A scSPRITE data completion method based on hypergraph random walk

Abstract

The present invention discloses a scSPRITE data completion method based on hypergraph random walk, comprising the following steps: S1, obtaining scSPRITE sequencing documents; S2, extracting gene sites in the scSPRITE sequencing documents, and obtaining a hyperedge set and a hypergraph node set; S3, calculating the degree of each hyperedge in the hyperedge set of each hypergraph and the degree of each node in the hypergraph node set; S4, determining the walk transfer probability of each node transferring to the next node, and obtaining the node vector of each hypergraph; S5, inputting the node vector of each hypergraph into a Node2Vec model for training, and outputting the node embedding vector of each hypergraph; S6, calculating the similarity of every two node embedding vectors in each hypergraph and screening the interaction matrix of each single cell. The single-cell three-dimensional genome structure completion method of the present invention can define the three-dimensional genome structure of a small number of group cells, and the single-cell genome structure can be accurately sequenced by a small number of group cells, thereby reducing the workload of sequencing and improving sequencing efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of bioinformatics, and in particular relates to a scSPRITE data completion method based on hypergraph random walk. Background Art

[0002] Despite the crucial importance of the three-dimensional genome for many aspects of nuclear function, it remains difficult to obtain sufficiently rich measurements of 3D genome structure at the single-cell level. Locally, the DNA sequences that wind and fold in the nucleus can be divided into open and closed chromatin regions, further subdivided into TADs, and local interactions (loops) between regulatory elements. The genome organization within individual nuclei influences various nuclear cell functions, including DNA replication, transcription, and RNA processing.

[0003] Currently, there are two main methods for measuring single-cell genome structure: scHi-C and microscopy. Microscopy can measure large-scale genomic interactions in single cells, but has a narrow field of view and cannot provide a genome-wide view. In contrast, scHi-C provides genome-wide nuclear organization in single cells, but requires specialized equipment and is limited to low resolution. In addition, because scHi-C relies on nearest neighbor connections to measure interactions, it is limited in capturing long-range and high-order interactions, such as the organization around nuclear bodies.

[0004] To address these issues, scSPRITE can be used to provide high-order, complex, genome-wide DNA structures from thousands of single cells. Since scSPRITE can measure multiple interactions in single cells, it can generate higher-resolution maps within single cells. However, the scSPRITE method is limited to traversing pairwise interactions for the measured multiple interactions. There are still many missing values ​​in scSPRITE sequencing that have not been observed. Although scSPRITE obtains higher paired reads than scHiC at low sequencing volumes, it is difficult to directly define the interaction structure in the three-dimensional genomic space under missing values. Therefore, a lot of sequencing work is required. Summary of the Invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a scSPRITE data completion method based on hypergraph random walk. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0006] A scSPRITE data completion method based on hypergraph random walks includes the following steps:

[0007] S1, obtain scSPRITE sequencing files;

[0008] S2, extracting gene loci from the scSPRITE sequencing document, and obtaining a hyperedge set and a hypergraph node set for each single cell based on the gene loci; the hyperedge set and hypergraph node set for each single cell constitute a hypergraph for each single cell;

[0009] S3, calculating the degree of each hyperedge in the hyperedge set of each hypergraph and the degree of each node in the hypergraph node set;

[0010] S4, determining a walk transition probability of each node transferring to the next node based on the degree of each hyperedge and the degree of each node of each hypergraph, and performing a biased random walk on each hypergraph based on the walk transition probability to obtain a node vector of each hypergraph;

[0011] S5, inputting the node vector of each hypergraph into the Node2Vec model for training, and outputting the node embedding vector of each hypergraph;

[0012] S6, calculating the similarity of the embedding vectors of every two nodes in each hypergraph and screening to generate the interaction matrix of each single cell.

[0013] In one embodiment of the present invention, step S2 includes:

[0014] S21, extracting all gene loci of the same chromosome field of the same single-cell multivariate interaction in the scSPRITE sequencing document, and constructing a preliminary hyperedge set based on all gene loci;

[0015] S22, mapping the gene locus to the chromosome sequence interval bin as a hypergraph node; the expression of the hypergraph node is:

[0016]

[0017] Among them, c i Indicates the value of the i-th bin, b j represents the jth multivariate interaction of the same single cell, w∈b j Indicates b j Gene loci in

[0018] S23, traversing the scSPRITE sequencing document according to step S21 and step S22 to obtain a preliminary hyperedge set and a preliminary hypergraph node set for each single cell;

[0019] S24, removing hyperedges whose number of nodes in the preliminary hypergraph node set is greater than a threshold ε, to obtain a hyperedge set and a hypergraph node set for each single cell.

