CN110060735A - A kind of biological sequence clustering method based on the segmentation of k-mer group - Google Patents

Abstract

The invention discloses a biological sequence clustering method based on k-mer group segmentation, which comprises the following steps: step 1, segment the sequences in the data set, and count the k-mer word frequencies after the segmentation; step 2, according to the sequence and The relationship between the k-mers, construct a bipartite graph; step 3, randomly group the k-mers, and calculate the importance of the sequences under each group of k-mers; step 4, sort the importance of the sequences in reverse order, and filter out the candidates sequence, and deduplicate it; step 5, cluster candidate sequences to find sequence centers; step 6, cluster all sequences. By constructing a bipartite graph model, the present invention performs cluster analysis on biological sequences, obtains deep information meaning and reliable conclusions from biological sequence data, and effectively solves the problems of high complexity of existing computing node weights and representative nodes of importance. The problem of insufficient sex and the importance of nodes is greatly affected by the length of the sequence.

Description

A biological sequence clustering method based on k-mer group segmentation

technical field

The invention relates to the field of bioinformatics, in particular to a biological sequence clustering method based on k-mer group segmentation.

Background technique

With the continuous development of next-generation sequencing technology, the scale of biological databases is expanding. The huge biological information database provides researchers with broad opportunities and challenges. How to mine useful information from the tens of billions of biological information databases, data mining provides a basic means for scientific research. The similarity of biological sequences often reflects the correlation of their functions, and cluster analysis is a commonly used technique in data mining.

In biology, sequence alignment identifies similar sequence regions that describe the function, structure, and evolutionary relationship between sequences by aligning biological sequences. Clustering is to divide similar sequences into the same group, and dissimilar sequences into different groups, so that the sequence distance between the same group is the smallest, and the distance between different groups is the largest. If two similar sequences are clustered into the same group, it indicates that the two have homology to a certain extent, which will greatly save the time and energy of re-determining the structure and function of unknown sequences. In addition, sequence alignment generally determines the analysis results of many bioinformatics techniques and programs, and affects the conclusions and biological interpretations of many sequence comparison studies. It is an important content in studies such as biological sequence cluster analysis.

Bipartite graph is one of the more widely used graph classes. The applications in real life include: how to arrange work to meet the needs of each person to the greatest extent, and to know the competence of each person's work; how to arrange courses to So that the conditions between classrooms, teachers and students are met. These all involve the matching problem of bipartite graphs, which can be solved by establishing a graph model.

Combined with graphical models, clustering research on biological sequences, and grouping functionally related sequences into one category, can help researchers quickly understand the function of biological sequences and clarify the diversity of their internal populations, thereby promoting biodiversity conservation. Reasonable Development and utilization of biological resources.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a biological sequence clustering method based on k-mer group segmentation.

The technical scheme adopted in the present invention is:

A biological sequence clustering method based on k-mer group segmentation, comprising the following steps:

Step 1: Set the size of the sliding window, segment the sequences in the dataset, and count the k-mers word frequency after segmentation;

Step 2: According to the relationship between the sequence and k-mers, construct a bipartite graph, and separately count the word frequencies of k-mers co-occurring between any sequence _si and other sequences;

Step 3: Divide the k-mers into t groups randomly and uniformly: g ₁ , g ₂ ,..., g _t , and calculate the importance of the sequence under each group of k-mers;

Step 4: Sort the importance of the sequences in reverse order: let k be the number of centers of m sequences, and screen out k sequences from the t group according to uniform intervals as candidate sequences; Candidate sequences are deduplicated to obtain non-repeated candidate sequences;

Step 5: Perform k-mers clustering on the candidate sequences; perform clustering on the DNA sequences based on the set sliding window size to obtain a k-mers set; in each cluster in the K-means clustering result, filter out the cluster with the current centroid. The closest point is used as the sequence center; and so on, k center sequences can be obtained;

Step 6: Perform clustering on m sequences S={s _i |i=1, 2, . . . , m} based on the k central sequences.

