Plasmid de novo assembly method based on nanopore sequencing

Abstract

The application provides a plasmid sequence de novo assembly method based on nanopore sequencing, and steps include: S1. Preprocessing of nanopore original sequencing reads; S2. Estimating the assembly plasmid size and average sequencing depth based on the read set; S3. Setting multiple target coverage degrees, sampling the read set respectively to obtain read sub-sample sets with different sequencing depths; S4. De novo assembling each sequencing depth sub-sample with multiple orthogonal algorithms respectively to obtain a candidate assembly set; S5. Clustering the candidate assembly set to obtain an optimal assembly; and S6. Polishing and finalizing the optimal assembly. The de novo assembly method uses sequencing depth gradient and multiple assemblers in double, and uses the assembly result repeatability to distinguish the correct signal, avoids the systematic error of fixed depth and single algorithm, so that the correct assembly can be generated with high probability, and the correct assembly can be identified from the results containing the correct assembly and numerous error assemblies.

Description

Technical Field

[0001] This invention relates to the field of genome assemblies and bioinformatics technology, specifically to a de novo method for assembling plasmid sequences based on nanopore sequencing. Background Technology

[0002] Plasmids are circular DNA molecules independent of chromosomes in bacteria and other microorganisms. In synthetic biology, genetic engineering, strain identification, and molecular biology experiments, obtaining their complete, accurate, and closed (circular) full-length sequences is often required. Single reads from Oxford Nanopore Technology can span the entire plasmid, theoretically suitable for full-length de novo plasmid assembly; however, its original reads have a high error rate and suffer from chimeric reads and uneven read length distribution, significantly hindering de novo assembly.

[0003] Currently, nanopore plasmid assembly typically employs a pipeline of "single sequencing depth + single assembler + polishing and error correction," which has the following drawbacks: (i) Assembly results are highly sensitive to the input sequencing depth: The same assembly software often gives different assembly lengths, different circularization judgments and different fragment numbers under different coverage. If the depth is too low, there will be insufficient information; if the depth is too high, noise and chimeric reads will accumulate. Moreover, there is no fixed depth that is optimal for all samples. (ii) Systematic biases exist in single assembly algorithms: Assembly software based on Repeat Graph / OLOC (such as Flye) is good at parsing repeats, but may incorrectly circularize small plasmids or incorrectly assemble monomers into two-body / multi-body concatenations. On the other hand, assembly software based on OLOC without error correction (such as miniasm) is fast but tends to produce short and fragmented results. Relying on either one alone will not allow the systematic errors inherent in each algorithm to be detected and corrected. (iii) Without a reference sequence, there is a lack of reliable criteria for judging the correct results. Traditional methods rely on manual experience for selection, which is difficult to scale up and has poor reproducibility.

[0004] Existing tools also fail to address the aforementioned issues: Flye, miniasm / minimap2, etc., are single assemblers; medaka, homopolish, etc., only perform consensus polishing and homology error correction, but do not solve the problem of selecting which assembly to use from among many candidates. If the input assembly itself is incorrect, polishing will only result in a refined erroneous sequence; plasmid workflows such as circuitSeq are essentially still single-path assembly under fixed parameters; Trycycler, which is closest to this invention, improves the quality of long-read assembly by clustering and harmonizing multiple assembly results, but it requires a large amount of manual interaction and judgment (manual confirmation of clustering, manual selection of representative sequences), does not perform systematic sequencing depth gradient sampling, and does not perform fully automatic optimization based on contig length cluster consensus, combined with the number of support for contigs and coverage uniformity of multiple assemblers.

[0005] Therefore, it is necessary to propose a reference-free, fully automated de novo plasmid sequence assembly method to address the shortcomings of existing technologies. Summary of the Invention

[0006] This invention proposes a de novo plasmid sequence assembly method based on nanopore sequencing, in order to at least solve one of the many problems existing in the prior art.

