A structural variation merging method based on third-generation sequencing data and considering sequence similarity and related device

Abstract

The application discloses a structural variation merging method based on third-generation sequencing data and considering sequence similarity and a related device, obtains single-sample variation data and multi-sample variation data, divides the single-sample variation data or the multi-sample variation data into different windows by using coarse clustering based on a sliding window, performs fine clustering based on sequence similarity in each window, merges fine clustering results, integrates the merging results, and obtains a consensus sequence. The application considers variation sequence similarity on the basis of length and position information of variations, can find more similar structural variations, and has higher merging precision.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of genome structural variation detection, and particularly relates to a structural variation merging method based on third-generation sequencing data and considering sequence similarity and a related device. BACKGROUND

[0002] Genome structural variation refers to a genomic rearrangement with a length of more than 50 bp, and usually includes deletion, insertion, inversion, duplication and translocation. Structural variation is closely related to each of us, mainly in aspects of human diseases, chromosome evolution and gene regulation. With the development of sequencing technology, especially the third-generation sequencing technology, which has the characteristics of ultra-long read length, high throughput and high accuracy, it has shown great advantages in many research fields, and the research on structural variation has gradually deepened. There are many structural variation detection tools based on third-generation sequencing data, such as Sniffles published in 2018, SVIM published in 2019, cuteSV published in 2020 and the like. However, in these structural variation detection tools, even the best comprehensive performance tool has a large number of false positive detection results in some high complexity regions of the genome. In addition, the objects detected by these tools are single samples, and the population-scale structural variation results cannot be directly obtained. Therefore, structural variation merging needs to be performed to process the detection results, so as to obtain a structural variation set with higher accuracy and meet the needs of population-scale structural variation research.

[0003] At present, there are some tools for structural variation merging. When merging structural variations, these tools mainly follow two different ideas. The first idea is to directly use the position information of the structural variation on the reference genome for merging, that is, by analyzing the alignment position and length of two structural variations on the genome, the degree of overlap between them is calculated and used as the basis for merging, and SURVIVOR is a representative tool using this merging method. The second idea is to convert the structural variation into a graph form, and then use a graph clustering method to realize merging, such as the merging tool Jasmine published in 2023. These two ideas essentially only use the position information of the variation, and cannot comprehensively describe the characteristics of the structural variation, limiting the performance of the merging. For example, SURVIVOR is prone to incorrectly merging different insertion and deletion variations at the same position, and the merging result is relatively rough. Although Jasmine uses a more detailed merging method, it often produces false judgments in high-repetition regions of the genome, affecting the accuracy of downstream analysis. Therefore, it is not enough to only use the position of the structural variation for merging, and more information needs to be considered as the basis for structural variation merging. SUMMARY

[0004] To overcome the technical problem of inaccurate detection results of genomic structural variations in the prior art, the present application aims to provide a structural variation merging method based on third-generation sequencing data and considering sequence similarity, and related devices, which can effectively reduce the false positive rate of detection results and improve the overall quality of the detection set in a single-sample single-detection set scenario, and can achieve better merging effect in a multi-sample scenario. At the same time, in a single-sample scenario, the present application also has the ability to identify part of the complex structural variations.

[0005] To achieve the above-mentioned purpose, the technical solutions adopted by the present application are as follows:

[0006] The first aspect of the present application provides a structural variation merging method based on third-generation sequencing data and considering sequence similarity, comprising the following steps:

[0007] Obtain single-sample variation data and multi-sample variation data;

[0008] Divide the single-sample variation data or multi-sample variation data into different windows using a rough clustering based on a sliding window;

[0009] Perform fine clustering based on sequence similarity in each window;

[0010] Merge the fine clustering results;

[0011] Integrate the merging results to obtain consensus sequences.

[0012] Further, for single-sample variation data, the fine clustering based on sequence similarity in each window comprises the following steps: for each window, first split the variation sequence into a k-mer set, then construct a sequence feature matrix in a high-dimensional space by counting the k-mer frequency, and finally calculate the structural variation sequence similarity in the same window according to the cosine distance between the feature matrices.

