Whole genome single cell differential methylation region identification method and device

Abstract

The invention discloses a differential methylation region identification method and device for whole genome single cell methylation data. The method comprises the following steps: preprocessing single cell whole genome methylation sequencing data to obtain a methylation signal value matrix; compressing the methylated signal value matrix, dividing the methylated signal value matrix into data blocks, and performing block storage by using an HDF5 format; establishing a regional index for the data structure in the HDF5 format, and establishing a dynamic updating mechanism; extracting the data block index information and loading the data block index information into a memory; the index only loads the currently calculated data block; the quality control efficiency is optimized, and GPU acceleration calculation is carried out; in combination with a dynamic optimization strategy, performing whole genome scanning to obtain a single cell methylation signal; the single cell methylation signal is parsed to identify a differential methylation region. According to the method disclosed by the invention, the processing speed is greatly improved, and the differential methylation region can be identified more efficiently and accurately.

Description

Technical Field

[0001] This invention relates to the fields of bioinformatics and data processing technology, specifically to a method and apparatus for identifying differentially methylated regions in single cells across the entire genome. Background Technology

[0002] With the rapid development of genomics, DNA methylation, as an important epigenetic modification, has gradually become a key research area for revealing gene expression regulation mechanisms, cell differentiation, developmental processes, and disease occurrence and development. DNA methylation typically occurs in the promoter regions of genes in cells, regulating cellular function by affecting gene transcriptional activity or genome stability. In recent years, breakthroughs in sequencing technology have enabled researchers to obtain increasingly high-resolution methylation data, providing crucial data support for a deeper understanding of genome function and epigenetic mechanisms. However, with the increase in data dimensionality and research complexity, the efficient processing and analysis of methylation data also faces many challenges.

[0003] Especially in single-cell methylation studies, the sheer volume and complexity of methylation data have increased dramatically, posing unprecedented challenges to data storage, computation, and analysis. The methylation information for each single cell is extremely fine-grained, involving millions of CpG sites, making the analysis process complex and computationally resource-intensive. While the detail of single-cell methylation data provides strong evidence for revealing gene regulatory mechanisms, how to efficiently process this massive and complex data has become a critical problem that urgently needs to be solved in the field of bioinformatics.

[0004] Currently, traditional methylation data analysis methods have several significant drawbacks: 1. Quality and Data Consistency Issues: Methylation data, especially single-cell methylation data, typically originates from high-throughput sequencing technologies and is susceptible to sample preparation, experimental bias, and technical noise. Since methylation signal acquisition relies on multiple experimental steps, data quality is easily affected by experimental errors, technical bias, and batch effects. Existing methods lack effective quality control mechanisms, failing to adequately identify and correct these potential quality issues early in the data processing stages. For example, sequencing errors or sample contamination can lead to data inconsistency, affecting subsequent identification of differentially methylated regions (DMRs) and the mining of methylation biomarkers. While some traditional data preprocessing methods attempt quality control and data cleaning, these methods often struggle to handle complex quality issues when dealing with extremely large-scale and highly complex single-cell methylation data. Existing technologies fail to effectively address the impact of experimental bias on data consistency, resulting in potentially significant noise and inconsistencies in the processed data, thus affecting the final analytical results.

[0005] 2. Large Storage Space Consumption Issue: Traditional methylation data storage methods typically rely on contiguous storage structures. While this structure works effectively for small-scale data analysis, its storage space consumption increases significantly when dealing with massive and complex whole-genome single-cell methylation data. When storing genome-wide methylation signals, traditional methods often fail to provide efficient data compression and storage management, resulting in huge storage requirements. This is especially true in single-cell methylation studies, where the methylation information for each single cell is highly granular and the data volume is enormous, leading to a surge in storage demands. Current storage solutions fail to consider optimizations for on-demand reading and writing. Data storage in traditional structures often results in excessive redundant data, which not only increases storage costs but also reduces data processing efficiency. More importantly, existing storage structures struggle to achieve efficient data compression and flexible access in practical applications, further increasing the system's storage space requirements.

[0006] 3. Memory Consumption Issues in Computational Sets: Memory consumption is a significant problem in existing methylation data analysis methods, especially for single-cell data processing. Due to the high dimensionality of methylation data, processing requires loading large amounts of data into memory, leading to high memory consumption and potentially memory overflow, thus affecting the smooth operation of the computation. Traditional methods typically process data in a contiguous manner, resulting in memory consumption and computational bottlenecks, especially when single-cell methylation datasets are very large. Furthermore, because each cell in a single-cell methylation dataset contains a large amount of information, and the methylation data matrix typically consists of millions of CpG sites, traditional memory management mechanisms cannot effectively optimize memory usage, easily leading to low computational efficiency and insufficient memory. In existing methods, the memory consumption problem of computational sets is often not effectively addressed; data needs to be loaded into memory all at once, greatly limiting the scale and efficiency of the analysis.

[0007] 4. Computational Speed ​​Issues: Due to the high dimensionality and complexity of methylation data, traditional whole-genome methylation data analysis methods require handling a large number of computational tasks. These tasks include the identification of differentially methylated regions (DMRs), the calculation of methylation signals, and the prediction of biomarkers based on these signals. However, existing computational methods often rely on computationally intensive operations, requiring significant computation time, especially when processing large-scale datasets, where computational speed is very slow. Currently, many traditional analysis methods have significant performance bottlenecks, particularly when dealing with data from millions of CpG sites, where computational speed drops sharply, resulting in excessively long processing times and even failing to meet the needs of real-time analysis. This computational speed bottleneck not only affects experimental efficiency but also limits the large-scale application of single-cell methylation data, especially in the field of genomics, where real-time performance and efficiency are essential.

[0008] The speed of methylation biomarker discovery is a critical issue in methylation data analysis. Identifying differentially methylated regions (DMRs) and further mining methylation biomarkers is a core task. However, existing methods suffer from significant time delays in locating and screening DMRs, hindering the speed of biomarker discovery for large-scale datasets. Traditional methods typically rely on static scanning strategies and cannot dynamically adjust computational regions and jump points, resulting in high latency and inefficiency in biomarker identification. This is particularly true for single-cell methylation data, where traditional methods often rely too heavily on coarse region segmentation and fail to dynamically optimize based on the actual distribution of methylation signals. This makes biomarker screening both slow and inaccurate. The inability of existing methods to fully utilize efficient dynamic optimization algorithms further hinders the timely discovery of potential DMRs and biomarkers, preventing efficient and accurate screening. Summary of the Invention

[0009] To address the shortcomings of the existing technologies, this invention provides a method and apparatus for identifying differentially methylated regions in single cells across the entire genome.