[0007] The technical solution of this invention is implemented as follows:

[0008] A de novo plasmid sequence assembly method based on nanopore sequencing, comprising the following steps: S1. Preprocess the raw nanopore sequencing reads to obtain a high-quality read set; S2. Estimating the assembled plasmid size and average sequencing depth based on read sets; S3. Based on the average sequencing depth and total number of reads, set multiple target coverage levels, sample the read sets separately, and obtain read subsample sets with different sequencing depths; S4. For each sequencing depth subsample, multiple orthogonal algorithms are used to assemble them from scratch to obtain a candidate assembly set; S5. Filter and cluster the candidate assemblies to obtain the optimal assembly, including:

[0009] Perform pre-screening on the candidate assembly set, and retain circular candidate contigs with a number below the threshold to enter the candidate pool;

[0010] Coarse clustering is performed on the length of the superimposed clusters in the candidate pool to obtain multiple clusters. Clusters with a number of members reaching the threshold and not all of them come from fragmented assembly are selected as qualified clusters.

[0011] Based on the assembly algorithm, the optimal cluster can be determined from qualified clusters by considering the number of species, the number of members, and the coverage uniformity.

[0012] Perform length-based fine clustering on the optimal cluster and select qualified sub-clusters based on the number of members;

[0013] Then, the optimal subcluster is determined based on the assembly algorithm's support for categories, number of members, and coverage uniformity indices.

[0014] The overlapping group with the best coverage uniformity within the optimal sub-cluster is selected as the optimal sample assembly.

[0015] S6. Polish and shape the optimal assembly to obtain a closed plasmid sequence.

[0016] Furthermore, in step S1, the preprocessing includes at least one of removing artificially introduced non-target sequences, removing chimera interference, and filtering low-quality and invalid short fragments.

[0017] Further, in step S2, the initial assembled contiguous group is first obtained by de novo assembly of the read set according to the plasmid pattern, and the plasmid size and average sequencing depth are estimated based on the contiguous group.

[0018] Furthermore, in step S3, the multi-level target coverage includes a sequencing depth between 30 and 500×;

[0019] And / or, the multi-level target coverage is 18 levels;

[0020] And / or, step S4, the de novo assembly algorithm, includes Flye and miniasm.

[0021] Further, in step S5, the qualified cluster screening conditions are: the number of cluster members reaches a preset threshold, the threshold is 3 when the total number of candidates in the candidate pool is greater than 10, the threshold is 2 when the total number of candidates in the candidate pool is not greater than 10, and at least one member in the cluster comes from the assembly result containing only one superimposed group.

[0022] Preferably, in step S5, the optimal cluster is selected by prioritizing the cluster that is supported by the most types of different assembly algorithms; when the supported types are the same, the cluster with the largest number of members is selected; if there is another qualified cluster whose average coverage uniformity index is 0.5 times or less than that of the cluster with the largest number of members, then the other qualified cluster is selected as the optimal cluster.

[0023] Preferably, in step S5, the qualified sub-cluster selection criteria are: the number of sub-cluster members reaches a preset threshold, the threshold is 2 when the total number of members of the optimal cluster is greater than 3, and the threshold is 1 when the total number of members of the optimal cluster is not greater than 3; the rule for determining the optimal sub-cluster is: the sub-cluster with the largest number of members is selected first, and if there is another sub-cluster whose average coverage uniformity index is 0.5 times or less than that of the sub-cluster with the largest number of members, then that other sub-cluster is selected as the optimal sub-cluster; the overlapping cluster with the smallest coverage variation coefficient CV among the optimal sub-clusters is selected as the optimal sample assembly.

[0024] Furthermore, the coverage uniformity index is the coverage variation coefficient, which is the ratio of the standard deviation of the sliding window sequencing depth to the mean of the window sequencing depth.

[0025] Further, in step S6, the polishing and shaping involves sequentially performing consensus polishing, sequence start-point rotation, re-consensus polishing, and homology error correction on the optimal assembly to obtain the final closed plasmid sequence.

[0026] This invention also proposes a de novo plasmid sequence assembly system based on nanopore sequencing, comprising:

[0027] Preprocessing module: preprocesses raw nanopore sequencing reads to obtain a high-quality read set;

[0028] Estimation module: Estimates the assembled plasmid size and average sequencing depth based on the read set;

[0029] Sampling module: Based on the average sequencing depth and total number of reads, multiple target coverage levels are set, and read sets are sampled separately to obtain read subsample sets with different sequencing depths;

[0030] Assembly module: For each sequencing depth subsample, multiple orthogonal algorithms are used to assemble them from scratch to obtain a candidate assembly set;

[0031] Clustering and filtering module: This module filters and clusters candidate assemblies to obtain the optimal assembly, including:

[0032] Perform pre-screening on the candidate assembly set, and retain circular candidate contigs with a number below the threshold to enter the candidate pool;

[0033] Coarse clustering is performed on the length of the superimposed clusters in the candidate pool to obtain multiple clusters. Clusters with a number of members reaching the threshold and not all of them come from fragmented assembly are selected as qualified clusters.

