Quick reading method for watermark superposition coding large-fragment DNA data storage

Abstract

The invention discloses a fast reading method for watermark superposition coding large-fragment DNA data storage, and belongs to the field of DNA data storage. The method comprises the following steps: firstly, breaking a large-fragment DNA (Deoxyribonucleic Acid) subjected to watermark superposition coding, establishing a library, and carrying out second-generation high-throughput sequencing to obtain corresponding sequencing data; then, carrying out sliding correlation on a watermark sequence hidden in the noise sequencing read section and a local known watermark sequence to calculate a cross-correlation peak value; and determining the position of the read segment in the large-fragment DNA sequence according to the corresponding position of the cross-correlation peak value, then generating a consistent sequence by adopting a majority voting decision algorithm, and further realizing the reliable recovery of the stored data through an efficient error correction and deletion correction decoding algorithm. The method has the advantages that the data DNA length from the Kb level to the Mb level is supported, read positioning is achieved through simple sliding correlation and sequence consensus, de novo assembly with high complexity is effectively avoided, non-coding DNA read interference can be eliminated, and original data error-free recovery can be achieved under the low sequencing coverage degree.

Description

[0001] The invention relates to the field of DNA data storage, and in particular to a method for quickly reading watermark superimposed coded large-segment DNA data. Background Art

[0002] With the rapid development of digitalization, networking, and intelligence in information technology, the amount of global data is growing rapidly, and the value of long-term data storage has been increasingly highlighted with the development of intelligent technologies such as large models. How to store massive amounts of data for a long time has become an important requirement for the development of information technology. With the explosive growth of data storage scale, traditional data storage media and storage technologies are facing challenges. Existing media face problems such as limited service life, high daily maintenance costs, and slow growth in storage density due to physical limits. In particular, long-term data storage often requires the use of data migration to achieve long-term data reliability, resulting in a significant increase in maintenance costs. Traditional storage media cannot meet the storage needs of rapidly growing massive data, and exploring new storage media and storage models has become one of the keys to the long-term and healthy development of information technology.

[0003] Compared with traditional data storage media, synthetic DNA molecules have the advantages of high storage density, low maintenance cost, and a large accumulation of high-throughput read and write technologies. They are becoming a new form of storage medium with great potential for future massive data archiving and storage. In recent years, data storage systems of various forms of synthetic DNA media have been evaluated, which can be mainly divided into: short fragment oligonucleotide pool storage method and large fragment DNA data storage method. Unlike the storage solution based on short fragment oligonucleotide pool, large fragment DNA data storage can be assembled and amplified through cell proliferation, with low data replication cost and high reliability, and has potential application value in large-scale data distribution scenarios; the processing of large fragment DNA can better utilize the processing mechanism in the body for DNA assembly and production, etc., with higher reliability and better flexibility.

[0004] The readout of large DNA fragments is similar to that of cell DNA sequencing, and can be assisted by traditional cell genome sequencing technology and data processing technology.

[0005] On the one hand, large DNA fragments can be sequenced using third-generation nanopore sequencing, but the error rate of third-generation sequencing is currently high (about 10%), and the error types are more complex, including base insertion and deletion errors that are more difficult to handle. The handling of such errors is usually challenging, and complex insertion and deletion processing algorithms are required to recover the stored data. For example, the research team of Tianjin University proposed a scheme using sparse LDPC codes and watermark sequence superposition encoding, designed and synthesized a yeast artificial chromosome with a length of 254,886 bases, and stored 37.8KB of data files. After third-generation nanopore sequencing, a decoding strategy that integrates genome assembly and efficient indel (Insertion and Deletion) error correction decoding algorithm was proposed, achieving error-free data recovery at 16.8× sequencing coverage. Researchers from Shanghai Jiaotong University and Tianjin University used 6 large-fragment uniform DNA sequences of 6,750 bases to store a 6.5KB image file, and restored the original data using 53.76× third-generation nanopore sequencing reads through the sequence alignment software MAFFT. The above methods all take advantage of the portability and rapidity of third-generation nanopore sequencing, but they also face the challenge of dealing with complex indel errors.

[0006] On the other hand, similar to genome sequencing, large DNA fragments can also be read out using the second-generation high-throughput sequencing technology based on the shotgun method to obtain a large number of short sequencing reads. Similar to genome de novo sequencing, the read assembly operation is complex, and it is impossible to obtain a "perfect" consistent sequence at low sequencing coverage, which poses a great challenge to low-cost error-free data recovery. BGI used a 54,520-base DNA fragment for intracellular data storage, and used the genome de novo assembly software SOAPdenovo to assemble the second-generation sequencing reads during the data readout process, and recovered the stored data files; the Tianjin University team performed second-generation high-throughput sequencing readout on the designed and synthesized yeast artificial chromosome, and used the second-generation sequencing read assembly software Velvet with the help of graph theory DNA assembly methods, further combined with the positioning of autonomous replicating sequences (ARS), to achieve error-free recovery of data with a sequencing coverage of 23.5×; at the same time, the researchers simulated a 2.5Mb circular chromosome through the design of a coding scheme with multiple extremely high-rate Reed-Solomon (RS) codewords and performed second-generation high-throughput sequencing readout, and used the de Brujin graph-based assembly software Velvet and ABySS to achieve data recovery at a sequencing coverage of 20×. Similar to de novo genome sequencing, when second-generation high-throughput sequencing is used to read data from large DNA fragments, the data processing complexity of read segment assembly is high, and the required sequencing coverage is generally high, usually requiring more than dozens of times.

[0007] Researchers have previously proposed a coding scheme based on the superposition of watermark sequences and sparse error correction codes for the storage of long-fragment DNA data. The watermark sequence is hidden in the sequencing read segment and can be used to identify the complex insertion and deletion errors introduced by the third-generation nanopore sequencing. Further combined with the substitution errors, the sequencing data based on the third-generation nanopore can be read out. For this large-fragment DNA with watermark superposition coding, the second-generation high-throughput sequencing readout is limited by the high-complexity de novo assembly of the read segments, and data recovery cannot be achieved at low sequencing coverage, which increases the cost of reading data from long-fragment DNA. Summary of the invention

[0008] The present invention provides a fast readout method for storing large-fragment DNA data with watermark superposition coding. The present invention supports data DNA lengths ranging from Kb to Mb, uses simple sliding correlation and sequence consensus to achieve segment positioning, effectively avoids highly complex de novo assembly, and can eliminate the interference of non-coding DNA reads, and can achieve error-free recovery of original data at lower sequencing coverage, as described below:

[0009] A method for quickly reading large-fragment DNA data stored by watermark superposition coding, the method comprising the following steps:

[0010] (1) The large DNA fragments encoded by watermark superposition are interrupted and the sequencing reads obtained by second-generation high-throughput sequencing are demapped into double-layer bit sequences according to the "base-bit" mapping rule, and the bit sequences are slidingly correlated with the locally known double-layer watermark sequence to calculate the cross-correlation peak, thereby determining the source of the sequencing reads and their starting positions on the large DNA fragment sequence;

[0011] (2) After placing the sequencing reads obtained by sliding correlation recognition at the correct position, a majority voting decision algorithm is used to obtain the consensus base sequence;

[0012] (3) The consistent base sequence is converted into a damaged bit sequence and then sent to a decoder for error correction and decoding to recover the data.