[0010] Firstly, a method for identifying differentially methylated regions in whole-genome single-cell methylation data is provided, including: Single-cell whole-genome methylation sequencing data are preprocessed to obtain a methylation signal value matrix; the preprocessing includes reordering, quality control, bias correction, methylation signal extraction, and format standardization. The methylation signal value matrix is ​​compressed and divided into data blocks, and then stored in blocks using the HDF5 format. A region index is built for the HDF5 format data structure, and a dynamic update mechanism is established; the data block index information is extracted and loaded into memory; the index only loads the data block currently being calculated; The data analysis process is optimized by adopting a lazy evaluation strategy, which improves the efficiency of quality control and enables GPU-accelerated computation. By combining dynamic optimization strategies, whole-genome scanning was performed to obtain single-cell methylation signals; a probability model and signal jump point index were constructed based on Gaussian density estimation to analyze single-cell methylation signals and identify differentially methylated regions.

[0011] In one possible implementation, the preprocessing of single-cell whole-genome methylation sequencing data to obtain a methylation signal value matrix includes: Single-cell whole-genome methylation sequencing data were reordered according to chromosome position; The sequencing depth and coverage of the reordered sequencing data are evaluated, the consistency and error rate of the data are calculated, and data with poor consistency and high error rate are removed to obtain the first sequencing data. Low-quality data were removed from the first sequencing data to obtain the second sequencing data; The second sequencing data is bias-corrected to obtain the third sequencing data; Methylation signals were extracted from the third sequencing data to obtain the fourth sequencing data; The fourth sequencing data was formatted to obtain a methylation signal value matrix.

[0012] In one possible implementation, the step of building a region index for the HDF5 format data structure and establishing a dynamic update mechanism; extracting the data block index information and loading it into memory includes: A hash table is constructed by mapping each data block to a unique hash value using a hash index; Constructing a regional index: Determine a non-overlapping window of a certain size as the basic unit; construct regional identifier keys using the "chromosome:start-end" format; select a data structure based on query requirements; the data structure is suitable for dynamic regional and overlapping queries; associate each regional key with a set of data block hash primary keys to establish a mapping relationship between key values ​​and primary key values, thereby constructing the regional index; Establish dynamic update mechanisms: data change perception and response mechanism; index dynamic maintenance mechanism; dynamic maintenance of index data structure.

[0013] In one possible implementation, the adoption of a lazy evaluation strategy to optimize the data analysis process, improve quality control efficiency, and perform GPU-accelerated computation includes: The computation process is broken down into multiple subtasks using the DelayedArray framework; each subtask corresponds to the operation of a data block. Optimize quality control efficiency: Use regional indexes for accurate location and load only gene regions and cell data blocks relevant to the current differential methylation region identification task; use hash indexes that uniquely identify data blocks for data quality control.

[0014] In one possible implementation, the method involves combining a dynamic optimization strategy to perform a whole-genome scan to obtain single-cell methylation signals; constructing a probability model and signal jump point index based on Gaussian density estimation to analyze single-cell methylation signals to identify differentially methylated regions, including: By combining dynamic optimization strategies, whole-genome scanning was performed to extract methylation signals at the single-cell level; The entire genome is divided into multiple small regions, and the methylation signal is estimated by Gaussian density within each small region to construct a region-level probability model. Based on the signal jump point index, the data is dynamically segmented according to the differences in methylation signals in different small regions to obtain dynamic segmentation results, thereby obtaining candidate difference region boundaries with clear structure. Based on the probability model and candidate differential methylation region boundaries, single-cell methylation signals are analyzed using a deep learning model to ultimately identify differentially methylated regions.

[0015] In one possible implementation, based on the signal jump point index and the differences in methylation signals across different small regions, the data is dynamically segmented to obtain dynamic segmentation results, thereby acquiring structurally clear candidate difference region boundaries, including: Preliminary region segmentation: using a methylation signal value matrix, and employing the formula... Find the index of the signal value jump point; where, Indicates the first The starting index of each region represents the first CpG site or genomic location in that region; Indicates the first The end index of each region represents the last CpG site or genomic location in that region; It is the total number of regions identified in a specific chromosome; It is a region matrix, where each row represents a region and each column represents a chromosome; when the average differential methylation value of two regions is greater than a threshold, they are identified as significantly different sites; boundary correction: significantly different sites with a spacing of <5 CpG are merged to obtain differentially methylated regions; Region resegmentation: In iterative analysis, the initial region is resegmented based on the calculation results of lazy loading to generate a new index structure, using the formula... Re-plan the jump points dynamically; among which, Indicates the first The starting index of a new region represents the first CpG site or genomic location in that region; Indicates the first The end index of a new region represents the last CpG site or genomic location in that region; It is the total number of new regions identified in a specific chromosome; It is a new region matrix, where each row represents a region and each column represents a chromosome; Based on the new index structure, a clearly defined boundary for candidate difference regions is obtained.

[0016] Secondly, a device for identifying differentially methylated regions from whole-genome single-cell methylation data is provided, comprising: The preprocessing module is used to preprocess single-cell whole-genome methylation sequencing data to obtain a methylation signal value matrix; the preprocessing includes reordering, quality control, bias correction, methylation signal extraction, and format unification; The block segmentation module is used to compress the methylation signal value matrix and divide it into data blocks, which are then stored in blocks using the HDF5 format. The module is used to build region indexes for HDF5 format data structures and establish a dynamic update mechanism; extract the data block index information and load it into memory; the index only loads the data block currently being calculated; The optimization and acceleration module is used to optimize the data analysis process using a lazy evaluation strategy, improve quality control efficiency, and perform GPU-accelerated computing. The identification module is used to perform whole-genome scanning to obtain single-cell methylation signals by combining dynamic optimization strategies; a probability model and signal jump point index are constructed based on Gaussian density estimation to analyze single-cell methylation signals and identify differentially methylated regions.

[0017] Thirdly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the differential methylation region identification method for whole-genome single-cell methylation data as described in the first aspect.

[0018] Fourthly, a non-transitory computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the differential methylation region identification method for whole-genome single-cell methylation data as described in the first aspect.

[0019] Fifthly, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the differential methylation region identification method for whole-genome single-cell methylation data as described in the first aspect.

[0020] The beneficial effects of this invention are as follows: The method described in this invention has significantly improved processing speed through algorithm optimization, especially for applications of data partitioning and parallel computing, and can identify differentially methylated regions more efficiently and accurately.

[0021] The method described in this invention employs an index management and lazy computation strategy, which significantly reduces memory and storage space usage, making single-cell methylation data analysis faster, more efficient, and lighter, especially suitable for large-scale data analysis and cloud computing environments.

[0022] The method described in this invention has significant advantages in accelerating the discovery of differentially methylated regions and improving research efficiency, and is particularly suitable for large-scale data analysis and precision medicine. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating the method for identifying differentially methylated regions in whole-genome single-cell methylation data provided in this embodiment of the invention.