Further, the data set in step 1 is a sequence set S={s _i |i=1, 2, ..., m} of length m, the sliding window size is L, and the segmented k-mers set is K={k _j |j=1,2,...,n}.

Further, the bipartite graph constructed in step 2 is G=(V, E), which is also referred to as a sequence-k-mers graph. G=(V, E) is an undirected graph model composed of node sets and edge sets, where V is the node set, and V can be decomposed into two subsets, namely V=S∪K, and S={s _i |i=1,2,...,m} is the sequence set, s _i is the i-th sequence, K={k _j |j=1,2,...,n} is k -mers set, k _j is the jth k-mers; E represents the set of edges formed by the interaction relationship between nodes, and the two endpoints of each edge in E are in subset S and subset K respectively, that is, E ={e(s _i , k _j )|s _i ∈S, k _j ∈K}, where e(s _i , k _j ) indicates that there is a membership relationship between sequence _si and k-mersk _j .

Further, the method for determining sequence importance in step 3 is:

Step 3.1: Calculate the weights of the edges. When two sequences vi and v _j have a common _k -mers, it is considered that vi and v _j are adjacent nodes, and there is an edge connected, and the weight of the edge w _ji is that the k-mers existing in the two sequences co _- occur The number of frequencies, that is: for any two sequences vi and v _j , if they have a common _k -mers, w _ji can be used to represent the undirected interaction between nodes.

Step 3.2: Calculate the weight of the node.

For any two nodes vi and v _j , the node v _i transmits the effect to the node v _j through the edge w _ji connecting them, and the size of the edge weight determines the _effect of v _i on v _j . When the node v _j has an edge relationship with multiple nodes, that is, the node v _j has multiple adjacent nodes, the weight of the node v _j at this time is the effect that the node v _j receives from other nodes. and.

Step 3.3: Iteratively calculate the weight of each node v _i to obtain the importance of the node v _i .

Further, in step 3.1, the weights of the edges of adjacent nodes vi and v _j are w _{ji ,} and w _ji can be calculated by the following formula:

Among them, kmer∈vi & _{kmer∈v j means} that _{k-mers exist in both node vi and node v j .} and Represent the frequency of the current k-mers in node v _i and node v _j , respectively. Since the interaction between nodes is undirected, there is w _ji =w _ij .

Further, w _j. in step 3.2 indicates that the node v _j receives the action from other nodes, and w _j. is calculated by the following formula:

Among them, w _j. represents the contribution of each node in the node set V to the node v _j .

Further, the importance of each node v _i in step 3.3 is WS(vi ₎ , and the SeqRank calculation formula corresponding to WS(vi ₎ is as follows:

Among them, d is the damping coefficient (0≤d≤1), which represents the probability of walking from one node to another at any time, that is, each node has a probability of (1-d) random walk to other nodes. v _j ∈ e(v _i , v _j ) indicates that node vi and node v _j have a common edge; in v _k ∈ e(v _j , v _k ), v _k is a common _edge with node v _j the edge nodes. w _ij (or w _ji ) represents the weight of the edge connecting the node v _i and the node v _j , that is, the sum of the common occurrence frequencies of the k-mers existing at the node v _i and the node v _j . denominator Represents the weighted sum of the weights of the edges of node v _j pointing to node v _k when v _k ∈ e(v _j , v _k ). WS(v _j ) is the importance of node v _j after the last iteration.

Since the weight of the node itself needs to be used when calculating the weight of the node, iterative calculation is required. If WS( _vi ) ^t is used to represent the importance of node _vi after t iterations, then formula (3) can be expressed as:

The SeqRank algorithm iteratively calculates the graph model until the convergence conditions are met.

Further, in step 3.3, d satisfies 0≤d≤1.

Further, the value of d is 0.85.

