Method and device for predicting multi-type miRNA-disease relation based on non-negative tensor decomposition of L2, 1 norm and self-paced learning

Abstract

The invention discloses a method and device for predicting a multi-type miRNA-disease relation through non-negative tensor decomposition based on L2, 1 norm and self-paced learning, and the method comprises the steps: obtaining a first relation tensor of miRNA-disease-type through non-negative tensor decomposition; obtaining an initial self-paced learning weight tensor through self-paced learning; constructing a hypergraph structure cost function through hypergraph learning; constructing an initial function of a target model according to the first relation tensor, the initial self-paced learning weight tensor and the hypergraph structure cost function in combination with the constraint of an L2, 1 norm; updating the initial function to obtain a target function; wherein the objective function is used for predicting the multi-type miRNA-disease relationship. According to the method, the accuracy of predicting the relationship between the multi-type miRNA and the disease can be improved, and the method can be widely applied to the technical field of bioinformatics.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics, and in particular to a method and device for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning. Background Art

[0002] At present, most computational methods only focus on the prediction of miRNA-disease binary relationships, resulting in the fact that the more complex underlying mechanisms between miRNA and disease are still not fully understood. In the existing scheme, miRNA functional similarity and disease semantic similarity are integrated into CP decomposition, and a tensor decomposition model with relational constraints is proposed to predict miRNA-disease relationships.

[0003] However, most of the existing models based on tensor decomposition face the following challenges: First, noisy samples make the model easy to fall into bad local optimal solutions, resulting in suboptimal results. Second, most models only consider pairwise relationships in homogeneous graphs, while ignoring high-order complex relationships in heterogeneous graphs. Moreover, these models rarely eliminate irrelevant disease pairs, making it difficult to improve the accurate identification of miRNAs associated with specific diseases. Finally, the prediction performance of the model may be affected by the high false negative rate (i.e., "there is a relationship, but we don't know it yet") between miRNAs and diseases. Summary of the invention

[0004] In view of this, the main purpose of the embodiments of the present invention is to provide a non-negative tensor decomposition based on L2,1 norm and self-paced learning to predict the relationship between multiple types of miRNA-disease, in order to solve at least one of the problems of the prior art. The present invention can improve the accuracy of predicting the relationship between multiple types of miRNA and diseases.

[0005] To achieve the above object, an embodiment of the present invention provides a method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning, the method comprising:

[0006] Obtain the first relationship tensor of miRNA-disease-type through non-negative tensor decomposition;

[0007] Through self-paced learning, the initial self-paced learning weight tensor is obtained;

[0008] Through hypergraph learning, a hypergraph structure cost function is constructed;

[0009] According to the first relationship tensor, the initial self-paced learning weight tensor and the hypergraph structure cost function, combined with L 2,1 Norm constraints, construct the initial function of the target model;

[0010] Updating the initial function to obtain the target function;

[0011] Wherein, the objective function is used to predict multi-type miRNA-disease relationships.

[0012] In some embodiments, obtaining a first relationship tensor of miRNA-disease-type by non-negative tensor decomposition includes the following steps:

[0013] Introducing a non-negative constraint into the CP decomposition to obtain the non-negative tensor decomposition;

[0014] According to the non-negative tensor decomposition, a first initial factor matrix of the miRNA pattern, a second initial factor matrix of the disease pattern, and a third initial factor matrix of the type pattern are obtained;

[0015] The first relationship tensor of miRNA-disease-type is obtained according to the first initial factor matrix, the second initial factor matrix and the third initial factor matrix.

[0016] In some embodiments, constructing a hypergraph structure cost function through hypergraph learning includes the following steps:

[0017] Based on the miRNA-miRNA similarity network, miRNA-disease relationship network and disease-disease similarity network, a miRNA-disease heterogeneous graph was constructed;

[0018] Based on the K-nearest neighbor algorithm, hypergraph learning is performed on multi-type miRNA hypergraphs and disease hypergraphs according to the miRNA-disease heterogeneous graph to obtain the hypergraph structure cost function.

[0019] In some embodiments, according to the first relationship tensor, the initial self-paced learning weight tensor and the hypergraph structure cost function, combined with L 2,1 The constraints of the norm and the construction of the initial function of the target model include the following steps:

[0020] Performing weighted adjustment on the first relationship tensor to obtain a second relationship tensor;

[0021] According to the hypergraph structure cost function, obtaining a first cost function of the miRNA pattern and a second cost function of the disease pattern;

[0022] According to the second relationship tensor, the initial self-paced learning weight tensor, the first cost function and the second cost function, combined with L 2,1 The constraint of the norm is used to obtain the initial function.

