Sequencing data error correction method and device, electronic equipment and storage medium
Error correction of nanopore sequencing data was performed using deep learning methods. By utilizing tandem repeat regions encoded by base recognition and language models, along with their upstream and downstream data, and combining them with deep models, the error correction problem of tandem repeat regions in nanopore sequencing was solved, thereby improving the accuracy of sequencing data and assembly results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HANGZHOU HUADA XUFENG TECHNOLOGY CO LTD
- Filing Date
- 2024-10-25
- Publication Date
- 2026-05-05
AI Technical Summary
Nanopore sequencing technology has errors in single-base tandem repeat regions that are difficult to correct during the sequencing process. Existing methods are not effective in correcting errors in tandem repeat regions, which limits the accuracy of sequencing data. Furthermore, the correction efficiency of base identification algorithms is not high.
A deep learning approach is employed to acquire tandem repeat regions and their upstream and downstream sequencing data from long-read sequencing data. These regions are then encoded using base recognition algorithms and language models, and input into a pre-trained correction model for error correction. Finally, deep models such as convolutional neural networks are used for further correction.
It achieves efficient and universal sequencing data error correction, improves the accuracy of sequencing data and subsequent assembly performance, and reduces systematic errors in tandem repeat regions.
Smart Images
Figure CN121983117A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of gene sequencing technology, and in particular to a sequencing data error correction method, apparatus, electronic device and storage medium. Background Technology
[0002] Nanopore sequencing, as a third-generation sequencing technology, measures the electrical signal changes as DNA molecules pass through a nanopore, and combines this with base recognition algorithms (such as Hidden Markov Models and neural network models) to convert the electrical signals into DNA sequences. Sequencing based on changes in current within the nanopore offers advantages such as real-time processing, high throughput, and low cost. However, despite these advantages, third-generation nanopore sequencing still has some limitations in practical applications. For example, it introduces errors during sequencing, which are typically concentrated in homopolymer regions rather than being uniformly distributed throughout the sequence. This makes it difficult to correct these errors using traditional methods of increasing sequencing depth. Introducing homopolymer regions introduces a large number of systematic errors, resulting in poor error correction effectiveness based on short-read data and limiting the accuracy of sequencing data. Furthermore, related technologies (such as neural network coding methods at the base recognition level) are detached from upstream base recognition technologies during sequencing, leading to low correction efficiency. Summary of the Invention
[0003] This disclosure provides a sequencing data error correction method, apparatus, electronic device, and storage medium.
[0004] According to a first aspect of this disclosure, a sequencing data error correction method is provided, the method comprising: acquiring long-read sequencing data, the long-read sequencing data including tandem repeat regions; encoding the tandem repeat regions and upstream and downstream sequencing data of the tandem repeat regions using a first encoding method to obtain base continuity features; encoding the upstream and downstream sequencing data of the tandem repeat regions using a second encoding method to obtain upstream and downstream features; and inputting the base continuity features and upstream and downstream features into a pre-trained correction model to correct errors in the tandem repeat regions of the long-read sequencing data.
[0005] In some embodiments of this disclosure, the first encoding method includes a base identification algorithm; the second encoding method includes a language model, the method comprising: using a base identification algorithm to encode tandem repeat regions and upstream and downstream sequencing data of tandem repeat regions to obtain base continuity features; and using a language model to encode upstream and downstream sequencing data of tandem repeat regions to obtain upstream and downstream features.
[0006] In some embodiments of this disclosure, encoding upstream and downstream sequencing data of tandem repeat regions using a language model to obtain upstream and downstream features includes: motifizing long-read sequencing data with a first preset length to obtain a motif table corresponding to the long-read sequencing data; motifizing is to segment the long-read sequencing data based on a language model to obtain multiple motifs; and encoding each motif in the motif table to obtain upstream and downstream features.
[0007] In some embodiments of this disclosure, base continuity features and upstream and downstream features are input into a pre-trained calibration model to correct errors in tandem repeat regions in long-read sequencing data. Previously, the method included: obtaining long-read sequencing data of a species similar to the species corresponding to the long-read sequencing data, and training the calibration model using the long-read sequencing data of the similar species to obtain a pre-trained calibration model.
[0008] In some embodiments of this disclosure, base continuity features and upstream and downstream features are input into a pre-trained correction model to correct tandem repeat regions in long-read sequencing data for errors. The method then includes: determining the number of corrected sequencing data after error correction, and storing the number, the sequencing name corresponding to the corrected sequencing data, and the corrected sequencing data in an output file.
[0009] In some embodiments of this disclosure, the calibration model employs a three-layer structure, which includes a convolutional layer, an activation layer, and a pooling layer.
[0010] In some embodiments of this disclosure, base continuity features and upstream and downstream features are input into a pre-trained correction model to correct tandem repeat regions in long-read sequencing data. Then, the method includes: determining the corrected long-read sequencing data corresponding to the corrected sequencing data after error correction, and determining the error evaluation parameters of the corrected long-read sequencing data; comparing the reference evaluation parameters with the error evaluation parameters to obtain the abnormal distribution data corresponding to the corrected long-read sequencing data.
[0011] According to a second aspect of this disclosure, a sequencing data error correction apparatus is provided, the apparatus comprising:
[0012] The acquisition unit is used to acquire long-read sequencing data, which includes tandem repeat regions.
[0013] The first coding unit is used to encode the tandem repeat region and the sequencing data upstream and downstream of the tandem repeat region using the first coding method to obtain the base continuity feature.
[0014] The second coding unit is used to encode the upstream and downstream sequencing data of the tandem repeat region using the second coding method to obtain upstream and downstream features;
[0015] The correction unit is used to input base continuity features and upstream and downstream features into a pre-trained correction model to correct errors in tandem repeat regions in long-read sequencing data.