[0025] Figure 2 This is a baseline data comparison chart between the present invention embodiment and the traditional method in terms of memory optimization, computing speed, and storage space usage.

[0026] Figure 3 This is a diagram showing the data quality control and result accuracy in an embodiment of the present invention.

[0027] Figure 4 This is a summary diagram of the results of viewing differential methylation signals during neural cell development in the IGV genome browser in this embodiment of the invention (each time point contains 3 biological replicates). Wherein: Figure 4 (a) shows the differential methylation signal regions of the CFAP77 gene at different time points (including day 6 and day 14); Figure 4 (b) shows the differential methylation signal regions of the CXXC5 gene at different time points (including day 6 and day 14); Figure 4 (c) shows the differential methylation signal regions of the ERGIC1 gene at different time points (including day 6 and day 14); Figure 4 (d) shows the differential methylation signal regions of the LAMA5 gene at different time points (including day 6 and day 14).

[0028] Figure 5 This is a graph showing the results of the differential methylation region identification speed analysis between the new invention and the traditional method in an embodiment of the present invention.

[0029] Figure 6 This is a schematic diagram of the device provided in an embodiment of the present invention.

[0030] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0031] To better illustrate the present invention, the technical solution will now be further described in conjunction with specific embodiments and the accompanying drawings.

[0032] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Current methylation data analysis methods suffer from the following problems: quality and data consistency issues, large output storage space requirements, computational set memory consumption issues, computational speed issues, and low efficiency issues.

[0035] In view of this, the present invention provides a method for identifying differentially methylated regions in single cells across the entire genome.

[0036] See Figure 1 This is a genome-wide single-cell differential methylation region identification method, including: S100. Preprocess the single-cell whole-genome methylation sequencing data to obtain a methylation signal value matrix; the preprocessing includes reordering, quality control, bias correction, methylation signal extraction and format unification.

[0037] It should be noted that single-cell methylation sequencing data are often affected by technical noise, sequencing errors, sample contamination and other factors, so rigorous data preprocessing is required to ensure the accuracy and reliability of subsequent analysis.

[0038] In one possible implementation, step S100 includes: S110. Obtain single-cell whole-genome methylation sequencing data from a high-throughput sequencing platform as the raw input for the analysis workflow. This raw data is typically stored in a post-alignment format (such as a BAM file) and contains methylation information for each single cell at various genomic locations. To improve the efficiency of subsequent signal processing, the data is reordered according to chromosomal location. Preferably, standard tools such as samtools sort can be used to perform the sorting operation of the BAM file. Alternatively, logical sorting can be performed based on gene regions or functional elements (such as promoters and CpG islands) to facilitate subsequent regional modeling and identification of differentially methylated regions.

[0039] It's important to note that in whole-genome single-cell methylation sequencing, data is typically generated in its original order, but this raw data may lack optimal structure and organization. The main benefits of data reordering include: First, improving computational efficiency, specifically through memory access optimization and parallel processing. Optimized memory access involves storing data from the same chromosome contiguously, reducing disk I / O and avoiding cache hits; parallel optimization allows for segmented computation of chromosomes for easier subsequent analysis. Second, ensuring biological consistency, including region association and annotation integration. Because methylation at adjacent CpG sites often exhibits synergy (methylation islands, promoter regions), sorting facilitates the detection of local methylation patterns; alignment with gene regulatory element coordinates facilitates the association of methylation signals with functional regions. Third, avoiding redundant operations. Unsorted data may lead to repeated traversal (a chromosome segment being scanned multiple times), while sorted data requires only one linear processing step.

[0040] S120. Evaluate the sequencing depth and coverage of the reordered sequencing data, calculate the consistency and error rate of the data, and remove data with poor consistency and high error rate to obtain the first sequencing data.

[0041] Furthermore, methods for calculating data consistency and error rate include: technical replication of the same cell (such as library-sequencing data), and calculating the Pearson correlation coefficient (R) of whole-genome methylation levels.

[0042] It should be noted that when the correlation R between technical replicates is less than 0.8, it is considered to be of poor consistency, which may be caused by experimental errors, data sorting problems, or other technical biases. Screening criteria: Retain sample data with R ≥ 0.8 to ensure the stability and representativeness of methylation signals.

[0043] S130. Low-quality data are removed from the first sequencing data to obtain the second sequencing data.

[0044] Furthermore, the removal methods include: using tools such as Trim Galore to trim the raw sequencing data, removing library adapter sequences and low-quality bases at the ends, in order to improve alignment efficiency and data reliability.

[0045] S140. The second sequencing data is biased to obtain the third sequencing data.

[0046] Furthermore, the deviation correction method includes: removing duplicate PCR, labeling and removing duplicate reads to correct uneven coverage caused by amplification deviation.

[0047] Specifically, the MarkDuplicates module in the Picard toolkit was used to remove PCR duplicates from the second sequencing data.

[0048] It should be noted that bias correction is necessary to eliminate systematic biases generated during sequencing and ensure the accuracy of methylation level assessment. This step helps improve the representativeness and comparability of sequencing data, providing a more reliable data foundation for subsequent analysis.

[0049] S150. Methylation signals are extracted from the third sequencing data to obtain the fourth sequencing data.

[0050] Furthermore, the method for extracting the methylation signal includes: S151. Use the MethylDackel tool to extract methylation level data for CpG sites; S152. Sort the extracted methylation level data using the sort command sort-k1,1-k2,2n to obtain sorted data; S153. Remove the data from the ENCODE Blacklist Regions and centromeric regions of the chromosome from the sequencing data to obtain the fourth sequencing data.

[0051] Understandably, data from blacklisted regions and centromeric regions may show abnormally high or low methylation signals due to non-biological reasons, leading to misleading analytical results. Therefore, these signals should be removed after extraction to avoid false positives and ensure the reliability of the extracted methylation signal matrix.

[0052] S160. The format of the fourth sequencing data is standardized to obtain the methylation signal value matrix.

[0053] In one possible implementation, S160 includes the following sub-steps: S161. Construct a fixed sliding window region as the reference coordinate system for analysis.

[0054] Specifically, using bedtools makewindows, sliding windows (tiles) are constructed based on the whole genome, with a step size of 1000 bp, to generate BED files in standard window format.

[0055] S162. Calculate window-level weighted methylation values: The bedtools map tool is used to map the fourth batch of sequencing data (methylation site data in bedGraph format) to a fixed sliding window region, and the weighted average methylation signal value of each window is calculated.

[0056] Furthermore, weighting methods need to consider the coverage of each site (such as the number of methylated reads) to improve the stability and representativeness of the signal.

[0057] S163. Form a methylation signal matrix in a unified format: The set of methylation level values ​​generated for each sample and corresponding one-to-one with the window are summarized to form a two-dimensional methylation signal value matrix of [cell × interval], which provides standard input for subsequent cluster analysis, dimensionality reduction or differential methylation region identification.

