence. The method is realized through the system, and the method includes the following steps: S1. Constructing virus-drug Associated adjacency matrix; S2. Calculate virus Gaussian distance similarity matrix and drug Gaussian distance similarity matrix; S3. Calculate virus gene sequence similarity matrix and drug chemical structure similarity matrix; S4. Use fast kernel learning method to integrate virus integration similarity matrix and Drug integration similarity matrix; S5. Use the sparse regularization joint projection method to construct a loss function; S6. Solve the loss function to obtain the virus-drug prediction score matrix; S7. Based on the virus-drug prediction score matrix, obtain the final prediction after screening and sorting result. The invention can effectively overcome the adverse influence of noise data, and accurately and efficiently screen effective drugs for treating viruses.

Description

Antiviral drug screening method, system and storage medium based on joint projection

technical field

The present invention relates to the interdisciplinary technical field of bioinformatics, computational biology and artificial intelligence, in particular to a joint projection-based antiviral drug screening method, system and storage medium.

Background technique

Studies have pointed out that the development of an effective clinical new drug basically requires billions of dollars, and it takes about 9-12 years on average to successfully introduce it to the market. Drug repurposing (also known as drug repositioning), that is, using existing drugs to find drugs for new indications, has become a promising solution to improve the productivity of new drug development. Exploratory experiments in the laboratory are often expensive and time-consuming. The use of modeling and calculation methods can obtain potential drug candidates with high accuracy in a short period of time. Using these optimized objects for "wet experiment" verification, you can Significantly reduce the workload and speed up the R&D process.

There are research results of virus-related drug screening, such as using molecular dynamics simulation and free energy calculation to simulate the virus invasion process and target binding status, and then deduce the optimal drug, and also identify potentially effective drugs by making point mutations based on structural information , these methods often have the shortcomings of being time-consuming or relying too much on manual intervention. In addition, the information available for newly discovered viruses is generally scarce, so the predictive performance of such models is also not ideal.

Contents of the invention

The invention provides an antiviral drug screening method, system and storage medium based on joint projection, which can accurately and efficiently predict antiviral related drugs according to virus-drug association pairs, virus genome sequence and drug chemical structure data.

The first aspect of the embodiment of this specification discloses a method for screening antiviral drugs based on joint projection, including the following steps:

S1. Construct the adjacency matrix of virus-drug association;

S2. Based on the adjacency matrix associated with the virus-drug, calculate the virus Gaussian distance similarity matrix and the drug Gaussian distance similarity matrix;

S3. Calculate the similarity matrix of the viral gene sequence based on the viral genome sequence, and calculate the similarity matrix of the chemical structure of the drug based on the chemical structure of the drug;

S4. Based on the virus Gaussian distance similarity matrix and the viral gene sequence similarity matrix, use a fast kernel learning method to integrate the virus integration similarity matrix; based on the drug Gaussian distance similarity matrix and the drug chemical structure similarity matrix, use a fast kernel learning method , to obtain the drug integration similarity matrix;

S5. Based on the adjacency matrix, virus integration similarity matrix and drug integration similarity matrix of the virus-drug association, use the sparse regularization joint projection method to construct a loss function;

S6. Solving the loss function to obtain a virus-drug prediction score matrix;

S7. Based on the virus-drug prediction score matrix, the score of the row where the target virus is located is screened out, and the final prediction result is obtained after sorting.

In the embodiment disclosed in this specification, in S1:

Input the known virus-drug association pairs to construct the adjacency matrix Y of virus-drug association;

If it is a known association pair, the corresponding position is 1, otherwise it is 0;

The number of rows of the adjacency matrix Y is the number of viruses nv, and the number of columns is the number of drugs nd.

In the embodiment disclosed in this specification, in S2:

If there is a relationship between the drug d(i) and a certain virus, the corresponding position is recorded as 1, otherwise it is recorded as 0, forming a vector of 0 or 1 with a size of 1×nv, which is recorded as the vector of drug d(i) Vector spectrum IP(d(i)), then calculate the Gaussian distance similarity between drugs d(i) and d(j):

；

;

In the above formula, IP(d(j)) is the vector spectrum of the drug d(j); the parameter γ _d is used to control the nuclear bandwidth, which is obtained by normalizing the new bandwidth parameter γ' _d :

；

;

Define the Gaussian distance similarity between viruses v(i) and v(j) in a similar way, and get a vector composed of 0 or 1 with a size of 1×nd, which is recorded as the vector spectrum IP(v (i)), calculate the Gaussian distance similarity between viruses v(i) and v(j):

；

;

IP(v(j)) is the vector spectrum of the virus v(j); the parameter γ _v is used to control the core bandwidth, which is obtained by normalizing the new bandwidth parameter γ' _v :

；

;

Both γ' _d and γ' _v above are constants.

In the embodiment disclosed in this specification, in S3:

Based on the viral genome sequence, the viral gene sequence similarity matrix is calculated using the multiple sequence alignment method;

Based on the chemical structure of the drug, the MACCS fingerprint of the drug is obtained, and the Tanimoto coefficient is used to calculate the similarity matrix of the drug chemical structure.

In the embodiment disclosed in this specification, in S4:

The semi-positive definite programming formula of described fast nuclear learning method is:

；

;

In the formula, the first item is the reconstruction loss norm item, which indicates the integration error of the similarity matrix; the second item is the regularization item, which is used to avoid overfitting; where Y is the virus-drug association adjacency matrix, S _j ^v (j=1,2) represent the virus Gaussian distance similarity matrix and virus gene sequence similarity matrix respectively, μ ^v is the regularization parameter, λ ^v ∈ R ^1×2 is the coefficient to be solved, and the virus integration similarity matrix is obtained by λ ^v :

；

;

Similarly, the integration parameter λ ^d ∈ R ^{1×2 of} the drug chemical structure similarity matrix and the drug Gaussian distance similarity matrix can be obtained according to the above, and then the drug integration similarity matrix is calculated:

；

;

Among them, S _j ^d (j=1,2) represent the drug Gaussian distance similarity matrix and the drug chemical structure similarity matrix, respectively.