[0013] Wherein, step (1) is:

[0014] (1.1) The large DNA fragment encoded with watermark is fragmented and sequenced by second-generation high-throughput sequencing to obtain a sequence read segment r with a length of n base pairs. According to the "base-bit" mapping rule {A→(00), T→(01), G→(10), C→(11)}, the double-layer bit sequence r with a length of n bits is demapped. 1 and r 2 ;

[0015] (1.2) The double-layer bit sequence r 1 and r 2And the known double-layer watermark sequence w with a length of L bits 1 and w 2 The "0" in is converted to "-1", and the bit "1" remains unchanged;

[0016] (1.3) Use a sliding window of length n to calculate the double-layer bit sequence r 1 and r 2 w 1 and w 2 The cross-correlation value of the i-th position is V i The calculation formula is expressed as:

[0017]

[0018] Get L-n+1 cross-correlation values; then compare the calculated cross-correlation values ​​of each position to obtain the cross-correlation peak value V peak =max(V 0 ,V 1 ,…V L-n ) and the corresponding positions;

[0019] (1.4) Using the cross-correlation peak threshold to screen the sequencing reads that meet the requirements, and record the peak position, that is, the starting position of the sequencing read on the large DNA sequence;

[0020] (1.5) The sequencing reads are retrieved according to the cross-correlation peak positions for recovery of different data.

[0021] Wherein, step (2) is:

[0022] (2.1) If the length of the data DNA sequence is l bases, construct a 5*l counting buffer corresponding to the number of the five bases {A, T, G, C, N} at each position;

[0023] (2.2) The starting positions of the sequencing reads obtained by sliding correlation recognition are different. The identified sequencing reads are placed at the corresponding starting positions according to the cross-correlation peak positions, and the counting buffers at the corresponding positions are added and counted according to the base types at each position of the reads to obtain a counting buffer containing the number of five bases at each position;

[0024] (2.3) Perform a base-by-base majority vote on the counting buffer and determine the base with the largest number in the buffer at the i-th position as the base at that position. If there is no sequencing data at that position or a majority vote cannot be performed, it is marked as "X", which means an erasure error, and a consensus sequence s with a length of l bases is obtained.

[0025] The consistent base sequence is converted into a damaged bit sequence and then sent to a decoder for error correction and decoding to recover the data. Step (3) is:

[0026] (3.1) Demap the consensus sequence s of length l bases into a double-layer bit sequence according to the "base-bit" mapping rule {A→(00), T→(01), G→(10), C→(11)}, and mark the base X as erasure;

[0027] (3.2) Perform bit-by-bit XOR operation on the fuzzy watermark sequence superimposed with the damaged error correction codeword sequence and the known watermark sequence, remove the superimposed watermark sequence, and obtain the damaged sparse error correction codeword sequence;

[0028] (3.3) Sparsely decode the damaged sparse error correction codeword sequence to obtain a damaged error correction codeword sequence, and erase the error position for marking;

[0029] (3.4) The damaged error correction codeword sequence is sent to the corresponding decoder, and the original data is restored after error correction and erasure decoding.

[0030] The double-layer bit sequence r is calculated using a sliding window of length n. 1 and r 2 w 1 and w 2 The cross-correlation values ​​of each position are compared to obtain the cross-correlation peak value V peak =max(V 0 ,V 1 ,…V L-n ) and the corresponding position, there are two different sliding correlation calculation cross-correlation peak schemes, namely:

[0031] (1) Sequencing reads are selected one by one from sequencing data containing a large number of reads, and after demapping, they are subjected to bit-by-bit sliding correlation with a known double-layer watermark sequence of length L bits, and the cross-correlation value corresponding to each position is sequentially obtained;

[0032] (2) Sequencing data containing a large number of sequencing read segments with a length of n bases are split to obtain multiple sequencing data sub-files; then a known double-layer watermark sequence with a length of L bits is segmented, and n bits of overlapping segments are maintained between adjacent short segments; multiple cross-correlation values ​​are calculated in parallel using a multi-threaded method at the same time, thereby accelerating the sliding correlation calculation process.

[0033] The use of the cross-correlation peak threshold to screen the sequencing reads that meet the requirements needs to determine and optimize the cross-correlation peak threshold based on whether the sequencing data contains DNA fragment sequencing data that is not superimposed with the preset watermark encoding:

[0034] (1) For application scenarios where DNA fragment sequencing data does not contain non-preset watermark superimposed encoding, sequencing read segment positioning is directly achieved based on the cross-correlation peak position;

[0035] (2) For application scenarios containing DNA fragment sequencing data with non-preset watermark superimposed encoding, including second-generation high-throughput sequencing data of host cell genomes and second-generation high-throughput sequencing data of DNA fragments with other encoding methods, different cross-correlation peak thresholds are first set to identify and screen sequencing reads; then, the identified sequencing reads are merged in large numbers to obtain the corresponding consistent sequences and calculate the base error rate; finally, the threshold corresponding to the minimum base error rate is selected as the optimal cross-correlation peak threshold to reduce the impact of interfering reads.