[0058] It should be noted that single-cell methylation sequencing data is inherently sparse. If only the raw site signals are used, they are easily affected by low coverage, resulting in data instability. By integrating the data within a fixed window using a sliding window method and using weighted methylation levels as a characterization, the problems of signal sparsity and noise interference can be effectively alleviated, improving the overall accuracy and comparability of the data.

[0059] For example, single-cell whole-genome methylation sequencing (scWGBS / scTAPS) data can be preprocessed according to the following steps to generate a high-quality methylation signal value matrix: First, raw FASTQ format data is obtained from a high-throughput sequencing platform. This data is then aligned to the human reference genome using alignment tools (such as Bismark / bwamem) to generate BAM format files. Subsequently, the BAM files are sorted using samtools sort (corresponding to S110), organizing the data by chromosome number and start position to improve memory access efficiency, support parallel computing, and facilitate the detection of local methylation patterns.

[0060] Next, quality control is performed (corresponding to S120 and S130). On the one hand, the consistency of sequencing data is assessed by technically replicating samples, and the Pearson correlation coefficient (R) of methylation levels is calculated. Samples with correlation below the threshold (e.g., R < 0.8) are removed. On the other hand, the Trim Galore tool is used to remove adapter contamination and prune low-quality bases from the raw FASTQ data, filtering out potentially low-quality data.

[0061] Based on this, bias correction (corresponding to S140) is performed. Picard's MarkDuplicates module is used to mark and remove PCR duplicate reads, correcting the bias caused by amplification and improving the representativeness of sequencing coverage.

[0062] Next, the methylation signal values ​​(corresponding to S150) were extracted from the aligned BAM file using the MethylDackel tool to obtain the methylation level of each CpG site, and the output was a standard bedGraph format file. To ensure consistent site information sorting and facilitate visualization and subsequent processing, the extraction results were sorted again (e.g., sort-k1, 1-k2, 2n). In addition, signal anomalous regions, including ENCODE Blacklist regions and centromere regions, were removed to avoid misleading signals interfering with the accuracy of the analysis.

[0063] Finally, the extracted and filtered methylation data were formatted uniformly (corresponding to S160). The bedtools tool was used to generate fixed-window (e.g., 1000 bp) whole-genome tiles, and the bedGraph methylation signals were mapped to these windows. The weighted methylation value for each window was calculated. Each sample formed a methylation signal vector organized by window, which was ultimately integrated into a unified [cell × region] methylation signal value matrix, serving as standard input for subsequent clustering, dimensionality reduction analysis, and differential methylation region identification.

[0064] It is understood that the method in step S100 of this embodiment can significantly improve the accuracy and reliability of the data, reduce the interference of technical noise on the analysis results, and ensure the high quality of the whole genome single-cell methylation data used, thereby providing reliable support for subsequent differential methylation region identification and other analyses.

[0065] S200. The methylation signal value matrix is ​​compressed and divided into data blocks, and then stored in blocks using the HDF5 format.

[0066] In one possible implementation, S200 includes: S210. Compress the methylation signal value matrix using a compression algorithm.

[0067] Furthermore, the compression algorithm includes the DEFLATE algorithm.

[0068] S220. Divide the compressed data into data blocks.

[0069] For example, partitioning refers to dividing the entire large dataset (a two-dimensional matrix of methylation signal values) into smaller blocks of a fixed size of 1,000,000 cells during the data storage process. Each block can be stored, read, and compressed independently.

[0070] S230. Divide and store the data blocks into blocks using HDF5 format.

[0071] Understandably, compression algorithms are used to compress the whole-genome methylation signal value matrix to reduce data storage space; HDF5 structure is used to chunk the single-cell methylation data, dividing the methylation signal matrix into fixed-size chunks and storing them separately in HDF5 files to support efficient data reading and management; distributed storage of data is achieved through chunk storage, supporting local access, improving I / O access efficiency, and supporting chunk-level compression to further optimize storage usage.

[0072] S300. Build a region index for the HDF5 format data structure and establish a dynamic update mechanism; extract the data block index information and load it into memory; the index only loads the data block currently being calculated.

[0073] In one possible implementation, S300 includes: S310. Use a hash index to map each data block to a unique hash value and build a hash table.

[0074] Furthermore, S310 includes: S310a, Data block identifier extraction, using metadata (row and column start and end positions) as unique data block identifier information; S310b, Hash value calculation, which assembles the data block identification information into a unique key. S310c: The methylation signal value is hashed using a hash function (Jenkins Hash) to generate a corresponding hash value, which serves as the primary key value of the data block; wherein, the primary key value is entirely dependent on the input of the key value, the key value is the original identifier, and the primary key value is the hash encoding result of the key value.

[0075] S310d, index structure construction, a hash table is built using the primary key value and the physical address information of the data block; S310e, based on a hash table, loads the index information of the primary key value that the current computing task needs to access into memory.

[0076] S320, Construct a region index.

[0077] Furthermore, the methods for constructing region indexes include: S320a. Determine a non-overlapping window of a certain size as the basic unit; S320a. Construct region identifier keys using the "chromosome:start-terminate" format; S320a. Select a data structure based on the query requirements; the data structure is suitable for dynamic regions and overlapping queries. S320a. Associate each region key with a set of data block hash primary keys to establish a mapping relationship between key value and primary key value in order to build a region index.

[0078] For example, the regional index uses a fixed 10kb non-overlapping window as the basic unit (e.g., chr1:100000-110000), and constructs regional identifier keys using the "chromosome:start-end" format (e.g., chr1:100000-200000). An efficient data structure is selected based on query requirements: suitable for dynamic regions and overlapping queries. Each regional key is associated with a set of data block hash primary keys generated by S310, establishing a key-value mapping relationship to achieve fast location and on-demand loading.

[0079] By constructing a regional index structure, the spatial distribution of methylation signal value matrix is ​​organized and managed, thereby enabling on-demand loading of regional-level data blocks, improving query efficiency and system response performance.

[0080] S330. Establish a dynamic update mechanism.

[0081] Furthermore, the S330 includes: S330a, Data Change Awareness and Response Mechanism: Real-time capture of data update events, automatically triggering the index maintenance process. Real-time capture of addition and modification events to data blocks in HDF5 files, calculation of new hash values ​​for modified data, comparison with the hashes in the original index, confirming whether the data has changed.

[0082] S330b, Dynamic Index Maintenance Mechanism: For newly added data records, the hash primary key value is calculated in real time and synchronously updated to the index table to ensure data and index consistency; for modified data blocks, the hash table is updated to keep the index synchronized; for deleted data, the relevant entries in the hash table are removed to avoid invalid index access.