In the embodiment disclosed in this specification, in S5:

The loss function constructed using the sparse regularized joint projection method is as follows:

；

;

为流形正则化项；W₁∈R^d1×nd和W₂∈R^d2×nv是待求系数矩阵；P∈R^d1×d2是投影矩阵；J(W₁,W₂,P)表示将等号右侧部分的值记为J，且J随变量W₁、W₂和P而变化；Y∈R^nv×nd是已知的病毒-药物关联矩阵，nv和nd分别表示病毒数量和药物数量；矩阵A和B是病毒整合相似矩阵S_v∈R^nv×nv和药物整合相似矩阵S_d∈R^nd×nd分解得到的低秩矩阵，即S_v≈AA^T，S_d≈BB^T，A∈R^nv×d1和B∈R^nd×d2表示病毒与药物的潜在特征，d₁和d₂分别表示矩阵S_v和S_d的秩；λ₁、λ₂、λ₃和α是正则化参数；||·||_F表示Frobenius范数，||·||_*表示核范数，||·||_2,1表示L2,1范数；L₁∈R^n×n和L₂∈R^m×m是图归一化拉普拉斯矩阵，计算方法为：In the formula

is the manifold regularization item; W ₁ ∈ R ^d1×nd and W ₂ ∈ R ^d2×nv are the coefficient matrices to be sought; P ∈ R ^d1×d2 is the projection matrix; J(W ₁ ,W ₂ ,P) represents the The value on the right side of the equal sign is denoted as J, and J varies with variables W ₁ , W ₂ and P; Y∈R ^nv×nd is a known virus-drug correlation matrix, and nv and nd represent the number of viruses and drugs, respectively Quantity; matrices A and B are low-rank matrices obtained by decomposing virus integration similarity matrix S _v ∈ R ^nv×nv and drug integration similarity matrix S _d ∈ R ^{nd ×nd} , that is, S _v ≈ AA ^T , S _d ≈ BB ^T , A∈R ^nv×d1 and B∈R ^nd×d2 denote the potential features of viruses and drugs, d ₁ and d ₂ denote the ranks of matrices S _v and S _d respectively; λ ₁ , λ ₂ , λ ₃ and α are the regularization Parameters; ||·|| _F represents the Frobenius norm, ||·|| _* represents the nuclear norm, ||·|| _2,1 represents the L2,1 norm; L ₁ ∈ R ^n×n and L ₂ ∈ R ^m×m is the graph normalized Laplacian matrix, the calculation method is:

；

;

；

;

where D _v ∈ ^Rnv×nv and D _d ∈ R ^nd×nd are diagonal matrices, and their element values are calculated as:

。

.

In the embodiment disclosed in this specification, in S6:

When solving the construction loss function, the matrix J is derived from the matrix W ₁ and W ₂ respectively, and then set to be equal to 0, the calculation formulas of the matrix W ₁ and W ₂ can be obtained as follows:

；

;

；

;

Among them, the matrices Z ₁ ∈ R ^d1×d1 and Z ₂ ∈ R ^d2×d2 are diagonal matrices, satisfying:

；

;

In the above formula, ||w ₁ ⁱ || ₂ and ||w ₂ ⁱ || ₂ represent the 2-norm of row i of matrix W ₁ and W ₂ respectively;

When solving the partial derivative of the nuclear norm of the matrix P, the singular value decomposition (SVD) of the matrix P is first performed, that is, P=UΣV ^T , where U∈R ^d1×d1 and V∈R ^d2×d2 ; then construct E∈R ^{d2× d1} , if d ₁ >d ₂ , then E is the first d ₂ rows of the identity matrix I∈R ^d1×d1 ; otherwise, take the first d ₁ rows of the identity matrix I∈R ^d2×d2 ; the matrix P is calculated by the following method:

；

;

The second aspect of the embodiment of this specification discloses an antiviral drug screening system based on joint projection, including:

Adjacency matrix building blocks for constructing adjacency matrices of virus-drug associations;

The Gaussian distance similarity matrix calculation module is used to calculate the virus Gaussian distance similarity matrix and the drug Gaussian distance similarity matrix based on the adjacency matrix associated with the virus-drug;

The viral gene sequence similarity matrix and medicinal chemical structure similarity matrix calculation module is used to calculate the viral gene sequence similarity matrix based on the viral genome sequence, and to calculate the medicinal chemical structure similarity matrix based on the medicinal chemical structure;

The integrated similarity matrix calculation module is used to integrate and obtain the virus integration similarity matrix based on the virus Gaussian distance similarity matrix and the viral gene sequence similarity matrix based on the virus Gaussian distance similarity matrix and the drug chemical structure similarity matrix; , use the fast kernel learning method to integrate the drug integration similarity matrix;

The loss function construction module is used to construct a loss function based on the adjacency matrix of the virus-drug association, the virus integration similarity matrix and the drug integration similarity matrix, using the sparse regularization joint projection method;

The loss function solving module is used to solve the loss function to obtain the virus-drug prediction score matrix;

The prediction module is used to screen out the score of the row where the target virus is located based on the virus-drug prediction score matrix, and obtain the final prediction result after sorting.

