Single cell Hi-C atlas interpolation method and device

Abstract

The invention relates to the technical field of biological computing, in particular to a single-cell Hi-C atlas interpolation method and device, and the method comprises the steps: obtaining single-cell sequencing library data and a reference genome, determining the position of the sequencing data in the genome, and carrying out the statistics of the interaction frequency to obtain a Hi-C contact matrix; performing interpolation on the Hi-C contact matrix of any chromatin of the single cell according to similarity of adjacent nodes to obtain an adjacent node interpolation map of the single chromatin of the single cell; calculating an interpolation map of each chromatin of each cell, performing representation learning by taking the interpolation map as an adjacent matrix, splicing vectors to obtain a cell pre-embedded vector, and performing dimensionality reduction to obtain a whole cell embedded vector matrix; a cellular level similarity matrix is calculated by using a single cell embedded vector matrix and converted into a Markov probability transfer matrix, and the Markov probability transfer matrix is used for correcting an adjacent node interpolation map to obtain a Hi-C interpolation map, so that the interpolation accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of biological computing technology, and in particular to a single-cell Hi-C map interpolation method and device. Background Art

[0002] In the nucleus of eukaryotic cells, genetic material DNA and proteins combine to form chromatin, which has a very complex three-dimensional spatial structure. These spatial structures play an important role in regulating cell life activities. High-throughput chromosome conformation capture (Hi-C) technology can capture the interaction information of chromatin fragments across the entire genome and is an important means to study the three-dimensional spatial structure of chromatin. Hi-C technology obtains a large number of paired chromatin interaction sequence fragments by high-throughput sequencing of chromatin. These sequence fragments are divided into segments (bins) of fixed length, and through statistical counting, a two-dimensional Hi-C contact matrix, namely Hi-C map, is finally obtained. The matrix subscript represents the chromatin segment number, and the value in the matrix represents the interaction frequency of the corresponding segment. Hi-C maps can be used to conveniently study the key three-dimensional spatial structures of chromatin, including A / B compartments, topologically associated domains (TADs), chromatin bands, and chromatin loops.

[0003] Hi-C technology is divided into group cell Hi-C and single cell Hi-C technology according to the number of cells in the sequencing sample. The Hi-C contact map generated by group cell Hi-C sequencing has rich data and can be used more conveniently for subsequent key structure identification. However, due to the limitation of the number of polymerase chain reaction (PCR) amplifications in sequencing, the Hi-C map of a single cell is highly sparse and has missing parts, which is mathematically manifested as a large number of "zero values" in the Hi-C contact matrix. The large number of "zero values" causes the single cell Hi-C map to fail to truly reflect the spatial interaction of chromatin fragments, seriously affecting the analysis of the key three-dimensional spatial structure of chromatin. Therefore, it is usually necessary to interpolate the single cell Hi-C map generated by the single cell Hi-C sequencing experiment, and infer the missing chromatin structure information during the sequencing process based on the known information in the Hi-C map.

[0004] Among the existing methods, scHiCluster, Higashi, and HiC-SGL are based on graph theory, modeling single-cell Hi-C maps as adjacency matrices of graphs, or modeling all single cells in the entire sequencing library as a hypergraph, and predicting the interaction frequencies between unknown or missing chromatin segments through random walks or graph neural networks, thereby achieving the purpose of interpolation. scDEC-Hi-C is based on a generative model and uses an automatic encoding and decoding structure to generate interpolated single-cell Hi-C contact maps. HiCImpute is based on statistical inference and introduces indicator variables to distinguish between structural deletions in chromatin itself and technical deletions in sequencing, achieving the purpose of Hi-C map interpolation by predicting technical deletions.

[0005] However, after investigation and experiments by the inventors, it was found that the existing single-cell Hi-C map interpolation method still has the following problems: the computational complexity is high and it is not suitable for deployment in general scenarios; the proximity of cells in developmental timing is not fully considered, and the interpolation accuracy is low. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a single-cell Hi-C map interpolation method and device in view of the deficiencies in the above-mentioned prior art, so as to solve the technical problems of high computational complexity and low accuracy of the existing single-cell Hi-C map interpolation methods.