[0083] S330c, Dynamic Maintenance of Index Data Structure: The index is managed using a data structure that supports dynamic insertion and deletion operations, and index updates are synchronized to the memory cache in real time to improve access efficiency.

[0084] It should be noted that the S330 proposes a dynamic update mechanism for real-time maintenance of data block indexes and supports updating newly added, modified, or deleted data blocks, avoiding a complete index reconstruction and improving the system's flexibility and response speed.

[0085] Understandably, the memory optimization based on the indexing mechanism involves building an index for the HDF5 data structure, extracting data block index information, and loading the index into memory to speed up data access. Based on the indexing mechanism, methylation signal data can be read on demand, loading only the data blocks required for the current calculation, reducing computer memory usage, and improving computational efficiency. This enables the computer to dynamically allocate memory, achieving more efficient single-cell methylation data interaction and computation.

[0086] The S400 employs a lazy evaluation strategy to optimize the data analysis process, improve quality control efficiency, and perform GPU-accelerated computation.

[0087] In one possible implementation, S400 includes: S410. The computation process is decomposed into multiple subtasks using the DelayedArray framework; each subtask corresponds to the operation of a data block.

[0088] The DelayedArray framework ensures that computation is triggered only when data blocks need to be accessed or processed, allowing for on-demand loading and calculation. During task decomposition, the dependencies and computation logic of each subtask are recorded to ensure that the computation results of each data block can be reused, preventing the same data block from being repeatedly computed in different tasks.

[0089] S420, Optimize quality control efficiency.

[0090] Furthermore, the method of S420 includes: S420a. Utilize region indexing for accurate localization and load only gene regions and cell data blocks relevant to the current differentially methylated region identification task; S420b utilizes hash indexes that uniquely identify data blocks to perform data quality control, thereby avoiding duplicate calculations of the same data block and significantly improving the utilization of computing resources and overall execution efficiency.

[0091] It should be noted that the two-dimensional methylation signal matrix is ​​stored using HDF5 as the backend, and combined with the DelayArray framework, fine-grained decomposition and dynamic scheduling of computational tasks are achieved, thereby optimizing quality control efficiency. Through S420, this embodiment of the application ensures efficient execution of the quality control process in large-scale single-cell methylation data analysis, while providing reliable data support for subsequent differential methylation region identification tasks.

[0092] The S430 combines GPU parallel computing and multi-threading optimization strategies. It schedules multiple subtasks to the GPU for batch processing through a latency computing mechanism and dynamically manages the allocation of computing resources based on the dependencies between data blocks to perform GPU-accelerated computing, thereby improving the efficiency and throughput of large-scale single-cell methylation data processing.

[0093] Understandably, S400 uses a dedicated backend (such as HDF5) for storage, loading only a portion of the data for each operation, avoiding the loading of entire blocks of data. The core idea of ​​DelayedArray is to break down the computation process into multiple subtasks, which load and process the relevant data as needed. For example, in single-cell methylation analysis, if only the methylation information of certain genomic regions is relevant, the system will only load the data for those relevant regions, rather than loading the entire genome into memory. This avoids processing large amounts of irrelevant data and significantly reduces memory and computational resource consumption.

[0094] Specifically, in this embodiment, the methylation signal matrix is ​​decomposed into multiple subtasks, which are then scheduled to the GPU for batch processing using a delayed computation mechanism. The allocation of computing resources is dynamically managed based on the dependencies between data blocks, ensuring parallel processing while minimizing memory usage. Furthermore, in tasks such as data preprocessing and file format conversion (e.g., BigWig generation), the system employs a multi-threaded optimization strategy, distributing independent subtasks (e.g., different chromosomes or regions) to multiple threads for parallel processing to improve the overall data loading and conversion throughput.

[0095] The multi-threaded scheduling logic in this embodiment automatically divides task boundaries based on data block indexes and combines a shared memory buffer pool to dynamically balance thread load, avoiding thread contention and resource blocking, thereby ensuring both execution efficiency and system stability. This design is particularly suitable for scenarios with frequent dynamic index updates and high data block reuse rates, effectively shortening computation paths and improving resource reuse rates.

[0096] S500, combined with a dynamic optimization strategy, performs whole-genome scanning to obtain single-cell methylation signals; based on Gaussian density estimation, a probability model and signal jump point index are constructed to analyze single-cell methylation signals to identify differentially methylated regions.

[0097] In one possible implementation, S500 includes: S510, combined with a dynamic optimization strategy, performs a whole-genome scan to extract methylation signals at the single-cell level.

[0098] Furthermore, the dynamic optimization strategy includes: (1) Dynamic resegmentation triggered by jump points: When a significant jump point is detected in the methylation signal in a specific region, the resegmentation of the local region is immediately triggered.

[0099] (2) Resource scheduling optimization: The priority of re-segmentation operation is dynamically adjusted, non-critical path tasks are executed in the background thread pool, and hot area segmentation results are cached in memory.

[0100] S520. Divide the entire genome into multiple small regions, and perform Gaussian density estimation based on local methylation signals within each small region to construct a region-level probability model.

[0101] Furthermore, the probability model includes a Gaussian distribution model and a Beta distribution model. In this embodiment, the probability model is a Gaussian distribution model.

[0102] Specifically, the Gaussian distribution is suitable for cases where methylation signal values ​​exhibit a symmetrical distribution, effectively describing the central tendency and dispersion of the data. The Beta distribution, on the other hand, is suitable for cases where methylation signal values ​​lie within the [0,1] interval and exhibit a skewed distribution, particularly well-suited for describing the distribution of methylation rates, as the methylation rate is essentially a probability between 0 and 1. By selecting an appropriate mathematical model for fitting, the statistical characteristics of methylation signals can be described more accurately, providing a theoretical basis for subsequent data analysis and differential methylation region identification. Furthermore, by combining dynamic optimization strategies and multi-level indexing mechanisms, the aforementioned probabilistic models can more efficiently handle complex patterns in large-scale methylation datasets, ensuring the reliability and accuracy of the analysis results.

[0103] S530. Based on the signal jump point index, the data is dynamically segmented according to the differences in methylation signals in different small regions to obtain dynamic segmentation results, thereby obtaining candidate difference region boundaries with clear structures.