[0023] In some embodiments, weighted adjustment is performed on the first relationship tensor to obtain a second relationship tensor, comprising the following steps:

[0024] Construct the initial adaptive weight tensor;

[0025] Performing weighted adjustment on the first relationship tensor by using the initial adaptive weight tensor to obtain a weighted adjustment tensor;

[0026] The first relationship tensor and the weighted adjustment tensor are summed to obtain the second relationship tensor.

[0027] In some embodiments, obtaining a first cost function of a miRNA pattern and a second cost function of a disease pattern according to the hypergraph structure cost function comprises the following steps:

[0028] Obtaining a first hypergraph Laplacian matrix of the miRNA pattern;

[0029] Obtaining the first cost function according to the hypergraph structure cost function and the first hypergraph Laplacian matrix;

[0030] obtaining a second hypergraph Laplacian matrix of the disease pattern;

[0031] The second cost function is obtained according to the hypergraph structure cost function and the second hypergraph Laplacian matrix.

[0032] In some embodiments, updating the initial function to obtain the target function comprises the following steps:

[0033] By using a first Lagrange multiplier, a first initial factor matrix in the initial function is updated to obtain a first target factor matrix;

[0034] By using a second Lagrange multiplier, a second initial factor matrix in the initial function is updated to obtain a second target factor matrix;

[0035] By using a third Lagrange multiplier, a third initial factor matrix in the initial function is updated to obtain a third target factor matrix;

[0036] Setting the partial derivative of the initial adaptive weight tensor in the initial function to zero to obtain a target adaptive weight tensor;

[0037] According to the first target factor matrix, the second target factor matrix and the third target factor matrix, the initial self-paced learning weight tensor in the initial function is updated to obtain a target weight tensor;

[0038] The objective function is obtained according to the first objective factor matrix, the second objective factor matrix, the third objective factor matrix, the objective adaptive weight tensor and the objective weight tensor.

[0039] To achieve the above object, another aspect of an embodiment of the present invention provides a device for predicting multi-type miRNA-disease relationships based on non-negative tensor decomposition of L2,1 norm and self-paced learning, the device comprising:

[0040] The first module is used to obtain the first relationship tensor of miRNA-disease-type through non-negative tensor decomposition;

[0041] The second module is used to obtain the initial self-paced learning weight tensor through self-paced learning;

[0042] The third module is used to construct a hypergraph structure cost function through hypergraph learning;

[0043] The fourth module is used to combine L 2,1 Norm constraints, construct the initial function of the target model;

[0044] The fifth module is used to update the initial function to obtain the target function;

[0045] Wherein, the objective function is used to predict multi-type miRNA-disease relationships.

[0046] To achieve the above-mentioned purpose, another aspect of an embodiment of the present invention provides an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the aforementioned method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning.

[0047] To achieve the above-mentioned purpose, another aspect of an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the aforementioned method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning.

[0048] To achieve the above-mentioned purpose, another aspect of an embodiment of the present invention provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the aforementioned method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning.

[0049] The embodiments of the present invention include at least the following beneficial effects: The present invention provides a method and device for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning. The scheme obtains a first relationship tensor of miRNA-disease-type by non-negative tensor decomposition; obtains an initial self-paced learning weight tensor by self-paced learning; constructs a hypergraph structure cost function by hypergraph learning; and constructs a hypergraph structure cost function based on the first relationship tensor, the initial self-paced learning weight tensor and the hypergraph structure cost function in combination with L 2,1 The invention can improve the accuracy of predicting the relationship between multiple types of miRNA and diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0051] Figure 1 It is a flowchart of a method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning provided by an embodiment of the present invention;

[0052] Figure 2a-2c is a workflow diagram of the target model provided by an embodiment of the present invention in predicting potential multiple types of miRNA-disease relationships;

[0053] Figure 3 It is a schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present invention, and they are only examples of devices and methods consistent with some aspects of the embodiments of the present invention as detailed in the attached claims.

[0055] It should be noted that, although the functional modules are divided in the system schematic diagram and the logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in a different order than the module division in the system or the flow chart. The terms "first / S100" and "second / S200" in the specification and claims and the above-mentioned drawings may be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiment of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determination".

[0056] The terms "at least one", "multiple", "each", "any", etc. used in the present invention, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0058] Before describing the embodiments of the present invention in detail, some nouns and terms involved in the embodiments of the present invention are first described. The nouns and terms involved in the embodiments of the present invention are subject to the following explanations.