Further, the determination method of the center sequence in step 5 is:

Step 5.1: Use the K-means algorithm to cluster the candidate sequences, the number of K-means centers is k, and the feature is the k-mers frequency of the candidate sequence;

Step 5.2: For each cluster, filter out the point closest to the current center as the sequence center.

Further, the determination method of sequence clustering in step 6 is:

Step 6.1: Mark the k center sequences as μ ₁ , μ ₂ , ..., μ _{k respectively} ;

Step 6.2: For each sequence _si in the sequence set S, use the following formula to calculate its predicted category:

Among them, pre _i represents the closest category of the sequence s _i to the k clusters, that is, the predicted category of the i-th sequence, ||w _i -μ _j || ² represents the importance value w _i of the sequence s _i , Calculate its Euclidean distance from each center point, Indicates that the center point closest to _wi is determined as the prediction category of the i-th sequence. From this, the prediction categories corresponding to the m sequences can be obtained.

The present invention adopts the above technical scheme, and by constructing a bipartite graph model, performs cluster analysis on biological sequences, attempts to obtain deep information meaning and reliable conclusions from biological sequence data at the level of cluster analysis, and effectively solves the problem of the prior art. There are problems such as high complexity of calculating node weights, insufficient representation of node importance, and node importance being greatly affected by sequence length.

Description of drawings

The present invention will be described in further detail below in conjunction with the accompanying drawings and specific embodiments;

1 is a schematic flowchart of a biological sequence clustering method based on k-mer group segmentation of the present invention;

Fig. 2 is the g1 group schematic diagram that the present invention carries out random and uniform grouping to k-mers;

Fig. 3 is the g2 group schematic diagram that the present invention carries out random and uniform grouping to k-mers;

4 is a bipartite schematic diagram of three sequences of XYZ of the present invention;

FIG. 5 is a schematic diagram of the result of performing K-means clustering on sequences according to the present invention.

Detailed ways

As shown in one of Figures 1-5, the present invention discloses a biological sequence clustering method based on k-mer group segmentation. According to the relationship between sequences and k-mers under different sliding window sizes, the invention constructs a bipartite graph, groups k-mers, calculates the importance of sequences under different groups, finds out sequence centers, and clusters them. It effectively solves the problems existing in the prior art, such as high complexity of calculating node weights, insufficient representation of node importance, and large influence of sequence length on node importance.

As shown in Figure 1, the present invention discloses a clustering algorithm based on k-mer group segmentation, which comprises the following steps:

Step 1: According to the sliding window size L, divide the given sequence set S={s _i |i=1, 2, ..., m}, and the segmented k-mers set is K={k _j | j=1, 2, .

Specifically, there are (|s _i |-L+1) elements of all n-Grams of length L in the sequence _si .

For each sequence s _i , count the frequency of occurrences of each k-mers. The set of k-mers between sequences is ∑ ^L , and the number of elements in the set is |∑| ^L , that is: for a DNA sequence with base type 4, the number of possible k-mers is 4 ^L ; for amino acid types is 20 protein sequences, the number of possible k-mers is 20 ^L.

For example, for three sequences X="ACAGT", Y="ACACG" and Z="CACGT", when the sliding window size of the k-mer element is set to 2, the elements with n-Gram length of 2 between all sequences in the three sequences are all 4, and the occurrence of k-mers in each sequence is shown in Table 1:

Table 1 The occurrence of k-mers between sequences

k-mers AC AG CA CG GT sum Sequence X 1 1 1 0 1 4 sequence Y 2 0 1 1 0 4 sequence Z 1 0 1 1 1 4

Step 2: According to the relationship between the sequence and k-mers, construct a bipartite graph G=(V, E). G=(V, E) is an undirected graph model consisting of node sets and edge sets. Among them, V is the set of nodes, E represents the set of edges formed by the interaction relationship between nodes, and E={e(s _i , k _j )|s _i ∈S, k _j ∈K}, where e(s _i , k _j ) means that there is a affiliation between the sequence _si and k-mersk _j .