In the embodiment disclosed in this specification, the antiviral drug screening system based on joint projection further includes:

The processor is respectively connected with the adjacency matrix construction module, Gaussian distance similarity matrix calculation module, virus gene sequence similarity matrix and medicinal chemical structure similarity matrix calculation module, integrated similarity matrix calculation module, loss function construction module, loss function solution module and prediction module connection;

The memory is connected to the processor and stores 455 proven human virus-drug interaction machine programs that can be run on the processor;

Wherein, when the processor executes the computer program, the processor controls the adjacency matrix construction module, the Gaussian distance similarity matrix calculation module, the viral gene sequence similarity matrix and drug chemical structure similarity matrix calculation module, and the integration similarity matrix The calculation module, the loss function construction module, the loss function solution module and the prediction module work to realize the joint projection-based antiviral drug screening method described in any one of the above.

The third aspect of the embodiments of this specification discloses a computer-readable storage medium, the storage medium stores computer instructions, and when the computer reads the computer instructions, the computer executes any one of the above-mentioned joint-based Projected Antiviral Drug Screening Methods.

In summary, the present invention has at least the following beneficial effects:

The present invention calculates virus Gaussian distance similarity matrix and drug Gaussian distance similarity matrix respectively by constructing virus-drug association adjacency matrix; uses virus genome sequence to calculate virus gene sequence similarity matrix, uses chemical structure information of medicine to calculate drug chemical structure similarity matrix; The virus-drug integration similarity matrix and drug integration similarity matrix were calculated using the fast kernel learning method; the loss function was constructed by combining matrix decomposition and sparse joint projection method, and the virus-drug association prediction score matrix was obtained by iterative solution, and the final result was obtained by screening and sorting. The invention can effectively overcome the adverse influence of noise data, quickly, accurately and efficiently screen effective drugs for virus treatment, and avoid the problems of long time-consuming and high cost of biomedical experiment methods.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. Those of ordinary skill in the art can also obtain other drawings based on these drawings without any creative effort.

FIG. 1 is a schematic diagram of the steps of the joint projection-based antiviral drug screening method involved in the present invention.

Fig. 2 is a schematic flowchart of the antiviral drug screening method based on joint projection involved in the present invention.

Fig. 3 is a comparison chart of the results of the five-fold cross-validation between the joint projection-based antiviral drug screening method involved in the present invention and the baseline method.

Fig. 4 is a schematic diagram of the antiviral drug screening system based on joint projection involved in the present invention.

Detailed ways

In the following, only some exemplary embodiments are briefly described. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. Accordingly, the drawings and descriptions are to be regarded as illustrative in nature and not restrictive.

The following disclosure provides many different implementations or examples for realizing different structures of the embodiments of the present invention. To simplify the disclosure of the embodiments of the present invention, components and arrangements of specific examples are described below. Of course, they are only examples and are not intended to limit the embodiments of the present invention. Furthermore, embodiments of the present invention may repeat reference numerals and/or reference letters in different instances, such repetition is for simplicity and clarity and does not in itself indicate a relationship between the various embodiments and/or arrangements discussed.

Embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.

It should be noted that the known human drug-virus association data used in the examples of this specification are collected from relevant literature, and text mining technology is used to sort out the experimentally verified drug-virus interaction pairs reported in the literature Finally, 455 confirmed human virus-drug interaction pairs were obtained, involving 34 viruses and 219 drugs (document DOI: 10.1016/j.asoc.2021.107135); the chemical structure of the drug was downloaded from the DrugBank database, and the nucleotides of the viral genome Sequences were obtained from the NCBI database of the National Center for Biotechnology Information.

As shown in Figure 1 and Figure 2, the first aspect of the embodiment of this specification discloses a method for screening antiviral drugs based on joint projection, including the following steps:

S1. Construct the adjacency matrix of virus-drug association.

Input the known virus-drug association pairs and construct the adjacency matrix Y:

；

;

The Y element of the obtained adjacency matrix is 0 or 1, the size is 34 rows×219 columns, and the value range of i and j satisfies 1≤i≤34, 1≤j≤219.

S2. Based on the adjacency matrix of the virus-drug association, calculate the virus Gaussian distance similarity matrix and the drug Gaussian distance similarity matrix.

If there is a relationship between the drug d(i) and a certain virus, the corresponding position is recorded as 1, otherwise it is recorded as 0, forming a vector of 0 or 1 with a size of 1×34, which is recorded as the vector of drug d(i) Vector spectrum IP(d(i)), then calculate the Gaussian distance similarity between drugs d(i) and d(j):

；

;

In the above formula, the vector spectrum IP(d(j)) of the drug d(j), the parameter γ _d is used to control the nuclear bandwidth, obtained by normalizing the new bandwidth parameter γ' _d :

；

;

Define the Gaussian distance similarity between viruses v(i) and v(j) in a similar way. If there is an association between a certain virus v(i) and a certain drug, the corresponding position is recorded as 1, otherwise it is recorded as 0 , forming a vector of 0 or 1 with a size of 1×219, which is recorded as the vector spectrum IP(v(i)) of the virus v(i), and the Gaussian distance between the virus v(i) and v(j) is calculated similarly sex:

；

;

The vector spectrum IP(v(j)) of the virus v(j), the parameter γ _v is used to control the core bandwidth, obtained by normalizing the new bandwidth parameter γ' _v :

；

;

The above γ' _d and γ' _v are both constants, take γ' _d = γ' _v = 1; where nv represents the number of viruses, which is 34 in this example, and nd represents the number of drugs, which is 219 in this example, this step After calculation, a symmetric matrix S ₁ ^v (virus Gaussian distance similarity matrix) with a size of 34×34 and a symmetric matrix S ₁ ^d (drug Gaussian distance similarity matrix) with a size of 219×219 are obtained, and the element values of these two matrices are all in Between 0 and 1.