[0059] miRNA: MicroRNA (miRNA) is a type of non-coding single-stranded RNA molecule encoded by endogenous genes, with a length of about 22-24 nucleotides, which can regulate the expression level of genes after transcription. The Chinese name of microRNA is microribonucleic acid.

[0060] CP decomposition: CANDECOMP / PARAFAC (CP) decomposition is a high-order tensor decomposition method used to decompose a high-dimensional tensor (multidimensional array) into the sum of multiple low-rank components. It was first proposed by Carroll and Chang (1970) and Harshman (1970) respectively, called CANDECOMP (CANonical DECOMPosition) and PARAFAC (PARAllelFACtor analysis), hence the name CP decomposition.

[0061] False negative samples: In the miRNA and disease relationship data, there are some samples in the labeled negative samples that "have interactions, but we don't know yet", which we call false negative samples. However, if false negative samples are selected to participate in iterative updates, it will mislead the training of the model and reduce the prediction performance.

[0062] K-Nearest Neighbors (KNN) is a simple and effective machine learning algorithm, widely used in classification and regression problems. The main feature of KNN is that it does not require explicit model training of data. It is an instance-based learning method. When given an unlabeled data point, the KNN algorithm will find its closest K neighbors in the training set and determine the category of the new data point or predict its value based on the labels of these neighbors.

[0063] like Figure 1 As shown, the embodiment of the present invention provides a method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning. The method may include but is not limited to steps S100 to S500:

[0064] Step S100, obtaining a first relationship tensor of miRNA-disease-type by non-negative tensor decomposition;

[0065] Step S200, obtaining an initial self-paced learning weight tensor through self-paced learning;

[0066] Step S300, constructing a hypergraph structure cost function through hypergraph learning;

[0067] Step S400, based on the first relationship tensor, the initial self-learning weight tensor and the hypergraph structure cost function, combined with L 2,1 Norm constraints, construct the initial function of the target model;

[0068] Step S500, updating the initial function to obtain the target function;

[0069] Wherein, the objective function is used to predict multi-type miRNA-disease relationships.

[0070] In some embodiments, by combining self-paced learning and hypergraph regularization with non-negative tensor decomposition, the problem that noisy samples make the model easily fall into bad local minima is alleviated, and high-order complex relationships in heterogeneous networks are better preserved. 2,1 The norm constrains the objective function, increases the sparsity of the disease matrix, and eliminates the interference of irrelevant diseases, which helps to more accurately predict the relationship between multiple types of miRNAs and diseases.

[0071] In step S100 of some embodiments, the prediction of the miRNA-disease-type triple relationship is converted into a third-order tensor completion task, and the miRNA-disease-type relationship tensor is obtained through non-negative tensor decomposition. CP (CANDECOMP / PARAFAC) decomposition is to decompose a tensor into a sum of a set of rank-one tensors. Through tensor decomposition, potential high-order relationships can be effectively extracted, thereby improving prediction performance. Since the association between miRNA and disease is non-negative, non-negative constraints are introduced in CP decomposition, so that the factor matrix obtained by CP decomposition and the final prediction results are more interpretable.

[0072] In some embodiments, step S100 may include but is not limited to steps S110 to S130:

[0073] Step S110, introducing a non-negative constraint into the CP decomposition to obtain the non-negative tensor decomposition;

[0074] Step S120, obtaining a first initial factor matrix of the miRNA pattern, a second initial factor matrix of the disease pattern, and a third initial factor matrix of the type pattern according to the non-negative tensor decomposition;

[0075] Step S130: obtaining the first relationship tensor of miRNA-disease-type according to the first initial factor matrix, the second initial factor matrix and the third initial factor matrix.

[0076] In steps S110 to S130 of some embodiments, a non-negative tensor decomposition method is obtained by introducing non-negative constraints in CP decomposition. By combining the first initial factor matrix M of the miRNA pattern, the second initial factor matrix D of the disease pattern, and the third initial factor matrix T of the type pattern, the first relationship tensor x of the miRNA-disease-type triple relationship can be obtained. Then, the non-negative tensor decomposition model of the first relationship tensor can be expressed as follows:

[0077]

[0078] Here, the symbol ο represents the outer product operation of a vector; R is a positive integer, which refers to the approximate tensor rank of the first vector m r belong The second vector d r belong The third vector t r belong r=1,2,…,R; represents the first relation tensor, represents a real number set; M represents the first initial factor matrix of the miRNA pattern, M=[m1,m2,…,mR ], D represents the second initial factor matrix of the disease model, D = [d1, d2, ..., d R ], T represents the third initial factor matrix of the type pattern, T = [t1, t2, ..., t R ], m, n, and t represent the sizes of the corresponding dimensions.