[0016] According to a third aspect of this disclosure, an electronic device is provided, comprising:
[0017] At least one processor; and
[0018] A memory communicatively connected to the at least one processor; wherein,
[0019] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect above.
[0020] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the first aspect above.
[0021] The sequencing data error correction method, apparatus, electronic device, and storage medium disclosed herein acquire long-read sequencing data, which includes tandem repeat regions. A first encoding method is used to encode the tandem repeat regions and their upstream and downstream sequencing data to obtain base continuity features. A second encoding method is used to encode the upstream and downstream sequencing data of the tandem repeat regions to obtain upstream and downstream features. The base continuity features and upstream and downstream features are input into a pre-trained correction model to correct errors in the tandem repeat regions of the long-read sequencing data. This achieves efficient and universal sequencing data error correction by combining the tandem repeat regions and their upstream and downstream sequencing data in a deep model, thereby improving sequencing data accuracy and subsequent assembly performance.
[0022] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description
[0023] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:
[0024] Figure 1 A schematic diagram illustrating a systematic error in nanopore sequencing provided in an embodiment of this disclosure;
[0025] Figure 2This disclosure provides a method for correcting sequencing sequence errors using the Quiver method in a related art.
[0026] Figure 3 This is a schematic flowchart of a sequencing data error correction method provided in an embodiment of this disclosure;
[0027] Figure 4a This is a schematic flowchart of another sequencing data error correction method provided in an embodiment of this disclosure;
[0028] Figure 4b This is a schematic flowchart of another sequencing data error correction method provided in an embodiment of this disclosure;
[0029] Figure 5 This is a schematic diagram illustrating a specific method for determining base continuity characteristics and upstream / downstream characteristics, provided as an embodiment of this disclosure.
[0030] Figure 6 This is a schematic diagram of the structure of a specific correction model provided in an embodiment of the present disclosure;
[0031] Figure 7 This is a schematic diagram illustrating how a specific correction model provided in this embodiment corrects upstream and downstream features and base continuity features.
[0032] Figure 8 This is a schematic diagram illustrating the performance of a sequencing data correction method provided in this embodiment on a human dataset (same species);
[0033] Figure 9 This is a schematic diagram of a correction result on Ecoli data provided in an embodiment of the present disclosure;
[0034] Figure 10 This is a schematic diagram illustrating the correction results of a pan-species model (without access to Ecoli data) provided in an embodiment of this disclosure on Ecoli data;
[0035] Figure 11 This is a schematic diagram of the structure of a sequencing data error correction device provided in an embodiment of the present disclosure;
[0036] Figure 12 A schematic block diagram of an example electronic device 1200 provided for embodiments of this disclosure. Detailed Implementation
[0037] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.
[0038] Third-generation sequencing technology uses nanopore sequencing, which sequences DNA molecules by passing them through tiny pores and measuring the change in electrical current as the DNA molecules pass through the pores. This method provides real-time sequencing data and can directly detect genetic material without PCR amplification or fragmentation. Compared to previous generations of methods, it offers advantages such as high throughput, high speed, and low cost.
[0039] Compared to traditional sequencing methods, nanopore technology can sequence multiple DNA molecules simultaneously, significantly improving sequencing efficiency through high throughput. Furthermore, nanopore technology offers fast sequencing speeds, enabling large-scale sequencing operations to be completed in a short time.
[0040] Secondly, nanopore technology is relatively inexpensive. Compared to other sequencing methods, the equipment and reagents used in nanopore technology are less expensive, making it accessible to more laboratories and research institutions. This opens up possibilities for broader genome research.
[0041] Furthermore, nanopore technology offers longer read lengths and lower GC content bias. Nanopore technology can read long DNA sequences, which is invaluable for understanding complex gene structures, including tandem repeats, structural variations, and alternative splicing events. At the same time, nanopore sequencing results also exhibit lower GC content bias, reducing the difficulty of data analysis and interpretation.
[0042] However, nanopore technology also faces some challenges. Single-molecule sequencing data has been widely used in numerous genomics studies. However, this technology suffers from drawbacks such as high sequencing error rates. Among these, nanopore technology presents difficulties when sequencing tandem repeat sequences. Due to the inherent errors in reading and parsing specific DNA sequences, the accuracy of processing tandem repeat sequences on the genome is limited.
[0043] In nanopore sequencing, base recognition refers to the process of converting detected electrical signals into DNA sequences. When DNA molecules pass through a nanopore, changes in current are captured by a sensor and converted into electrical signals. These electrical signals contain information about the DNA sequence. Using base recognition algorithms, these electrical signals can be translated into the corresponding base sequences, thus obtaining the DNA sequencing results. Base recognition is the core step in the entire sequencing process, and its accuracy directly affects the reliability of the sequencing results. The basic principles and commonly used algorithms for base recognition include Hidden Markov Models and Neural Network Models.
[0044] Compared to traditional sequencing methods, nanopore sequencing offers several significant advantages in base recognition. First, it provides real-time sequencing capabilities because electrical signals are instantly converted into DNA sequences, eliminating the need to wait for the entire sequencing process to complete. This makes nanopore sequencing a promising candidate for rapid sequencing applications, such as clinical diagnostics and virus surveillance.
[0045] Secondly, nanopore sequencing eliminates the need for DNA amplification, thus avoiding biases and errors introduced by amplification. This makes nanopore sequencing exhibit higher accuracy and reliability in sequencing low-volume or low-quality samples.
[0046] However, base recognition in nanopore sequencing also faces some challenges and drawbacks. Firstly, because the electrical signals within the nanopores are affected by various factors, such as the velocity of DNA movement and the shape of the nanopore, electrical signal noise, DNA modification, and sequencing error rates contribute to the challenges. Therefore, the accuracy and error rate of base recognition remain key issues requiring further research and improvement. While current base recognition algorithms have made significant progress, a certain error rate still exists.