Specifically, it can be seen from Table 1 that for sequence X and sequence Y, there are common k-mers between the two sequences: AC and CA, then it is considered that sequence X and sequence Y are connected by an edge; for sequence X and sequence Z, the two sequences If there are common k-mers between the two sequences: AC, CA and GT, it is considered that there is an edge connecting the sequence X and the sequence Z; for the sequence Y and the sequence Z, there are common k-mers between the two sequences: AC, CA and CG, it is considered that Sequence Y and sequence Z are connected by an edge.

Step 3: Calculate the importance of m sequences.

Step 3.1: Divide the n k-mers into _t groups evenly and randomly: g ₁ , g ₂ , ..., gt , as shown in Figure 2, the sequence s ₁ and the sequence s _i have a common k-mersk ₁ , the sequence There is a common k-mersk ₃ between s ₁ and the sequence s _m , that is, the sequence s ₁ and the sequence s _i , and the sequence s ₁ and the sequence s _m have an interaction relationship. At this time, with sequences and k-mers as nodes, a graph model G=(V, E) can be constructed.

Step 3.2: Calculate the weights of the edges. When two sequences vi and v _j have a common _k -mers, it is considered that vi and v _j are adjacent nodes, and there is an edge connected, and the weight of the edge w _ji is that the k-mers existing in the two sequences co _- occur The number of frequencies, that is: for any two sequences vi and v _j , if they have common _k -mers, w _ji can be used to represent the undirected interaction between nodes, and w _ji is calculated by the following formula:

Specifically, for the three sequences X="ACAGT", Y="ACACG" and Z="CACGT", when the sliding window size of the k-mer element is set to 2, the bipartite graph of the three sequences is shown in Figure 3 :

The frequencies of k-mers for the three sequences of X, Y and Z are shown in Table 2.

Table 2 The occurrence of k-mers of the three sequences

k-mers AC AG CA CG GT Sequence X 1 1 1 0 1 sequence Y 2 0 1 1 0 sequence Z 1 0 1 1 1

It can be seen from Table 2 that for sequence X and sequence Y, there are common k-mers between the two sequences: AC and CA. Since the weight of the edge is the number of co-occurrence frequencies of k-mers existing in the two sequences, the relationship between sequence X and sequence Y in Table 2 can be simplified as:

Table 3 Co-occurrence frequency of k-mers in X and Y sequences

k-mers AC CA Sequence X 1 1 sequence Y 2 1

At this time, the interaction w _YX between the sequence X and the sequence Y can be calculated by the following formula:

w _YX = min(|AC _X |, |AC _Y |)+min(|CA _X |, |CA _Y |)=1+1=2

Step 3.3: Calculate the weight of the node. For any two nodes vi and v _j , the node v _i transmits the effect to the node v _j through the edge w _ji connecting them, and the size of the edge weight determines the _effect of v _i on v _j . When there is an edge relationship between node v _j and multiple nodes, that is, node v _j has multiple adjacent nodes, w _j. indicates that node v _j receives action from other nodes, and w is calculated by the following formula _j .:

For Table 3, by calculating the co-occurrence frequency of k-mers existing in the three sequences X, Y and Z, the following relationship matrix M can be obtained:

Table 4 Co-occurrence frequency of k-mers in the three sequences of X, Y and Z

Sequence X sequence Y sequence Z Sequence X 0 2 3 sequence Y 2 0 3 sequence Z 3 3 0

In Table 4, the size of the matrix M is |V|*|V|, where |V| represents the number of nodes. The values in the matrix represent the interaction between the two sequences, for example, M[1, 3] means that the interaction between sequence X and sequence Z is 3.