[0013] Further, for multi-sample variation data, the fine clustering based on sequence similarity in each window comprises the following steps: encode the k-mer in quaternary, then use the encoding and frequency as a key-value pair, store the k-mer frequency in a hash table and perform KLD distance calculation to obtain a distance matrix, and according to the distance matrix, use multidimensional scaling transformation to map the variations to a Euclidean space while maintaining the relative distance.

[0014] Further, for single-sample variation data, the fine clustering results are merged, comprising the following steps:

[0015] According to the fine clustering results, based on the distance matrix between the structural variations in the window, a bottom-up hierarchical clustering method is used for iterative clustering and matrix updating until a set distance threshold is reached.

[0016] Further, for multi-sample variant data, the fine clustering results are merged, including the following steps:

[0017] According to the fine clustering results, the density clustering optimized by the KD tree is used for merging.

[0018] Further, for single-sample variant data, the merging results are integrated to obtain a consensus sequence, including the following steps: according to the merging results, a multi-sequence alignment is performed by using a tree structure of hierarchical clustering to guide the path of a double-sequence alignment, and a consensus sequence is generated.

[0019] Further, for multi-sample variant data, the merging results are integrated to obtain a consensus sequence, including the following steps:

[0020] According to the merging results, a high-dimensional matrix is used to record the scores of the multi-sequence alignment, a partial-order alignment graph is constructed, and a maximum likelihood path is solved, so as to generate a consensus sequence.

[0021] The second aspect of the present application provides a structural variant merging system based on third-generation sequencing data and considering sequence similarity, comprising:

[0022] A data acquisition module is configured to acquire single-sample variant data and multi-sample variant data.

[0023] A coarse clustering module is configured to divide the single-sample variant data or the multi-sample variant data into different windows by using a coarse clustering based on a sliding window.

[0024] A fine clustering module is configured to perform fine clustering based on sequence similarity in each window.

[0025] A merging module is configured to merge fine clustering results.

[0026] An integration module is configured to integrate the merging results to obtain a consensus sequence.

[0027] The third aspect of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to realize the structural variant merging method based on third-generation sequencing data and considering sequence similarity.

[0028] The fourth aspect of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the structural variant merging method based on third-generation sequencing data and considering sequence similarity.

[0029] Compared with the prior art, the present application has the beneficial effects that:

[0030] (1) The present invention considers the sequence similarity of variants in addition to the length and position information of the variants, and can discover more similar structural variants with higher merging accuracy;

[0031] (2) In the single-sample, single-test set scenario, potential complex structural variations can be discovered by merging different types of variations, providing a certain reference for subsequent related research;

[0032] (3) The merging ability is stable, and the performance improvement of a single detection set is effective under different supporting reads, and it still has good merging ability in complex regions of the genome;

[0033] (4) Even with the addition of sequence similarity calculation, it still has high efficiency;

[0034] (5) In multi-sample scenarios, it has stronger merging capabilities compared to mainstream merging tools.

[0035] This invention can provide researchers with a more efficient way to merge structural variations in both single-sample and multi-sample scenarios, thereby obtaining higher-quality variation sets at both individual and population scales. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is the overall framework of the structural variation merging algorithm proposed in this invention;

[0037] Figure 2 Schematic diagram of the sliding window coarse clustering method;

[0038] Figure 3 Schematic diagram of the similarity calculation method based on k-mer frequency;

[0039] Figure 4 Schematic diagram of the optimization method for sequence feature representation of quaternary encoded hash;

[0040] Figure 5 Schematic diagram of the mapping between KLD distance and MDS Euclidean space;

[0041] Figure 6 Schematic diagram of the AGNES hierarchical clustering merging method; (a) is the distance matrix, and (b) is the merging process;

[0042] Figure 7 Schematic diagram of the density clustering merging method optimized for KD tree;

[0043] Figure 8 Schematic diagram of the length splicing algorithm based on support reads;

[0044] Figure 9 Schematic diagram of the consensus sequence generation algorithm based on tree alignment;

[0045] Figure 10 Schematic diagram of consensus sequence generation algorithm based on partial order comparison

[0046] Figure 11 Comparison chart of the results of the method of the present application and the results of existing detection tools. Wherein (a) is SVIM, (b) is cuteSV, (c) is Sniffles2.