[0020] In one embodiment of the present invention, the calculation formula of the degree of each node is:

[0021]

[0022] Where E represents the set of hyperedges, ω represents the weight of each hyperedge e, and h(v,e) represents that node v exists in hyperedge e. If node v is on a hyperedge e, then h(v,e) = 1, otherwise h(v,e) = 0.

[0023] The calculation formula of the degree of the hyperedge is:

[0024]

[0025] Wherein, V represents the hypergraph node set.

[0026] In one embodiment of the present invention, step S4 includes:

[0027] S41, determining a first-order transition probability of each node transferring to the next node according to the degree of each hyperedge and the degree of each node of each hypergraph;

[0028] S42, determining the second-order transition probability of each node transferring to the next node according to the first-order transition probability and the preset search bias;

[0029] S43, normalizing the second-order transition probability to obtain a wandering transition probability of each node transferring to the next node;

[0030] S44: Perform a biased random walk on each hypergraph according to the walk transition probability to obtain a node vector of each hypergraph.

[0031] In one embodiment of the present invention, the expression of the first-order transition probability is:

[0032]

[0033] Where π1(x|v) represents the first-order transition probability of node v transferring to the next node x, h(v,e) and h(x,e) respectively indicate that node v and node x exist in hyperedge e.

[0034] In one embodiment of the present invention, the expression of the second-order transition probability is:

[0035] π2(x|v,u)=α(x|v,u)π1(x|v);

[0036] Wherein, u represents the neighbor node of node v, and α represents the preset search bias;

[0037]

[0038] In one embodiment of the present invention, step S6 includes:

[0039] S61, calculating the similarity between each two node embedding vectors in each hypergraph; wherein the calculation formula for the similarity between each two node embedding vectors is:

[0040]

[0041] Wherein, f(X, Y) represents the cosine similarity between one node embedding vector X and another node embedding vector Y, and n represents the dimension of the node embedding vector;

[0042] S62, taking the similarity greater than the threshold β as the interaction value, and generating the interaction matrix of each single cell.

[0043] Beneficial effects of the present invention:

[0044] The single-cell three-dimensional genome structure completion method of the present invention includes a hypergraph random walk algorithm. Before training the model and predicting missing values, the hypergraph random walk algorithm performs multiple biased random walks on each node in the hypergraph to obtain multiple word vectors, which can obtain the interaction information between bins in different hyperedges and the feature information of the bins in the hyperedges, thereby completing the missing values ​​through the flow of information between different hyperedges to obtain the final interaction matrix. The three-dimensional genome structure of a small number of cell groups can be defined, and the single-cell genome structure can be accurately sequenced from a small number of cell groups, thereby reducing the workload of sequencing and improving sequencing efficiency.

[0045] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic flow chart of a scSPRITE data completion method based on hypergraph random walks provided by an embodiment of the present invention;

[0047] Figure 2 A scSPRITE data completion method based on hypergraph random walks provided by an embodiment of the present invention and a benchmark experiment line chart of the prior art. DETAILED DESCRIPTION

[0048] The present invention will be further described in detail below with reference to specific examples, but the embodiments of the present invention are not limited thereto.

[0049] Example 1

[0050] like Figure 1 As shown, a scSPRITE data completion method based on hypergraph random walk includes the following steps:

[0051] S1, obtain scSPRITE sequencing files.

[0052] S2, extract the gene loci in the scSPRITE sequencing document, and obtain the hyperedge set and hypergraph node set of each single cell based on the gene loci; the hyperedge set and hypergraph node set of each single cell constitute the hypergraph of each single cell.

[0053] In this step, step S2 includes S21-S24:

[0054] S21, extract all gene loci of the same chromosome field of the same single-cell multivariate interaction in the scSPRITE sequencing document, and construct a preliminary hyperedge set based on all gene loci.