By iteratively calculating the weight of each node v _i , the importance WS(vi ₎ of the node v _i can be obtained, and the SeqRank calculation formula corresponding to WS(vi ₎ is:

Among them, d is the damping coefficient (0≤d≤1), which represents the probability of walking from one node to another at any time, that is, each node has a probability of (1-d) random walk to other nodes. v _j ∈ e(v _i , v _j ) indicates that node vi and node v _j have a common edge; in v _k ∈ e(v _j , v _k ), v _k is a common _edge with node v _j the edge nodes. w _ij (or w _ji ) represents the weight of the edge connecting the node v _i and the node v _j , that is, the sum of the common occurrence frequencies of the k-mers existing at the node v _i and the node v _j . denominator Represents the weighted sum of the weights of the edges of node v _j pointing to node v _k when v _k ∈ e(v _j , v _k ). WS(v _j ) is the importance of node v _j after the last iteration.

Since the weight of the node itself needs to be used when calculating the weight of the node, iterative calculation is required. If WS( _vi ) ^t is used to represent the importance of node _vi after t iterations, then formula (3) can be expressed as:

Step 4: Sort the sequences in reverse order of importance. The importance of m sequences under different groups of k-mers is represented by a t×m-dimensional matrix I, which is shown in Table 5.

Table 5 Importance of m sequences under t group

Specifically, in matrix I, the rows of the matrix represent t groups of k-mers, the columns of the matrix represent m sequences, and I[p, q] in the matrix represents the k-mers in the pth (1≤p≤t) group Below, the importance value of the q (1≤q≤m)-th sequence, such as I[1, ] represents the importance of m sequences under group g ₁ , I[, q] represents the calculated value under all groups The importance of the qth sequence.

It is worth noting that the same sequence under different groups will show different importance, such as the sequence s ₁ in matrix I, the importance under group g ₁ is 0.9696823, and the importance under group g ₂ is 1.040769; on the contrary, the maximum value obtained by the importance calculation under different groups can be different sequences, for example, the most important sequence under group g ₁ is s _m , and the most important sequence under group g _p is s ₁ .

Taking the group as a unit, the importance of the sequence is sorted in reverse order, and the result is shown in Table 6.

Table 6 sorts the importance in reverse order

The difference from Table 5 is that in Table 6, the values in the matrix I represent the sequence numbers sorted in reverse order. For example, I[2, 1]=7 represents: the most important value calculated under the group g ₂ is a 7-sequence; I[ _p ,m]=7 means that the 7-sequence is considered to be the least important sequence under group gp. The most important sequence numbers calculated under different groups of k-mers are different.

Assuming that k (k≤m) is the number of centers of m sequences, the sequence number SR ₁ under group g ₁ shall prevail, and k sequences are screened from each of the t groups according to uniform intervals, as shown in the shaded part of Table 7. Show.

Table 7 Screening for Importance

For each group of k sequence numbers screened out, the sequence in which it is located is taken as a candidate sequence, and the size of the candidate sequence set is t*k. In _t *k sequences, there may be repeated sequences. For example, sequences 5 and 7 in group g ₁ are also in the candidate sequence set of group _gt ; sequence 78 in group gt is also a candidate sequence in group g ₂ . in the collection. Therefore, we need to deduplicate the selected t*k candidate sequences to obtain n (n≤t*k) non-repeated candidate sequences.

Step 5: Cluster the candidate sequences with K-means algorithm. The number of K-means centers is k, and the feature is the k-mers frequency of the candidate sequence. When the sliding window size is L, for DNA sequences, the set of k-mers that may appear between sequences is ∑ ^L . Assuming L = 2, there is the following matrix O:

Table 8 Frequency of occurrence of k-mers in candidate sequences

The size of the matrix O is n*|∑| ^L , where n is the number of candidate sequences, and ∑ ^L is the set of k-mers corresponding to the sequence when the current sliding window size is L. When L=2, for a DNA sequence, Σ ^L = {AA, AC, AG, . . . , TT}. In matrix O, 0[s _i , ] represents the frequency of occurrence of each k-mers in the i-th sequence, namely: 140, 122, 200, ..., 101.