[0079] Furthermore, the optimization problem of non-negative tensor decomposition can be simply considered as follows:

[0080]

[0081] stM≥0,D≥0,T≥0

[0082] Among them, ||·|| F represents the Frobenius norm of the tensor. In addition, the rank of the tensor is usually set to a small constant to obtain a low-rank approximation.

[0083] In step S200 of some embodiments, the problem of the model falling into a bad local optimal solution is alleviated by self-paced learning (SPL). In SPL, by introducing a regularizer (SP-regularizer), the model can automatically select samples that are easy to classify (high-confidence samples) and give them higher weights, thereby gradually guiding the model to learn samples from easy to difficult, so that the model can learn more smoothly and reduce the risk of falling into a local optimal solution. Compared with hard weighting, soft weighting is an important weight allocation strategy in SPL, which can better reflect the potential importance of samples in training and effectively improve the prediction performance of the model. Exemplarily, given a training data set The function with soft-weighted self-paced learning can be defined as follows:

[0084]

[0085] Where L(·,·) is the training error calculated by the loss function; w=[w1,w2,…,w N ] represents a weight variable; is the learning step size parameter; N represents the number of samples, i = 1, 2, ..., N; f(x i ,θ) represents the prediction function, where θ is the parameter of the prediction function; x i represents the feature vector of the i-th sample; y i represents the label or category of the i-th sample; γ is used to control the strength of the weight assigned to the selected sample.

[0086] Furthermore, the optimal weight variable w * It can be calculated as follows:

[0087]

[0088] Among them, γ>0 is used to control the strength of the weight assigned to the selected sample; is the loss value; w * is the optimal weight variable, i=1,2,…,N; is the learning step size parameter. As increases, more and more samples will be selected until all samples are selected.

[0089] In some embodiments, step S300 may include but is not limited to steps S310 to S320:

[0090] Step S310, constructing a miRNA-disease heterogeneous graph based on the miRNA-miRNA similarity network, the miRNA-disease relationship network, and the disease-disease similarity network;

[0091] Step S320, based on the K nearest neighbor algorithm, according to the miRNA-disease heterogeneous graph, hypergraph learning is performed on the multi-type miRNA hypergraph and the disease hypergraph to obtain the hypergraph structure cost function.

[0092] In some embodiments, miRNA-miRNA similarity networks, miRNA-disease relationship networks, and disease-disease similarity networks are used to construct Figure 2a The miRNA-disease heterogeneous graph shown in the figure shows that when different types of miRNA-disease relationship networks are integrated, the relationship value greater than 1 is assigned 1. Further, the K nearest neighbor algorithm (KNN) is used to learn the hypergraph of miRNAs and diseases based on the constructed heterogeneous graph. Assume that G = (V, E, W) is a weighted hypergraph, is a finite set of vertices in the hypergraph, is the set of hyperedges of the hypergraph, each hyperedge e is a subset of V, represents the weight of the hyperedge, which is a diagonal matrix. The relationship between vertices and hyperedges in the hypergraph can be represented by an association matrix H, the size of which is |V|×|E|. For example, the association matrix H of the hypergraph G can be defined as:

[0093]

[0094] Among them, e is a hyperedge; v is a vertex; H(v,e) is the association degree between vertex v and hyperedge e.

[0095] Assigning weights to each hyperedge in the hypergraph, the affinity matrix A of the hypergraph can be calculated as follows:

[0096]

[0097] Among them, A ij represents the (i, j)th element of the affinity matrix; σ represents the standard deviation distance between all vertices; v i 、v j For each vertex. Then, the weight of each hyperedge is W i It can be expressed as:

[0098]

[0099] Each vertex v j The degree of ∈V is to calculate v j The sum of the weights of the hyperedges to which it belongs. Each hyperedge e i The degree of ∈E refers to the number of vertices contained in E. E (e) and d) V (v) represents the degree of the hyperedge and the degree of the vertex, and W(e) represents the weight of the hyperedge e, then we have the following expression:

[0100]

[0101] like Figure 2c Schematic diagram of the similarity of multiple types of miRNAs (if miRNAm i and miRNAm j belong to the same hyperedge, and the hyperedge weight is high, then M(m i ) and M(m j ) will be very similar, i.e. the nodes have the same color). According to the similarity hypothesis, if the similarity between two biological entities is higher, then their distance in the low-dimensional embedding representation should be closer. Let F be the low-dimensional embedding representation of miRNA (or disease) learned by the model. The cost function is defined as the objective to minimize the difference measure between similar miRNAs (or diseases), so that the high-order complex relationships of the hypergraph can be preserved in the low-dimensional space. The following hypergraph structure cost function can be obtained:

[0102]

[0103] Where, L hyper =D V –S, and

[0104] Where Ω represents the hypergraph structure cost function; H(v i ,e) represents the i-th vertex vi The degree of association with the hyperedge e; H(v j ,e) represents the jth vertex v j The degree of association with hyperedge e; d E (e) represents the degree of the hyperedge; F(v i ) is about vertex v i The model learns the low-dimensional embedding representation of miRNA (or disease); F(v j ) is about vertex v j The low-dimensional embedding representation of miRNA (or disease) learned by the model; F is the low-dimensional embedding representation of miRNA (or disease) learned by the model; (·) T Represents the transpose operation; D v represents a diagonal matrix whose elements are the degrees of the vertices d V (v); D E represents another diagonal matrix whose elements are the degree d of the hyperedge E (e); S represents the intermediate parameter; L hyper represents the hypergraph Laplacian matrix.

[0105] In some embodiments, step S400 may include but is not limited to steps S410 to S430:

[0106] Step S410, performing weighted adjustment on the first relationship tensor to obtain a second relationship tensor;

[0107] In some optional embodiments, although in the first relation tensor There are many "unknown" relationships in the relationship tensor, but the high false negative rate in these relationships will affect the prediction performance of the model. In order to alleviate the impact of false negative samples on the prediction performance of the model, the relationship tensor Decompose into two parts and innovatively design an adaptive weight tensor (like Figure 2b The first relation tensor of the third-order association shown Construction and adaptive weight tensor use).

[0108] In some embodiments, step S410 may include but is not limited to steps S411 to S413:

[0109] Step S411, constructing an initial adaptive weight tensor;

[0110] Step S412, performing weighted adjustment on the first relationship tensor by using the initial adaptive weight tensor to obtain a weighted adjustment tensor;

[0111] Step S413: sum the first relationship tensor and the weighted adjustment tensor to obtain the second relationship tensor.

[0112] In some optional embodiments, such as Figure 2b As shown, by constructing the initial adaptive weight tensor For a tensor with all elements set to 1 With the first relationship tensor The difference between them is weighted and adjusted, and the weighted adjustment tensor can be obtained as: in, express and The difference between This difference is represented by the adaptive weight tensor Adjusted by, where ⊙ represents the multiplication of the corresponding elements of the tensor. Decompose into two parts for summation, one part is the original first relation tensor The other part is the weighted adjustment tensor, then the sum of the two parts can get the second relationship tensor

[0113] Step S420, obtaining a first cost function of a miRNA pattern and a second cost function of a disease pattern according to the hypergraph structure cost function;

[0114] In some embodiments, step S420 may include but is not limited to steps S421 to S424:

[0115] Step S421, obtaining a first hypergraph Laplacian matrix of the miRNA pattern;

[0116] Step S422, obtaining the first cost function according to the hypergraph structure cost function and the first hypergraph Laplacian matrix;

[0117] Step S423, obtaining a second hypergraph Laplacian matrix of the disease pattern;

[0118] Step S424: Obtain the second cost function according to the hypergraph structure cost function and the second hypergraph Laplacian matrix.

[0119] In some optional embodiments, for the miRNA pattern, its first hypergraph Laplacian matrix is ​​L hm =D vm –S m Similarly, for the disease model, its second hypergraph Laplacian matrix is ​​L hd =D vd –S dAccording to the hypergraph structure cost function and the first hypergraph Laplacian matrix, the first cost function can be obtained as Tr(M T L hm M); According to the hypergraph structure cost function and the second hypergraph Laplacian matrix, the second cost function can be obtained as Tr(D T L hd D) Among them, the first intermediate parameter The second intermediate parameter D vm represents the first diagonal matrix; D vd represents the second diagonal matrix.

[0120] Step S430, according to the second relationship tensor, the initial self-paced learning weight tensor, the first cost function and the second cost function, combined with L 2,1 The constraint of the norm is used to obtain the initial function.

[0121] In step S430 of some embodiments, an initial function may be constructed according to the second relationship tensor, the initial self-paced learning weight tensor, the first cost function, and the second cost function. 2,1 The factor matrix of diseases is constrained by norm to increase the sparsity of the disease matrix and eliminate irrelevant disease pairs.