[0036] The beneficial effects of the technical solution provided by the present invention are:

[0037] 1. The present invention proposes a fast readout method for storing large-fragment DNA data with watermark superposition coding. The proposed method first performs second-generation high-throughput sequencing on the large-fragment DNA sequence with watermark superposition coding to obtain the corresponding sequencing data; then uses the fuzzy watermark sequence hidden in the noise sequencing read segment and the locally known watermark sequence to perform sliding correlation to calculate the cross-correlation peak; according to the cross-correlation peak and the corresponding position, a sequence consensus algorithm is used to obtain the consistent base sequence, and then an efficient error correction and erasure decoding algorithm is used to achieve efficient and reliable recovery of the stored data;

[0038] 2. The present invention supports the storage of data DNA lengths ranging from Kb to Mb, uses simple sliding correlation and sequence consensus to achieve read segment positioning, effectively avoids highly complex de novo assembly, and can eliminate the interference of non-coding DNA read segments. It can restore data stored in DNA sequences without errors at lower sequencing coverage, thereby reducing data readout costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A system block diagram for implementing the solution in the present invention;

[0040] Figure 2 A flow chart of recovering data from NGS noisy reads in the present invention;

[0041] Figure 3 It is a flow chart of the read segment alignment between the sequencing read segment and the reference watermark sequence in the present invention;

[0042] Figure 4 A flowchart for merging large numbers to generate a consistent sequence in the present invention;

[0043] Figure 5 A flowchart of transcoding a consistent sequence into a damaged codeword sequence for error correction decoding in the present invention;

[0044] Figure 6 It is a flow chart of the parallel sliding related scheme in the present invention;

[0045] Figure 7 This is a distribution diagram of specific data recovery conditions in specific embodiment 1 of the present invention;

[0046] Figure 8 This is a base error distribution diagram of the majority voting decision process in specific embodiment 1 of the present invention;

[0047] Fig. 9 is the acceleration performance of the parallel sliding related solution in the specific embodiment 1 of the present invention;

[0048] Fig.10 This is a distribution diagram of mixed sequencing data in specific embodiment 2 of the present invention;

[0049] Fig.11 This is a distribution diagram of specific data recovery conditions in specific embodiment 2 of the present invention;

[0050] Fig.12 is the acceleration performance of the parallel sliding related solution in the specific embodiment 2 of the present invention;

[0051] Fig.13 This is a base error distribution diagram of the majority voting decision process in specific embodiment 2 of the present invention;

[0052] Fig.14 The base coverage distribution under the minimum sequencing coverage is restored without error in the present invention;

[0053] Fig.15 The data recovery scheme proposed in the present invention is compared with the traditional read segment de novo assembly scheme.

[0054] Table 1 shows the actual second generation sequencing data parameters used in the present invention;

[0055] Table 2 shows the experimental results of parallel data recovery of different sequencing samples in specific embodiment 1 of the present invention;

[0056] Table 3 shows the experimental results of parallel data recovery of different sequencing samples in specific embodiment 2 of the present invention;

[0057] Table 4 is a comparison between the present invention and other large-fragment DNA data storage solutions. DETAILED DESCRIPTION

[0058] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention are described in further detail below.

[0059] The present invention belongs to the field of DNA data storage, and proposes a fast readout method for watermark superposition coding large-fragment DNA data storage. The present invention first breaks the large-fragment DNA with watermark superposition coding and then builds a library for second-generation high-throughput sequencing to obtain corresponding sequencing data; then the watermark sequence hidden in the noise sequencing read segment is used to calculate the cross-correlation peak with the locally known watermark sequence by sliding correlation; the position of the read segment in the large-fragment DNA sequence is determined according to the corresponding position of the cross-correlation peak, and then the consensus sequence is generated by the majority voting judgment algorithm, and then the reliable recovery of the stored data is realized through an efficient error correction and erasure decoding algorithm. The advantage of the present invention is that it supports data DNA lengths from Kb to Mb, uses simple sliding correlation and sequence consensus to realize read segment positioning, effectively avoids highly complex de novo assembly, and can exclude interference from non-coding DNA read segments, and can realize error-free recovery of original data at a lower sequencing coverage, greatly reducing the cost of data reading.

[0060] Attached Figure 1 The system block diagram of the present invention is shown in FIG. Figure 2 The flowchart of recovering data from NGS noisy reads is shown. The scheme of the present invention includes the following steps:

[0061] (1) The large DNA fragments encoded by watermark superposition are interrupted and the sequencing reads obtained by second-generation high-throughput sequencing are demapped into double-layer bit sequences according to the "base-bit" mapping rule, and the bit sequences are slidingly correlated with the locally known double-layer watermark sequence to calculate the cross-correlation peak, thereby determining the source of the sequencing reads and their starting positions on the large DNA fragment sequence;

[0062] (2) After placing the sequencing reads obtained by sliding correlation recognition at the correct position, a majority voting decision algorithm is used to obtain the consensus base sequence;

[0063] (3) The consistent base sequence is converted into a damaged bit sequence and then sent to a decoder for error correction and decoding to recover the data.

[0064] Among them, as attached Figure 3 As shown, in step (1) of the present invention, the large DNA fragments encoded by watermark superposition are interrupted and the sequencing reads obtained by the second-generation high-throughput sequencing are demapped into a double-layer bit sequence according to the "base-bit" mapping rule, and the bit sequence is slidingly correlated with the locally known double-layer watermark sequence, and the cross-correlation peak is calculated to determine the source of the sequencing reads and their starting positions on the large DNA fragment sequence. The specific steps are:

[0065] (1.1) The large DNA fragment encoded with watermark is fragmented and sequenced by second-generation high-throughput sequencing to obtain a sequence read segment r with a length of n base pairs. According to the "base-bit" mapping rule {A→(00), T→(01), G→(10), C→(11)}, the double-layer bit sequence r with a length of n bits is demapped. 1 and r 2 ;

[0066] (1.2) The double-layer bit sequence r 1 and r 2 And the known double-layer watermark sequence w with a length of L bits 1 and w 2 The "0" in is converted to "-1", and the bit "1" remains unchanged;

[0067] (1.3) Use a sliding window of length n to calculate the double-layer bit sequence r 1 and r 2 w 1 and w 2 The cross-correlation value of the i-th position is V i The calculation formula is expressed as:

[0068]

[0069] Get L-n+1 cross-correlation values; then compare the calculated cross-correlation values ​​of each position to obtain the cross-correlation peak value V peak =max(V 0 ,V 1 ,…V L-n ) and the corresponding positions;

[0070] (1.4) Using the cross-correlation peak threshold to screen the sequencing reads that meet the requirements, and record the peak position, that is, the starting position of the sequencing read on the large DNA sequence;

[0071] (1.5) The sequencing reads are retrieved according to the cross-correlation peak positions for recovery of different data.

[0072] Among them, as attached Figure 4 As shown, in step (2) of the present invention, after placing the sequencing read segments obtained by sliding correlation identification at the correct position, a consensus base sequence is obtained by using a majority voting decision algorithm. The specific steps are:

[0073] (2.1) If the length of the data DNA sequence is l bases, construct a 5*l counting buffer corresponding to the number of the five bases {A, T, G, C, N} at each position;

[0074] (2.2) The starting positions of the sequencing reads obtained by sliding correlation recognition are different. The identified sequencing reads are placed at the corresponding starting positions according to the cross-correlation peak positions, and the counting buffers at the corresponding positions are added and counted according to the base types at each position of the reads to obtain a counting buffer containing the number of five bases at each position;

[0075] (2.3) Perform a base-by-base majority vote on the counting buffer and determine the base with the largest number in the buffer at the i-th position as the base at that position. If there is no sequencing data at that position or a majority vote cannot be performed, it is marked as "X", which means an erasure error, and a consensus sequence s with a length of l bases is obtained.