[0055] S22, maps the gene loci to the chromosome sequence interval bin as a hypergraph node; the expression of the hypergraph node is:

[0056]

[0057] Among them, c i Indicates the value of the i-th bin, b j represents the jth multivariate interaction of the same single cell, w∈b j Indicates b j Gene loci in

[0058] In the above steps, each genomic spatial cluster entry in the scSPRITE sequencing document is constructed as a hyperedge, with the bins mapped to the corresponding genomic locations in the entry as nodes. This means that each single cell is constructed as a hypergraph, with the genome divided into bins at different resolutions and serving as nodes in the hypergraph. Preferably, high-order interactions are read for 1,000 single cells containing genomic locations in the scSPRITE sequencing document.

[0059] S23, traversing the scSPRITE sequencing document according to step S21 and step S22 to obtain a preliminary hyperedge set and a preliminary hypergraph node set for each single cell;

[0060] S24, removing hyperedges whose number of nodes in the preliminary hypergraph node set is greater than a threshold ε, and obtaining a hyperedge set and a hypergraph node set for each single cell.

[0061] S3, calculates the degree of each hyperedge in the hyperedge set of each hypergraph and the degree of each node in the hypergraph node set.

[0062] Specifically, a hypergraph is defined as G(V,E), where V is a set of nodes and E is a set of hyperedges. A hyperedge e∈E represents a relationship between some nodes (≥2), that is, a subset of V. A weighted hypergraph is defined as G(V,E,ω), where ω represents the weight of each hyperedge e. The degree of a node in the hypergraph is defined as:

[0063]

[0064] For a hyperedge e∈E, the degree of the hyperedge is defined as the number of nodes contained in the hyperedge:

[0065] δ(e)=|e|

[0066] Define two square matrices D v 、D e , respectively representing the order of the nodes and the order of the hyperedges of the hypergraph. Similar to the matrix representation of ordinary graphs, hypergraphs can also be represented by constructing a |V|×|E|-dimensional point-edge incidence matrix H. h(v,e) indicates that node v exists in hyperedge e. In the incidence matrix, if a node v is on a hyperedge e, then h(v,e) = 1, otherwise h(v,e) = 0. According to the definition of the incidence matrix, the order of nodes (node ​​degree) and the order of hyperedges (hyperedge degree) in the hypergraph can be further expressed as:

[0067]

[0068]

[0069] S4, determine the walk transition probability of each node transferring to the next node according to the degree of each hyperedge of each hypergraph and the degree of each node, perform a biased random walk on each hypergraph according to the walk transition probability, and obtain the node vector of each hypergraph.

[0070] Specifically, step S4 includes S41-S44:

[0071] S41, determining the first-order transition probability of each node transferring to the next node according to the degree of each hyperedge of each hypergraph and the degree of each node; the expression of the first-order transition probability is:

[0072]

[0073] Where π1(x|v) represents the first-order transition probability of node v transferring to the next node x, h(v,e) and h(x,e) respectively indicate that node v and node x exist in hyperedge e.

[0074] S42, determining the second-order transition probability of each node transferring to the next node based on the first-order transition probability and the preset search bias; the expression of the second-order transition probability is:

[0075] π2(x|v,u)=α(x|v,u)π1(x|v);

[0076] Where u represents the neighbor node of node v, and α represents the preset search bias;

[0077]

[0078] S43, normalize the second-order transition probability to obtain the wandering transition probability of each node to the next node; in this step, the normalization is the second-order transition probability of each node to the next node divided by the sum of the second-order transition probabilities of the node to other nodes in the hypergraph.

[0079] S44, performing biased random walk on each hypergraph according to the walk transition probability to obtain the node vector of each hypergraph.

[0080] The above steps S41 to S44 can also be understood as constructing a single-cell three-dimensional genome structure completion (SpriteHyper2vec) model O based on hypergraph random walk:

[0081] The first step is to build the SpriteHyper2vec model:

[0082] To maintain the high-order proximity of the hypergraph network constructed from scSPRITE data, we propose an effective and scalable biased second-order random walk model under the Skip-gram framework.