S3. Calculate the similarity matrix of the viral gene sequence based on the viral genome sequence, and calculate the similarity matrix of the chemical structure of the drug based on the chemical structure of the drug.

Input the viral genome sequence, use the multiple sequence alignment tool MAFFT to calculate the similarity matrix S ₂ ^v of the viral gene sequence; input the chemical structure of the drug represented by the SMILES code, and then use the cheminformatics software RDKit or Open Babel to obtain the molecular access system fingerprint of the drug ( MACCS), and then use the R package RxnSim to calculate the Tanimoto similarity to obtain the drug chemical structure similarity matrix S ₂ ^d , the specific calculation method is, for two drugs d(i) and d(j), the MACCS fragments of the two drugs The string sets expressed in binary are respectively recorded as D(i) and D(j), and the similarity S ^d _ij value between d(i) and d(j) can be calculated by the following formula:

；

;

Based on the virus Gaussian distance similarity matrix and the virus gene sequence similarity matrix, use a fast kernel learning method to integrate the virus integration similarity matrix; based on the drug Gaussian distance similarity matrix and the drug chemical structure similarity matrix, use a fast kernel learning method to integrate Get the drug integration similarity matrix.

The semi-positive definite programming formula of described fast nuclear learning method is:

；

;

In the formula, the first item is the reconstruction error item, indicating the size of the integration error of the similarity matrix; the second item is the regularization item, which is used to avoid over-fitting; where Y is the virus-drug association adjacency matrix, S _j ^v (j =1, 2) represent the virus Gaussian distance similarity matrix and the virus gene sequence similarity matrix respectively, μ ^v is the regularization parameter, λ ^v ∈ R ^1×2 is the coefficient to be solved, and the CVX toolbox in Matlab software can be used to solve this As an optimization problem, the virus integration similarity matrix is obtained through ^λv :

；

;

Similarly, the integration parameter λ ^d ∈ R ^1×2 of the drug structure similarity matrix and the drug Gaussian distance similarity matrix can be obtained according to the above, and then the drug integration similarity matrix is calculated:

；

;

Among them, S _j ^d (j=1,2) represent the drug Gaussian distance similarity matrix and the drug chemical structure similarity matrix, respectively.

S5. Based on the virus-drug association adjacency matrix, virus integration similarity matrix, and drug integration similarity matrix, a loss function is constructed using a sparse regularization joint projection method.

The loss function constructed using the sparse regularized joint projection method is as follows:

；

;

为流形正则化项；W₁∈R^d1×nd和W₂∈R^d2×nv是待求系数矩阵；P∈R^d1×d2是投影矩阵；J(W₁,W₂,P)表示将等号右侧部分的值记为J，且J随变量W₁，W₂和P而变化；Y∈R^nv×nd是已知的病毒-药物关联矩阵，nv和nd分别表示病毒数量和药物数量；矩阵A和B是病毒整合相似矩阵S_v∈R^nv×nv和药物整合相似矩阵S_d∈R^nd×nd分解得到的低秩矩阵，即S_v≈AA^T，S_d≈BB^T，A∈R^nv×d1和B∈R^nd×d2表示病毒与药物的潜在特征，d₁和d₂分别表示矩阵S_v和S_d的秩；λ₁、λ₂、λ₃和α是正则化参数；||·||_F表示Frobenius范数，||·||_*表示核范数，||·||_2,1表示L_2,1范数；L₁∈R^n×n和L₂∈R^m×m是图归一化拉普拉斯矩阵，计算方法为：In the formula

is the manifold regularization item; W ₁ ∈ R ^d1×nd and W ₂ ∈ R ^d2×nv are the coefficient matrices to be sought; P ∈ R ^d1×d2 is the projection matrix; J(W ₁ ,W ₂ ,P) represents the The value on the right side of the equal sign is denoted as J, and J varies with variables W ₁ , W ₂ and P; Y∈R ^nv×nd is a known virus-drug correlation matrix, and nv and nd represent the number of viruses and drugs, respectively Quantity; matrices A and B are low-rank matrices obtained by decomposing virus integration similarity matrix S _v ∈ R ^nv×nv and drug integration similarity matrix S _d ∈ R ^{nd ×nd} , that is, S _v ≈ AA ^T , S _d ≈ BB ^T , A∈R ^nv×d1 and B∈R ^nd×d2 denote the potential features of viruses and drugs, d ₁ and d ₂ denote the ranks of matrices S _v and S _d respectively; λ ₁ , λ ₂ , λ ₃ and α are the regularization Parameters; ||·|| _F represents the Frobenius norm, ||·|| _* represents the nuclear norm, ||·|| _2,1 represents the L _2,1 norm; L ₁ ∈ R ^n×n and L ₂ ∈R ^m×m is the graph normalized Laplacian matrix, calculated as:

；

;

；

;

where D _v ∈ ^Rnv×nv and D _d ∈ ^{R nd×nd} are diagonal matrices, and their element values are calculated as:

。

.

The loss function is solved to obtain a virus-drug prediction score matrix.

When solving the construction loss function, derivate J with respect to W ₁ and W ₂ respectively, and then make it equal to 0, the calculation formulas of matrix W ₁ and W ₂ can be obtained as follows:

；

;

；

;

Among them, the matrices Z ₁ ∈ R ^d1×d1 and Z ₂ ∈ R ^d2×d2 are diagonal matrices, satisfying:

；

;

In the above formula, ||w ₁ ⁱ || ₂ and ||w ₂ ⁱ || ₂ represent the 2-norm of row i of matrix W ₁ and W ₂ respectively;

When solving the partial derivative of the nuclear norm of the matrix P, the singular value decomposition (SVD) of the matrix P is first performed, that is, P=UΣV ^T , where U∈R ^d1×d1 and V∈R ^d2×d2 ; then construct E∈R ^{d2× d1} , if d ₁ >d ₂ , then E is the first d ₂ rows of the identity matrix I∈R ^d1×d1 ; otherwise, take the first d ₁ rows of the identity matrix I∈R ^d2×d2 ; the matrix P is calculated by the following method:

；

;

S7. Based on the virus-drug prediction score matrix, the score of the row where the target virus is located is screened out, and the final prediction result is obtained after sorting.