[0104] Furthermore, S530 includes: S531. Preliminary Region Division: Using the methylation signal value matrix obtained in S100, and applying the formula... Find the index of the signal value jump point; where, Indicates the first The starting index of each region represents the first CpG site or genomic location in that region; Indicates the first The end index of each region represents the last CpG site or genomic location in that region; It is the total number of regions identified in a specific chromosome; This is a region matrix, where each row represents a region and each column represents a chromosome. The genome is divided into contiguous initial regions using a jump point index (each region is defined by adjacent jump points). A weighted average methylation value is calculated for each initial region. Sites with a mean differential methylation value greater than a threshold (default 0.1) are identified as significantly different sites. Boundary correction: significantly different sites with a distance < 5 CpG are merged to obtain differentially methylated regions. In practical applications, this method allows researchers to effectively segment large-scale single-cell methylation data into smaller, more manageable regions, significantly reducing memory usage and improving data processing efficiency. Furthermore, this region segmentation can be combined with subsequent multi-level indexing and dynamic optimization strategies to further accelerate access to and analysis of specific region data, ensuring flexibility and efficiency when handling complex, large-scale datasets. This method not only helps improve the accuracy of methylation status identification but also provides a solid foundation for in-depth analysis of subsequent differentially methylated regions. In summary, finding the signal value jump point index using a formula is a crucial step in achieving efficient single-cell methylation data analysis, significantly improving the accuracy and efficiency of research.

[0105] S532, Region Resegmentation: In iterative analysis, the initial region is resegmented based on the calculation results of lazy loading to generate a new index structure, using the formula... Re-plan the jump points dynamically; among which, Indicates the first The starting index of a new region represents the first CpG site or genomic location in that region; Indicates the first The end index of a new region represents the last CpG site or genomic location in that region; It is the total number of new regions identified in a specific chromosome; It is a new region matrix, where each row represents a region and each column represents a chromosome.

[0106] S533. Based on the new index structure, obtain the boundaries of candidate difference regions with clear structure.

[0107] Understandably, the new index structure ( The matrix is ​​the final result of the dynamic partitioning, where each (a) m ,b m Determine the boundaries of the difference regions.

[0108] It should be noted that the candidate differential region boundary refers to the genomic coordinate interval (a) ultimately determined through the dynamic segmentation process. m , b m Its core definition and function include: a biologically defined boundary region containing a continuous group of CpG sites with significant methylation differences, with a clear epigenetic state transition on both sides of the boundary. Computational features: providing an accurate genomic coordinate framework for downstream differential analysis. Ensuring differential signal detection is performed within biologically homogeneous units.

[0109] S540: Based on the probability model and candidate differential methylation region boundaries, single-cell methylation signals are analyzed using a deep learning model to ultimately identify differential methylation regions.

[0110] In one possible implementation, S540 includes: S541. Data Preprocessing and Feature Engineering: Based on the index information of candidate differential region boundaries, extract gene coordinates and sequence information of the target region using the bedtools tool, and integrate epigenetic annotation and sequence feature information.

[0111] S542. Use the Transformer deep learning model architecture to process sequence feature information, locate key CpG sites through multi-head attention mechanism; output differential methylation region index and corresponding region attention score; identify differential methylation regions through attention score.

[0112] It's important to note that differentially methylated regions (DMRs) are potential sources of biomarkers: by comparing diseased and healthy samples, screened DMRs may contain clinically valuable methylation biomarkers. Biomarkers are typically validated DMRs; not all DMRs qualify as biomarkers. Only those validated in independent cohorts and significantly associated with disease mechanisms or phenotypes are identified as biomarkers. Biomarkers are biological indicators that can objectively measure and reflect normal physiological processes, pathological states, or responses to treatment interventions. In methylation studies, biomarkers are those where the methylation pattern (e.g., high / low methylation) of a specific genomic region (e.g., CpG islands) is significantly associated with disease. Transformers predict the probability or confidence level (e.g., 0.9) that a differentially methylated region belongs to a disease (e.g., "cancer-related"). Multi-head attention mechanisms are used to assign weights to CpG sites within the region, reflecting their importance in classification decisions. Higher values ​​indicate greater confidence.

[0113] S540: By using deep learning to analyze methylation patterns and combining dynamic segmentation boundaries with a probability model to achieve closed-loop optimization, biologically validated differentially methylated regions are finally identified.

[0114] Furthermore, the deep learning model is a BERT model based on the Transformer architecture. This model can perform contextual modeling on the identified regions, thereby more accurately capturing the nonlinear relationships and potential regulatory patterns between regions and improving the accuracy of methylation state prediction. To intuitively verify the identified methylation regions, after modeling, the methylation signal data can be converted to BigWig format and visualized using a genome browser (such as IGV). This method can be used to visualize raw methylation signals, differentially methylated region identification, etc., helping researchers quickly determine the accuracy and changing trends of methylation state.

[0115] It's worth noting that the BERT model, when processing single-cell methylation data, can identify interactions between different genomic regions and trends in methylation patterns. This method is particularly suitable for processing large-scale, high-dimensional single-cell methylation datasets, significantly improving data processing speed without sacrificing prediction accuracy. Furthermore, BERT can be combined with other mathematical models (such as Gaussian and Beta distributions) to further optimize the modeling and prediction of methylation signals. In this way, researchers can more accurately identify differentially methylated regions (DMRs), providing strong support for the discovery of potential differentially methylated regions and disease mechanisms. This approach not only enhances the flexibility of single-cell methylation data analysis but also improves the interpretability and practicality of research results, contributing to technological advancements and development in this field.

[0116] Figure 2 The test results of the embodiments of this application compared with traditional methods in terms of data object construction time, hard disk storage space usage, and memory management are presented. The test results demonstrate the efficiency and optimization potential of the method described in the embodiments of this application in single-cell methylation data analysis. Among them, the traditional methods specifically include dataframe format, single-cell data format (anndata), sequencing data dispersion shrinking format (BSseq / DSS), methylation analysis toolkit format (methylKit), and chromatin single-cell analysis toolkit format (signac)).

[0117] 1. Comparison of data object construction time. Figure 2The first line shows a comparison of the data object construction time between this invention and traditional methods at different resolutions. Experimental results show that this invention completes data object construction in just 5 seconds at single-base resolution, while traditional methods take a maximum of 13 seconds and a maximum of 834 seconds. As the resolution increases, the construction time of this invention at 1kb, 3kb, 5kb, and 10kb resolutions does not exceed 0.5 seconds, demonstrating a significant time advantage. This performance improvement is attributed to an efficient indexing mechanism, block-level processing, and parallel computing optimization, resulting in a substantial increase in data processing speed.

[0118] 2. Comparison of Hard Disk Storage Space. The second line shows the hard disk storage space usage of the present invention and traditional methods at different resolutions. The present invention requires only 296MB to store the whole genome methylation data of 6 cells at single-base resolution, while the traditional method requires as much as 1500MB, a difference of over 80%. At higher resolutions, the storage requirements of the present invention are also significantly lower than those of the traditional method, even reduced to 5.3MB at 10kb resolution. Through indexed compression storage, HDF5 format, and interval-level data storage technology, the present invention achieves effective storage optimization.

