CN107784196A - Method based on Artificial Fish Swarm Optimization Algorithm identification key protein matter - Google Patents

Abstract

The invention discloses a method for identifying key proteins based on an artificial fish swarm optimization algorithm, transforming a protein interaction network into an undirected graph, constructing a purified protein interaction network, obtaining ribonucleic acid gene expression values corresponding to proteins, and GO annotation information As well as the degree of protein in the known complex, the purified protein interaction network edges and nodes are processed, the known key proteins are selected as the initial artificial fish, the artificial fish performs foraging behavior, random behavior, tail chasing behavior, aggregation Group behavior and production of key proteins. The method of the present invention can accurately identify key proteins; the simulation experiment results show that the performance of indicators such as sensitivity, specificity, positive predictive value, and negative predictive value is better; compared with other key protein identification methods, the optimized characteristics of artificial fish swarm Combined with the topological features of the protein interaction network to realize the identification process of key proteins and improve the identification accuracy of key proteins.

Description

A Method for Identifying Key Proteins Based on Artificial Fish Swarm Optimization Algorithm

technical field

The invention belongs to the field of biological information, and in particular relates to a method for identifying key proteins based on an artificial fish swarm optimization algorithm.

Background technique

Key proteins are the products of key genes and are an essential part of organisms to maintain life activities. The absence of key proteins will lead to the failure of life activities and even the death of organisms. The prediction and identification of key proteins is a research work of great significance. On the one hand, it helps to study the growth regulation process related to cells; on the other hand, it also has far-reaching significance for disease diagnosis and drug design. Initially, the identification of key proteins was mainly through biological experimental methods, such as single gene knockout and RNA interference. Although the identification of key proteins through these experimental techniques is accurate and effective, it is costly and inefficient. Therefore, in the field of bioinformatics, identifying key proteins through computational methods has become a research hotspot and focus.

At present, there are mainly two methods to realize the identification of key proteins through computational methods: the node centrality method based on network topology, and the method of combining PPI network and biological information data.

The "centrality-lethality" law proposed by Jeong et al. in 2001 points out that the criticality of a protein is closely related to the topological properties of the protein in the protein interaction network, that is, the absence of a protein with more neighbor nodes is more likely to affect the entire network. topology. In short, in the protein network, protein nodes with higher degrees tend to be more critical, and the absence of such proteins is more likely to cause the loss of body functions and produce lethal effects. This law lays the foundation for key protein identification based on network topology. After that, a series of key protein identification methods based on topological centrality were proposed, including degree centrality (Degree Centrality, DC), betweenness centrality (Betweenness Centrality, BC), closeness centrality (Closeness Centrality, CC), feature Vector Centrality (Eigenvector Centrality, EC), Information Centrality (Information Centrality, IC), Subgraph Centrality (Subgraph Centrality, SC). These methods identify key proteins by scoring and sorting the centrality value of all protein nodes in the protein interaction network. However, these centrality methods are highly dependent on the reliability of the protein interaction network, because the protein interaction network is obtained through high-throughput biological experiments, which contains a large number of false positives, which largely affects the identification of key proteins. Accuracy.

Aiming at the shortcomings of the centrality method in identifying key proteins, the researchers proposed some new identification methods to improve the accuracy of identifying key proteins. For example, the PeC key protein identification method combines the protein interaction network with the gene expression profile, the ION key protein identification method combines the homology characteristics of the protein with the protein interaction network, and the UDoNC key protein identification method combines the protein domain and the protein interaction. The interaction network, SCP key protein identification method combines subcellular localization information and protein interaction network. In addition, there are some methods for identifying key proteins based on prior knowledge, such as CPPK and CEPPK, which use some known key proteins as prior knowledge, and judge the key protein of the protein by the closeness of other proteins in the network to the prior knowledge. sex.

Numerous studies have shown that there is a close connection between protein criticality and protein complexes. Hart et al. found through research experiments that the criticality of proteins is not determined by a single protein, but often depends on the function of protein complexes. And the experimental data show that key proteins are often enriched in some complexes. Therefore, a large number of key protein identification methods based on protein complexes and functional modules have been proposed.

Although with the development of bioinformatics, researchers have conducted in-depth research on the identification of key proteins, the accuracy of current identification methods based on network topology is still low, and most methods use a small number of parameters in isolation or piecemeal Or feature analysis of key proteins, lack of overall and global grasp of nodes. In addition, since the protein interaction data obtained through high-throughput technology contains a large number of false positives, it cannot represent the real protein network, so constructing a protein interaction network that can more realistically mimic organisms can help further improve the accuracy of key protein identification .

Based on the defects of the key protein identification methods mentioned above, the main reason is that the reliability of the protein interaction network is not considered, only some features are considered, and the overall and overall grasp is lacking, and the accuracy of key protein identification is low.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, provide a method for identifying key proteins based on an artificial fish swarm optimization algorithm, construct a purified protein interaction network, and identify key proteins with high accuracy.