[0007] The objective of the present invention is achieved by the following technical solutions: In a first aspect, the present invention provides a single cell Hi-C map interpolation method, comprising: S1: Obtain the original sequencing data of a single cell in the single-cell sequencing library and the corresponding reference genome, determine the position of the original sequencing data in the reference genome, and count the interaction frequency of DNA fragments in each segment according to the position of the original sequencing data on the reference genome to obtain the Hi-C contact matrix; S2: For the Hi-C contact matrix of any chromatin in the single cell in the sequencing library, the similarity between adjacent nodes is used to interpolate, so as to obtain an adjacent node interpolation map of a single chromatin in a single cell; S3: Calculate the neighboring node interpolation map of each chromatin of each cell in the sequencing library, use the neighboring node interpolation map as an adjacency matrix for representation learning, obtain the vector representation of the same chromatin of each cell, splice the vector representations of different chromatins of the same cell to obtain the pre-embedded vector of the cell, and obtain the embedded vector matrix of all cells after the pre-embedded vector is reduced in dimension; S4: A cell-level similarity matrix is ​​calculated using the embedding vector matrix of a single cell, the similarity matrix is ​​converted into a Markov probability transfer matrix, and the graph of neighboring node interpolation is corrected using the probability transfer matrix to obtain a Hi-C interpolation graph.

[0008] As a further improvement of the present invention, the step S2 specifically includes: A weighted average operator is used to perform a convolution operation on the Hi-C contact matrix, and a smoothing operation is performed to obtain a smoothed Hi-C map; The Laplacian edge extraction operator is used to extract the grayscale gradient information in the smoothed Hi-C map; the grayscale gradient information biologically includes TADs, A / B compartments, and chromatin stripe information; Perform a restarted random walk operation on the Hi-C contact matrix to obtain the probability matrix of interactions between each node; After combining the obtained smoothed Hi-C map, grayscale gradient information and the probability matrix, a neighboring node interpolation map is obtained.

[0009] As a further improvement of the present invention, the restart random walk operation on the Hi-C contact matrix specifically includes: Calculate a normalized matrix of the Hi-C contact matrix, where the sum of elements in each row of the normalized matrix is ​​1; According to the set number of steps, the normalized matrix, the unit matrix and the restart probability are subjected to random walk calculation in an iterative manner until the first condition is met to obtain a probability matrix of interaction between the nodes; the first condition is that the number of iteration steps reaches the set number of steps, or the sum of the absolute values ​​of all element changes in the matrix calculated by the current number of steps is less than 0.1% of the sum of the elements of the matrix calculated by the previous number of steps.

[0010] As a further improvement of the present invention, the step S3 specifically includes: Sample and label all subgraphs from the weighted undirected graph corresponding to the neighboring node interpolation graph; Train a prediction model based on sampling and labeling all subgraphs; According to the trained prediction model, the existence probability of the subgraph in the input neighboring node interpolation graph is predicted to obtain an embedding layer vector; the prediction model adopts the skip-gram model; The embedding layer vectors representing different chromatin of the same cell obtained by learning are spliced ​​to obtain the pre-embedding vector of the cell, and the pre-embedding vector of the cell is reduced in dimension to obtain the final embedding vector matrix of all cells in the sequencing library, where each column of the embedding vector matrix represents a cell.

[0011] As a further improvement of the present invention, the specific method of reducing the dimension of the cell pre-embedded vector is truncated singular value decomposition, and the truncated singular value decomposition specifically includes:

[0012] In the formula, is the pre-embedded vector, is the dimension of the vector after dimensionality reduction, which is used to control the embedding vector to retain the similarity information between cells. is the left singular vector matrix, is the right singular vector matrix, is a diagonal matrix of singular values.

[0013] As a further improvement of the present invention, in the step S4, the specific calculation method of the cell-level similarity matrix is:

[0014] Where: is the cell-level similarity matrix, Y ·i The matrix Y represents i column vector, and Y is the cell embedding vector matrix.

[0015] As a further improvement of the present invention, the cell-level similarity matrix is ​​a normalized matrix, specifically comprising: The normalization formula is:

[0016] In the formula, is the normalized similarity matrix, D is the degree matrix of S, the degree matrix refers to the diagonal matrix composed of the sum of elements in each row of the similarity matrix, and S is the similarity matrix at the cell level.