[0083] The first-order random walk strategy determines the probability of moving to the next node based on the current node. To generalize random walks in hypernetworks, the random walk strategy in ordinary graphs is generalized. Given the current node v, we first obtain the set of hyperedges E1 associated with v based on the weight of e, and then randomly select the next node based on the degree d(x) of the node x∈e in each hyperedge e. The unnormalized transition probability of each walk in the first-order model can be given as follows:

[0084]

[0085] When performing prediction tasks, it is feasible to introduce a second-order random walk model to navigate between homogeneity and structural equivalence. The neighbors of a node v are divided into three categories: the previous node u, the neighbors of the previous node u, and others. The second-order search bias is defined as follows:

[0086]

[0087] The unnormalized transition probability of the second-order random walk is defined as π2(x|v,u)=α(x|v,u)π1(x|v). The relationship between the parameters p and q guides a depth-first search (DFS) that encourages outward jumps and a breadth-first search (BFS) that retrieves the local neighborhood of the starting node. The final embedding result is generated by the Skip-gram model. Skip-gram is a language model that maximizes the co-occurrence probability between words that appear within a sentence window. In this paper, this is replaced by a hypergraph random walk.

[0088] The second step is to define the loss function L of the SpriteHyper2vec model:

[0089]

[0090] Among them, K represents the number of negative samples, u i The feature vector representing word i, v c The feature vector representing the central word is used. Finally, the stochastic gradient descent algorithm (SGD) is used for iterative training.

[0091] S5, input the node vector of each hypergraph into the Node2Vec model for training, and output the node embedding vector of each hypergraph.

[0092] S6, calculate the similarity of the embedding vectors of every two nodes in each hypergraph and filter to generate the interaction matrix of each single cell. Specifically, step S6 includes S61-S62:

[0093] S61, calculating the similarity between the embedding vectors of every two nodes in each hypergraph; wherein the calculation formula for the similarity between the embedding vectors of every two nodes is:

[0094]

[0095] Where f(X,Y) represents the cosine similarity between a node embedding vector X and another node embedding vector Y, and n represents the dimension of the node embedding vector;

[0096] S62, taking the similarity greater than the threshold β as the interaction value, and generating the interaction matrix of each single cell.

[0097] The trained bin embeddings are used to calculate the cosine similarity between each pair, and the similarities above a certain threshold are retained. All bin pairs are traversed to generate the interaction matrix.

[0098] The single-cell three-dimensional genome structure completion model constructed by the present invention includes a hypergraph random walk algorithm. Before training the model and predicting missing values, the hypergraph random walk algorithm performs multiple biased random walks on each node in the hypergraph to obtain multiple word vectors. It can obtain the interaction information between bins in different hyperedges and the characteristic information of bins in the hyperedges, thereby completing the missing values ​​through the flow of information between different hyperedges to obtain the final interaction matrix, avoiding the existing decomposition method that only uses the interaction information of bins in the hyperedges, resulting in the inability to define the three-dimensional genome structure of a small number of cell groups.

[0099] The hypergraph random walk method proposed in this paper has the advantages of low memory usage and high speed compared to conventional random walks. During the random walk of each node in a hyperedge, nodes with the same degree in the same hyperedge are stored in a shared transition probability vector. This significantly reduces memory pressure compared to maintaining a separate second-order transition probability vector for each node. It also speeds up the preprocessing of transition probability tables and the querying of second-order transition nodes, which in turn speeds up the algorithm.

[0100] The single-cell 3D genome structure completion model constructed by this invention can complete the local A\B compartmentalization of the 3D genome of a small number of cell swarms and make all A\B compartmentalizations on a chromosome more prominent. By calculating the frequency-based compartment scores of a small number of cell swarms under the swarm, it can be seen that the model can significantly improve the compartmental scores.

[0101] Example 2

[0102] An embodiment of the present invention provides a scSPRITE data completion method based on hypergraph random walk, comprising the following steps:

[0103] Step 1: Obtain the single-cell SPRITE dataset:

[0104] Obtain the 1000 mouse embryonic stem cell SPRITE sequencing data published by Mary V. Arrastia et al. in 2022 from GSE154353 in the Gene Expression Omnibus (GSE);

[0105] Step 2: Build model input:

[0106] A single-cell hypergraph was constructed using the acquired SPRITE sequencing data from 1000 cells. The sequence bin size was set to 100KB, resulting in an original hypergraph adjacency matrix for 1000 different chromosomes. The matrix size was determined by the sequencing data.