Iteratively calculate the matrix in the above steps until convergence, and get the virus-drug prediction score matrix:

；

;

The scores of rows corresponding to specific viruses are screened out, and the final prediction results are obtained after sorting.

In the above example, the regularization parameters λ ₁ =1, λ ₂ =32, λ ₃ =2 ^-1 and α=2 ^-5 are selected after preliminary optimization, and the parameters d ₁ =20, d ₂ =30; using Matlab When programming to implement the above algorithm, the matrix W ₁ is randomly initialized to a matrix of 20 rows × 219 columns, W ₂ is randomly initialized to a matrix of 30 rows × 34 columns, and the matrix P is randomly initialized to a matrix of 20 rows × 30 columns. All the elements of the matrix are in the (0,1) interval range; the matrix inversion all uses the pseudo-inverse operation function pinv(); when the number of times is set to 5, the iteration is exited, and the predicted score matrix Y ^* is obtained after the loop runs, and the program ends. Filter and sort to get the final prediction result.

Validity verification of the present invention:

The antiviral drug screening method based on joint projection as shown in Figure 1 and Figure 2 uses five-fold cross-validation for predictive performance evaluation. The specific implementation method is: first, all known drug-virus associations are randomly and evenly divided into 5 groups, Then, each of the 5 groups is set as a test sample in turn, and the other groups are used as training samples (when the selection of test samples is different, the Gaussian distance similarity matrix calculated by relying on the test samples will also change accordingly). Predictions are obtained using the training samples as input to the method, and finally the predicted score of each test sample in the set is compared with the score of the candidate samples. In order to reduce the impact of random division on the results in the process of generating test samples, 100 times of 5-fold cross-validation was performed.

The following data were obtained after calculation using Matlab programming. Figure 3 shows the comparison of the AUROC (area under the ROC curve) value between SRJPVDA of this method and several virus-drug screening models that have been reported. This method achieved an AUROC value of 0.9075±0.0056 in the five-fold cross-validation, showing better predictive performance than several classical models.

On the other hand, for a specific virus, such as the new coronavirus (SARS-CoV-2), use this method to make predictions, and filter the row corresponding to SARS-CoV-2 in the scoring matrix Y ^* to obtain the prediction score of the new crown-related drugs , 16 of the top 20 drugs in descending order can be supported by the reported literature.

The following table shows the top 20 drug names and supporting literature PMID numbers in the prediction results.

to sort drug name PMID number of the supporting document 1 Ribavirin 33689451 2 Chloroquine 33906514 3 Nitazoxanide 36332361 4 N4-Hydroxycytidine 35492218 5 Camostat 35692220 6 Amantadine 35390511 7 Niclosamide 34664162 8 Mizoribine 17336519 9 Mycophenolic Acid 32579258 10 Gemcitabine 32432977 11 Betulinic Acid not yet confirmed 12 Glycyrrhizic Acid 33041173 13 Berberine 36183284 14 Remdesivir 35221670 15 Alisporivir 32376613 16 Umifenovir 36245851 17 Favipiravir 35692220, 36332361 18 Memantine 32828269 19 Lopinavir 32251767 20 Artemisinin not yet confirmed

In summary, the advantages of the present invention:

1. By introducing the L _2,1 norm constraint term, the loss function generates a sparse solution, which can effectively reduce the influence of the inherent noise in the training data set, and thus make the virus-drug association prediction results more robust and accurate;

2. The joint projection method effectively integrates the virus similarity and drug similarity information, making the model more scalable and robust, thus obtaining better prediction results;

3. It integrates the manifold learning theory by integrating the Laplacian term, which is a semi-supervised model and can efficiently use negative sample information to improve the prediction performance of unknown correlation pairs.

As shown in Figure 4, the second aspect of the embodiment of this specification discloses an antiviral drug screening system based on joint projection, including:

Adjacency matrix building blocks for constructing adjacency matrices of virus-drug associations;

The Gaussian distance similarity matrix calculation module is used to calculate the virus Gaussian distance similarity matrix and the drug Gaussian distance similarity matrix based on the adjacency matrix associated with the virus-drug;

The viral gene sequence similarity matrix and medicinal chemical structure similarity matrix calculation module is used to calculate the viral gene sequence similarity matrix based on the viral genome sequence, and to calculate the medicinal chemical structure similarity matrix based on the medicinal chemical structure;

The integrated similarity matrix calculation module is used to integrate and obtain the virus integration similarity matrix based on the virus Gaussian distance similarity matrix and the viral gene sequence similarity matrix based on the virus Gaussian distance similarity matrix and the drug chemical structure similarity matrix; , use the fast kernel learning method to integrate the drug integration similarity matrix;

The loss function construction module is used to construct a loss function based on the adjacency matrix of the virus-drug association, the virus integration similarity matrix and the drug integration similarity matrix, using the sparse regularization joint projection method;

The loss function solving module is used to solve the loss function to obtain the virus-drug prediction score matrix;

The prediction module is used to screen out the score of the row where the target virus is located based on the virus-drug prediction score matrix, and obtain the final prediction result after sorting.