The results obtained by K-means are shown in Figure 4. In Figure 4, n candidate sequences are clustered into k categories: Cluster1, Cluster2, ..., Clusterk. For each cluster, the point closest to the current centroid is selected as the sequence center, that is, for Cluster2, after a certain distance measurement, the closest point to Cluster2 is point A, and we think that the sequence where point A is located is Cluster2. Center sequence. By analogy, k central sequences can be obtained.

Step 6: Perform clustering on m sequences S={s _i |i=1, 2, . . . , m}.

Specifically, when the size of the sliding window is L, the set of k-mers under m sequences is K={k _j |j=1, 2, ..., |∑| ^L }, the k-mers constructed at this time The size of the frequency matrix O is m*|∑| ^L .

For k centers, the feature is the frequency matrix O representing k-mers in m sequences. The description of the clustering process of the K-means algorithm is as follows:

(1) Mark the selected k centers as μ ₁ , μ ₂ , . . . , μ _{k respectively} ;

(2) For each point wi _,j in matrix O, use formula (5) to calculate the category to which it belongs:

In formula (3), pre _i represents the closest category of the sequence s _i to the k clusters, that is, the predicted category of the i-th sequence, and ||w _i -μ _j || ² represents the importance of the sequence s _i property value w _i , calculate its Euclidean distance from each center point, Indicates that the center point closest to _wi is determined as the prediction category of the i-th sequence. From this, the prediction categories corresponding to the m sequences can be obtained.

The present invention adopts the above technical scheme, and by constructing a bipartite graph model, performs cluster analysis on biological sequences, attempts to obtain deep information meaning and reliable conclusions from biological sequence data at the level of cluster analysis, and effectively solves the problem of the prior art. There are problems such as high complexity of calculating node weights, insufficient representation of node importance, and node importance being greatly affected by sequence length.

Claims (10)