[0107] Step 3: Build the scSPRITE data completion model O based on hypergraph random walk:

[0108] (3a) Construct the structure of the scSPRITE data completion model O based on hypergraph random walk:

[0109] The scSPRITE data completion model based on hypergraph random walks is divided into a hypergraph random walk module and a Skip-gram algorithm training embedding module.

[0110] The hypergraph random walk module consists of two parts: hypergraph node transition probability initialization and hypergraph random walk. Hypergraph node transition probability initialization initializes the transition probabilities between nodes in the hypergraph into a lookup table. To speed up initialization and reduce the memory usage of the lookup table, nodes of the same degree in the same hyperedge share a common transition probability vector. For second-order biased transition probabilities, the input and output parameters p and return parameters q are set to 0.5 and 0.25 respectively. The hypergraph random walk performs five random walks on each node, and the length of the random walk node sequence l is 9.

[0111] (3b) Define the loss function L of the scSPRITE data completion model O:

[0112]

[0113] Among them, K represents the number of negative samples, u i Represents the word vector of word i, v c The word vector representing the central word. Finally, the stochastic gradient descent (SGD) algorithm is used for iterative training, and the embedding dimension during training is 32.

[0114] Step 4: Iteratively train the scSPRITE data completion model O:

[0115] The node sequence L set obtained from the random walk of the hypergraph is used as the input of the Skip-gram algorithm, and the Skip-gram algorithm outputs the embedding vector of the node sequence.

[0116] Step 5: Get the interaction matrix of paired nodes:

[0117] Traverse all pairs of nodes and use the trained node embedding vectors to calculate the cosine similarity between pairs of nodes. The similarity above the threshold β is used as the interaction value to generate the interaction matrix.

[0118] Example 3

[0119] The effectiveness of the completion methods of the first and second embodiments can be verified by the following simulation method:

[0120] The goal of this experiment was to verify the effectiveness of SpriteHyper2vec completion by calculating the similarity between a small number of cell populations and baseline data at different resolutions and cell numbers, and conducting comparative experiments with other methods. A compartment scoring criterion was set to verify that model completion significantly improved compartment observations for a small number of cell populations. The experiment also calculated scores for all compartments on single-cell chromosomes and compared the score distributions.

[0121] The geodesic adjusted correlation coefficient (SCC) of the interaction matrix of a small number of swarm cells and the interaction matrix of swarm cells is calculated and compared with the decomposition-based method, which is implemented as follows:

[0122] (a) Aggregate all single-cell interaction matrices, perform z-score normalization on the aggregated interaction matrix, and use the processed interaction matrix as the interaction matrix of the group cells;

[0123] (b) All single-cell interaction matrices are sampled multiple times, and then the sampled single-cell interaction matrices are aggregated and z-score normalized. The processed interaction matrices are used as the interaction matrices of a small number of cell groups;

[0124] (c) Calculate the geodesic adjusted correlation coefficient (SCC) of the interaction matrix of a small number of cells and the interaction matrix of the group cells;

[0125] (d) Comparison of similarity coefficients of a small number of swarm cell-swarm cell interaction matrices between SpriteHyper2vec and decomposition-based methods;

[0126] (e) In the simulation experiment, the number of sampled single cells is 5, 10, 50, and 100, the sample size for each sampling is 20 groups, and the resolution is set to 100kb, 40kb, and 10kb, respectively.

[0127] Compute frequency-based compartmental scores for the interaction matrix of single cells and compare them with decomposition-based methods as follows:

[0128] (a) All single-cell interaction matrices were z-score normalized;

[0129] (b) Obtain the compartmental division file of the cell population from GSE154353 in the Gene Expression Omnibus (GSE);

[0130] (c) For each ABA or BAB slide compartment, calculate a frequency-based compartment score using the following formula:

[0131]

[0132] where c o represents the observed or completed intra-compartment interaction, cb represents the observed or completed inter-compartmental interaction, c a represents all possible compartment interactions, including intra-compartment interactions and inter-compartment interactions.

[0133] (d) The mean compartment scores of 1000 single cells were calculated.