[0034] Based on the assembly algorithm, the optimal cluster can be determined from qualified clusters by considering the number of species, the number of members, and the coverage uniformity.

[0035] Perform length-based fine clustering on the optimal cluster and select qualified sub-clusters based on the number of members;

[0036] Then, the optimal subcluster is determined based on the assembly algorithm's support for categories, number of members, and coverage uniformity indices.

[0037] The overlapping group with the best coverage uniformity within the optimal sub-cluster is selected as the optimal sample assembly.

[0038] Polishing and error correction module: Polishes and shapes the optimal assembly to obtain a closed plasmid sequence.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0040] The de novo assembly method described in this invention uses both sequencing depth gradient and multiple assemblers, and judges the correct signal by the repeatability of the assembly result. This avoids the systematic errors of fixed depth and single algorithm, so that the correct assembly is likely to be generated. Furthermore, it can identify the correct assembly from the results that include both correct assembly and many incorrect assembly (such as incorrect circularization, length drift, fragmentation, etc.).

[0041] The de novo assembly method described in this invention can automatically correct the systematic artifacts of a single assembler: when an assembler outputs an incorrect length for a cascade of two or more bodies for some samples, this method automatically determines the true length of a single body by using cross-assembler support and length clustering.

[0042] The assembly method provided by this invention achieves repeatable results while maintaining both accuracy and robustness: optimal selection is based on objective clustering and CV criteria, ensuring consistent results with the same input; two-level clustering first eliminates outliers with a wide radius and then locks in the precise consensus length with a narrow radius, while CV further guarantees that the selected sequence has uniform coverage. The entire process requires no reference sequence or manual selection, directly outputting the optimal closed plasmid sequence, and can be processed in batches on a large scale. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a schematic flowchart of the de novo plasmid sequence assembly method described in this invention;

[0045] Figure 2 This is a schematic diagram of candidate assembly length clustering and cross-assembler support optimization of samples in Embodiment 1 of the present invention;

[0046] Figure 3 This is a coverage distribution diagram of the final assembled sequence of the sample in Embodiment 1 of the present invention. Detailed Implementation

[0047] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0048] In one embodiment, a de novo plasmid sequence assembly method based on nanopore sequencing is proposed, the process of which is as follows: Figure 1 As shown, the specific steps are as follows:

[0049] 1. Segment preprocessing

[0050] Sequencing adapters are removed from the raw nanopore sequencing reads, and the middle of chimeric reads is cut off. Then, quality and length filtering is performed to obtain the filtered read set R. The specific steps are as follows: 1) Adapter removal: Use Porechop (https: / / github.com / rrwick / Porechop) to remove sequencing adapters from the raw nanopore reads and cut off the middle of the chimeric reads. The parameters are --format fastq --end_threshold50 --extra_end_trim 50 --discard_middle --middle_threshold 80; 2) Quality control filtering: NanoFilt (https: / / github.com / wdecoster / nanofilt) is used for filtering with parameters -q 15 -l 500 (retaining reads with an average quality Q ≥ 15 and a length ≥ 500bp), resulting in a filtered read set R. Quality control statistics can be generated using NanoPlot (https: / / github.com / wdecoster / NanoPlot) and seqkit.

[0051] 2. Foundation Assembly and Parameter Estimation

[0052] Baseline assembly and parameter estimation: A sample of reads R is subjected to de novo plasmid assembly in a plasmid pattern. The resulting contig serves as a basic assembly result used in the initial stage, and the plasmid size G and average sequencing depth D of the sample are estimated based on this result. Specifically, this includes: 1) Use Rasusa (https: / / github.com / mbhall88 / rasusa) to sample reads to a preset number of bases (rasusa reads -s 123 --bases <number of bases>); 2) Use Flye (https: / / github.com / fenderglass / Flye) plasmid mode for base assembly: flye --plasmids --nano-raw <reads>-i 1 --min-overlap 1000; 3) Parse fields such as Total length and Mean coverage in the Flye log to estimate the genome (plasmid) size G and average sequencing depth D of the sample, and write the prediction information (such as JSON).