[0119] 3. Memory Usage Comparison. The third line shows the memory usage of this invention and traditional methods when loading whole-genome methylation data. At single-base resolution, the memory usage of this invention is only 5KB, while the memory usage of traditional methods ranges from 3140MB to 6030MB, a difference of thousands of times. Even at higher resolutions, the memory requirements of this invention are still significantly lower than those of traditional methods. By employing lazy computation and memory mapping techniques, this invention can achieve extremely low memory usage in large-scale data analysis, improving overall data processing efficiency.

[0120] Figure 3 This invention demonstrates how innovative quality control strategies and visualization tools can significantly improve the accuracy and efficiency of single-cell whole-genome methylation data analysis, providing important support for epigenetic research and disease mechanism exploration.

[0121] 1. Scatter plot analysis Figure 3 (a): Genome coverage and methylation signal intensity. Figure 3 (a) A scatter plot shows the distribution of methylation data from 1170 single-cell genome-wide samples between genome coverage and average methylation signal. The horizontal axis represents cell coverage, and the vertical axis represents cell methylation value. Different neuronal types are distinguished by color to facilitate observation of their characteristic differences in methylation data. The results show that cells with high coverage are mainly concentrated in specific neuronal subpopulations, suggesting that these cells may have higher data integrity. Conversely, cells with low coverage may be affected by sequencing depth or technical noise, highlighting the importance of methylation modification in neurobiology.

[0122] 2. Methylation patterns in gene functional regions Figure 3 (b) Figure 3 (b) The relationship between methylation levels and coverage in different gene functional regions (such as promoters and exons) was analyzed. The horizontal axis represents the number of cells with cellular characteristic values, and the vertical axis represents the methylation value. The results showed that the methylation level in promoter regions was low, consistent with previous findings that high methylation usually inhibits gene expression. The methylation levels in exons and introns were relatively uniform, while the 3'UTR and 5'UTR regions showed different methylation characteristics. This analysis helps to reveal the role of epigenetic modifications in gene regulation.

[0123] 3. Distribution of differentially methylated regions Figure 3 (c) Figure 3 (c) Describes the differential methylation values ​​of the differentially methylated regions identified in this invention within the Adarb2 gene region. The upper horizontal axis represents the location of differential methylation in the genomic region. Dots represent differential methylation sites, and the vertical axis represents methylation signal values.

[0124] 4. UMAP dimensionality reduction and cell population analysis Figure 3 (d). Figure 3 (d) The distribution of 1170 single-cell methylation data in low-dimensional space was displayed using the UMAP dimensionality reduction method, showing that different neuron types form significant clusters in the UMAP space. The horizontal axis represents the UMAP horizontal axis coordinate, and the vertical axis represents the UMAP vertical axis coordinate. This result indicates that there is significant differentiation in the epigenetic characteristics of cell types, which helps researchers in cell type identification and the discovery of disease-related epigenetic patterns. The UMAP method provides a powerful tool for in-depth research on cell subpopulations and their functions.

[0125] Figure 4 Visual analysis using the IGV genome browser revealed changes in the methylation of key genes and their biological significance in cell differentiation and neural development, providing important evidence for understanding epigenetic regulation and disease mechanisms.

[0126] 1. Application of the IGV Genome Viewer and Key Gene Methylation Changes. This paper illustrates the use of the IGV (Integrative Genomics Viewer) genome browser to view methylation results, allowing for intuitive observation of methylation changes in multiple key genes at different experimental time points (D6 and D14). Genes such as GFAP, CXXC5, LAMA5, and ERGIC1 showed significant methylation changes at these two time points, demonstrating their important regulatory roles in physiological or pathological processes such as cell differentiation, neural development, and tumorigenesis.

[0127] 2. Biological Significance of Methylation Changes. Analysis of the methylation patterns of these genes revealed that the hypermethylation of GFAP in group D6 may inhibit its expression and participate in the protection of the nervous system; demethylation of CXXC5 may activate the Wnt / β-catenin signaling pathway, promoting neuronal differentiation; restoration of LAMA5 demethylation contributes to extracellular matrix remodeling; and the hypermethylation of ERGIC1 in group D14 may be related to its intracellular functional inhibition. These changes not only reflect the regulation of cellular function but may also provide new insights and potential identification of differentially methylated regions for the study of nervous system diseases.

[0128] See Figure 5 This invention significantly improves data processing speed compared to traditional methods when identifying differentially methylated regions, and shows a clear efficiency advantage, especially in large-scale sample analysis.

[0129] 1. Speed ​​comparison analysis. Figure 5 This paper presents the speed and memory usage results of the present invention and conventional methods in finding differentially methylated regions, where DSS represents the sequence data dispersion shrinking format, methyTracer represents the method described in this application, and data_frame represents the data frame format. By comparing the running time of the two methods on the same dataset, the results highlight the significant advantage of the present invention in data processing speed. The present invention, employing efficient algorithms and an optimized computational framework, can significantly reduce computation time in the screening and analysis of massive amounts of data, especially when processing large-scale samples, where the efficiency difference is even more pronounced.

[0130] 2. Traditional methods typically rely on relatively simple computational processes, resulting in slow processing speeds when handling complex biological data, especially in multidimensional genomic data analysis where computational costs are high. This leads to traditional methods requiring significantly more time than the method of this invention when searching for differentially methylated regions. Through data partitioning and parallel computing optimization, the method described in this invention can more efficiently screen for potential differentially methylated regions, thereby accelerating the discovery of methylation biomarkers, significantly improving research efficiency, and is particularly suitable for the field of precision medicine.

[0131] The differential methylation region identification device for whole-genome single-cell methylation data provided by the present invention is described below. The differential methylation region identification device for whole-genome single-cell methylation data described below can be referred to in correspondence with the differential methylation region identification method for whole-genome single-cell methylation data described above.

[0132] Figure 6 This is a schematic diagram of the differential methylation region identification device for whole-genome single-cell methylation data provided in an embodiment of the present invention, as shown below. Figure 6As shown, it includes: a preprocessing module 61, a segmentation module 62, a construction module 63, an optimization and acceleration module 64, and a recognition module 65, wherein: Preprocessing module 61 is used to preprocess single-cell whole-genome methylation sequencing data to obtain a methylation signal value matrix; the preprocessing includes reordering, quality control, bias correction, methylation signal extraction and format unification; Blocking module 62 is used to compress the methylation signal value matrix and divide it into data blocks, and store the blocks using HDF5 format; Module 63 is used to build a region index for the HDF5 format data structure and establish a dynamic update mechanism; extract the data block index information and load it into memory; the index only loads the data block currently being calculated; The optimization acceleration module 64 is used to optimize the data analysis process by adopting a lazy evaluation strategy, improve the efficiency of quality control, and perform GPU-accelerated computing. The identification module 65 is used to perform whole-genome scanning to obtain single-cell methylation signals by combining dynamic optimization strategies; and to analyze single-cell methylation signals to identify differentially methylated regions by constructing a probability model and signal jump point index based on Gaussian density estimation.