[0076] Among them, as attached Figure 5 As shown, step (3) of the present invention converts the consistency sequence into a damaged codeword sequence and then sends it to the decoder for error correction decoding to recover the data. The specific steps are:

[0077] (3.1) Demap the consensus sequence s of length l bases into a double-layer bit sequence according to the "base-bit" mapping rule {A→(00), T→(01), G→(10), C→(11)}, and mark the base X as erasure;

[0078] (3.2) Perform bit-by-bit XOR operation on the fuzzy watermark sequence superimposed with the damaged error correction codeword sequence and the known watermark sequence, remove the superimposed watermark sequence, and obtain the damaged sparse error correction codeword sequence;

[0079] (3.3) Sparsely decode the damaged sparse error correction codeword sequence to obtain a damaged error correction codeword sequence, and erase the error position for marking;

[0080] (3.4) The damaged error correction codeword sequence is sent to the corresponding decoder, and the original data is restored after error correction and erasure decoding.

[0081] In step (1.3) of the present invention, a sliding window of length n is used to calculate the double-layer bit sequence r 1 and r 2 w 1 and w 2 The cross-correlation values ​​of each position are compared to obtain the cross-correlation peak value V peak =max(V 0 ,V 1 ,…V L-n ) and the corresponding position, there are two different sliding correlation calculation cross-correlation peak schemes, namely:

[0082] (1) After demapping, the sequencing reads are selected one by one from the sequencing data containing a large number of reads, and then bit-by-bit sliding correlation is performed with a known double-layer watermark sequence of length L bits to sequentially obtain the cross-correlation value corresponding to each position;

[0083] (2) Sequencing data containing a large number of sequencing read segments with a length of n bases are split to obtain multiple sequencing data sub-files; then a known double-layer watermark sequence with a length of L bits is segmented, and n bits of overlapping segments are maintained between adjacent short segments; multiple cross-correlation values ​​are calculated in parallel using a multi-threaded method at the same time, thereby accelerating the sliding correlation calculation process.

[0084] Among them, as attached Figure 6 As shown, the present invention splits sequencing data containing a large number of sequencing read segments with a length of n bases to obtain multiple sequencing data subfiles; then segments a known double-layer watermark sequence with a length of L bits, and keeps n-bit overlapping segments between adjacent short segments; uses multi-threading mode to parallelly calculate sliding correlation to obtain multiple cross-correlation values ​​in the same time, thereby accelerating the sliding correlation calculation process. The specific steps are:

[0085] (4.1) Split the second-generation high-throughput sequencing data containing m reads of length n bases into q sub-files, each containing m / q sequencing reads;

[0086] (4.2) The known double-layer watermark sequence w={w 1 ,w 2} to obtain k short watermark segments {w (0) ,w (1) ,…w (k-1)}, n bits of overlap are maintained between adjacent short segments;

[0087] (4.3) Start q*k threads, which are used for sliding correlation between the sequenced reads in each sub-file and each short watermark fragment, and calculate q*k cross-correlation values ​​in parallel at the same time.

[0088] Among them, step (1.4) of the present invention uses the cross-correlation peak threshold to screen the sequencing reads that meet the requirements, and it is necessary to optimize and determine the cross-correlation peak threshold based on whether the sequencing data contains non-coding DNA sequencing reads:

[0089] (1) For application scenarios that do not contain non-coding DNA sequencing reads, sequencing read positioning is achieved directly based on the cross-correlation peak position;

[0090] (2) For application scenarios containing non-coding DNA sequencing reads, including the second-generation high-throughput sequencing data of the host cell genome and the second-generation high-throughput sequencing data of other coding DNA fragments, different cross-correlation peak thresholds are first set to identify and screen the sequencing reads; then, the identified sequencing reads are merged in large numbers to obtain the corresponding consensus sequence and calculate the base error rate; finally, the threshold corresponding to the minimum base error rate is selected as the optimal cross-correlation peak threshold to reduce the impact of interfering reads. Specific embodiments

[0092] The present invention designs a fast readout method for storing large-fragment DNA data with watermark superposition coding, performs second-generation high-throughput sequencing on large-fragment DNA sequences with watermark superposition coding, and then uses the fuzzy watermark sequence hidden in the noise sequencing read segment to calculate the cross-correlation peak value to achieve read segment alignment, and can use sequence consensus algorithm and efficient error correction and erasure decoding to restore stored data without error, effectively avoiding the high-complexity de novo assembly, and can exclude the interference of non-coding DNA sequencing read segments, and can achieve error-free recovery of data stored in DNA sequences at lower sequencing coverage, reducing data readout costs. Two specific embodiments are given below to illustrate the feasibility of the present invention. Combined with the given embodiments, the purpose, features and advantages of the invention can be further understood.

[0093] Example 1

[0094] The first specific embodiment of the present invention is given below. This embodiment is aimed at the application scenario that does not contain other non-coding DNA sequencing reads. Based on the 254,886-base yeast artificial chromosome with a sparse low-density parity check (LDPC) code and a watermark sequence superposition coding scheme proposed in the previous work, the second-generation high-throughput sequencing technology is used for data reading. The nucleic acid is directly extracted and the library is constructed from the yeast cells, and finally the second-generation high-throughput sequencing data with a length of 150 bases is obtained by interruption and second-generation high-throughput sequencing, which includes the second-generation high-throughput sequencing data of the artificial chromosome and the second-generation high-throughput sequencing data of the host cell genome.

[0095] Among them, two different encoding methods are used to construct the corresponding information blocks in the yeast artificial chromosome with a length of 254,886 bases, including:

[0096] (1) A non-binary LDPC (64,512,2,256) code with a code rate of 1 / 2 is used to encode the data. After the sparse operation and the watermark sequence are superimposed and transcoded, an information sub-block with a length of 40,320 bases is obtained, which is used to store the 4KB image 2 file;

[0097] (2) The data is encoded using a non-binary LDPC (64,800,54,000) code with a code rate of 5 / 6. After the sparse operation and watermark sequence superposition transcoding, five information sub-blocks with a length of 40,500 bases are obtained, of which one information sub-block is used to store a 6.5KB image 1 file, and the remaining four information sub-blocks are used to store a 25.5KB video file;

[0098] (3) The data is encoded using a non-binary LDPC (64,800,16,200) code rate of 2 / 3. After sparse operation and watermark sequence superposition transcoding, an information sub-block with a length of 40,500 bases is obtained, which is used to store the 5.4KB image 3 file.