[0053] 3. Gradient multi-depth sampling

[0054] Based on the average sequencing depth D and the total number of reads, the same set of reads is sampled according to a set of target coverage to obtain a series of read subsamples corresponding to different sequencing depths.

[0055] For each of the 18 target coverage levels (30, 40, 45, 50, 55, 60, 70, 90, 100, 120, 135, 150, 165, 200, 250, 300, 400, 500, unit ×), the target sampling base quantity was calculated: Target base quantity = (Target depth / D) × Total base quantity of the read segment (the upper limit is taken when it exceeds the actual usable base quantity). Filtlong (https: / / github.com / rrwick / Filtlong) was used for quality-weighted sampling: filtlong --mean_q_weight 10 -t <target base quantity> --max_length <1.2×G> <reads>Eighteen depth subsamples were obtained. Different coverage depths are particularly important for plasmid samples of different lengths, so the coverage range needs to be large enough to ensure that the true length can be correctly assembled at certain depths.

[0056] 4. Parallel assembly by multiple assemblers

[0057] For each depth subsample, at least two independent ab initio assembly algorithms are used to assemble it, resulting in a candidate assembly set consisting of numerous candidate contigs.

[0058] In this implementation, two independent assembly algorithms are run for each depth subsample, as follows:

[0059] 1) Flye (https: / / github.com / fenderglass / Flye): flye --plasmids --nano-raw <subsample> -i 1 --min-overlap 1000.

[0060] 2) miniasm (https: / / github.com / lh3 / miniasm): First, use minimap2 (https: / / github.com / lh3 / minimap2) to perform full alignment of read segments: `minimap2 -x ava-ont <subsample> <subsample>.

[0061] 3) Take overlaps.paf and then use miniasm -o 500 -f <subsample> overlaps.paf >overlaps.gfa` to extract the sequence contigs from GFA. This yields approximately 18×2 candidate assemblies for each sample. Each candidate may contain one or more contigs, and its length, whether it is circ., and the total number of contigs in its assembly are recorded.

[0062] 5. Coverage uniformity calculation

[0063] For each candidate contig, the corresponding subsamples are read back and the depth is statistically analyzed according to the sliding window, and the coefficient of variation (CV) of its coverage depth (the ratio of the standard deviation of the window depth to the mean) is calculated.

[0064] For each candidate contig, use minimap2 <contig><subsample> -a | samtools sort(samtools, https: / / github.com / samtools / samtools) compares and sorts the samples, calculates the average depth of each window by a 100bp sliding window, and calculates the mean coverage (cov) and the coefficient of variation (CV) of coverage (CV = standard deviation of window depth / mean of window depth).

[0065] 6. Automatic Optimization of Length-Based Multi-Level Clustering

[0066] This step is the core of the present invention and includes the following steps: (1) Pre-screening: The assembly results obtained by each assembler are pre-screened. They must be circular and contain no more than 3 contigs.

[0067] (2) Coarse clustering: First, coarse clustering is performed on all contig lengths in the candidate pool, with a neighborhood radius eps=500bp. Those with similar lengths are grouped into one cluster. (3) Select qualified clusters: Only clusters with no less than the threshold number of members and at least one member from an assembly containing only one contig are retained as qualified clusters; (4) Select the optimal cluster of coarse clustering: count how many different assembly algorithms support each qualified cluster, and prioritize the cluster with the most supporting algorithms. If they are tied, select the cluster with the largest number of members. However, if there is another cluster whose average CV is significantly higher than the average CV of the cluster, then select the cluster with the lower CV. (5) Fine clustering: Perform fine clustering on the length within the selected optimal cluster and set the neighborhood radius to form subclusters. (6) Screening qualified subclusters: Only subclusters with no less than the threshold number of members (usually 2, but this threshold is reduced to 1 when the cluster used for fine clustering has only 3 or less) are retained as qualified subclusters; (7) Select the optimal sub-cluster of fine clustering: select the sub-cluster with the largest number of members, but if there is another cluster whose average CV is significantly higher than the average CV of the largest cluster, then select the sub-cluster with the lower CV. (8) Take the contig with the smallest CV in the sub-cluster as the optimal assembly of the sample.