[0134] The complement effect and compartment division accuracy of the present invention were simulated respectively, and the results are as follows: Figure 2 , as shown in Table 1, Table 2, and Table 3:

[0135] Simulation 1: In order to simulate the three-dimensional genome interaction of swarm cells and compare the similarity coefficients of the interaction matrix between a small number of swarm cells and swarm cells in the present invention with those in the prior art, in the simulation experiment:

[0136] (1a) Aggregate the interaction matrices of all single cells, perform z-score normalization on the aggregated interaction matrix, and use the processed interaction matrix as the interaction matrix of the group cells;

[0137] (1b) All single cells are sampled multiple times, and then the sampled single-cell interaction matrix is ​​aggregated and then z-score normalized. The processed interaction matrix is ​​used as the interaction matrix of a small number of group cells;

[0138] (1c) Calculate the geodesic adjusted correlation coefficient (SCC) of the interaction matrix of a small number of cells and the interaction matrix of the cells in the group;

[0139] (1d) Comparison of similarity coefficients of a small number of group cell-group cell interaction matrices between SpriteHyper2vec and decomposition-based methods;

[0140] (1e) In the simulation experiment, the number of sampled single cells is 5, 10, 50, and 100, the sample size for each sampling is 20 groups, and the resolution is set to 100kb, 40kb, and 10kb, respectively.

[0141] Reference Figure 2 , Figure 2 This is a line graph of SCC for SpriteHyper2vec and decomposition-based methods as the number of cells gradually increases, where the solid line represents SpriteHyper2vec and the dotted line represents the decomposition-based method. Figure 2 It can be seen that the similarity between a small number of cell groups and a small number of cell groups at the same resolution with different cell numbers is higher than that of the existing method, which proves the effectiveness of the model in completing scSPRITE data.

[0142] The present invention and the prior art perform multiple samplings at the same resolution with different cell numbers to obtain the average SCC values ​​as shown in Table 1.

[0143] Table 1 SCC of 2 methods in 4 different cell numbers and cell populations

[0144]

[0145] The present invention and the prior art perform multiple samplings at different resolutions with the same cell number to obtain the average SCC numbers, as shown in Table 2.

[0146] Table 2 SCC table of two methods at three different resolutions and cell groups

[0147]

[0148] Combining Tables 1 and 2, it can be seen that under different sampled cell numbers or different resolutions, the SCC mean of a small number of cell groups completed by SpriteHyper2vec is better than that of the existing method. Then, a T test was performed on the SCC distribution of multiple groups of samples (p < 0.0001), indicating that the scSPRITE data completed by the present invention enriches the structural information contained in a small number of cell groups and can better reflect the overall structure of the three-dimensional genome, verifying the effectiveness of the model completion.

[0149] Simulation 2: In order to simulate the compartment distribution of single cells and compare the compartment division of the single cell interaction matrix generated by the present invention with that of the prior art, in the simulation experiment:

[0150] (2a) All single-cell interaction matrices were z-score normalized;

[0151] (2b) Obtain the compartmental division file of the cell population from GSE154353 in the Gene Expression Omnibus (GSE);

[0152] (2c) For each ABA or BAB sliding compartment, calculate a frequency-based compartment score;

[0153] (2d) Calculate the mean compartment score of 1000 single cells.

[0154] The compartment scoring results obtained according to the present invention and the prior art verify the effectiveness of the complementation of the present invention.

[0155] The frequency-based compartment scores of the present invention and the prior art are shown in Table 3.

[0156] Table 3 Frequency-based scores of the two methods in five different compartments

[0157]

[0158]

[0159] As can be seen from Table 3, for the six compartments defined by swarm cells on chromosome 1, SpriteHyper2vec achieves higher compartment scores than decomposition-based methods. Comparing the compartment score distributions for all single cells, a T-test p-value less than 0.0001 confirms that this improvement is significant. This demonstrates that the present invention can effectively complement scSPRITE data and make compartmental delineation more distinct even for small swarm cells.

[0160] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0161] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature identified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0162] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they may refer to fixed connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0163] In the present invention, unless otherwise expressly specified or limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features being in contact not directly but through another feature between them. Furthermore, a first feature being "above," "above," and "above" a second feature may include the first feature being directly above or obliquely above the second feature, or may simply mean that the first feature is higher in level than the second feature. A first feature being "below," "below," and "below" a second feature may include the first feature being directly below or obliquely below the second feature, or may simply mean that the first feature is lower in level than the second feature.