In the embodiment disclosed in this specification, the antiviral drug screening system based on joint projection further includes:

The processor is respectively connected with the adjacency matrix construction module, Gaussian distance similarity matrix calculation module, virus gene sequence similarity matrix and medicinal chemical structure similarity matrix calculation module, integrated similarity matrix calculation module, loss function construction module, loss function solution module and prediction module connection;

a memory connected to the processor and storing a computer program operable on the processor;

Wherein, when the processor executes the computer program, the processor controls the adjacency matrix construction module, the Gaussian distance similarity matrix calculation module, the viral gene sequence similarity matrix and drug chemical structure similarity matrix calculation module, and the integration similarity matrix The calculation module, the loss function construction module, the loss function solution module and the prediction module work to realize the joint projection-based antiviral drug screening method described in any one of the above.

The third aspect of the embodiments of this specification discloses a computer-readable storage medium, the storage medium stores computer instructions, and when the computer reads the computer instructions, the computer executes any one of the above-mentioned joint-based Projected Antiviral Drug Screening Method

The above-mentioned embodiments are used to illustrate the present invention, not to limit the present invention, so the change of the numerical value or the replacement of equivalent elements should still belong to the scope of the present invention.

From the above detailed description, those skilled in the art can understand that the present invention can indeed achieve the above-mentioned purpose, and it is in compliance with the provisions of the patent law.

While preferred embodiments of the invention have been described, additional changes and modifications to these embodiments can be made by those skilled in the art once the basic inventive concept is appreciated. Therefore, it is intended that the appended claims be construed to cover the preferred embodiment as well as all changes and modifications which fall within the scope of the invention. The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. It should be noted that any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention should include Within the protection scope of the present invention.

It should be noted that, the above descriptions about the process are only for illustration and description, and do not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to the flow under the guidance of this specification. However, such modifications and changes are still within the scope of this specification.

The basic concepts have been described above. Obviously, for those of ordinary skill in the art after reading this application, the above invention disclosure is only an example and does not constitute a limitation to this application. Although not expressly stated herein, various modifications, improvements, and amendments to this application may occur to those skilled in the art. Such modifications, improvements, and amendments are suggested in this application, so such modifications, improvements, and amendments still belong to the spirit and scope of the exemplary embodiments of this application.

Meanwhile, the present application uses specific words to describe the embodiments of the present application. For example, "one embodiment", "an embodiment", and/or "some embodiments" means a certain feature, structure or characteristic related to at least one embodiment of the present application. Therefore, it should be emphasized and noticed that "an embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more in different positions in this specification do not necessarily refer to the same embodiment. In addition, certain features, structures or characteristics of one or more embodiments of the present application may be properly combined.

Furthermore, those of ordinary skill in the art will appreciate that aspects of the present application may be illustrated and described in several patentable categories or circumstances, including any new and useful process, machine, product, or combination of matter, or any new and useful improvements to it. Therefore, various aspects of the present application may be fully implemented by hardware, may be fully implemented by software (including firmware, resident software, microcode, etc.), or may be implemented by a combination of hardware and software. The above hardware or software may be referred to as a "unit", "module" or "system". Furthermore, aspects of the present application may take the form of a computer program product embodied on one or more computer-readable media, with computer-readable program code embodied therein.

The computer program codes required for the operation of each part of the application can be written in any one or more programming languages, including object-oriented programming languages such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc. Programming language, such as C programming language, VisualBasic, Fortran2103, Perl, COBOL2102, PHP, ABAP conventional programming language, such as Python, Ruby and Groovy dynamic programming language or other programming languages, etc. The program code may run entirely on the user's computer, or as a stand-alone software package, or partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. In the latter case, the remote computer can be connected to the user computer through any form of network, such as a local area network (LAN) or wide area network (WAN), or to an external computer (such as through the Internet), or in a cloud computing environment, or as a service Use software as a service (SaaS).

In addition, unless explicitly stated in the claims, the order of processing elements and sequences described in the application, the use of numbers and letters, or the use of other designations are not used to limit the order of the flow and methods of the application. While the foregoing disclosure has discussed by way of various examples some embodiments of the invention that are presently believed to be useful, it should be understood that such detail is for illustrative purposes only and that the appended claims are not limited to the disclosed embodiments, but rather, the claims The claims are intended to cover all modifications and equivalent combinations that fall within the spirit and scope of the embodiments of the application. For example, while the implementation of the various components described above could be embodied in a hardware device, it could also be implemented as a pure software solution, for example, as an installation on an existing server or mobile device.

In the same way, it should be noted that in order to simplify the expression disclosed in the present application and help the understanding of one or more embodiments of the invention, in the foregoing descriptions of the embodiments of the present application, sometimes multiple features are combined into one embodiment , drawings or descriptions thereof. This method of application, however, is not to be interpreted as reflecting an intention that the claimed subject matter requires more features than are expressly recited in each claim. Rather, the subject matter of the invention should have fewer features than a single embodiment described above.

Claims (9)

1. The antiviral drug screening method based on the combined projection is characterized by comprising the following steps of:

s1, constructing an adjacency matrix of virus-drug association;

s2, calculating a virus Gaussian distance similarity matrix and a drug Gaussian distance similarity matrix based on the adjacent matrix of the virus-drug association;

s3, calculating a virus gene sequence similarity matrix based on a virus genome sequence, and calculating a drug chemical structure similarity matrix based on a drug chemical structure;

s4, based on the viral Gaussian distance similarity matrix and the viral gene sequence similarity matrix, integrating by using a fast kernel learning method to obtain a viral integration similarity matrix; based on the Gaussian distance similarity matrix of the medicine and the chemical structure similarity matrix of the medicine, a rapid kernel learning method is used for integrating to obtain a medicine integration similarity matrix;

s5, constructing a loss function by using a sparse regularization joint projection method based on the adjacent matrix, the virus integration similarity matrix and the drug integration similarity matrix of the virus-drug association;