1. a biological sequence clustering method based on k-mer group segmentation, is characterized in that: it comprises the following steps: Step 1: Obtain the data set of the sequence set to be processed, and divide the sequence in the data set according to the set sliding window size to obtain the k-mers set; Step 2: Construct a bipartite graph according to the relationship between the sequence and the k-mers set, and count the word frequencies of the k-mers co-occurring between any sequence _si and other sequences respectively; Step 3: Divide the k-mers into _t groups randomly and uniformly: g ₁ , g ₂ , ..., gt , and calculate the importance of the sequence under each group of k-mers; Step 4: Sort the importance of the sequences in reverse order: let k be the number of centers of m sequences, and screen out k sequences from the t group according to uniform intervals as candidate sequences; Candidate sequences are deduplicated to obtain non-repeated candidate sequences; Step 5: Perform k-mers clustering on the candidate sequences; perform clustering on the DNA sequences based on the set sliding window size to obtain a k-mers set; in each cluster in the K-means clustering result, filter out the cluster with the current centroid. The closest point is used as the sequence center; and so on, k center sequences can be obtained; Step 6: Clustering m sequences S={s _i |i=1, 2, . 2. a kind of biological sequence clustering method based on k-mer group segmentation according to claim 1 is characterized in that: the data set in step 1 is a sequence set S={s _i |i with a length of m =1, 2,...,m}, the size of the sliding window is L, and the k-mers set after segmentation is K={k _j |j=1,2,...,n}. 3. a kind of biological sequence clustering method based on k-mer group segmentation according to claim 1, is characterized in that: the bipartite graph constructed in step 2 is G=(V, E), also make sequence- k-mers graph; G=(V, E) is an undirected graph model composed of node sets and edge sets, where V is the set of nodes and V can be decomposed into two subsets, namely V=S∪K, and S={s _i |i=1,2,...,m} is the sequence set, s _i is the i-th sequence, K={k _j |j=1,2,...,n} is k -mers set, k _j is the jth k-mers; E represents the set of edges formed by the interaction relationship between nodes and the two endpoints of each edge in E are in subset S and subset K respectively, that is, E={e(s _i , k _j )|s _i ∈S , k _j ∈ K}, where e(s _i , k _j ) means that there is a membership relationship between the sequence _si and k-mers k _j . 4. a kind of biological sequence clustering method based on k-mer group segmentation according to claim 1, is characterized in that: the determination method of sequence importance in step 3 is: Step 3.1: Calculate the weight of the edge: When there is a common _k -mers between the two sequences vi and v _j , it is considered that vi and v _j are adjacent nodes, and there is an edge connected, and the weight of the edge w _ji is two _. the number of co-occurrence frequencies of k-mers present in the sequence, Step 3.2: Calculate the weight of the node: For any two nodes vi and v _j , the node vi transmits the effect to the node v _j through the edge _{w ji} connecting them _, and the size of the edge weight determines the effect of _{vi on v j ;} When the node v _j has an edge relationship with multiple nodes, that is, the node v _j has multiple adjacent nodes, the weight of the node v _j at this time is the effect that the node v _j receives from other nodes. and; Step 3.3: Iteratively calculate the weight of each node v _i to obtain the importance of the node v _i . 5. a kind of biological sequence clustering method based on k-mer group segmentation according to claim 4, it is characterized in that: in step 3.1, the weight of adjacent node v _i and v _j side is w _ji , by the following The formula calculates w _ji : Among them, kmer∈vi & _{kmer∈v j means} that _{k-mers exist in both node vi and node v j ;} and respectively represent the frequency of the current _{k-mers in the node vi and the node v j ,} and w _ji =w _ij . 6. a kind of biological sequence clustering method based on k-mer group segmentation according to claim 4, is characterized in that: in step 3.2, w _j. represents that node v _j receives the effect from other nodes, through The following formula calculates w _j. : Among them, w _j. represents the contribution of each node in the node set V to the node v _j . 7. a kind of biological sequence clustering method based on _k -mer group segmentation according to claim 4, is characterized in that: the importance of each node vi in step 3.3 is WS( _vi ), WS( vi) The corresponding _SeqRank calculation formula is as follows: Among them, d is the damping coefficient and 0≤d≤1, it represents the probability of walking from one node to another node at any time, v _{j ∈} e(vi , v _j ) represents the node v _i and the node v _j has a common edge; in v _k ∈ e(v _j , v _k ), v _k is a node that has a common edge with node v _j ; w _ij or w _ji means connecting the node v _i and the node The weight of the edge of the point v _j , that is, the sum of the common occurrence frequencies of the k-mers existing at the node v _i and the node v _j ; the denominator Represents the weighted sum of the weights of the edges of the node v _j pointing to the node v _k when v _k ∈ e(v _j , v _k ), WS(v _j ) is the importance of the node v _j after the last iteration; Using WS( _vi ) ^t to represent the importance of node _vi after t iterations, formula (3) can be expressed as: The SeqRank algorithm iteratively calculates the graph model until the convergence conditions are met. 8 . The biological sequence clustering method based on k-mer group segmentation according to claim 7 , wherein the value of d is 0.85. 9 . 9. a kind of biological sequence clustering method based on k-mer group segmentation according to claim 1, is characterized in that: the determination method of central sequence in step 5 is: Step 5.1: Use the K-means algorithm to cluster the candidate sequences, the number of K-means centers is k, and the feature is the k-mers frequency of the candidate sequence; Step 5.2: Screen out the closest point to the current center for each cluster as the sequence center. 10. a kind of biological sequence clustering method based on k-mer group segmentation according to claim 1, is characterized in that: the determination method of sequence clustering in step 6 is: Step 6.1: Mark the k center sequences as μ ₁ , μ ₂ , ..., μ _{k respectively} ; Step 6.2: For each sequence _si in the sequence set S, use the following formula to calculate its predicted category: Among them, pre _i represents the closest category of the sequence s _i to the k clusters, that is, the predicted category of the i-th sequence, ||w _i -μ _j || ² represents the importance value w _i of the sequence s _i , Calculate its Euclidean distance from each center point, Indicates that the center point closest to _wi is determined as the prediction category of the i-th sequence.