[0047] Figure 12 Comparison chart of the results of the method of the present application and the results of existing merging tools.

[0048] Figure 13 Schematic diagram of the structural variation merging system based on three-generation sequencing data and considering sequence similarity of the present application. DETAILED DESCRIPTION

[0049] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings. The preferred embodiments of the present application are shown in the drawings. However, the present application can be realized in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0050] Reference Figure 1 A structural variation merging method SeqSV based on three-generation sequencing data and considering sequence similarity, which is divided into two merging modes containing single sample and multiple samples, respectively, single sample structural variation merging algorithm SeqSV-S and multiple sample structural variation merging algorithm SeqSV-M, both of which have similar processes and contain four main steps of data preprocessing, coarse clustering, fine clustering and merging information integration. However, different implementation methods are used in some steps according to the characteristics of single sample and multiple sample scenes, mainly including the following steps:

[0051] In the single-sample structural variant merging algorithm SeqSV-S, first, a rough clustering based on sliding windows is used to divide the variants into different windows, each window is independent in the subsequent merging process, thereby reducing the granularity of the merging task and improving the degree of concurrency of the program. Moreover, this process also allows different types of variants to exist in the same window, thereby supporting the discovery of potentially complex structural variants. Subsequently, within each window, the algorithm performs fine clustering based on sequence similarity, converts the sequence distance into a high-dimensional sparse matrix of k-mer frequency as a feature representation, uses cosine distance to obtain the distance matrix between variants, and uses AGNES hierarchical clustering based on the distance matrix to merge the variants. In the information integration stage, SeqSV-S uses different basic information integration methods for different types of variants. For the integration of sequence information, SeqSV-S performs multiple sequence alignment by guiding the path of the double sequence alignment with the tree structure of the hierarchical clustering, and generates a consensus sequence as the merged sequence. Experiments on real data sets prove that for the current mainstream structural variant detection tools, SeqSV-S can not only effectively reduce the false positive rate of a single detection set and significantly improve the overall performance of the detection results, but also can obtain higher quality detection results by merging multiple detection sets from different tools.

[0052] The multi-sample structural variant merging algorithm SeqSV-M uses multiple optimization methods for the larger data size in the group size scenario, including: using a quaternary encoding hash table instead of the high-dimensional sparse matrix in the single-sample structural variant merging algorithm, optimizing the k-mer extraction and storage process, and improving the calculation efficiency of the cosine distance; propose KLD distance instead of sequence similarity-based variant distance measurement in single sample, meet the needs of structural variant merging when sequence is missing in multi-sample scenario, and map the variants to Euclidean space through multidimensional scaling transformation; use the optimized DBSCAN density clustering of KD tree instead of hierarchical clustering, improve the merging efficiency in the multi-sample scenario, and adapt to the irregular distribution shape of similar variants in the Euclidean space.

[0053] Embodiment 1

[0054] The structural variant merging method SeqSV based on three-generation sequencing data and considering sequence similarity of the application has the following steps:

[0055] (1) Rough clustering method based on sliding window: sliding windows are used in single-sample and multi-sample modes, and the chromosomes and positions of the structural variants are divided by a fixed step (default 1kbp), and each window is used as a basic unit for merging. Among them, the windows are not distinguished by variant type in the single-sample single-detection set scenario to support the discovery of potentially complex structural variants, such as Figure 2 indicated;

[0056] (2) Sequence similarity calculation based on k-mer frequency: in single-sample mode, for the sequence in each window, first split the variant sequence into a set of k-mers, then construct a sequence feature matrix in high-dimensional space by counting the frequency of k-mers (see Figure 3 ), and finally calculate the similarity of structural variant sequences in the same window according to the cosine distance between the feature matrices (the calculation process is a known technique in the art), as shown in Figure 3 : Take K = 3 as an example, the dimension of the matrix is also 3, each base in the 3-mer corresponds to its coordinates in the corresponding dimension, then each 3-mer is mapped to the specified position of the feature matrix, and the final value of the matrix is obtained by counting the frequency;