[0047] Secondly, the base recognition algorithms for nanopore sequencing are still under development and improvement. Because nanopore sequencing is a relatively new technology, different manufacturers and laboratories may use different base recognition algorithms, leading to variations in the consistency and comparability of results. Therefore, selecting a suitable base recognition algorithm is also an important consideration in nanopore sequencing.
[0048] In summary, while nanopore sequencing offers rapid, real-time sequencing, it also faces challenges in improving accuracy and consistency. With continuous advancements in technology and algorithms, nanopore sequencing holds promise as a more reliable and widely applicable sequencing method.
[0049] Currently, the following formula is used in related technologies to assess the error of base sequencing:
[0050]
[0051] The above formula confirms that third-generation nanopore sequencing technology has certain errors during the sequencing process, as detailed below:
[0052] Insertion / deletion errors: Insertion and deletion errors (i.e., insertion and deletion in the formula) also exist in nanopore sequencing. While the incidence of insertion / deletion errors is low, they do occur, especially in the sequencing of long DNA molecules.
[0053] Mismatch error: Nanopore sequencing may also result in base mismatch error, which is the misidentification of one base as another (i.e., mismatch in the formula). The occurrence rate of mismatch error is relatively low, but it may be more likely to occur in some specific regions, such as tandem repeat sequences or GC-rich regions.
[0054] Correction Error: The correction process in nanopore sequencing can also introduce certain errors. Correction refers to the process of converting the raw electrical signal into a base sequence, which involves model-based algorithms and the selection of correction parameters. Different correction methods and parameter choices may affect the final sequencing results, leading to the introduction of errors.
[0055] Most errors are random and can be corrected by increasing sequencing depth. However, systematic errors exist in tandem repeat regions that are difficult to correct, and errors generated during nanopore sequencing are typically concentrated in tandem repeat regions (homopolymer regions) rather than being uniformly distributed throughout the sequence. This makes it difficult to correct these errors by increasing sequencing depth. Therefore, related technologies propose using deep learning to correct errors in nanopore sequencing data, such as... Figure 1 The diagram illustrates the systematic errors in nanopore sequencing. The left side shows the error distribution for specific sequences, while the right side shows the error rate at specific positions (ESB), which includes specific errors from some Homopolymers (GGGGG, CCCCC, GGGGGGG). The left side also shows the distribution density of general ESBs. The left image shows that the first-order error rate reaches 50% in a 5-length C-Homopolymer and 75% in a 6-length Homopolymer. Furthermore, the general error distribution rate in the original experimental data is concentrated around 10%. The right image shows the error distribution of the original data versus the error distribution excluding the Homopolymer region. Nanopore sequencing may also exhibit some sequence-specific biases. For example, certain specific sequences can lead to inaccurate sequencing results. The right image concludes that compared to errors without Homopolymers (HP-compressed), introducing Homopolymer regions introduces a significant amount of base substitution and deletion errors.
[0056] However, related technologies often rely on large KMER databases for error correction of sequencing sequences. In solving the problem of tandem repeat sequences, the length of KMER is inflexible, and when the length of KMER is shorter than that of the tandem repeat sequence, it cannot correct the error across the repeat region, and may even introduce new errors.
[0057] The relevant technologies have the following drawbacks in the error correction process of sequencing sequences:
[0058] 1) Third-generation nanopore sequencing has errors caused by the addition or deletion of bases. Existing error correction methods using short-read data perform poorly in repetitive regions (with ambiguous alignments) and regions with extreme GC content (due to reduced coverage caused by biases in short-read sequencing). This patent aims to correct the errors caused by the addition or deletion of bases in tandem repeat sequences, which are mainly systematic errors.
[0059] 2) Base recognition cannot resolve systematic errors.
[0060] 3) The neural network encoding method (embedding) at the base recognition level is detached from the upstream base recognition technology and cannot efficiently correct this error.
[0061] 4) Other methods for error correction using neural networks exist, and a base recognition encoding method is employed. When trained and corrected in E. coli, the error is reduced from 19.55% to 14.22%, which is better than the above method. However, this method trains on existing reference sequence data and then corrects on subsequent data of the same species, and cannot perform pan-species correction.
[0062] For example, among related technologies, PacBio's long-read sequencing method is CLR (Single-Molecule Real-Time sequencing). This method has lower single-read accuracy compared to Illumina's short-read accuracy (greater than 99.9%), but its error patterns exhibit significant randomness. Therefore, genome polishing tools (such as Quiver and Arrow) can be used to correct these errors. Figure 2 The image shows a method in related technologies that uses the Quiver approach to correct sequencing sequence errors. These tools use CLR alignment results and underlying raw pulse information to infer the true sequence of the region based on sequence consistency.
[0063] Compared to the less than 1% error rate of second-generation Illumina sequencing, the error rate of third-generation nanopore sequencing technology is relatively high. However, with technological advancements and algorithm improvements, the error rate of nanopore sequencing is gradually decreasing. Current error rates are typically between 10% and 15%, but studies and improvements have reduced error rates to below 2% in some applications (improved data correction algorithms, model training and updates, data quality filtering, optimized nanopore preparation and regulation, optimized sequencing enzymes, and the combination of multiple algorithms and alignment strategies). Various errors are particularly pronounced in tandem repeat fragments, and numerous studies have highlighted this error problem. Furthermore, ample evidence suggests a close link between tandem repeat regions and variant diseases, which are currently unexplained and cannot be accurately measured.
[0064] With further research and technological improvements, nanopore sequencing is expected to achieve higher accuracy and reliability. Therefore, it is crucial to develop self-correcting algorithms to correct sequencing system errors in tandem repeat regions, especially Homopolymer regions, and to address errors that may be introduced by base detection algorithms.