[0099] In order to meet the application scenarios that do not contain other non-coding DNA sequencing reads, different sequencing data are separated by preprocessing the sequencing data. The specific steps are to use the sequence alignment software BWA to align the sequencing reads with the known host cell genome sequence, and screen out the second-generation high-throughput sequencing reads belonging to the host cell genome in the alignment results to obtain the second-generation high-throughput sequencing data containing only watermark-superimposed large-fragment DNA encoding.

[0100] The sequencing reads obtained by interrupting the large DNA fragments encoded by watermark superposition and second-generation high-throughput sequencing are demapped into double-layer bit sequences according to the "base-bit" mapping rule, and the bit sequences are slidingly correlated with the locally known double-layer watermark sequences, and the peak of the cross-correlation value is calculated to determine the source of the sequencing reads and their starting positions on the large DNA fragment sequence. The specific steps are as follows:

[0101] (1.1) The large DNA fragments encoded with watermarks are fragmented and sequenced by second-generation high-throughput sequencing to obtain a 150-base sequencing read r according to the "base-bit" mapping rule {A→(00), T→(01), G→(10), C→(11)} to demap the 150-bit double-layer bit sequence r 1 and r 2 ;

[0102] (1.2) The double-layer bit sequence r 1 and r 2 and a known double-layer watermark sequence w with a length of 254,886 bits 1 and w 2 The "0" in is converted to "-1", and the bit "1" remains unchanged;

[0103] (1.3) Use a sliding window of length 150 bases to calculate the double-layer bit sequence r 1 and r 2 w 1 and w2 The cross-correlation values ​​of the calculated positions are compared to obtain the cross-correlation peak and the corresponding position;

[0104] (1.4) The sequencing reads are retrieved according to the cross-correlation peak positions for recovery of different data.

[0105] Among them, a sliding window with a length of 150 bases is used to calculate the double-layer bit sequence r 1 and r 2 w 1 and w 2 The cross-correlation values ​​of the calculated positions are compared to obtain the cross-correlation peak and the corresponding position. There are two different sliding correlation schemes for calculating the cross-correlation peak, which are:

[0106] (1) After demapping, sequencing reads are selected one by one from sequencing data containing a large number of 150-base reads, and then bit-by-bit sliding correlation is performed with a known double-layer watermark sequence of 254,886 bits in length to sequentially obtain the cross-correlation value corresponding to each position;

[0107] (2) Sequencing data containing a large number of 150-base read segments were split to obtain multiple sequencing data sub-files. Then, a known double-layer watermark sequence with a length of 254,886 bits was segmented, and an overlap of 150 bits was maintained between adjacent short segments. Multi-threaded sliding correlation was used to calculate multiple cross-correlation values ​​in the same time, thereby accelerating the sliding correlation calculation process.

[0108] The sequencing data containing a large number of 150-base read segments were split to obtain multiple sequencing data subfiles; the known double-layer watermark sequence with a length of 254,886 bits was segmented, and 150 bits of overlap was maintained between adjacent short segments; multiple cross-correlation values ​​were calculated in parallel using multi-threaded sliding correlation at the same time. The specific steps to accelerate the sliding correlation calculation process are as follows:

[0109] (2.1) Split the second-generation high-throughput sequencing data containing m reads of 150 bases in length into q sub-files, each containing m / q sequencing reads;

[0110] (2.2) The known double-layer watermark sequence w = {w 1 ,w 2} to obtain k short watermark segments {w (0) ,w (1) ,…w (k-1)}, a 150-bit overlap is maintained between adjacent short segments;

[0111] (2.3) Start q*k threads, which are used for sliding correlation between the sequenced reads in each sub-file and each short watermark fragment, and calculate q*k cross-correlation values ​​in parallel at the same time.

[0112] After placing the sequencing reads obtained by sliding correlation recognition at the correct position, the specific steps of using the majority voting algorithm to obtain the consensus base sequence are as follows:

[0113] (3.1) If the length of the data DNA sequence is l bases (l = 40,320 for information sub-blocks with a code rate of 1 / 2, l = 40,500 for information sub-blocks with a code rate of 5 / 6), a 5*l counting buffer is constructed corresponding to the number of the five bases {A, T, G, C, N} at each position;

[0114] (3.2) The starting positions of the sequencing reads obtained by sliding correlation recognition are different. The identified sequencing reads are placed at the corresponding starting positions according to the cross-correlation peak positions, and the counting buffers at the corresponding positions are added and counted according to the base types at each position of the reads to obtain a counting buffer containing the number of five bases at each position;

[0115] (3.3) Perform a base-by-base majority vote on the counting buffer and determine the base with the largest number in the buffer at the i-th position as the base at that position. If there is no sequencing data at that position or a majority vote cannot be performed, it is marked as "X", which means an erasure error, and a consensus sequence s with a length of l bases is obtained.

[0116] The specific steps of converting the obtained consistency sequence s with a length of l bases into a damaged LDPC codeword sequence and then sending it to the corresponding decoder for error correction decoding and data recovery are as follows:

[0117] (4.1) Demap the consensus sequence s of length l bases into a double-layer bit sequence according to the "base-bit" mapping rule {A→(00), T→(01), G→(10), C→(11)}, and mark the base X as erasure;

[0118] (4.2) Perform bit-by-bit XOR operation on the fuzzy watermark sequence superimposed with the damaged error correction codeword sequence and the known watermark sequence, remove the superimposed watermark sequence, and obtain the damaged sparse error correction codeword sequence;

[0119] (4.3) Sparsely decode the damaged sparse error correction codeword sequence to obtain a damaged error correction codeword sequence, and erase the error position for marking;

[0120] (4.4) The damaged error correction codeword sequence is sent to the corresponding decoder, and the original data is restored after error correction and erasure decoding.