[0057] (3) Sequence feature representation optimization of quaternary coding hash: in view of the larger data size in the multi-sample scenario, first encode the k-mer according to the quaternary, then use the encoding and frequency as a key-value pair, store the k-mer frequency in a hash table and perform distance calculation, optimize memory usage and calculation efficiency, as shown in Figure 4 ;

[0058] (4) KLD distance and MDS Euclidean space mapping: in the multi-sample mode, KLD distance is proposed as a measure of the distance between variants, taking into account the length and sequence similarity of the variant, and using multidimensional scaling (MDS) to map the variant to the Euclidean space while preserving the relative distance, as shown in Figure 5 : Take sv1 ~ sv4 as an example, where figure (a) is the distance matrix of the four SV sequences obtained by KLD distance calculation, based on the matrix, MDS can be used to map the Sv to the Euclidean space (figure (b)), and the Euclidean distance between each SV is consistent with the distance matrix as much as possible;

[0059] (5) Merging method based on AGNES hierarchical clustering: in single-sample mode, based on the distance matrix between structural variants in the window, a bottom-up hierarchical clustering method is used for iterative clustering and matrix updating, and the process is terminated when the specified distance threshold is reached, as shown in Figure 6 : Based on the distance matrix of SV1 ~ SV4, each SV is a cluster at the beginning, and the two SVs with the shortest distance are merged, and the distance between the merged cluster and other clusters is updated using the full chain method until the shortest distance is greater than the threshold;

[0060] (6) KD tree optimized DBSCAN density clustering merging method: see Figure 7 , use KD tree optimized density clustering for merging in multi-sample mode, which improves the efficiency of merging while adapting to the irregular distribution of similar variants after mapping in the Euclidean space;

[0061] (7) Basic information integration: In single sample merging, different types of variation before merging are marked as potential complex structural variations, and the length of insertion and deletion variation is calculated using a length splicing algorithm based on support reads. In other scenarios, the average method is used to calculate. As shown in Figure 8 , the process of calculating the length of insertion and deletion variation using a length splicing algorithm based on support reads is as follows: a hash table is maintained for each cluster to record all support reads of structural variations before merging and the splicing length corresponding to each read. When traversing the structural variations in the cluster, the unique identifier of all support reads of the variation and the length of the variation are sequentially put into the hash table as keys and values, respectively. If the current read identifier already exists, the corresponding value is modified to be the sum of the original value and the current variation length, that is, length splicing. Finally, sort the values corresponding to all reads, and take the upper quartile as the length after merging. The position after merging is also taken as the upper quartile of all positions.

[0062] (8) Consensus sequence generation based on tree alignment: Referring to Figure 9 , the tree alignment algorithm Tree Align is used to guide the path of sequence pairwise alignment using the cluster tree in the single sample merging process, to gradually generate a multiple sequence alignment result and generate a consensus sequence as the sequence after merging.

[0063] (9) Consensus sequence generation based on partial order alignment: Referring to Figure 10 , in the multi-sample mode, a high-dimensional matrix is used to record the scores of multiple sequence alignment, a partial order alignment graph is constructed, and the maximum likelihood path is solved to generate a consensus sequence, as shown in Table 1.

[0064] Table 1 Before and after merging results

[0065]

[0066] Experiments on HG002 family data show that in the single sample single dataset scenario, SeqSV-S can improve the F1 value of the results of the current mainstream detection tool through merging (as shown in Table 2 and Figure 11 ). And in the single sample multi-dataset scenario, the F1 of the high-quality result obtained is better than that of any single detection set (as shown in Table 2). In the multi-sample scenario, compared with the two most mainstream structural variation merging tools, the number of de-nova variation after merging of the multi-sample structural variation merging algorithm SeqSv-M is 4699, which is more accurate than the current mainstream merging tools Jasmine (5746) and SURVIVOR (6377) (see Figure 12 ).

[0067] Table 2 Comparison of results of the present application and existing detection tools

[0068]

[0069] The method of the present application adds sequence similarity calculation based on k-mer frequency on the basis of structural variation position and length information, which can not only effectively reduce the false positive rate of detection results and improve the overall quality of the detection set in the single sample single detection set scene, but also can achieve better merging effect in the multi-sample scene. At the same time, in the single sample scene, the method also has the ability to identify part of the complex structural variation.