[0068] The specific steps are as follows:

[0069] Clustering was performed using DBSCAN from scikit-learn (https: / / scikit-learn.org / , `DBSCAN(eps,min_samples=1)`):

[0070] 1) Pre-screening: Only candidate contigs with circ.=Y, total number of contigs in their assembly ≤3, and coverage ≥30× are retained in the candidate pool.

[0071] 2) Coarse clustering: DBSCAN clustering is performed on all contig lengths in the candidate pool, with a neighborhood radius eps=500. Those with similar lengths are clustered into one cluster. Only clusters with no less than the adaptive threshold (3 when the number of candidates is >10, otherwise 2) and where not every member in the cluster comes from a cluster with multiple contigs are retained as qualified clusters.

[0072] 3) Select the optimal cluster: Count how many different assembly algorithms support each qualified cluster (i.e., the number of supporters that simultaneously contain Flye and miniasm sources is 2); prioritize the cluster with the most supporting algorithms; if there is a tie, select the cluster with the largest number of members as the optimal cluster, but if the average CV of another cluster is less than 0.5 times the average CV of the largest cluster, then select the cluster with the lower CV.

[0073] 4) Fine clustering: Perform DBSCAN fine clustering on the length within the selected optimal cluster, with a neighborhood radius eps=30; apply similar criteria to the selection of the optimal cluster: ①. Member number threshold (the threshold is reduced to 1 when the number of candidate subclusters is ≤3); ② Take the cluster with the largest number of members as the optimal subcluster, but if the average CV of another cluster is lower than 0.5 times the average CV of the largest subcluster, then take the subcluster with the lower CV.

[0074] 5) Select the optimal assembly: Take the contig with the smallest coverage variation coefficient (CV) in the subcluster as the optimal assembly result for the sample, and extract the corresponding contig sequence based on its assembly ID.

[0075] In this invention, the at least two independent assembly algorithms are not limited to Flye and miniasm, but can be any two or more de novo assembly algorithms; the clustering is not limited to DBSCAN, but can be OPTICS, hierarchical clustering, or any density or distance clustering, and can be single-level or multi-level; the coverage uniformity is not limited to the coefficient of variation (CV), but can be any uniformity measure such as depth variance, Gini coefficient, maximum-minimum depth ratio; the sequencing platform is not limited to nanopores, but is also applicable to other long-read sequencing such as PacBio; the object is not limited to plasmids, but can be extended to the de novo assembly and optimization of other small circular DNA molecules such as phage genomes and organelle (mitochondrial / chloroplast) genomes.

[0076] 7. Polishing and Shaping

[0077] 1) One-time polishing: Use medaka (https: / / github.com / nanoporetech / medaka) to perform consensus polishing on the optimal assembly: medaka_consensus -i <filtered reads> -d <optimal assembly> -m r941_min_hac_g507.

[0078] 2) Rotation shaping: Rotate the starting point of the polished sequence to the middle of the sequence, so that the original assembly's beginning and end seams fall into the sequence, which facilitates subsequent repair.

[0079] 3) Secondary polishing: Polish the rotated sequence again with medaka (model as above) to repair the polishing blind spots at the joints.

[0080] 4) Homology correction: Homopolish (https: / / github.com / ythuang0522 / homopolish) is used for homology correction: homopolish polish -a <sequence> -l <reference plasmid database> -m R9.4.pkl, to fix systematic errors such as homopolymers and obtain the final closed plasmid sequence.

[0081] 8. Quality Assessment

[0082] Use quast (https: / / github.com / ablab / quast) to evaluate the final sequence: quast.py <sequence> -r <sequence> --min-contig 0; and plot the original filtered read back ratio and coverage distribution in 100bp windows to confirm coverage uniformity and assembly correctness.

[0083] Example 1

[0084] The total read volume of a certain sample of nanopore sequencing data after filtering was approximately 2,347,856 bp. After sampling the reads, a base was constructed using the Flye plasmid model, estimating an average depth of approximately 83× and a length of approximately 9,865 bp. It is worth noting that this single base assembly yielded an artifact length of two-monomer cascades (9,865 bp ≈ twice the actual monomer length), and directly accepting this result would be erroneous. Sampling was performed on the same read set at 18 target coverage levels (30…500×), and each subsample was assembled using both Flye and miniasm. After pre-screening (circ.=Y, total number of contigs ≤3, cov≥30×), 31 candidate closed contigs were obtained, and their lengths showed a clear three-group distribution, as shown in Table 1.