[0017] As a further improvement of the present invention, the Markov probability transfer matrix M in step S4 specifically includes:

[0018] Where M is the corresponding probability transfer matrix. t After diffusion, the corresponding probability transfer matrix is , is the normalized similarity matrix.

[0019] As a further improvement of the present invention, the diffusion Markov probability transfer matrix is ​​used to correct the neighboring node interpolation Hi-C map, specifically including:

[0020] The Hi-C map matrix after interpolation of all cells in the sequencing library is flattened into a matrix composed of one-dimensional row vectors, and is the probability transfer matrix corresponding to diffusion T times.

[0021] In a second aspect, the present invention provides a single-cell Hi-C map interpolation device, which is used to implement the above-mentioned single-cell Hi-C map interpolation method, comprising: The data acquisition module is used to obtain the sequence pairs obtained from the original Hi-C sequencing experiment and the reference genome of the corresponding species; Hi-C map generation module, which is used to align sequencing sequences, determine the locations of interacting DNA fragments on the reference genome, segment chromatin and count the frequency of chromatin interactions in each segment, and assemble into a Hi-C contact matrix; A neighboring node interpolation module is used to interpolate the Hi-C contact matrix of any chromatin in a single cell in the sequencing library using the similarity between neighboring nodes to obtain a neighboring node interpolation map of a single chromatin in a single cell; The whole-graph embedding module is used to calculate the neighboring node interpolation map of each chromatin of each cell in the sequencing library, and perform whole-graph embedding based on the neighboring node interpolation map to obtain the cell-level embedding representation of the Hi-C contact matrix of each chromatin; The neighboring cell interpolation module is used to calculate the cell-level similarity matrix using the embedded vector matrix of a single cell, convert the similarity matrix into a Markov probability transfer matrix, and use the probability transfer matrix to correct the neighboring node interpolation map to obtain the Hi-C interpolation map.

[0022] In a third aspect, the present invention provides a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, enable the computing device to perform the above-mentioned single-cell Hi-C map interpolation method.

[0023] In a fourth aspect, the present invention provides a computing device, comprising: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include steps for executing the above-mentioned single-cell Hi-C map interpolation method.

[0024] The beneficial effects of the present invention are as follows: a single-cell Hi-C map interpolation method provided by the present invention obtains the original sequencing data of a single cell and compares it with the reference genome, accurately counts the frequency of DNA fragment interactions to obtain a Hi-C contact matrix, provides basic data for subsequent analysis, and interpolates using the similarity of neighboring nodes, which can effectively make up for the missing parts in the data and obtain a more complete neighboring node interpolation map of a single cell single chromatin. The interpolation map is used as an adjacency matrix for representation learning and splicing vectors. The embedded vector matrix obtained after dimensionality reduction can better represent cell characteristics and help to deeply analyze the differences and similarities between cells. In addition, all Hi-C maps in the sequencing library of the present invention are interpolated in parallel, and the Hi-C maps of all cells in the library can be interpolated in one operation, which reduces redundant calculations and speeds up the operation efficiency; the neighboring relationship in the cell development order is considered, and the Hi-C contact matrix of cells with closer developmental timing is used to estimate the Hi-C contact matrix of the target cell, including more information, and improving the interpolation accuracy.

[0025] Furthermore, by performing convolution and smoothing operations on the Hi-C contact matrix to obtain a smooth Hi-C map, the noise interference in the data can be effectively reduced, making subsequent analysis more accurate and reliable. The Laplacian edge extraction operator is used to extract the grayscale gradient information in the smooth Hi-C map, which can accurately capture important biological structural information such as TADs (topologically associated domains), A / B compartments, and chromatin stripes. The Hi-C contact matrix is ​​restarted and random walks are performed to obtain the probability matrix of the interaction between each node, which provides a quantitative indicator for analyzing the connection strength and dynamic changes between different chromosome regions. The probability matrix can reflect the potential connections and interaction patterns between chromosome regions, providing a new perspective for in-depth research on chromosome structure and function.