[0070] Referring to Figure 13 Another embodiment of the present application provides a structural variation merging system based on third-generation sequencing data and considering sequence similarity, comprising a data acquisition module, a coarse clustering module, a fine clustering module, a merging module and an integration module.

[0071] The data acquisition module is used to acquire single sample variation data and multi-sample variation data.

[0072] The coarse clustering module is used to divide the single sample variation data or the multi-sample variation data into different windows by using coarse clustering based on a sliding window.

[0073] The fine clustering module is used to perform fine clustering based on sequence similarity in each window.

[0074] The merging module is used to merge the fine clustering results.

[0075] The integration module is used to integrate the merging results to obtain a consensus sequence.

[0076] Another embodiment of the present application provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to realize the structural variation merging method based on third-generation sequencing data and considering sequence similarity.

[0077] Another embodiment of the present application provides a computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to realize the structural variation merging method based on third-generation sequencing data and considering sequence similarity.

[0078] The above only describes the best embodiments of the present application, but cannot be understood as the limitation of the claims. The present application is not limited to the above embodiments, and the specific structure allows changes. Any changes made within the protection scope of the independent claims of the present application are within the protection scope of the present application.

[0079] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

Claims

1. A method for merging structural variations based on third-generation sequencing data and considering sequence similarity, characterized in that: The following steps are involved: Obtain single-sample variation data and multi-sample variation data; Divide single-sample variation data or multi-sample variation data into different windows using coarse clustering based on sliding windows; Fine clustering based on sequence similarity is performed within each window; Merge the fine clustering results; The merged results were integrated to obtain the consensus sequence; For single-sample variant data, fine-grained clustering based on sequence similarity is performed within each window, including the following steps: for each sequence within the window, the variant sequence is first split into k-mer sets, then a sequence feature matrix is ​​constructed in a high-dimensional space by counting the k-mer frequencies, and finally the structural variant sequence similarity within the same window is calculated based on the cosine distance between the feature matrices; For multi-sample variant data, fine clustering based on sequence similarity is performed within each window, including the following steps: encoding k-mers according to quaternary system, then using the code and frequency as key-value pairs, storing the k-mer frequencies in a hash table and performing KLD distance calculation to obtain a distance matrix. Based on the distance matrix, multidimensional scaling is used to map the variants to Euclidean space while maintaining relative distances; For single-sample variation data, merging the fine clustering results includes the following steps: According to the fine clustering results, based on the distance matrix between structural variations within the window, a bottom-up hierarchical clustering method is used to iteratively cluster and update the matrix until the set distance threshold is reached; For multi-sample variation data, merging the fine clustering results includes the following steps: Based on the fine clustering results, the KD tree optimized density clustering is used for merging; For single-sample variant data, the merge results are integrated to obtain a consensus sequence, including the following steps: based on the merge results, a hierarchical clustering tree structure is used to guide the path of the two-sequence alignment to perform multiple sequence alignment and generate a consensus sequence; For multi-sample variant data, the merge results are integrated to obtain a consensus sequence, including the following steps: Based on the merge results, a high-dimensional matrix is ​​used to record the scores of the multiple sequence alignments, a partial alignment graph is constructed, and the maximum likelihood path is solved to generate a consensus sequence.

2. A structural variation merging system based on third-generation sequencing data and considering sequence similarity, used in the method of claim 1, characterized in that: include: Data acquisition module, used to obtain single-sample variation data and multi-sample variation data; Coarse clustering module, used to divide single-sample variation data or multi-sample variation data into different windows using coarse clustering based on sliding windows; Fine clustering module, used to perform fine clustering based on sequence similarity within each window; Merging module, used to merge fine clustering results; The integration module is used to integrate the merging results to obtain the consensus sequence.

3. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the structural variation merging method based on third-generation sequencing data and considering sequence similarity as claimed in claim 1 is implemented.

4. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for merging structural variations based on third-generation sequencing data and considering sequence similarity as claimed in claim 1 is implemented.

Images