[0121] First, this example conducts multiple independent repeated experiments based on the proposed data error correction and readout scheme at different sequencing coverages. The experimental data parameters are shown in Table 1 to explore the reliability of the proposed scheme and the minimum sequencing coverage required for error-free data recovery. Figure 7 As shown in the figure, error-free recovery of different data stored in artificial chromosomes can be achieved at 1.1×-5× sequencing coverage, and the required sequencing coverage is smaller than the current large-fragment DNA data recovery scheme based on read segment assembly. Secondly, in order to explore the specific impact of sequencing coverage on the error distribution of the majority voting decision process, Figure 8 By statistically analyzing the corresponding base error rate distribution at different sequencing coverages, it can be found that the base error rate gradually decreases with the increase of sequencing coverage, and mainly includes base erasure errors and base substitution errors. This embodiment effectively avoids complex de novo assembly operations through sliding correlation, and with the help of a simple sequence consensus algorithm and a reliable error correction coding algorithm, it can achieve error-free data recovery under low sequencing coverage in application scenarios that do not contain interference from other non-coding large fragment DNA sequencing data. In order to prove the stability of the proposed scheme, multiple parallel data recovery experiments were performed using the sequencing data samples in Table 1, and the results were consistent, as shown in Table 2.

[0122] Then, this embodiment analyzes and verifies the two sliding correlation methods proposed in the solution of the present invention. For the test of the parallel sliding correlation solution, 60 (6×10) threads are selected, the sequencing data is divided into 6 sub-files, and the reference watermark sequence is divided into 10 segments, and 150 bits of overlap are maintained between adjacent segments. The two schemes are used to perform sliding correlation operations on different numbers of sequencing reads. Fig. 9 The running time comparison of two sliding correlation schemes under different numbers of sequencing reads and the speed improvement effect of the parallel scheme are given. Fig. 9 It can be seen that in the application scenario that does not include other non-coding DNA sequencing reads, the number of reads corresponding to the effective sequencing coverage is small, and the parallel acceleration solution can improve the read alignment process by about 20 times under 60 threads. The effectiveness of the different sliding correlation schemes proposed in the scheme of the present invention is verified.

[0123] Example 2

[0124] The second specific embodiment of the present invention is given below. This embodiment is aimed at application scenarios including other non-coding large-fragment DNA sequencing data. Based on the 254,886-base yeast artificial chromosome with a sparse low-density parity check (LDPC) code and watermark sequence superposition coding scheme proposed in the previous work, the second-generation high-throughput sequencing technology is used for data reading. Nucleic acid is directly extracted and library is constructed for yeast cells, and finally, second-generation high-throughput sequencing data with a length of 150 bases is obtained by fragmentation and second-generation high-throughput sequencing, which includes the second-generation high-throughput sequencing data of the artificial chromosome and the second-generation high-throughput sequencing data of the host cell genome.

[0125] Among them, two different encoding methods are used to construct the corresponding information blocks in the yeast artificial chromosome with a length of 254,886 bases, including:

[0126] (1) A non-binary LDPC (64,512,32,256) code with a code rate of 1 / 2 is used to encode the data. After the sparse operation and the watermark sequence are superimposed and transcoded, an information sub-block with a length of 40,320 bases is obtained, which is used to store the 4KB image 2 file;

[0127] (2) The data is encoded using a non-binary LDPC (64,800,54,000) code with a code rate of 5 / 6. After the sparse operation and watermark sequence superposition transcoding, five information sub-blocks with a length of 40,500 bases are obtained, of which one information sub-block is used to store a 6.5KB image 1 file, and the remaining four information sub-blocks are used to store a 25.5KB video file;

[0128] The sequencing reads obtained by interrupting the large DNA fragments encoded by watermark superposition and second-generation high-throughput sequencing are demapped into double-layer bit sequences according to the "base-bit" mapping rule, and the bit sequences are slidingly correlated with the locally known double-layer watermark sequences, and the peak of the cross-correlation value is calculated to determine the source of the sequencing reads and their starting positions on the large DNA fragment sequence. The specific steps are as follows:

[0129] (1.1) The large DNA fragments encoded with watermarks are fragmented and sequenced by second-generation high-throughput sequencing to obtain a 150-base sequencing read r according to the "base-bit" mapping rule {A→(00), T→(01), G→(10), C→(11)} to demap the 150-bit double-layer bit sequence r 1 and r 2 ;

[0130] (1.2) The double-layer bit sequence r 1 and r 2 and a known double-layer watermark sequence w with a length of 254,886 bits 1 and w2 The "0" in is converted to "-1", and the bit "1" remains unchanged;

[0131] (1.3) Use a sliding window of length 150 bases to calculate the double-layer bit sequence r 1 and r 2 w 1 and w 2 The cross-correlation values ​​of the calculated positions are compared to obtain the cross-correlation peak and the corresponding position;

[0132] (1.4) Using the cross-correlation peak threshold to screen the sequencing reads that meet the requirements, and record the peak position, that is, the starting position of the sequencing read on the large DNA sequence;

[0133] (1.5) The sequencing reads are retrieved according to the cross-correlation peak positions for recovery of different data.

[0134] Among them, a sliding window with a length of 150 bases is used to calculate the double-layer bit sequence r 1 and r 2 w 1 and w 2 The cross-correlation values ​​of the calculated positions are compared to obtain the cross-correlation peak and the corresponding position. There are two different sliding correlation schemes for calculating the cross-correlation peak, which are:

[0135] (1) After demapping, sequencing reads are selected one by one from sequencing data containing a large number of 150-base reads, and then bit-by-bit sliding correlation is performed with a known double-layer watermark sequence of 254,886 bits in length to sequentially obtain the cross-correlation value corresponding to each position;

[0136] (2) Sequencing data containing a large number of 150-base read segments were split to obtain multiple sequencing data sub-files. Then, a known double-layer watermark sequence with a length of 254,886 bits was segmented, and an overlap of 150 bits was maintained between adjacent short segments. Multi-threaded sliding correlation was used to calculate multiple cross-correlation values ​​in the same time, thereby accelerating the sliding correlation calculation process.

[0137] The sequencing data containing a large number of 150-base read segments were split to obtain multiple sequencing data subfiles; the known double-layer watermark sequence with a length of 254,886 bits was segmented, and 150 bits of overlap was maintained between adjacent short segments; multiple cross-correlation values ​​were calculated in parallel using multi-threaded sliding correlation at the same time. The specific steps to accelerate the sliding correlation calculation process are as follows:

[0138] (2.1) Split the second-generation high-throughput sequencing data containing m reads of 150 bases in length into q sub-files, each containing m / q sequencing reads;

[0139] (2.2) The known double-layer watermark sequence w = {w 1 ,w 2} to obtain k short watermark segments {w (0) ,w (1) ,…w (k-1)}, a 150-bit overlap is maintained between adjacent short segments;

[0140] (2.3) Start q*k threads, which are used for sliding correlation between the sequenced reads in each sub-file and each short watermark fragment, and calculate q*k cross-correlation values ​​in parallel at the same time.