[0026] Furthermore, the normalized matrix is ​​calculated so that the sum of the elements in each row is 1, thus achieving data standardization. This allows the interaction strengths of different nodes to be compared at the same scale, avoiding analytical bias caused by differences in the magnitude of the original data. The iterative calculation method, combined with the set number of steps and strict first conditions, ensures the accuracy and stability of the calculation. Through step-by-step iteration, the algorithm can fully explore the data space and find the most reasonable interaction probability matrix for each node. At the same time, strict convergence conditions avoid stopping too early or falling into a local optimal solution. Random walk calculations simulate the dynamic interaction process between chromosome regions in reality. This simulation is more in line with the actual biological situation, and can capture the dynamic changes in chromosome structure and potential connection patterns, providing a more realistic model for studying the three-dimensional structure and function of chromosomes. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0028] Figure 1 4 is a flow chart of the single-cell Hi-C map interpolation method provided in the first embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of the principle of single-cell Hi-C image representation learning provided by the first embodiment of the present invention.

[0030] Figure 3 4 is a system block diagram of a single-cell Hi-C atlas interpolation device provided in the second embodiment of the present invention.

[0031] Figure 4 It is a heat map comparison diagram of the present invention applied to the data before and after interpolation of the GM12878 single cell Hi-C map; Figure 5 It is a schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose and technical solution of the present invention clearer and easier to understand, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0033] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments, wherein the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.

[0034] Example 1 like Figure 1-Figure 5 As shown, this embodiment provides a single-cell Hi-C map interpolation method, and the specific implementation method is as follows.

[0035] S1: Obtain the original sequencing data of single cells in the single-cell sequencing library and the corresponding reference genome, determine the position of the original sequencing data in the reference genome, and count the interaction frequency of DNA fragments in each segment according to the position of the original sequencing data on the reference genome to obtain the Hi-C contact matrix.

[0036] Specifically, for a given input single-cell sequencing library, the original sequencing data, i.e., sequence, of each single cell in the library is obtained, and the sequence is aligned to the reference genome to determine its position on the reference genome, where the reference genome is consistent with the species of the input single-cell library; the reference genome is divided according to the chromatin number and the specified bin size, and the interaction frequency of DNA fragments in each bin is counted according to the position of the sequence on the reference genome to form a Hi-C contact matrix (i.e., Hi-C contact matrix).

[0037] S2: The Hi-C contact matrix of any chromatin in a single cell in the sequencing library is interpolated using the similarity between neighboring nodes to obtain a neighboring node interpolation map of a single chromatin in a single cell.

[0038] Wherein, for the input Hi-C map matrix A, a weighted average operator is used to perform a convolution operation on the Hi-C contact matrix, and a smoothing operation is performed to obtain a smoothed Hi-C map A1; The Laplacian edge extraction operator is used to extract the grayscale gradient information A2 in the smoothed Hi-C map; the grayscale gradient information biologically includes structural information such as TADs, A / B compartments, and chromatin stripes; Perform a Random Walk with Restart (RWR) operation on the Hi-C contact matrix to obtain the probability matrix A3 of the interaction between each node; After combining the obtained smoothed Hi-C map, grayscale gradient information and probability matrix, we get the neighboring node interpolation map, where the neighboring node interpolation map is: B=A1+A2+A3.

[0039] The weighted average operator in this implementation is: ; The Laplacian edge extraction operator in this implementation is ; The restarted random walk process in this implementation is as follows. First, the row normalization matrix Q of the matrix A is calculated to ensure that the sum of the elements in each row is 1:

[0040] P t For passing t The matrix after the random walk, P0=I, I is the unit matrix, and the restart probability is r , then P t It can be calculated using the following formula:

[0041] This process is iterated until t Reaching the maximum setting value or The sum of the absolute values ​​of the changes of all elements in is less than 0.1% of the sum of the elements, A3 is when the iteration stops .

[0042] S3: Calculate the neighboring node interpolation map of each chromatin of each cell in the sequencing library, use the neighboring node interpolation map as an adjacency matrix for representation learning, obtain the vector representation of the same chromatin of each cell, splice the vector representations of different chromatins of the same cell to obtain the pre-embedded vector of the cell, and obtain the embedded vector matrix of all cells after dimensionality reduction of the pre-embedded vector.

[0043] Among them, all subgraphs are sampled and labeled from the weighted undirected graph corresponding to the neighboring node interpolation graph; The prediction model is trained by sampling and labeling all subgraphs; the prediction model adopts the skip-gram model, which includes an input layer, an embedding layer, and an output layer, such as Figure 2 shown.