[0122] In some embodiments, step S500 uses a multiplication update rule to update the initial function to obtain a target function. Optionally, step S500 may include but is not limited to steps S510 to S560:

[0123] Step S510, updating the first initial factor matrix in the initial function by using the first Lagrange multiplier to obtain a first target factor matrix;

[0124] Step S520, updating the second initial factor matrix in the initial function by a second Lagrange multiplier to obtain a second target factor matrix;

[0125] Step S530, updating the third initial factor matrix in the initial function by a third Lagrange multiplier to obtain a third target factor matrix;

[0126] Step S540, setting the partial derivative of the initial adaptive weight tensor in the initial function to zero, and obtaining the target adaptive weight tensor;

[0127] Step S550, updating the initial self-paced learning weight tensor in the initial function according to the first target factor matrix, the second target factor matrix and the third target factor matrix to obtain a target weight tensor;

[0128] Step S560, obtaining the objective function according to the first objective factor matrix, the second objective factor matrix, the third objective factor matrix, the target adaptive weight tensor and the target weight tensor.

[0129] In step S510 of some embodiments, according to the strategy of the multiplication update rule, when other variables are fixed, the terms involving M in the initial function can be updated as follows:

[0130]

[0131] By introducing the first Lagrangian multiplier Φ, the following first Lagrangian function can be constructed:

[0132]

[0133] right Taking partial derivatives of the first initial factor matrix M in , we get the following equation:

[0134]

[0135] in, represents the adaptive weighting of the first-mode flattened matrix of the tensor; the parameter α is used to control the influence of the hypergraph Laplace regularization term; (·) (1) represents the flattening matrix of the first mode of the tensor; λ t represents the first regularization coefficient; Represents the self-paced learning weight tensor; is the Kronecker product.

[0136] For the (i,j)th element, according to the Karush-Kuhn-Tucker (KKT) condition, let Φ ij M ij =0, we can get:

[0137]

[0138] After the above calculations, the iterative formula for the first initial factor matrix M is:

[0139]

[0140] Among them, M′ ij is the (i,j)th element of the first target factor matrix.

[0141] In step S520 of some embodiments, according to the strategy of the multiplication update rule, when other variables are fixed, the term involving D in the initial function can be re-expressed as follows:

[0142]

[0143] By introducing the second Lagrangian multiplier Θ, the following second Lagrangian function can be constructed:

[0144]

[0145] right Taking partial derivatives of the second initial factor matrix D in, we get the following equation:

[0146]

[0147] in, Adaptively weight the second-mode flattened matrix of the tensor; (·) (2) represents the second-mode flattening matrix of the tensor; β is used to control the influence of the hypergraph Laplace regularization term; ‖·‖ 2,1 For L 2,1 norm; A is a diagonal matrix, [A] ii =1 / 2||D i ||2.

[0148] For the (i,j)th element, according to the KKT condition, let Θ ij M ij =0, we can get:

[0149]

[0150] After the above calculations, the iterative formula for the second initial factor matrix D is:

[0151]

[0152] Among them, D′ ij is the (i,j)th element of the second target factor matrix.

[0153] In step S530 of some embodiments, according to the strategy of the multiplication update rule, when other variables are fixed, the terms involving T in the initial function can be re-expressed as follows:

[0154]

[0155] in,(·) (3) Represents the third-mode flattening of the tensor.

[0156] The third Lagrangian multiplier Ψ is introduced to constrain T ≥ 0. Then, the third Lagrangian function The definition is as follows:

[0157]

[0158] The partial derivatives with respect to the third initial factor matrix T can be expressed as follows:

[0159]

[0160] in, Represents adaptive weighting of the 3rd-mode flattened matrix of the tensor.

[0161] For the (i,j)th element, according to the KKT condition, let Ψ ij T ij =0, we can get:

[0162]

[0163] After the above calculations, the update rule for the third initial factor matrix T is:

[0164]

[0165] Among them, T′ ij is the (i,j)th element of the third target factor matrix.

[0166] In step S540 of some embodiments, according to the strategy of the multiplication update rule, when other variables are fixed, The function can be expressed as follows:

[0167]

[0168] make and Then the above formula can be written as follows:

[0169]

[0170] Since the Hadamard product operation is involved, for the (i, j, k)th element in the adaptive weight tensor It can be rewritten as follows:

[0171]

[0172] Then, by setting the partial derivatives to zero, we can solve

[0173]

[0174] in, is the (i,j,k)th element of the self-learning weight tensor.

[0175] Finally, the following solution formula can be obtained:

[0176]

[0177] Among them, λ c represents the second regularization coefficient.