[0141] The sequencing data obtained in this embodiment contains interference from other non-coding large-fragment DNA sequencing data, so it is necessary to optimize and determine the cross-correlation peak threshold to screen the sequencing reads that meet the requirements. The specific steps for determining the optimal cross-correlation peak are:

[0142] (3.1) Selecting values ​​in the set {66, 72, 78, …, 150} one by one as the cross-correlation peak threshold, using different thresholds to identify and screen the sequencing reads, and obtaining the sequencing reads whose cross-correlation peaks meet the threshold requirements;

[0143] (3.2) The sequence reads obtained under different sequencing coverages are merged to obtain the corresponding consensus sequence, and the base error rate in the sequence is calculated;

[0144] (3.3) Select the threshold with the smallest base error rate as the optimal cross-correlation peak threshold.

[0145] The sliding correlation calculation is performed using the second-generation high-throughput sequencing reads and the known watermark sequence to obtain the cross-correlation peak and position. After further identification and screening of the cross-correlation peak threshold, the interference of the second-generation high-throughput sequencing data of the host cell genome is reduced to obtain relatively pure artificial chromosome sequencing data. Then, the consensus sequence is obtained by majority voting based on the peak position. The specific steps are as follows:

[0146] (4.1) If the length of the data DNA sequence is l bases (the code rate is 1 / 2 information block, l = 40,320bp, the code rate is 5 / 6 information block, l = 40,500bp), construct a 5*l counting buffer corresponding to the number of the five bases {A, T, G, C, N} at each position;

[0147] (4.2) The starting positions of the sequencing reads obtained by sliding correlation recognition are different. The identified sequencing reads are placed at the corresponding starting positions according to the cross-correlation peak positions, and the counting buffers at the corresponding positions are added and counted according to the base types at each position of the reads to obtain a counting buffer containing the number of five bases at each position;

[0148] (4.3) Perform a base-by-base majority vote on the counting buffer and determine the base with the largest number in the buffer at the i-th position as the base at that position. If there is no sequencing data at that position or a majority vote cannot be performed, it is marked as "X", which means an erasure error, and a consensus sequence s with a length of l bases is obtained.

[0149] The specific steps of converting the obtained consistency sequence s with a length of l bases into a damaged LDPC codeword sequence and then sending it to the corresponding decoder for error correction decoding and data recovery are as follows:

[0150] (5.1) Demap the consensus sequence s of length l bases into a double-layer bit sequence according to the "base-bit" mapping rule {A→(00), T→(01), G→(10), C→(11)}, and mark the base X as erasure;

[0151] (5.2) Perform bit-by-bit XOR operation on the fuzzy watermark sequence superimposed with the damaged error correction codeword sequence and the known watermark sequence, remove the superimposed watermark sequence, and obtain the damaged sparse error correction codeword sequence;

[0152] (5.3) Sparsely decode the damaged sparse error correction codeword sequence to obtain a damaged error correction codeword sequence, and erase the error position for marking;

[0153] (5.4) The damaged error correction codeword sequence is sent to the corresponding decoder, and the original data is restored after error correction and erasure decoding.

[0154] The sequencing data obtained in this embodiment includes the second-generation high-throughput sequencing data of the host cell genome, such as Fig.10 As shown in the figure, the proportion of artificial chromosome sequencing data is only 3%. In the absence of a host genome, all noise reads come from artificial chromosomes carrying data. In order to determine the optimal cross-correlation peak recognition threshold, the values ​​in the set {66, 72, 78, …, 150} are selected as the cross-correlation peak threshold, which are used to identify and screen the host genome sequencing data reads, and the large number of screening sequencing reads are merged under different sequencing coverage, and the average error rate of bases in the corresponding consensus sequence is counted. According to the error rate distribution under different thresholds, the six information sub-blocks are analyzed, and finally the optimal cross-correlation peak thresholds are determined to be 108, 108, 102, 100, 102, and 102, respectively.

[0155] First, based on the proposed data retrieval and readout scheme, multiple independent repeated experiments were conducted at different sequencing coverages to explore the reliability of the proposed scheme in different application scenarios and the minimum sequencing coverage required for error-free data recovery. Fig.11 As shown in the figure, error-free recovery of different data stored in artificial chromosomes can be achieved at 1.2×-8.4× sequencing coverage, and the required sequencing coverage is less than the current large-fragment DNA data recovery scheme based on read segment assembly. This further verifies that the scheme of the present invention avoids complex de novo assembly through sliding correlation, sequence consensus algorithm and reliable error correction and erasure decoding, and achieves accurate data recovery at low coverage, thereby reducing the complexity of biochemical operations and the workload of data processing. In order to prove the stability of the proposed scheme, multiple parallel data recovery experiments were carried out using different sequencing data samples, and the results were consistent, as shown in Table 3.

[0156] Secondly, this embodiment analyzes and tests the parallel sliding correlation method proposed in the solution of the present invention. The number of threads selected in the test of the parallel sliding correlation solution is 60 (6×10), the sequencing data is divided into 6 sub-files, and the reference watermark sequence is divided into 10 segments, and 150 bits of overlap are maintained between adjacent segments. Two schemes are used to perform sliding correlation processing on different numbers of sequencing reads. Fig.12 The running time comparison of two sliding correlation schemes under different numbers of sequencing reads and the speed improvement effect of the parallel scheme are given. Fig.12 It can be seen that in the application scenario containing other non-coding DNA fragment sequencing data, the same effective sequencing coverage corresponds to a large number of second-generation high-throughput sequencing reads, and the scale of the read alignment operation based on sliding correlation is large. The use of parallel acceleration can increase the speed of large-scale read alignment based on sliding correlation by about 45 times under 60 threads.

[0157] Fig.13 The base error distribution diagram of the majority voting decision process in Example 2 is given. Since the interfering data cannot be completely eliminated, the base error rate increases at the same sequencing coverage. Fig.14The distribution of base coverage under the minimum sequencing coverage for error-free data recovery in Example 2 was explored. It was found that when the minimum sequencing coverage was 1.1×, the original data could be successfully recovered without the host genome. Compared with the current large-fragment DNA data recovery scheme based on de novo assembly of reads, the advantage of the proposed data recovery method is that the known watermark sequence at the read-out end is similar to the reference genome sequence in genome resequencing, which supports direct and accurate positioning and identification of sequencing reads, and effectively avoids de novo assembly of reads; further, during the sequence reconstruction process, this method can efficiently connect short sequencing reads at different starting positions through a simple majority voting decision algorithm, avoiding complex graph construction, optimization and other operations, and eliminating the need to obtain long continuous fragments; with the help of a reliable error correction coding scheme, error-free data recovery at low sequencing coverage can be achieved, such as Fig.15 shown.