[0044] According to the trained prediction model, the existence probability of the subgraph in the input neighboring node interpolation graph is predicted to obtain the embedding layer vector; The embedding layer vectors of different chromatin of the same cell obtained by graph representation learning are spliced ​​to obtain the pre-embedding vector of the cell. The pre-embedding vector of the cell is reduced in dimension to obtain the final embedding vector matrix of all cells in the sequencing library. Each column of the embedding vector matrix represents a cell.

[0045] Specifically, the specific method of reducing the dimension of the cell pre-embedded vector is truncated singular value decomposition, which specifically includes:

[0046] In the formula, is the pre-embedded vector, is the dimension of the vector after dimensionality reduction. The dimension is used to control the embedding vector to retain the similarity information between cells without introducing redundant calculations. is the left singular vector matrix, is the right singular vector matrix, is a diagonal matrix of singular values.

[0047] Furthermore, the embedding vector matrix Y of all cells in the sequencing library can be expressed as:

[0048] Each column vector of Y represents the embedding vector of a single cell.

[0049] S4: The cell-level similarity matrix is ​​calculated using the single-cell embedding vector matrix, and the similarity matrix is ​​converted into a Markov probability transfer matrix. The probability transfer matrix is ​​used to correct the neighboring node interpolation map to obtain the Hi-C interpolation map.

[0050] Among them, the cell-level similarity matrix is ​​calculated using the embedding vector matrix of the single cell, and the proximity measurement between all single cells in the sequencing library is established; the similarity measurement used in this implementation is cosine similarity, and the specific calculation method is:

[0051] Where: is the cell-level similarity matrix, Y ·i The matrix Y represents i A column vector.

[0052] Furthermore, in this embodiment, the cell-level similarity matrix is ​​regarded as the adjacency matrix of the cell-level graph, and degree normalization is applied to balance the contribution of each cell to the interpolation of the developmentally adjacent cells to obtain a degree-normalized similarity matrix. The specific method of degree normalization is as follows:

[0053] Where: is the normalized similarity matrix, D is the degree matrix of S, the degree matrix refers to the diagonal matrix composed of the sum of elements in each row of the similarity matrix, and S is the cell-level similarity matrix.

[0054] The cell similarity matrix is ​​processed by row random normalization to convert it into a Markov probability transfer matrix, where each row is a probability distribution and the sum is equal to 1. The probability transfer matrix is ​​diffused multiple times to further enhance the weights of nodes corresponding to more similar cells in the matrix, while smoothing the matrix and suppressing background noise.

[0055] The specific calculation method of the Markov probability transfer matrix M in this implementation is:

[0056] M is the corresponding probability transfer matrix, after t After diffusion, the corresponding probability transfer matrix is , is the normalized similarity matrix.

[0057] The process of probability matrix diffusion corresponds to a random walk on the graph, which is a power operation in mathematical representation, that is, after t After diffusion, the corresponding probability transfer matrix is The diffusion time is determined according to the change amplitude of the interpolated data. When the ratio of the sum of the absolute values ​​of the changes of all elements in the Hi-C spectrum matrix after interpolation to the sum of the absolute values ​​of all elements in the Hi-C spectrum matrix is ​​less than 0.1%, the diffusion stops. T , then the probability transfer matrix after diffusion is .

[0058] The diffused Markov probability transfer matrix is ​​used to correct the neighboring node interpolation Hi-C map. The corrected map is the Hi-C map that considers the interpolation of node and cell proximity in the library at the same time, and is also the final Hi-C interpolation map output by this method.

[0059] In this embodiment, the specific method of applying the diffused probability transfer matrix to correct the neighboring nodes to interpolate the Hi-C map B is as follows: The upper triangular matrices of the Hi-C contact maps of all cells in the sequencing library are flattened into one-dimensional row vectors, and these vectors are stacked row by row to construct a row vector matrix H. The probability transfer matrix after diffusion is used to interpolate H, that is:

[0060] In the formula, The Hi-C profile matrix after interpolation for all cells in the library is flattened into a matrix consisting of one-dimensional row vectors.

[0061] Furthermore, if it is necessary to obtain the Hi-C map after interpolation of any cell in the library, the corresponding row vector is extracted and the reshape function in the numpy scientific computing package of Python is applied to reorganize it into a square matrix.