[0178] In step S550 of some embodiments, after all the above parameters are updated, the target model SPLDHAWL 2,1 -NTF(Self-Paced Learning with Dual Hypergraph regularized and AdaptiveWeighted L 2,1 norm-Nonnegative Tensor Factorization) starts updating the initial self-paced learning weight tensor Involved The function can be expressed as follows:

[0179]

[0180] For the n-th mode (n=1,2,3) of the tensor, the optimal flattening matrix is It can be calculated as follows:

[0181]

[0182] Among them, corresponding to the factor matrix M, the loss function for Corresponding factor matrix D, loss function for Corresponding factor matrix T, loss function for

[0183] In step S560 of some embodiments, the final objective function may be obtained according to the first objective factor matrix, the second objective factor matrix, the third objective factor matrix, the target adaptive weight tensor, and the target weight tensor:

[0184]

[0185] in, is a tensor with all elements set to 1. α and β are used to control the influence of the hypergraph Laplace regularization term. t and λ c are two regularization coefficients, L hm The first hypergraph Laplacian matrix representing the miRNA pattern; L hd The second hypergraph Laplacian matrix representing the disease pattern, ⊙ represents the Hadamard product. To prevent overfitting, the L2 norm constraint factor matrix and adaptive weight tensor are used

[0186] The embodiment of the present invention also provides a device for predicting the relationship between multiple types of miRNA-diseases based on non-negative tensor decomposition of L2,1 norm and self-paced learning, which can implement the above-mentioned method for predicting the relationship between multiple types of miRNA-diseases based on non-negative tensor decomposition of L2,1 norm and self-paced learning. The device includes:

[0187] The first module is used to obtain the first relationship tensor of miRNA-disease-type through non-negative tensor decomposition;

[0188] The second module is used to obtain the initial self-paced learning weight tensor through self-paced learning;

[0189] The third module is used to construct a hypergraph structure cost function through hypergraph learning;

[0190] The fourth module is used to combine L 2,1 Norm constraints, construct the initial function of the target model;

[0191] The fifth module is used to update the initial function to obtain the target function;

[0192] Wherein, the objective function is used to predict multi-type miRNA-disease relationships.

[0193] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0194] The embodiment of the present invention also provides an electronic device, which includes a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, the method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning is implemented. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0195] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0196] refer to Figure 3 , Figure 3 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0197] The processor 601 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention;

[0198] The memory 602 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device or a random access memory (RAM). The memory 602 can store an operating system and other applications. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 602, and the processor 601 calls and executes a method for predicting multi-type miRNA-disease relationships based on non-negative tensor decomposition based on L2,1 norm and self-paced learning in the embodiment of the present invention;

[0199] Input / output interface 603, used to implement information input and output;

[0200] Communication interface 604, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0201] Bus 605 , which transmits information between various components of the device (e.g., processor 601 , memory 602 , input / output interface 603 , and communication interface 604 );

[0202] The processor 601 , the memory 602 , the input / output interface 603 and the communication interface 604 are connected to each other in communication within the device via a bus 605 .

[0203] An embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning.

[0204] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiments, the functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0205] The embodiment of the present invention also provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above-mentioned method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning.

[0206] In summary, the above-mentioned method and device for predicting multi-type miRNA-disease relationships based on L2,1 norm and self-paced learning based on non-negative tensor decomposition in the embodiment of the present invention has the following advantages:

[0207] 1. The embodiment of the present invention integrates self-paced learning to alleviate the problem of noisy samples and make the proposed SPLDHAWL 2,1 -NTF objective model avoids falling into bad local minima.

[0208] 2. Embodiments of the present invention integrate hypergraph regularization into SPLDHAWL 2,1 -NTF target model, which can better preserve high-order complex relationships in heterogeneous networks.

[0209] 3. The embodiment of the present invention decomposes the tensor into two parts and innovatively designs an adaptive weight tensor. Reducing false negative samples has an important impact on the proposed SPLDHAWL 2,1 -The impact of NTF model prediction performance.

[0210] 4. The embodiment of the present invention uses L 2,1 Norm constraints on the factor matrix D representing the disease increase the sparsity of the disease matrix and eliminate the interference of irrelevant diseases, which helps to more accurately predict the relationship between multiple types of miRNAs and diseases. This will accelerate the research and development process and provide effective preliminary screening for traditional wet experiments.

[0211] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0212] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features described may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0213] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0214] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0215] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0216] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0217] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0218] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0219] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A method for predicting multi-type miRNA-disease relationships based on non-negative tensor decomposition of L2,1 norm and self-paced learning, characterized in that: The following steps are involved: Obtain the first relationship tensor of miRNA-disease-type through non-negative tensor decomposition; Through self-paced learning, the initial self-paced learning weight tensor is obtained; Through hypergraph learning, a hypergraph structure cost function is constructed; According to the first relationship tensor, the initial self-paced learning weight tensor and the hypergraph structure cost function, combined with L 2,1 Norm constraints, construct the initial function of the target model; Updating the initial function to obtain the target function; Wherein, the objective function is used to predict multi-type miRNA-disease relationships.