[0164] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean 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 representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification.

[0165] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, and all of these should be considered to fall within the scope of protection of the present invention.

Claims

1. A scSPRITE data completion method based on hypergraph random walk, characterized in that: The following steps are involved: S1, obtain scSPRITE sequencing files; S2, extracting gene loci from the scSPRITE sequencing document, and obtaining a hyperedge set and a hypergraph node set for each single cell based on the gene loci; the hyperedge set and hypergraph node set for each single cell constitute a hypergraph for each single cell; S3, calculating the degree of each hyperedge in the hyperedge set of each hypergraph and the degree of each node in the hypergraph node set; S4, determining a walk transition probability of each node transferring to the next node based on the degree of each hyperedge and the degree of each node of each hypergraph, and performing a biased random walk on each hypergraph based on the walk transition probability to obtain a node vector of each hypergraph; S5, inputting the node vector of each hypergraph into the Node2Vec model for training, and outputting the node embedding vector of each hypergraph; S6, calculating the similarity of the embedding vectors of every two nodes in each hypergraph and screening to generate the interaction matrix of each single cell.

2. The scSPRITE data completion method based on hypergraph random walk according to claim 1, characterized in that: The step S2 comprises: S21, extracting all gene loci of the same chromosome field of the same single-cell multivariate interaction in the scSPRITE sequencing document, and constructing a preliminary hyperedge set based on all gene loci; S22, mapping the gene locus to the chromosome sequence interval bin as a hypergraph node; the expression of the hypergraph node is: Among them, c i Indicates the value of the i-th bin, b j represents the jth multivariate interaction of the same single cell, w∈b j Indicates b j Gene loci in S23, traversing the scSPRITE sequencing document according to step S21 and step S22 to obtain a preliminary hyperedge set and a preliminary hypergraph node set for each single cell; S24, removing hyperedges whose number of nodes in the preliminary hypergraph node set is greater than a threshold ε, to obtain a hyperedge set and a hypergraph node set for each single cell.

3. The scSPRITE data completion method based on hypergraph random walk according to claim 1, characterized in that: The calculation formula of the degree of each node is: Where E represents the set of hyperedges, ω represents the weight of each hyperedge, and h(,e) represents that node v exists in hyperedge e. If node v is on a hyperedge e, then h(,e) = 1, otherwise h(,e) = 0; The calculation formula of the degree of the hyperedge is: Wherein, V represents the hypergraph node set.

4. The scSPRITE data completion method based on hypergraph random walk according to claim 3, characterized in that: The step S4 comprises: S41, determining a first-order transition probability of each node transferring to the next node according to the degree of each hyperedge and the degree of each node of each hypergraph; S42, determining the second-order transition probability of each node transferring to the next node according to the first-order transition probability and the preset search bias; S43, normalizing the second-order transition probability to obtain a wandering transition probability of each node transferring to the next node; S44: Perform a biased random walk on each hypergraph according to the walk transition probability to obtain a node vector of each hypergraph.

5. The scSPRITE data completion method based on hypergraph random walk according to claim 4, characterized in that: The expression of the first-order transition probability is: Where π1(x|v) represents the first-order transition probability of node v transferring to the next node x, h(v,e) and h(x,e) respectively indicate that node v and node x exist in hyperedge e.

6. The scSPRITE data completion method based on hypergraph random walk according to claim 5, characterized in that: The expression of the second-order transition probability is: π2(x|v,u)=α(x|v,u)π1(x|v); Wherein, u represents the neighbor node of node v, and α represents the preset search bias; 7. The scSPRITE data completion method based on hypergraph random walk according to claim 1, characterized in that: The step S6 comprises: S61, calculating the similarity between each two node embedding vectors in each hypergraph; wherein the calculation formula for the similarity between each two node embedding vectors is: Wherein, f(X, Y) represents the cosine similarity between one node embedding vector X and another node embedding vector Y, and n represents the dimension of the node embedding vector; S62, taking the similarity greater than the threshold β as the interaction value, and generating the interaction matrix of each single cell.

Images