[0133] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include a processor 710, a communications interface 720, a memory 730, and a communication bus 740, wherein the processor 710, communications interface 720, and memory 730 communicate with each other via the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute a differential methylation region identification method for whole-genome single-cell methylation data.

[0134] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0135] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute the differential methylation region identification method for whole genome single-cell methylation data provided by the above methods.

[0136] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for identifying differentially methylated regions in whole-genome single-cell methylation data provided by the methods described above.

[0137] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0138] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying differentially methylated regions in whole-genome single-cell methylation data, characterized in that, include: Single-cell whole-genome methylation sequencing data are preprocessed to obtain a methylation signal value matrix; the preprocessing includes reordering, quality control, bias correction, methylation signal extraction, and format standardization. The methylation signal value matrix is ​​compressed and divided into data blocks, and then stored in blocks using the HDF5 format. A region index is built for the HDF5 format data structure, and a dynamic update mechanism is established; the data block index information is extracted and loaded into memory; the index only loads the data block currently being calculated; The data analysis process is optimized by adopting a lazy evaluation strategy, which improves the efficiency of quality control and enables GPU-accelerated computation. By combining dynamic optimization strategies, whole-genome scanning was performed to obtain single-cell methylation signals; Based on Gaussian density estimation, a probability model and signal jump point index are constructed to analyze single-cell methylation signals to identify differentially methylated regions.

2. The method for identifying differentially methylated regions according to claim 1, characterized in that, The preprocessing of single-cell whole-genome methylation sequencing data to obtain a methylation signal value matrix includes: Single-cell whole-genome methylation sequencing data were reordered according to chromosome position; The sequencing depth and coverage of the reordered sequencing data are evaluated, the consistency and error rate of the data are calculated, and data with poor consistency and high error rate are removed to obtain the first sequencing data. Low-quality data were removed from the first sequencing data to obtain the second sequencing data; The second sequencing data is bias-corrected to obtain the third sequencing data; Methylation signals were extracted from the third sequencing data to obtain the fourth sequencing data; The fourth sequencing data was formatted to obtain a methylation signal value matrix.

3. The method for identifying differentially methylated regions according to claim 1, characterized in that, The aforementioned data structure in HDF5 format is indexed regionally, and a dynamic update mechanism is established. Extracting the data block index information and loading it into memory includes: A hash table is constructed by mapping each data block to a unique hash value using a hash index; Constructing a regional index: Determine a non-overlapping window of a certain size as the basic unit; construct regional identifier keys using the "chromosome:start-end" format; select a data structure based on query requirements; the data structure is suitable for dynamic regional and overlapping queries; associate each regional key with a set of data block hash primary keys to establish a mapping relationship between key values ​​and primary key values, thereby constructing the regional index; Establish dynamic update mechanisms: data change perception and response mechanism; index dynamic maintenance mechanism; dynamic maintenance of index data structure.

4. The method for identifying differentially methylated regions according to claim 1, characterized in that, The optimization of the data analysis process using a lazy evaluation strategy, the improvement of quality control efficiency, and the GPU-accelerated computation include: The computation process is broken down into multiple subtasks using the DelayedArray framework; each subtask corresponds to the operation of a data block. Optimize quality control efficiency: Use regional indexes for accurate location and load only gene regions and cell data blocks relevant to the current differential methylation region identification task; use hash indexes that uniquely identify data blocks for data quality control.

5. The method for identifying differentially methylated regions according to claim 1, characterized in that, The aforementioned dynamic optimization strategy is used to perform a whole-genome scan to obtain single-cell methylation signals. Based on Gaussian density estimation, a probability model and signal jump point index are constructed to analyze single-cell methylation signals to identify differentially methylated regions, including: By combining dynamic optimization strategies, whole-genome scanning was performed to extract methylation signals at the single-cell level; The entire genome is divided into multiple small regions, and the methylation signal is estimated by Gaussian density within each small region to construct a region-level probability model. Based on the signal jump point index, the data is dynamically segmented according to the differences in methylation signals in different small regions to obtain dynamic segmentation results, thereby obtaining candidate difference region boundaries with clear structure. Based on the probability model and candidate differential methylation region boundaries, single-cell methylation signals are analyzed using a deep learning model to ultimately identify differentially methylated regions.

6. The method for identifying differentially methylated regions according to claim 1, characterized in that, Based on the signal jump point index, and according to the differences in methylation signals in different small regions, the data is dynamically segmented to obtain dynamic segmentation results, thereby obtaining clearly structured candidate difference region boundaries, including: Preliminary region segmentation: using a methylation signal value matrix, and employing the formula... Find the index of the signal value jump point; where, Indicates the first The starting index of each region represents the first CpG site or genomic location in that region; Indicates the first The end index of each region represents the last CpG site or genomic location in that region; It is the total number of regions identified in a specific chromosome; It is a region matrix, where each row represents a region and each column represents a chromosome; when the average differential methylation value of two regions is greater than a threshold, they are identified as significantly different sites; boundary correction: significantly different sites with a spacing of <5 CpG are merged to obtain differentially methylated regions; Region resegmentation: In iterative analysis, the initial region is resegmented based on the calculation results of lazy loading to generate a new index structure, using the formula... Re-plan the jump points dynamically; among which, Indicates the first The starting index of a new region represents the first CpG site or genomic location in that region; Indicates the first The end index of a new region represents the last CpG site or genomic location in that region; It is the total number of new regions identified in a specific chromosome; It is a new region matrix, where each row represents a region and each column represents a chromosome; Based on the new index structure, a clearly defined boundary for candidate difference regions is obtained.

7. A device for identifying differentially methylated regions in whole-genome single-cell methylation data, characterized in that, include: The preprocessing module is used to preprocess single-cell whole-genome methylation sequencing data to obtain a methylation signal value matrix; the preprocessing includes reordering, quality control, bias correction, methylation signal extraction, and format unification; The block segmentation module is used to compress the methylation signal value matrix and divide it into data blocks, which are then stored in blocks using the HDF5 format. The module is used to build region indexes for HDF5 format data structures and establish a dynamic update mechanism; extract the data block index information and load it into memory; the index only loads the data block currently being calculated; The optimization and acceleration module is used to optimize the data analysis process using a lazy evaluation strategy, improve quality control efficiency, and perform GPU-accelerated computing. The identification module is used to perform whole-genome scanning to obtain single-cell methylation signals in conjunction with a dynamic optimization strategy; Based on Gaussian density estimation, a probability model and signal jump point index are constructed to analyze single-cell methylation signals to identify differentially methylated regions.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the differential methylation region identification method for whole-genome single-cell methylation data as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the differential methylation region identification method for whole-genome single-cell methylation data as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the differential methylation region identification method for whole-genome single-cell methylation data as described in any one of claims 1 to 6.

Images