[0085] Table 1: Grouping Statistics of Candidate Assembly Length Monomer clusters 4,912–4,933 21 Flye 8 + miniasm 13 Largest number of members, cross-assembler support Selected Two-body clusters 9,515–10,027 9 Flye 9 Supported by a single assembler only Eliminate The Three-Body Problem 14,618 1 Flye 1 Isolated, insufficient members Eliminate

[0086] Coarse clustering (DBSCAN, eps=500bp) was performed on the above candidates, and each of the three groups clustered into a separate cluster. The single-unit cluster, supported by both Flye and miniasm assemblers and having the most members (21), was selected as the optimal cluster. The two-unit cluster, although having a considerable number of members (9), was excluded because it was supported only by the Flye assembler. This highlights the crucial role of the cross-assembler support criterion in automatically eliminating the systematic artifacts of a single assembler (two-unit clusters). Figure 2 As shown in Table 2. Then, fine clustering (DBSCAN, eps=30bp) was performed within the single-monomer clusters. The candidate lengths from 4,912 to 4,933bp were all less than 30bp apart, linking them into a sub-cluster with consistent length. The assembly with the smallest CV was selected as the optimal assembly. That is, this method automatically corrected the incorrect binary length (9,865bp) to the true single-monomer length of 4,932bp. The final sequence was closed after medaka polishing, rotation, secondary polishing, and homopolish error correction. Its coverage was uniform along the entire length (approximately 470×), with only a few sites showing regression, as shown in Table 2. Figure 3 As shown, this is consistent with the obtained low CV (0.027).

[0087] Table 2: Final Sample Selection and Assembly Results Selected Assembly ID A3_flye_subsample_17_contig_1 Final length 4,932 bp Is it cyclic? Yes (circ.=Y) Average coverage Approximately 470× Coverage coefficient of variation (CV) 0.027 Estimated length of single assembly for foundation construction 9,865bp (two-body artifact, corrected to 4,932bp for single-body by this method)

[0088] Example 2 (36 sample batches)

[0089] The assembly method of this invention was applied to a sample containing 36 plasmids. All 36 samples automatically generated a single closed circular plasmid sequence (circ.=Y, total number of contigs=1). The length of the selected sequences ranged from 4,894 to 23,988 bp, and the median coefficient of variation (CV) was 0.046 (range 0.020-0.295). Among them, the optimal assembly of 31 samples was ultimately selected from miniasm, and 5 were selected from Flye, demonstrating the practical role of cross-verification across assemblers. The results of representative samples are shown in Table 3.

[0090] Table 3: Representative Assembly Results of 36 Sample Batches (Excerpt)

[0091] 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 within the protection scope of the present invention.< / contig> < / reads> < / reads>

Claims

1. A method for de novo assembly of plasmid sequences based on nanopore sequencing, characterized in that the steps include... include: S1. Preprocess the raw nanopore sequencing reads to obtain a high-quality read set; S2. Estimating the assembled plasmid size and average sequencing depth based on read sets; S3. Based on the average sequencing depth and total number of reads, set multiple target coverage levels, sample the read sets separately, and obtain read subsample sets with different sequencing depths; S4. For each sequencing depth subsample, multiple orthogonal algorithms are used to assemble them from scratch to obtain a candidate assembly set; S5. Filter and cluster the candidate assemblies to obtain the optimal assembly, including: Perform pre-screening on the candidate assembly set, and retain circular candidate contigs with a number below the threshold to enter the candidate pool; Coarse clustering is performed on the length of the superimposed clusters in the candidate pool to obtain multiple clusters. Clusters with a number of members reaching the threshold and not all of them come from fragmented assembly are selected as qualified clusters. Based on the assembly algorithm, the optimal cluster can be determined from qualified clusters by considering the number of species, the number of members, and the coverage uniformity. Perform length-based fine clustering on the optimal cluster and select qualified sub-clusters based on the number of members; Then, the optimal subcluster is determined based on the assembly algorithm's support for categories, number of members, and coverage uniformity indices. The overlapping group with the best coverage uniformity within the optimal sub-cluster is selected as the optimal sample assembly. S6. Polish and shape the optimal assembly to obtain a closed plasmid sequence.