[0065] Some self-calibration methods in related technologies include PBCR, HGAP-Sprai, and LoRMA-Vechat. These methods perform graph-based self-calibration of the homogeneous region in the assembly problem, or introduce short-read data for polishing, such as the quiver in PbCR, rather than self-calibration based on single sequencing reads. Graph-based calibration methods can effectively correct homogeneous sequences only if the reads are shorter, the error accumulation is smaller, assembly is complete, and homogeneous sequence information is available. However, they cannot solve the specific problem of high errors (50%-75%) in homopolymers (1 bp tandem repeats).
[0066] To address the problems in related technologies, this disclosure proposes a sequencing data error correction method based on deep learning, which analyzes upstream and downstream features to correct tandem repeat fragments. This achieves an efficient, species-wide long-read self-correction method to improve sequencing data accuracy, reduce systematic errors, and provide a reliable data foundation for downstream application analysis.
[0067] The sequencing data error correction method, apparatus, electronic device, and storage medium of this disclosure are described below with reference to the accompanying drawings.
[0068] Figure 3 This is a schematic flowchart illustrating a sequencing data error correction method provided in an embodiment of this disclosure. Figure 3 As shown, the method includes:
[0069] Step 101: Obtain long read sequencing data, which includes tandem repeat regions.
[0070] In some embodiments, this disclosure can sequence a species to obtain initial sequencing data corresponding to the target species. The initial sequencing data includes short-read sequencing data and long-read sequencing data. The correction model can perform error correction on the short-read and long-read sequencing data; in this embodiment, error correction on long-read sequencing data is used as an example for illustration.
[0071] In some embodiments, long-read sequencing data are typically stored in file formats such as FASTQ or FASTA. These data contain base sequence information from the species' genome or transcriptome.
[0072] Tandem repeat regions include tandem repeat sequencing data, which refers to continuously repeating DNA sequence segments in the genome. These tandem repeat sequencing data can be simple (e.g., AAAA), complex (e.g., ABCABCABC), or composed of repeat units of varying lengths.
[0073] Step 102: Encode the tandem repeat region and the upstream and downstream sequencing data of the tandem repeat region using the first encoding method to obtain the base continuity feature.
[0074] Step 103: Encode the upstream and downstream sequencing data of the tandem repeat region using the second encoding method to obtain upstream and downstream features.
[0075] In some embodiments, during sequencing, this disclosure reads not only the tandem repeat sequencing data itself, but also the sequencing data surrounding these data, i.e., upstream and downstream sequencing data. This upstream and downstream sequencing data provides important information about the location and context of the tandem repeat sequences within the genome.
[0076] In some embodiments, this disclosure can identify tandem repeat regions in long-read sequencing data. For each tandem repeat region, sequencing data of a certain length are extracted upstream and downstream, i.e., upstream and downstream sequencing data. These upstream and downstream sequencing data can provide contextual information about the repeat region, which helps in subsequent encoding and correction to convert the long-read sequencing data into digital vectors.
[0077] After obtaining the tandem repeat region and the upstream and downstream sequencing data of the tandem repeat region in the long read sequencing data, this disclosure can use the first encoding method and the second encoding method to encode them respectively, thereby obtaining the encoded base continuity feature and upstream and downstream feature.
[0078] Step 104: Input the base continuity features and upstream and downstream features into the pre-trained calibration model to correct errors in tandem repeat regions in long-read sequencing data.
[0079] In some embodiments, the pre-trained correction model in this disclosure is a deep model, which can be used for error correction, such as a convolutional neural network (CNN), a recurrent neural network (RNN), or a variant thereof (e.g., LSTM, GRU).
[0080] In this disclosure, a calibration model can be trained using known error-free or low-error sequencing data. During training, the calibration model learns how to identify and correct potential sequencing errors from the input numeric vectors, resulting in error-corrected sequencing data. Specifically, the calibration model identifies potential sequencing errors from base continuity features and upstream / downstream features, and performs error correction on these errors.
[0081] Furthermore, this disclosure can also perform result analysis on the corrected sequencing data to analyze the reasons for errors in long-read sequencing data.
[0082] It should be noted that this disclosure can utilize the corrected group sequencing data to iteratively optimize the deep model or training process, thereby further improving the accuracy and efficiency of error correction.
[0083] In summary, the technical solution provided in this disclosure acquires long-read sequencing data, which includes tandem repeat regions; encodes the tandem repeat regions and their upstream and downstream sequencing data using a first encoding method to obtain base continuity features; encodes the upstream and downstream sequencing data of the tandem repeat regions using a second encoding method to obtain upstream and downstream features; and inputs the base continuity features and upstream and downstream features into a pre-trained correction model to correct errors in the tandem repeat regions of the long-read sequencing data. This achieves efficient and universal sequencing data error correction by combining the tandem repeat regions and their upstream and downstream sequencing data in a deep model, thereby improving sequencing data accuracy and subsequent assembly performance.
[0084] Figure 4a This is a schematic flowchart of another sequencing data error correction method provided in an embodiment of this disclosure. Figure 4a based on Figure 1 The illustrated embodiment further defines step 103 for the second encoding method. Figure 4a In the illustrated embodiment, step 103 includes step 201, as follows: Figure 4a As shown, the method may include:
[0085] Step 201: Use a language model to encode the upstream and downstream sequencing data of the tandem repeat region to obtain upstream and downstream features.
[0086] In some embodiments, this disclosure can utilize a language model in the calibration model to motifize long-read sequencing data at a first preset length, obtaining a motif table corresponding to the long-read sequencing data. Motifization involves segmenting the long-read sequencing data based on the language model to obtain multiple motifs. The motif table includes motifs corresponding to upstream and downstream features of tandem repeat regions in the long-read sequencing data. The motifs in the motif table are encoded to obtain encoded motif-encoded feature data. The first preset length in this disclosure can be set according to factors such as genome size, read length, and error rate distribution of the long-read sequencing data. The first preset length in this disclosure can be a preset length representing a single data point, or a preset length range formed by multiple data points. In embodiments of this disclosure, the first preset length can be set to a preset length range less than or equal to 30.