[0062] Taking the interpolation of the Hi-C map of chromatin No. 1 of the GM12878 single cell as an example, the single cell Hi-C map interpolation method of this embodiment is introduced.

[0063] Input the single-cell sequencing library, use the samtools sequence alignment tool to align the sequencing sequence to the human reference genome hg38, and divide the genome into 50kb segments. Count the original uninterpolated single-cell Hi-C contact matrix, as shown below: Figure 4 As shown, the left subfigure shows the corresponding single-cell Hi-C map.

[0064] The weighted average operator is applied to smooth the single-cell Hi-C map, the Laplacian operator is used to extract the edge, the random walk is restarted to infer the interaction probability between two segments, and the single-cell Hi-C map with neighboring node interpolation is obtained after normalization and summation. Figure 4 As shown in the middle sub-figure, it can be seen that after interpolation of neighboring nodes, the detailed information of the single-cell Hi-C map is preliminarily restored.

[0065] Whole-graph representation learning is applied to calculate the low-dimensional representation vector of each single cell, and the representation vector is used to construct the similarity matrix between cells, and degree normalization is applied to balance the number of effective neighbors of each cell.

[0066] The Markov probability transfer matrix is ​​further constructed according to the cell similarity matrix, and random walk diffusion is performed.

[0067] The diffused probability transfer matrix is ​​applied to the single-cell Hi-C map interpolated by neighboring nodes, and the Hi-C map is further modified by considering the proximity of cells. The modified single-cell Hi-C map is as follows Figure 4 As shown in the right subfigure, it can be seen that after cell proximity correction, the information of the single-cell Hi-C map is richer, especially the long-range chromatin interactions far away from the diagonal are also restored.

[0068] Example 2 like Figure 3 As shown, this embodiment provides a single-cell Hi-C map interpolation device, which is used to implement the single-cell Hi-C map interpolation method in the above embodiment 1, including: The data acquisition module is used to obtain the sequence pairs obtained from the original Hi-C sequencing experiment and the reference genome of the corresponding species; Hi-C map generation module, which is used to align sequencing sequences, determine the locations of interacting DNA fragments on the reference genome, segment chromatin and count the frequency of chromatin interactions in each segment, and assemble into a Hi-C contact matrix; A neighboring node interpolation module is used to interpolate the Hi-C contact matrix of any chromatin in a single cell in the sequencing library using the similarity between neighboring nodes to obtain a neighboring node interpolation map of a single chromatin in a single cell; The whole-graph embedding module is used to calculate the neighboring node interpolation map of each chromatin of each cell in the sequencing library, and perform whole-graph embedding based on the neighboring node interpolation map to obtain the cell-level embedding representation of the Hi-C contact matrix of each chromatin; The neighboring cell interpolation module is used to calculate the cell-level similarity matrix using the embedded vector matrix of a single cell, convert the similarity matrix into a Markov probability transfer matrix, and use the probability transfer matrix to correct the neighboring node interpolation map to obtain the Hi-C interpolation map.

[0069] In addition, the device also includes a visualization module for displaying heat maps of interpolated single-cell Hi-C contact maps, from which key chromatin structures such as TADs can be discovered.

[0070] Example 3 In another embodiment of the present invention, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device, and can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that more specific examples (non-exhaustive list) of the computer-readable storage medium here include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0071] Computer readable storage media also include data signals propagated in baseband or as part of a carrier wave, which carry readable program codes. Such propagated data signals can take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable storage medium can also be any readable medium other than a readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, device, or device. The program code contained on the readable storage medium can be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0072] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0073] The processor may load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the single-cell Hi-C profile interpolation method in the above embodiment; the processor may load and execute the following steps: S1: Obtain the original sequencing data of a single cell in the single-cell sequencing library and the corresponding reference genome, determine the position of the original sequencing data in the reference genome, and count the interaction frequency of DNA fragments in each segment according to the position of the original sequencing data on the reference genome to obtain the Hi-C contact matrix; S2: For the Hi-C contact matrix of any chromatin in the single cell in the sequencing library, the similarity between adjacent nodes is used to interpolate, so as to obtain an adjacent node interpolation map of a single chromatin in a single cell; S3: Calculate the neighboring node interpolation map of each chromatin of each cell in the sequencing library, use the neighboring node interpolation map as an adjacency matrix for representation learning, obtain the vector representation of the same chromatin of each cell, splice the vector representations of different chromatins of the same cell to obtain the pre-embedded vector of the cell, and obtain the embedded vector matrix of all cells after the pre-embedded vector is reduced in dimension; S4: A cell-level similarity matrix is ​​calculated using the embedding vector matrix of a single cell, the similarity matrix is ​​converted into a Markov probability transfer matrix, and the graph of neighboring node interpolation is corrected using the probability transfer matrix to obtain a Hi-C interpolation graph.