s6, solving the loss function to obtain a virus-drug prediction score matrix;

s7, screening out the scores of the rows of the target viruses based on the virus-medicine prediction score matrix, and sequencing to obtain a final prediction result;

in S5:

the construction loss function using sparse regularized joint projection method is as follows:

wherein R (A, B, W) ₁ ,W ₂ )＝tr((AW ₁ ) ^T L ₁ (AW ₁ ))+tr((BW ₂ ) ^T L ₂ (BW 2)) is a manifold regularization term; w (W) ₁ ∈R ^d1×nd And W is ₂ ∈R ^d2×nv Is a coefficient matrix to be solved; p epsilon R ^d1×d2 Is a projection matrix; j (W) ₁ ,W ₂ P) denotes that the value of the right part of the equal sign is recorded as J, and J follows the variable W ₁ 、W ₂ And P varies; y εR ^nv×nd Is a known virus-drug association matrix, nv and nd represent the number of viruses and the number of drugs, respectively; the matrices A and B are virus integration similarity matrices S _v ∈R ^nv×nv And drug integration similarity matrix S _d ∈R ^{nd ×nd} Low rank matrix obtained by decomposition, i.e. S _v ≈AA ^T ，S _d ≈BB ^T ，A∈R ^nv×d1 And B.epsilon.R ^nd×d2 Representing potential characteristics of virus and drug, d1 and d2 respectively represent matrix S _v And S is _d Rank of (c); lambda (lambda) ₁ 、λ ₂ 、λ ₃ And α is a regularization parameter; I.I _F Representing the Frobenius norms, |·|| _* The number of kernels is represented by a kernel norm, I.I _2,1 Represents L _2,1 A norm; l (L) ₁ ∈R ^n×n And L ₂ ∈R ^m×m The method is a graph normalized Laplace matrix, and comprises the following steps:

wherein D is _v ∈R ^nv×nv And D _d ∈R ^nd×nd Is a diagonal matrix, and the element values are calculated in the following way:

2. the method for screening antiviral drugs based on joint projection according to claim 1, wherein in S1:

inputting a known virus-drug association pair to construct a adjacency matrix Y of the virus-drug association;

if the correlation pair is known, the corresponding position is 1, otherwise, the correlation pair is 0;

the row number of the adjacent matrix Y is the virus number nv, and the column number is the medicine number nd.

3. The method for screening antiviral drugs based on joint projection according to claim 1, wherein in S2:

if the association exists between the medicine d (i) and a certain virus, the corresponding position is marked as 1, otherwise, the corresponding position is marked as 0, a vector formed by 0 or 1 with the size of 1 Xnv is formed, the vector spectrum is marked as the vector spectrum IP (d (i)) of the medicine d (i), nv is the number of viruses, and then the Gaussian distance similarity between the medicines d (i) and d (j) is calculated:

S ^d (d(i)，d(j))＝exp(-γ _d ||IP(d(i))-IP(d(j))|| ² )；

in the above formula, IP (d (j)) is the vector spectrum of the drug d (j); parameter gamma _d For controlling the nuclear bandwidth by normalizing the new bandwidth parameter gamma' _d Obtaining:

in a similar manner, defining the Gaussian distance similarity between viruses v (i) and v (j), obtaining a vector consisting of 0 or 1 with the size of 1×nd, namely a vector spectrum IP (v (i)) of the virus v (i), nd being the number of medicines, and calculating the Gaussian distance similarity between the viruses v (i) and v (j):

S ^v (v(i)，v(j))＝exp(-γ _v ||IP(v(i))-IP(v(j))|| ² )；

IP (v (j)) is the vector spectrum of virus v (j); parameter gamma _v For controlling the nuclear bandwidth by normalizing the new bandwidth parameter gamma' _v Obtaining:

above gamma' _d And gamma' _v Are constant.

4. The method for screening antiviral drugs based on joint projection according to claim 1, wherein in S3:

based on the viral genome sequence, calculating a viral gene sequence similarity matrix by using a multi-sequence alignment method;

based on the chemical structure of the medicine, the MACS fingerprint of the medicine is obtained, and the valley coefficient is adopted to calculate the chemical structure similarity matrix of the medicine.

5. The method for screening antiviral drugs based on joint projection according to claim 1, wherein in S4:

the semi-positive programming formula of the fast kernel learning method is as follows:

wherein, the first term is a reconstruction loss norm term and represents the magnitude of the integration error of the similarity matrix; the second term is a regularization term, which is used to avoid overfitting; wherein Y is a virus-drug association adjacency matrix, S _j ^v (j=1, 2) respectively represent a viral Gaussian distance similarity matrix and a viral gene sequence similarity matrix, μ ^v For regularization parameters, lambda ^v ∈R ^1×2 For the coefficients to be solved, by lambda ^v Obtaining a virus integration similarity matrix S _v ：

Similarly, the integration parameter lambda of the drug structure similarity matrix and the drug Gaussian distance similarity matrix can be obtained according to the above ^d ∈R ^1×2 Then calculate the drug integration similarity matrix S _d ：

Wherein S is _j ^d (j=1, 2) represents a pharmaceutical gaussian distance similarity matrix and a pharmaceutical chemical structure similarity matrix, respectively.