[0087] The language model can be the DNABERT2 model, which is a BERT model trained from multi-species sequences. BERT (Bidirectional Encoder Representations from Transformers) is a Transformer-based Natural Language Processing (NLP) model. It is a pre-trained model that learns word and sentence representations through unsupervised pre-training on large-scale text data, thus enabling its use in downstream tasks of various NLP tasks.
[0088] In Natural Language Processing (NLP), a token is the smallest unit of text, which can be a word, a character, or a subword. The process of dividing text into tokens is called tokenization. In linguistics such as English, punctuation and spaces can be used to segment the smallest semantic units (tokens). In genetics, the KMER method can be used to segment tokens for subsequent training to solve NLP tasks. Its applications are very wide, including various NLP tasks such as text classification, named entity recognition, question answering systems, and machine translation. Because the BERT model has strong semantic understanding capabilities, it often yields better results. This disclosure utilizes the BERT model to extract upstream and downstream features from tandem repeat sequencing data in tandem repeat regions from different species (Human, Yeast, Mouse, Rice, Virus), thus obtaining upstream and downstream features.
[0089] Figure 4b This is a schematic flowchart of another sequencing data error correction method provided in an embodiment of this disclosure. Figure 4bbased on Figure 1 The illustrated embodiment further defines step 102 for the first encoding method. Figure 4b In the illustrated embodiment, step 102 includes step 301, as follows: Figure 4b As shown, the method may include:
[0090] Step 301: Encode the tandem repeat region and the upstream and downstream sequencing data of the tandem repeat region using a base recognition algorithm to obtain base continuity features.
[0091] In some embodiments, this disclosure can utilize a base recognition algorithm to extract tandem repeat regions at a second preset length, determine the upstream and downstream sequencing data of the tandem repeat regions, and encode the tandem repeat regions and the upstream and downstream sequencing data to obtain base continuity features. In this disclosure, the second preset length can be set according to factors such as genome size, read length of long-read sequencing data, and tandem repeat error rate distribution. The second preset length in this disclosure can also be a preset length represented by a single data point, or a preset length range formed by multiple data points. In embodiments of this disclosure, the second preset length can be set to a preset length range less than 14.
[0092] like Figure 5 As shown, this disclosure provides a specific schematic diagram for determining base continuity characteristics and upstream and downstream characteristics, with reference to... Figure 5 This disclosure can transform upstream and downstream sequencing data on both sides of a Homopolymer (a tandem repeat region of 1 bp in length). The upper layer is a process of encoding upstream and downstream features using a language model, and the lower layer is a process of encoding base continuity features using a base recognition algorithm.
[0093] For example, in this embodiment, the base recognition algorithm can encode the tandem repeat region and the upstream and downstream sequencing data of the tandem repeat region by directly encoding ATCG as the numbers 0-3, which is not limited here.
[0094] It is understood that after the long read sequencing data is corrected for errors using the correction model, the corrected sequencing data can be obtained.
[0095] In some embodiments, such as Figure 6 As shown in the diagram, this disclosure provides a specific structural schematic of a correction model. The deep model disclosed herein can be a simple three-layer CNN model, including convolutional layers, activation layers, and pooling layers. Its specific model parameters are shown in Table 1.
[0096]
[0097]
[0098] Table 1
[0099] The depth model disclosed herein may also use models such as VGG16. The specific model structure is subject to actual conditions and is not limited in the embodiments of this disclosure.
[0100] like Figure 7 As shown, this disclosure provides a schematic diagram of a specific calibration model for correcting upstream and downstream features and base continuity features. The upper layer shows the database real value generation process; the lower layer shows the calibration model's calibration process for long-read sequencing data; and the right side shows the result analysis. It is important to note that before inputting base continuity features and upstream and downstream features into the pre-trained calibration model to correct errors in tandem repeat regions in the long-read sequencing data, this disclosure requires obtaining long-read sequencing data from similar species of the species corresponding to the long-read sequencing data and using this data to train the calibration model, thereby obtaining a pre-trained calibration model and improving the accuracy of the calibration.
[0101] It should be noted that when performing error correction, you can directly use the above-mentioned trained correction model for error correction, or you can obtain the above-mentioned trained correction model, use a portion of the long-read sequencing sequences in the long-read sequencing data as training data for the deep model, and use the training data to train the correction model trained on the long-read sequencing data of similar species, so as to use the correction model after secondary training for error correction.
[0102] In addition, base continuity features and upstream and downstream features are input into a pre-trained correction model to correct errors in tandem repeat regions in long-read sequencing data. After that, this disclosure also includes: determining the number of corrected sequencing data after error correction, and storing the number, the sequencing name corresponding to the corrected sequencing data, and the corrected sequencing data in an output file.
[0103] Simultaneously, base continuity features and upstream / downstream features are input into a pre-trained calibration model to correct errors in tandem repeat regions of long-read sequencing data. This disclosure further includes: determining the calibrated long-read sequencing data corresponding to the calibrated sequencing data, and determining error evaluation parameters for the calibrated long-read sequencing data; comparing the reference evaluation parameters with the error evaluation parameters to obtain the anomaly distribution data corresponding to the calibrated long-read sequencing data. Here, the evaluation parameter refers to the ESB value, an indicator used to evaluate the quality of sequencing data or the corrected sequencing data. This disclosure, by comparing the error evaluation parameters and the reference evaluation parameters, can identify differences between the reference sequencing data and the calibrated sequencing data. These differences can include single nucleotide variants (SNPs), insertions / deletions (InDels), structural variations (SVs), etc. By analyzing these differences, anomaly distribution data of the calibrated sequencing data can be obtained, such as addition / deletion bias performance analysis, error distribution before and after at HP bases, error distribution before and after at different bases, error distribution before and after at different lengths, model training set accuracy, model test set accuracy, etc. The reference evaluation parameters are mainly derived from public databases, genome projects, laboratory-in-house sequencing, and data integration and sharing, and are used to compare error evaluation parameters. This disclosure utilizes Minimap2 technology when comparing sequencing data.