[0074] Example 4 See also Figure 5 , the terminal device is a computer device, and the computer device 60 of this embodiment includes: a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When the computer program 63 is executed by the processor 61, the single-cell Hi-C atlas interpolation method in the embodiment is implemented. To avoid repetition, it is not described one by one here. Alternatively, when the computer program 63 is executed by the processor 61, the functions of each model / unit in the single-cell Hi-C atlas interpolation component computing system in the embodiment are implemented. To avoid repetition, it is not described one by one here.

[0075] The computer device 60 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will appreciate that Figure 5This is only an example of the computer device 60 and does not constitute a limitation of the computer device 60. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, etc.

[0076] The processor 61 may be a central processing unit (CPU), or other general-purpose processors, central processing units, graphics processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, data processing logic based on quantum computing, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0077] The memory 62 may be an internal storage unit of the computer device 60, such as a hard disk or memory of the computer device 60. The memory 62 may also be an external storage device of the computer device 60, such as a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 60.

[0078] Furthermore, the memory 62 may include both an internal storage unit of the computer device 60 and an external storage device. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 may also be used to temporarily store data that has been output or is to be output.

[0079] Any reference to a memory, database or other medium used in the embodiments provided in this application may include at least one of a non-volatile and a volatile memory. Non-volatile memory may include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical memory, a high-density embedded non-volatile memory, a resistive random access memory (ReRAM), a magnetic random access memory (MRAM), a ferroelectric random access memory (FRAM), a phase change memory (PCM), a graphene memory, etc. Volatile memory may include a random access memory (RAM) or an external cache memory, etc. As an illustration and not limitation, RAM may be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM), etc.

[0080] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

Claims

1. A single cell Hi-C map interpolation method, characterized in that: include: S1: Obtain the original sequencing data of a single cell in the single-cell sequencing library and the corresponding reference genome, determine the position of the original sequencing data in the reference genome, and count the interaction frequency of DNA fragments in each segment according to the position of the original sequencing data on the reference genome to obtain the Hi-C contact matrix; S2: For the Hi-C contact matrix of any chromatin in the single cell in the sequencing library, the similarity between adjacent nodes is used to interpolate, so as to obtain an adjacent node interpolation map of a single chromatin in a single cell; S3: Calculate the neighboring node interpolation map of each chromatin of each cell in the sequencing library, use the neighboring node interpolation map as an adjacency matrix for representation learning, obtain the vector representation of the same chromatin of each cell, splice the vector representations of different chromatins of the same cell to obtain the pre-embedded vector of the cell, and obtain the embedded vector matrix of all cells after the pre-embedded vector is reduced in dimension; S4: A cell-level similarity matrix is ​​calculated using the embedding vector matrix of a single cell, the similarity matrix is ​​converted into a Markov probability transfer matrix, and the graph of neighboring node interpolation is corrected using the probability transfer matrix to obtain a Hi-C interpolation graph.

2. The single cell Hi-C profile interpolation method according to claim 1, characterized in that: The S2 step specifically includes: A weighted average operator is used to perform a convolution operation on the Hi-C contact matrix, and a smoothing operation is performed to obtain a smoothed Hi-C map; The Laplacian edge extraction operator is used to extract the grayscale gradient information in the smoothed Hi-C map; the grayscale gradient information biologically includes TADs, A / B compartments, and chromatin stripe information; Perform a restarted random walk operation on the Hi-C contact matrix to obtain the probability matrix of interactions between each node; After combining the obtained smoothed Hi-C map, grayscale gradient information and the probability matrix, a neighboring node interpolation map is obtained.