6. The antiviral drug screening method based on joint projection as claimed in claim 1, wherein in S6:

matrix W ₁ And W is ₂ The calculation formula of (2) is as follows:

W ₁ ＝(2A ^T A+2αA ^T L ₁ A+λ ₁ Z ₁ ) ^-1 2A ^T Y；

W ₂ ＝(2B ^T B+2αB ^T L ₂ B+λ ₂ Z ₂ ) ^-1 2B ^T Y ^T ；

wherein L is ₁ ∈R ^n×n And L ₂ ∈R ^m×m Is a graph normalized Laplace matrix, Y epsilon R ^nv×nd Is a known virus-drug association matrix, and matrices A and B are virus integration similarity matrices S _v ∈R ^nv×nv And drug integration similarity matrix S _d ∈R ^nd×nd Low rank matrix obtained by decomposition, i.e. S _v ≈AA ^T ，S _d ≈BB ^T ，A∈R ^nv×d1 And B.epsilon.R ^nd×d2 Representing potential characteristics of virus and drug, d1 and d2 respectively represent matrix S _v And S is _d Rank of (c); w (W) ₁ ∈R ^d1×nd And W is ₂ ∈R ^d2×nv Is a coefficient matrix to be solved; nv and nd represent the number of viruses and the number of drugs, respectively; lambda (lambda) ₁ 、λ ₂ 、λ ₃ And α is a regularization parameter; matrix Z ₁ ∈R ^d1×d1 And Z ₂ ∈R ^d2×d2 Is a diagonal matrix, satisfying:

the above formula is Zhongzhi w ₁ ⁱ || ₂ And W ₂ ⁱ || ₂ Respectively represent matrix W ₁ And W is ₂ 2 norms of row i;

when solving the partial derivative of the kernel norm of the projection matrix P, first the matrix P is decomposed with singular values, i.e. p=uΣv ^T Wherein U is E R ^{d1 ×d1} And V.epsilon.R ^d2×d2 The method comprises the steps of carrying out a first treatment on the surface of the Then construct E.epsilon.R ^d2×d1 If d ₁ >d ₂ E is the identity matrix I epsilon R ^d1×d1 D front of (2) ₂ A row; otherwise, taking the identity matrix I E R ^d2×d2 D front of (2) ₁ A row; the matrix P is calculated by:

7. an antiviral drug screening system based on joint projection, comprising:

the adjacency matrix construction module is used for constructing an adjacency matrix of virus-drug association;

the Gaussian distance similarity matrix calculation module is used for calculating a viral Gaussian distance similarity matrix and a drug Gaussian distance similarity matrix by using the adjacent matrix of the virus-drug association;

the virus gene sequence similarity matrix and pharmaceutical chemical structure similarity matrix calculation module is used for calculating a virus gene sequence similarity matrix by using a virus genome sequence and calculating a pharmaceutical chemical structure similarity matrix by using a pharmaceutical chemical structure;

the integration similarity matrix calculation module is used for integrating the virus Gaussian distance similarity matrix and the virus gene sequence similarity matrix by using a fast kernel learning method to obtain a virus integration similarity matrix; integrating the drug Gaussian distance similarity matrix and the drug chemical structure similarity matrix by using a rapid kernel learning method to obtain a drug integration similarity matrix;

the loss function construction module is used for constructing a loss function by using a sparse regularization joint projection method based on the adjacent matrix, the virus integration similarity matrix and the drug integration similarity matrix of the virus-drug association;

the loss function solving module is used for solving the loss function to obtain a virus-medicine prediction score matrix; the prediction module is used for screening out the scores of the rows of the target viruses based on the virus-medicine prediction score matrix, and sequencing to obtain a final prediction result;

the construction loss function by using the sparse regularization joint projection method is as follows:

in the middle of

Regularizing the term for the manifold; w (W) ₁ ∈R ^d1×nd And W is ₂ ∈R ^d2×nv Is a coefficient matrix to be solved; p epsilon R ^d1×d2 Is a projection matrix; j (W) ₁ ,W ₂ P) denotes that the value of the right part of the equal sign is noted as J, and J followsVariable W ₁ 、W ₂ And P varies; y εR ^nv×nd Is a known virus-drug association matrix, nv and nd represent the number of viruses and the number of drugs, respectively; the matrices A and B are virus integration similarity matrices S _v ∈R ^nv×nv And drug integration similarity matrix S _d ∈R ^nd×nd Low rank matrix obtained by decomposition, i.e. S _v ≈AA ^T ，S _d ≈BB ^T ，A∈R ^nv×d1 And B.epsilon.R ^nd×d2 Representing potential characteristics of virus and drug, d1 and d2 respectively represent matrix S _v And S is _d Rank of (c); lambda (lambda) ₁ 、λ ₂ 、λ ₃ And α is a regularization parameter; I.I _F Representing the Frobenius norms, |·|| _* The number of kernels is represented by a kernel norm, I.I _2,1 Represents L _2,1 A norm; l (L) ₁ ∈R ^n×n And L ₂ ∈R ^m×m The method is a graph normalized Laplace matrix, and comprises the following steps:

wherein D is _v ∈R ^nv×nv And D _d ∈R ^nd×nd Is a diagonal matrix, and the element values are calculated in the following way:

8. the joint projection-based antiviral drug screening system of claim 7, further comprising:

the processor is respectively connected with the adjacent matrix construction module, the Gaussian distance similarity matrix calculation module, the viral gene sequence similarity matrix and pharmaceutical chemical structure similarity matrix calculation module, the integration similarity matrix calculation module, the loss function construction module, the loss function solving module and the prediction module;

a memory coupled to the processor and storing a computer program executable on the processor; wherein when the processor executes the computer program, the processor controls the adjacency matrix construction module, the gaussian distance similarity matrix calculation module, the viral gene sequence similarity matrix and pharmaceutical chemistry structure similarity matrix calculation module, the integration similarity matrix calculation module, the loss function construction module, the loss function solving module and the prediction module to work so as to realize the antiviral drug screening method based on joint projection according to any one of claims 1 to 6.

9. A computer readable storage medium storing computer instructions which, when read by a computer, perform the joint projection based antiviral drug screening method of any one of claims 1 to 6.

Images