[0104] In summary, this disclosure improves the high sequencing error rate of tandem repeat fragments by using artificial intelligence algorithms, combining tandem repeat regions in long-read sequencing data with upstream and downstream sequencing data of tandem repeat regions, and applying supervised learning neural networks. This achieves efficient and universal error correction for long-read sequencing data, improves data accuracy, and enhances subsequent assembly performance.
[0105] As one possible implementation, based on the above embodiments, this disclosure provides a specific method for correcting sequencing data errors. Taking the correction of long-read sequencing data of Ecoli as an example, this disclosure uses HG002 from the 1000 Genomes Study as a training sample source to construct a model training database. This disclosure can extract 58,961 long-read base sequences from the nanopore sequencing data of chromosome 10 of HG002, and use the alignment results with the HG002 reference dataset as the training dataset. Using the Ecoli nanopore sequencing test data, a test dataset different from the training set species is finally generated, and the correction results are analyzed, including average read length, average error rate, etc. Specifically, the simulated data characteristics are as follows: For the HG002 training dataset, the distribution of pre-training read length and single-base error rate is analyzed using Minimap2 alignment results. The average read length is 30 kbp, and the average error rate is 15%. A total of 58,961 reads are obtained. The Ecoli test dataset was used to analyze the distribution of pre-trained read length and single-base error rate using Minimap2 alignment results. The average read length was 30kbp, and the average error rate was 15%. A total of 2,421,999 reads were recorded.
[0106] In other words, this disclosure can collect nanopore sequencing test data of target species (such as Escherichia coli) and reference datasets (such as human sample HG002) as training samples.
[0107] The DNABERT2 model was used to perform motif processing on the TGS Cyclone data from the reference dataset (HG002) and the test dataset (Ecoli). Considering factors such as human genome size, read length of long reads, and error rate distribution, a first preset length of less than or equal to 30 was determined, i.e., the motif length was less than or equal to 30. Upstream and downstream sequencing data from tandem repeat regions were extracted to generate a motif table, and the upstream and downstream sequencing data were encoded to generate upstream and downstream feature datasets. The final feature sequence information obtained was a feature set with a total length of 60 from upstream and downstream sequences. A total of 21,787,485 samples were collected in the test set and 138,364 samples were collected in the training set.
[0108] For tandem repeat regions, tandem repeat regions were extracted from the TGS Cyclone data of HG002 and ATCC_8739. Considering factors such as the size of the human genome, the length of long reads, and the distribution of tandem repeat error rates, a second preset length of less than or equal to 14 was determined. This means that sequencing data from tandem repeat regions with a total length less than 14 (tandem repeat regions of length 1) were used. The tandem repeat regions and their upstream and downstream sequencing data were encoded according to the size of the upstream and downstream feature layers of the motif. A base continuity feature dataset was generated using a base recognition encoding method. The final feature sequence information was a feature set with a total length of 60 from upstream and downstream sequences. A total of 21,787,485 samples were collected for the test set and 138,364 samples were collected for the training set.
[0109] In this context, motifization refers to the token segmentation mechanism in the DNABERT2 model for processing DNA sequence data. The DNABERT2 model segments the DNA sequence into a series of tokens based on a preset motif length (a first preset length), with each token representing a continuous DNA sequence fragment. This segmentation method allows the model to capture local patterns or motifs within the DNA sequence, which may possess specific biological functions. When performing base recognition encoding, reference is made to... Figure 5 As shown, this disclosure can use the token segmentation mechanism in the DNABERT2 model to process the motif size (i.e., the first preset length) determined by token segmentation in DNA sequence data, and set the coding unit size of the base recognition part. This means that during the encoding process, this disclosure can divide the tandem repeat region and its upstream and downstream sequencing data according to the motif size, and encode each divided unit (i.e., token). In this way, this disclosure ensures that the encoded feature sequence can reflect both the characteristics of the tandem repeat region itself and the contextual information of its upstream and downstream sequences, and that the granularity of this information is consistent with the token granularity when the model processes the DNA sequence.
[0110] This disclosure allows the use of partial sequencing data from current long-read sequencing data as input for secondary training of the repetition error in tandem repeat regions of long-read sequences. This training enables the creation of a pre-trained calibration model (a deep model, i.e., a neural network model) to predict and correct errors in tandem repeat regions of the initial sequencing data. Alternatively, the pre-trained calibration model can be directly used for error detection in long-read sequencing data. In terms of deep learning interpretability, in two layers of data, upstream and downstream features can be used to find the relationship between specific motifs and tandem repeat region errors through model training; base continuity features can enable the model to correct errors introduced by the instability of continuous electrical signals. Furthermore, during the training process of this disclosure, different calibration methods can be applied to different base recognition models, and separate training of the two parts can protect the stability of base recognition.
[0111] This disclosure applies a trained calibration model to upstream and downstream features and base continuity features in long-read sequencing data to calibrate the long-read sequencing data, resulting in calibrated sequencing data.
[0112] The specific training / pre-training steps include:
[0113] By providing base recognition to generate FASTA / ...
[0114] Then, upstream and downstream sequencing data of the tandem repeat sequencing data are read, and the tandem repeat region and the upstream and downstream sequencing data of the tandem repeat region are encoded using a base recognition algorithm to obtain base continuity features; the upstream and downstream sequencing data of the tandem repeat region are encoded using a language model to obtain upstream and downstream features.