3. The single cell Hi-C profile interpolation method according to claim 2, characterized in that: The restart random walk operation on the Hi-C contact matrix specifically includes: Calculate a normalized matrix of the Hi-C contact matrix, where the sum of elements in each row of the normalized matrix is ​​1; According to the set number of steps, the normalized matrix, the unit matrix and the restart probability are subjected to random walk calculation in an iterative manner until the first condition is met to obtain a probability matrix of interaction between the nodes; the first condition is that the number of iteration steps reaches the set number of steps, or the sum of the absolute values ​​of all element changes in the matrix calculated by the current number of steps is less than 0.1% of the sum of the elements of the matrix calculated by the previous number of steps.

4. The single cell Hi-C profile interpolation method according to claim 1, characterized in that: The S3 step specifically includes: Sample and label all subgraphs from the weighted undirected graph corresponding to the neighboring node interpolation graph; Train a prediction model based on sampling and labeling all subgraphs; According to the trained prediction model, the existence probability of the subgraph in the input neighboring node interpolation graph is predicted to obtain an embedding layer vector; the prediction model adopts the skip-gram model; The embedding layer vectors representing different chromatin of the same cell obtained by learning are spliced ​​to obtain the pre-embedding vector of the cell, and the pre-embedding vector of the cell is reduced in dimension to obtain the final embedding vector matrix of all cells in the sequencing library, where each column of the embedding vector matrix represents a cell.

5. The single cell Hi-C profile interpolation method according to claim 4, characterized in that: The specific method of reducing the dimension of the cell pre-embedded vector is truncated singular value decomposition, which specifically includes: In the formula, is the pre-embedded vector, is the dimension of the vector after dimensionality reduction, which is used to control the embedding vector to retain the similarity information between cells. is the left singular vector matrix, is the right singular vector matrix, is a diagonal matrix of singular values.

6. The single cell Hi-C profile interpolation method according to claim 1, characterized in that: In the above-mentioned step S4, the specific calculation method of the cell-level similarity matrix is ​​as follows: Where: is the cell-level similarity matrix, Y ·i The matrix Y represents i column vector, and Y is the cell embedding vector matrix.

7. The single cell Hi-C profile interpolation method according to claim 6, characterized in that: The cell-level similarity matrix is ​​a normalized matrix, specifically including: The normalization formula is: In the formula, is the normalized similarity matrix, D is the degree matrix of S, the degree matrix refers to the diagonal matrix composed of the sum of elements in each row of the similarity matrix, and S is the similarity matrix at the cell level.

8. The single cell Hi-C profile interpolation method according to claim 7, characterized in that: The Markov probability transfer matrix M in step S4 specifically includes: Where M is the corresponding probability transfer matrix. t After diffusion, the corresponding probability transfer matrix is , is the normalized similarity matrix.

9. The single cell Hi-C profile interpolation method according to claim 1, characterized in that: Apply the diffused Markov probability transfer matrix to correct the neighboring nodes and interpolate the Hi-C map, including: The Hi-C map matrix after interpolation of all cells in the sequencing library is flattened into a matrix composed of one-dimensional row vectors, and is the probability transfer matrix corresponding to diffusion T times.

10. A single cell Hi-C map interpolation device, used to implement the single cell Hi-C map interpolation method according to any one of claims 1 to 9, characterized in that: include: The data acquisition module is used to obtain the sequence pairs obtained from the original Hi-C sequencing experiment and the reference genome of the corresponding species; Hi-C map generation module, which is used to align sequencing sequences, determine the locations of interacting DNA fragments on the reference genome, segment chromatin and count the frequency of chromatin interactions in each segment, and assemble into a Hi-C contact matrix; A neighboring node interpolation module is used to interpolate the Hi-C contact matrix of any chromatin in a single cell in the sequencing library using the similarity between neighboring nodes to obtain a neighboring node interpolation map of a single chromatin in a single cell; The whole-graph embedding module is used to calculate the neighboring node interpolation map of each chromatin of each cell in the sequencing library, and perform whole-graph embedding based on the neighboring node interpolation map to obtain the cell-level embedding representation of the Hi-C contact matrix of each chromatin; The neighboring cell interpolation module is used to calculate the cell-level similarity matrix using the embedded vector matrix of a single cell, convert the similarity matrix into a Markov probability transfer matrix, and use the probability transfer matrix to correct the neighboring node interpolation map to obtain the Hi-C interpolation map.

Images