2. According to claim 1, a method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning, characterized in that: The method of obtaining the first relationship tensor of miRNA-disease-type by non-negative tensor decomposition includes the following steps: Introducing a non-negative constraint into the CP decomposition to obtain the non-negative tensor decomposition; According to the non-negative tensor decomposition, a first initial factor matrix of the miRNA pattern, a second initial factor matrix of the disease pattern, and a third initial factor matrix of the type pattern are obtained; The first relationship tensor of miRNA-disease-type is obtained according to the first initial factor matrix, the second initial factor matrix and the third initial factor matrix.

3. According to claim 1, a method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning, characterized in that: The method of constructing a hypergraph structure cost function through hypergraph learning includes the following steps: Based on the miRNA-miRNA similarity network, miRNA-disease relationship network and disease-disease similarity network, a miRNA-disease heterogeneous graph was constructed; Based on the K-nearest neighbor algorithm, hypergraph learning is performed on multi-type miRNA hypergraphs and disease hypergraphs according to the miRNA-disease heterogeneous graph to obtain the hypergraph structure cost function.

4. According to claim 1, a method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning, characterized in that: The first relationship tensor, the initial self-paced learning weight tensor and the hypergraph structure cost function are combined with L 2,1 The constraints of the norm and the construction of the initial function of the target model include the following steps: Performing weighted adjustment on the first relationship tensor to obtain a second relationship tensor; According to the hypergraph structure cost function, obtaining a first cost function of the miRNA pattern and a second cost function of the disease pattern; According to the second relationship tensor, the initial self-paced learning weight tensor, the first cost function and the second cost function, combined with L 2,1 The constraint of the norm is used to obtain the initial function.

5. According to claim 4, a method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning, characterized in that: The step of performing weighted adjustment on the first relationship tensor to obtain a second relationship tensor comprises the following steps: Construct the initial adaptive weight tensor; Performing weighted adjustment on the first relationship tensor by using the initial adaptive weight tensor to obtain a weighted adjustment tensor; The first relationship tensor and the weighted adjustment tensor are summed to obtain the second relationship tensor.

6. According to claim 4, a method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning, characterized in that: The method of obtaining a first cost function of a miRNA pattern and a second cost function of a disease pattern according to the hypergraph structure cost function comprises the following steps: Obtaining a first hypergraph Laplacian matrix of the miRNA pattern; Obtaining the first cost function according to the hypergraph structure cost function and the first hypergraph Laplacian matrix; obtaining a second hypergraph Laplacian matrix of the disease pattern; The second cost function is obtained according to the hypergraph structure cost function and the second hypergraph Laplacian matrix.

7. The method for predicting multi-type miRNA-disease relationships by non-negative tensor decomposition based on L2,1 norm and self-paced learning according to claim 1, characterized in that: The updating of the initial function to obtain the target function comprises the following steps: By using a first Lagrange multiplier, a first initial factor matrix in the initial function is updated to obtain a first target factor matrix; By using a second Lagrange multiplier, a second initial factor matrix in the initial function is updated to obtain a second target factor matrix; By using a third Lagrange multiplier, a third initial factor matrix in the initial function is updated to obtain a third target factor matrix; Setting the partial derivative of the initial adaptive weight tensor in the initial function to zero to obtain a target adaptive weight tensor; According to the first target factor matrix, the second target factor matrix and the third target factor matrix, the initial self-paced learning weight tensor in the initial function is updated to obtain a target weight tensor; The objective function is obtained according to the first objective factor matrix, the second objective factor matrix, the third objective factor matrix, the objective adaptive weight tensor and the objective weight tensor.

8. A device for predicting multi-type miRNA-disease relationships based on non-negative tensor decomposition of L2,1 norm and self-paced learning, characterized in that: include: The first module is used to obtain the first relationship tensor of miRNA-disease-type through non-negative tensor decomposition; The second module is used to obtain the initial self-paced learning weight tensor through self-paced learning; The third module is used to construct a hypergraph structure cost function through hypergraph learning; The fourth module is used to combine L 2,1 Norm constraints, construct the initial function of the target model; The fifth module is used to update the initial function to obtain the target function; Wherein, the objective function is used to predict multi-type miRNA-disease relationships.

9. An electronic device, characterized in that: including a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The storage medium stores a program, and the program is executed by a processor to implement the method according to any one of claims 1 to 7.

Images