[0115] During the calibration process, a pre-trained calibration model can be used to directly read long-read sequencing data in "FASTA" or "FASTQ" format. The trained calibration model corrects tandem repeat regions in the long-read sequencing data based on upstream and downstream features and base continuity characteristics. Alternatively, a portion of the long-read sequencing data can be randomly selected, trained on the pre-trained calibration model, and then calibrated; or calibration can be performed directly, and the number of calibrated sequencing data points can be obtained. The calibrated single sequencing data points (calibrated sequencing data), along with the original single sequencing data point name (the corresponding sequencing name), and the number, are stored in an output file in "FASTA" format. The pre-trained calibration model refers to a calibration model trained using long-read sequencing data from species similar to the species corresponding to the long-read sequencing data, to address the problem of other models' inability to generalize to multiple species.
[0116] Additionally, this disclosure can also use minimap2 to label tandem repeat sequence parts, perform supervised training to obtain error estimates of tandem repeat sequences under motif units and upstream and downstream features, and use specialized evaluation parameters (such as ESB parameters) for evaluation.
[0117] Reference Figure 1 To analyze the accuracy of specific tandem repeat regions, related studies used the ESB value as the evaluation parameter. This disclosure also uses the same evaluation method, and the evaluation results are as follows: Figure 8 , Figure 9 As shown. Figure 8 This diagram illustrates the performance of a sequencing data correction method disclosed herein on a human dataset (same species). Figure 9 This is a schematic diagram of the calibration results on Ecoli data provided in this disclosure. The deep model was trained using a portion of human long-length sequencing data.
[0118] Figure 8 The error change of ESB before and after correction of the original single-base long sequence of human HG002 is shown. It can be seen that the model trained on the human database can significantly reduce the error of tandem repeat regions on the human test set, taking a 6-base long T-Homopolymer as an example. Figure 8 First row, left Figure 1 (ii) The first-order error decreased from about 70% to 50%, and the average error decreased from 60% to 32%.
[0119] Figure 9 The model, trained on a human dataset, shows the changes before and after correction for a single original E. coli read. The left image in Fig. 13 shows a reduction in error. The two right images show that the length error distribution of the Homopolymer region becomes more convergent before and after correction. This demonstrates that the model has good generalization ability and adaptability to data from diverse species, making it more reliable and effective in practical applications.
[0120] Figure 10 This is a schematic diagram of the correction results of a pan-species model (without access to Ecoli data) provided in this disclosure on Ecoli data. The comparison was performed using minimap2. The left figure shows the confusion matrix representation of the original tandem repeat regions before correction, and the right figure shows the confusion matrix distribution of the tandem repeat regions after correction.
[0121] In summary, this disclosure uses the DNABERT2 model to extract upstream and downstream sequencing data of tandem repeat regions for encoding, extracts flexible-length and variable-length sequence features for model training, and does not rely on species-specific KMER tables as training data. Simultaneously, a base recognition algorithm is used to extract tandem repeat regions, and base encoding is performed on both the tandem repeat regions and their upstream and downstream sequencing data. These two methods are combined as input to a pre-trained calibration model. This ensures the interpretability of error correction introduced by the base recognition process, as well as the interpretability of errors in tandem repeat regions caused by structure, significantly reducing the error rate and improving the accuracy of sequence analysis to achieve accurate correction of tandem repeat regions. Furthermore, this disclosure does not rely on a specified large reference genome database, but uses pan-species sequence features to perform self-calibration of tandem repeat regions on arbitrary sequencing data. This has a positive impact on subsequent genome assembly, overcoming the problem of poor assembly quality of tandem repeat regions, improving integrity, reducing the dependence of tandem repeat regions on short-read data, reducing time and economic costs, solving the problem of extremely high errors in tandem repeat data in consensus fragments, and improving the continuity of contiguous groups.
[0122] Corresponding to the sequencing data error correction method described above, this invention also proposes a sequencing data error correction device. Since the device embodiments of this invention correspond to the method embodiments described above, details not disclosed in the device embodiments can be referred to in the method embodiments described above, and will not be repeated here.
[0123] Figure 11 This is a schematic diagram of the structure of a sequencing data error correction device provided in an embodiment of this disclosure, as shown below. Figure 11 As shown, the device includes:
[0124] Acquisition unit 1110 is used to acquire long read sequencing data, which includes tandem repeat regions.
[0125] The first coding unit 1120 is used to encode the tandem repeat region and the upstream and downstream sequencing data of the tandem repeat region using the first coding method to obtain the base continuity feature.
[0126] The second coding unit 1130 is used to encode the upstream and downstream sequencing data of the tandem repeat region using the second coding method to obtain upstream and downstream features;
[0127] The correction unit 1140 is used to input base continuity features and upstream and downstream features into a pre-trained correction model to correct errors in tandem repeat regions in long-read sequencing data.
[0128] In some embodiments of this disclosure, the first encoding method includes a base recognition algorithm; the second encoding method includes a language model; the first encoding unit 1120 is used to: encode the tandem repeat region and the upstream and downstream sequencing data of the tandem repeat region using the base recognition algorithm to obtain base continuity features; the second encoding unit 1130 is used to encode the upstream and downstream sequencing data of the tandem repeat region using the language model to obtain upstream and downstream features.
[0129] In some embodiments of this disclosure, the second encoding unit is used to: motifize the long read sequencing data with a first preset length to obtain a motif table corresponding to the long read sequencing data; motifization is to segment the long read sequencing data based on a language model to obtain multiple motifs; and to encode each motif in the motif table to obtain upstream and downstream features.
[0130] In some embodiments of this disclosure, the apparatus 1100 includes a training unit for: before inputting base continuity features and upstream and downstream features into a pre-trained calibration model to correct errors in tandem repeat regions in long-read sequencing data, acquiring long-read sequencing data of a species similar to the species corresponding to the long-read sequencing data, and training the calibration model using the long-read sequencing data of the similar species to obtain a pre-trained calibration model.
[0131] In some embodiments of this disclosure, the correction unit 1130 is configured to: input base continuity features and upstream and downstream features into a pre-trained correction model to perform error correction on tandem repeat regions in long-read sequencing data; then, determine the number of corrected sequencing data after error correction; and store the number, the sequencing name corresponding to the corrected sequencing data, and the corrected sequencing data in an output file.