[0158] Finally, Table 4 compares other schemes using large DNA fragments for data storage. It can be seen that compared with other schemes for large DNA fragment reading, the recovery method of the scheme of the present invention has lower requirements for the coverage of error-free data recovery, which is only 1.1 times or 5 times, thereby effectively reducing the reading cost.

[0159] Table 1

[0160]

[0161] Table 2

[0162]

[0163] Table 3

[0164]

[0165]

[0166] Table 4

[0167]

[0168] Those skilled in the art will appreciate that the accompanying drawing is only a schematic diagram of a preferred embodiment, and the serial numbers of the embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0169] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for rapid reading of large-fragment DNA data stored by watermark superposition coding, characterized in that: The method comprises the following steps: (1) The large DNA fragments encoded by watermark superposition are interrupted and the sequencing reads obtained by second-generation high-throughput sequencing are demapped into double-layer bit sequences according to the "base-bit" mapping rule, and the bit sequences are slidingly correlated with the locally known double-layer watermark sequence to calculate the cross-correlation peak, thereby determining the source of the sequencing reads and their starting positions on the large DNA fragment sequence; (2) After placing the sequencing reads obtained by sliding correlation recognition at the correct position, a majority voting decision algorithm is used to obtain the consensus base sequence; (3) The consistent base sequence is converted into a damaged bit sequence and then sent to a decoder for error correction and decoding to recover the data.

2. According to claim 1, a method for rapid reading of watermark superposition coded large-fragment DNA data storage, characterized in that: The step (1) is: (1.1) The large DNA fragments encoded with watermarks are fragmented and sequenced by second-generation high-throughput sequencing to obtain a sequence read segment r with a length of n base pairs. According to the "base-bit" mapping rule {A→(00), T→(01), G→(10), C→(11)}, the double-layer bit sequences r1 and r2 with a length of n bits are demapped; (1.2) Convert the bit "0" in the double-layer bit sequences r1 and r2 and the known double-layer watermark sequences w1 and w2 with a total length of L bits to "-1", and keep the bit "1" unchanged; (1.3) Use a sliding window of length n to calculate the cross-correlation values ​​between the two-layer bit sequences r1 and r2 and the known two-layer watermark sequences w1 and w2, where the cross-correlation value V at the i-th position is i The calculation formula is expressed as: Get L-n+1 cross-correlation values; then compare the calculated cross-correlation values ​​of each position to obtain the cross-correlation peak value V peak =max(V0,V1,…V L-n ) and the corresponding positions; (1.4) Using the cross-correlation peak threshold to screen the sequencing reads that meet the requirements, and record the peak position, that is, the starting position of the sequencing read on the large DNA sequence; (1.5) The sequencing reads are retrieved according to the cross-correlation peak positions for recovery of different data.

3. According to claim 1, a method for fast reading of watermark superposition coded large-fragment DNA data storage is characterized in that: The step (2) is: (2.1) If the length of the data DNA sequence is l bases, construct a 5*l counting buffer corresponding to the number of the five bases {A, T, G, C, N} at each position; (2.2) The starting positions of the sequencing reads obtained by sliding correlation recognition are different. The identified sequencing reads are placed at the corresponding starting positions according to the cross-correlation peak positions, and the counting buffers at the corresponding positions are added and counted according to the base types at each position of the reads to obtain a counting buffer containing the number of five bases at each position; (2.3) Perform a base-by-base majority vote on the counting buffer and determine the base with the largest number in the buffer at the i-th position as the base at that position. If there is no sequencing data at that position or a majority vote cannot be performed, it is marked as "X", which means an erasure error, and a consensus sequence s with a length of l bases is obtained.

4. According to claim 1, a method for rapid reading of watermark superposition coded large-fragment DNA data storage is characterized in that: The step (3) is: (3.1) Demap the consensus sequence s of length l bases into a double-layer bit sequence according to the "base-bit" mapping rule {A→(00), T→(01), G→(10), C→(11)}, and mark the base X as erasure; (3.2) Perform bit-by-bit XOR operation on the fuzzy watermark sequence superimposed with the damaged error correction codeword sequence and the known watermark sequence, remove the superimposed watermark sequence, and obtain the damaged sparse error correction codeword sequence; (3.3) Sparsely decode the damaged sparse error correction codeword sequence to obtain a damaged error correction codeword sequence, and erase the error position for marking; (3.4) The damaged error correction codeword sequence is sent to the corresponding decoder, and the original data is restored after error correction and erasure decoding.

5. According to claim 2, a method for fast reading of watermark superimposed coded large-fragment DNA data storage, characterized in that: The sliding window with a length of n is used to calculate the mutual correlation values ​​between the double-layer bit sequences r1 and r2 and the known double-layer watermark sequences w1 and w2, and obtain L-n+1 mutual correlation values; then the calculated mutual correlation values ​​of each position are compared to obtain the mutual correlation peak value V peak =max(V0,V1,…V L-n ) and the corresponding position, there are two different sliding correlation calculation cross-correlation peak schemes, namely: (1) After demapping, the sequencing reads are selected one by one from the sequencing data containing a large number of reads, and then bit-by-bit sliding correlation is performed with a known double-layer watermark sequence of length L bits to sequentially obtain the cross-correlation value corresponding to each position; (2) Sequencing data containing a large number of sequencing read segments with a length of n bases are split to obtain multiple sequencing data sub-files; then a known double-layer watermark sequence with a length of L bits is segmented, and n bits of overlapping segments are maintained between adjacent short segments; multiple cross-correlation values ​​are calculated in parallel using a multi-threaded method at the same time, thereby accelerating the sliding correlation calculation process.

6. A method for rapid reading of watermark superimposed coded large-fragment DNA data storage according to claim 2, characterized in that: The cross-correlation peak threshold is used to screen the sequencing reads that meet the requirements. It is necessary to determine and optimize the cross-correlation peak threshold according to whether the sequencing data contains non-coding DNA sequencing reads: (1) For application scenarios that do not contain non-coding DNA sequencing reads, sequencing read positioning is achieved directly based on the cross-correlation peak position; (2) For application scenarios containing non-coding DNA sequencing reads, including the second-generation high-throughput sequencing data of the host cell genome and the second-generation high-throughput sequencing data of other coding DNA fragments, different cross-correlation peak thresholds are first set to identify and screen the sequencing reads; then, the identified sequencing reads are merged in large numbers to obtain the corresponding consensus sequence and calculate the base error rate; finally, the threshold corresponding to the minimum base error rate is selected as the optimal cross-correlation peak threshold to reduce interfering reads.

Images