2. The assembly method as described in claim 1, characterized in that, In step S1, the preprocessing includes at least one of removing artificially introduced non-target sequences, removing chimera interference, and filtering low-quality and invalid short fragments.

3. The assembly method as described in claim 1, characterized in that, In step S2, the initial contiguous group is first assembled from scratch using the read set according to the plasmid pattern, and the plasmid size and average sequencing depth are estimated based on the contiguous group.

4. The assembly method as described in claim 1, characterized in that, In step S3, the multi-level target coverage includes a sequencing depth between 30 and 500×; And / or, the multi-level target coverage is 18 levels; And / or, step S4, the de novo assembly algorithm, includes Flye and miniasm.

5. The assembly method as described in claim 1, characterized in that, In step S5, the qualified cluster screening conditions are: the number of cluster members reaches a preset threshold, the threshold is 3 when the total number of candidates in the candidate pool is greater than 10, the threshold is 2 when the total number of candidates in the candidate pool is not greater than 10, and at least one member in the cluster comes from the assembly result containing only 1 superimposed group.

6. The method according to claim 5, characterized in that, In step S5, the optimal cluster is selected by prioritizing the cluster that is supported by the most different assembly algorithms; if the supported types are the same, the cluster with the largest number of members is selected; if there is another qualified cluster whose average coverage uniformity index is 0.5 times or less than that of the cluster with the largest number of members, then the other qualified cluster is selected as the optimal cluster.

7. The method according to claim 6, characterized in that, In step S5, the qualified sub-cluster selection criteria are: the number of sub-cluster members reaches a preset threshold, the threshold is set to 2 when the total number of members of the optimal cluster is greater than 3, and the threshold is set to 1 when the total number of members of the optimal cluster is not greater than 3; the rule for determining the optimal sub-cluster is: the sub-cluster with the largest number of members is selected first, and if there is another sub-cluster whose average coverage uniformity index is 0.5 times or less than that of the sub-cluster with the largest number of members, then that other sub-cluster is selected as the optimal sub-cluster; the overlapping cluster with the smallest coverage variation coefficient CV among the optimal sub-clusters is selected as the optimal sample assembly.

8. The method according to claim 1, characterized in that, The coverage uniformity index is the coverage variation coefficient, which is the ratio of the standard deviation of the sliding window sequencing depth to the mean of the window sequencing depth.

9. The assembly method as described in claim 1, characterized in that, In step S6, the polishing and shaping process involves sequentially performing consensus polishing, sequence start-point rotation, re-consensus polishing, and homology error correction on the optimal assembly to obtain the final closed plasmid sequence.

10. A de novo plasmid sequence assembly system based on nanopore sequencing, characterized in that, include: Preprocessing module: preprocesses raw nanopore sequencing reads to obtain a high-quality read set; Estimation module: Estimates the assembled plasmid size and average sequencing depth based on the read set; Sampling module: Based on the average sequencing depth and total number of reads, multiple target coverage levels are set, and read sets are sampled separately to obtain read subsample sets with different sequencing depths; Assembly module: For each sequencing depth subsample, multiple orthogonal algorithms are used to assemble them from scratch to obtain a candidate assembly set; Clustering and filtering module: This module filters and clusters candidate assemblies to obtain the optimal assembly, including: Perform pre-screening on the candidate assembly set, and retain circular candidate contigs with a number below the threshold to enter the candidate pool; Coarse clustering is performed on the length of the superimposed clusters in the candidate pool to obtain multiple clusters. Clusters with a number of members reaching the threshold and not all of them come from fragmented assembly are selected as qualified clusters. Based on the assembly algorithm, the optimal cluster can be determined from qualified clusters by considering the number of species, the number of members, and the coverage uniformity. Perform length-based fine clustering on the optimal cluster and select qualified sub-clusters based on the number of members; Then, the optimal subcluster is determined based on the assembly algorithm's support for categories, number of members, and coverage uniformity indices. The overlapping group with the best coverage uniformity within the optimal sub-cluster is selected as the optimal sample assembly. Polishing and error correction module: Polishes and shapes the optimal assembly to obtain a closed plasmid sequence.

Images