[0132] In some embodiments of this disclosure, the calibration model employs a three-layer structure, which includes a convolutional layer, an activation layer, and a pooling layer.
[0133] In some embodiments of this disclosure, the correction unit 1130 is configured to: input base continuity features and upstream and downstream features into a pre-trained correction model to perform error correction on tandem repeat regions in long-read sequencing data; then, determine the corrected long-read sequencing data corresponding to the error-corrected sequencing data, and determine the error evaluation parameters of the corrected long-read sequencing data; compare the reference evaluation parameters with the error evaluation parameters to obtain the abnormal distribution data corresponding to the corrected long-read sequencing data.
[0134] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of this embodiment, and the principle is the same, so it is not limited in this embodiment.
[0135] According to embodiments of this disclosure, this disclosure also provides an electronic device and a readable storage medium.
[0136] Figure 12 A schematic block diagram of an example electronic device 1200 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.
[0137] like Figure 12 As shown, device 1200 includes a computing unit 1201, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 1202 or loaded from storage unit 1208 into RAM (Random Access Memory) 1203. RAM 1203 may also store various programs and data required for the operation of device 1200. The computing unit 1201, ROM 1202, and RAM 1203 are interconnected via bus 1204. I / O (Input / Output) interface 1205 is also connected to bus 1204.
[0138] Multiple components in device 1200 are connected to I / O interface 1205, including: input unit 1206, such as keyboard, mouse, etc.; output unit 1207, such as various types of monitors, speakers, etc.; storage unit 1208, such as disk, optical disk, etc.; and communication unit 1209, such as network card, modem, wireless transceiver, etc. Communication unit 1209 allows device 1200 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0139] The computing unit 1201 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1201 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 1201 performs the various methods and processes described above, such as sequencing data error correction methods. For example, in some embodiments, the sequencing data error correction method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1208. In some embodiments, part or all of the computer program may be loaded and / or installed on device 1200 via ROM 1202 and / or communication unit 1209. When the computer program is loaded into RAM 1203 and executed by the computing unit 1201, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 1201 may be configured to perform the aforementioned sequencing data error correction method by any other suitable means (e.g., by means of firmware).
[0140] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.
[0141] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.
[0142] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
[0143] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).
[0144] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.
[0145] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service ecosystem, addressing the shortcomings of traditional physical hosts and VPS (Virtual Private Server, or simply "VPS") services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.
[0146] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.
[0147] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.
[0148] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.
Claims
1. A method for correcting sequencing data errors, characterized in that, The method includes: Acquire long-read sequencing data, wherein the long-read sequencing data includes tandem repeat regions; The tandem repeat region and the upstream and downstream sequencing data of the tandem repeat region are encoded using a first encoding method to obtain base continuity characteristics; The upstream and downstream sequencing data of the tandem repeat region are encoded using a second encoding method to obtain upstream and downstream features; The base continuity feature and the upstream and downstream features are input into a pre-trained correction model to correct errors in tandem repeat regions in the long-read sequencing data.
2. The method according to claim 1, characterized in that, The first encoding method includes a base recognition algorithm; the second encoding method includes a language model, and the method includes: The tandem repeat region and the upstream and downstream sequencing data of the tandem repeat region are encoded using a base identification algorithm to obtain base continuity features; The upstream and downstream sequencing data of the tandem repeat region are encoded using a language model to obtain upstream and downstream features.
3. The method according to claim 2, characterized in that, The process of encoding upstream and downstream sequencing data of the tandem repeat region using a language model to obtain upstream and downstream features includes: The long read sequencing data is motifized at a first preset length to obtain a motif table corresponding to the long read sequencing data; the motifization is to segment the long read sequencing data based on the language model to obtain multiple motifs; Each motif in the motif table is encoded to obtain the upstream and downstream features.
4. The method according to claim 1, characterized in that, Before inputting the base continuity feature and the upstream and downstream features into a pre-trained correction model to correct errors in tandem repeat regions in the long-read sequencing data, the method includes: Obtain long-read sequencing data of similar species to the species corresponding to the long-read sequencing data, and use the long-read sequencing data of the similar species to train the calibration model to obtain a pre-trained calibration model.
5. The method according to claim 1, characterized in that, The method involves inputting the base continuity features and the upstream and downstream features into a pre-trained correction model to correct errors in tandem repeat regions in the long-read sequencing data. The method then includes: Determine the number of corrected sequencing data after error correction, and store the number, the sequencing name corresponding to the corrected sequencing data, and the corrected sequencing data in an output file.
6. The method according to claim 1, characterized in that, The correction model adopts a three-layer structure, which includes a convolutional layer, an activation layer, and a pooling layer.
7. The method according to claim 1, characterized in that, The method involves inputting the base continuity features and the upstream and downstream features into a pre-trained correction model to correct errors in tandem repeat regions in the long-read sequencing data. The method then includes: Determine the corrected long read sequencing data corresponding to the corrected sequencing data after error correction, and determine the error evaluation parameters of the corrected long read sequencing data; By comparing the reference evaluation parameters with the error evaluation parameters, the abnormal distribution data corresponding to the corrected long read sequencing data are obtained.
8. A sequencing data error correction device, characterized in that, The device is applied to a depth model, and the device includes: An acquisition unit is used to acquire long read sequencing data, wherein the long read sequencing data includes tandem repeat regions; The first coding unit is used to encode the tandem repeat region and the upstream and downstream sequencing data of the tandem repeat region using a first coding method to obtain base continuity features. The second coding unit is used to encode the upstream and downstream sequencing data of the tandem repeat region using a second coding method to obtain upstream and downstream features; The correction unit is used to input the base continuity feature and the upstream and downstream features into a pre-trained correction model to correct errors in the tandem repeat regions in the long read sequencing data.
9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.