To achieve the above object, the present invention adopts the following technical solutions:

Include the following steps:

(1) Transform the protein interaction network into an undirected graph:

Transform the protein interaction network into an undirected graph G=(V, E), where V={v _i ,i=1,2,...,n} is the set of nodes v _i , E is the edge e Set, the node v _i represents the protein, and the edge e represents the interaction between proteins;

(2) Construct a purified protein interaction network:

At time point t, if the gene expression value Ep _it of node v _i is greater than gene expression activity threshold Active_Th(i), then node v _i is considered active at time point t, otherwise it is considered that the node is not active at time point t is active; if any two different nodes v and u in V are active at the time point t at the same time, it is considered that the nodes v and u are co-expressed at the time point t; The edges corresponding to the protein interactions that are not co-expressed are deleted to construct a purified protein interaction network;

(3) Process the edges and nodes of the purified protein interaction network: calculate the aggregation coefficient ECC of the edges, the Pearson correlation coefficient PCC of the edges, the GO functional similarity of the edges, and the degree of the nodes in the protein complex;

(4) Select the known key protein composition initial artificial fish:

Let N be the artificial fish population size, m be the number of known key proteins contained in each artificial fish; randomly select m known key proteins from the currently known key proteins to form an artificial fish with prior knowledge; Fish (k) represents the set of known key proteins contained in the k initial artificial fish, k=1,2...N; Cn is the number of candidate key proteins;

(5) Foraging behavior:

Find all the neighbor proteins of the protein in each artificial fish to form a neighbor protein node set Neighbor(k), and the proteins in the set Neighbor(k) and the set Neighbor(l) are different from each other, k=1,2...N ,l=1,2...N,k≠l; for each node v _i in Neighbor(k), according to the formula score1(i)=fitness1(v _i ,Fish(k)) determine the merged artificial fish Fish( The possibility in k), sort the nodes in the neighbor protein node set Neighbor(k) in descending order according to their score1 score, add the protein node with the highest score1 value to Fish(k), and add it to the set at the same time In Add(k); the foraging behavior is repeated Tn times, and Tn protein nodes are added to the initial artificial fish;

(6) Rear-end behavior:

After the foraging behavior is executed, determine the artificial fish in the optimal state according to the formula Score2(k)=fitness2(Add(k)) for each artificial fish, sort all artificial fish in descending order according to their Score2 scores, and the value of Score2 is the highest The artificial fish is the optimal artificial fish Fish(p), p∈[1,N], add the protein node in the set Add(p) corresponding to the optimal artificial fish Fish(p) to the set Candidate;

(7) Group behavior:

Except for the set Add(p) corresponding to the optimal artificial fish Fish(p), the nodes v _i in the set Add(k) corresponding to the remaining artificial fish Fish(k) are calculated according to the formula Score3(i)=fitness3(v _i ) to calculate the score, where k≠p; sort all v _i in descending order according to their Score3 scores, let δ be the crowding factor, and select the first δ protein nodes to add to the set Candidate;

(8) Produce key proteins:

Output the protein nodes in the set Candidate obtained in step (7) as key proteins.

Further, the gene expression threshold Active_Th(i) is obtained from formula (1):

Active_Th(i)=μ(i)+3σ(i)(1-F(i)) Formula (1)

In formula (1), μ(i) is the average gene expression value of node v _i , σ(i) is the standard deviation of gene expression value; F(i)=1/(1+σ ² ) is the weight function.

Further, in step (3), calculate the edge clustering coefficient according to formula (2):

In the formula, N _i , N _j represent the neighbor node sets of nodes v _i , v _j respectively;

Calculate the Pearson correlation coefficient of the side according to formula (3):

In the formula, Ep _it and Ep _jt represent the gene expression values of nodes v _i and v _j at time point t respectively, and μ(i) and μ(j) are the average gene expression values of nodes v _i and v _j , T is the maximum value at the time point;

Calculate the GO functional similarity of edges according to formula (4):

In the formula, GO _i and GO _j represent the GO terms of annotation node v _i and node v _j respectively;

Calculate the degree of node v _i inside the protein complex according to formula (5):

In the formula, V(|C|) represents the set of nodes contained in the protein complex, C _vi represents the protein complex containing node v _i , D _in (v _i , C _vi ) represents the node v _i in the protein complex The degree in the compound C _vi , v _j is the neighbor node of vi _.

Further, the probability fiitness1 of adding node v _i in the set Neighbor(k) to the artificial fish Fish(k) in step (5) is obtained by formula (6):

In the formula, v _j is the protein node in the artificial fish Fish(k), ECC is the aggregation coefficient of the edge between node v _i and node v _j , PCC is the edge between node v _i and node v _j Pearson correlation coefficient of edge, GO_sim is the functional similarity between node v _i and node v _j .

Further, in step (5), if there is no suitable protein node added to the artificial fish during the execution of the foraging behavior, a random behavior is performed, and a protein node is randomly selected to be added to the neighbor protein node set Neighbor(k )middle.

Further, in step (6), the possibility of determining that the artificial fish is in the optimal state fitness2 is obtained from formula (7):

In the formula, Add(k) represents the set of protein nodes added by the kth artificial fish after Tn foraging behaviors.

Further, the set Add(k) is determined in step (7), and the score fitness3 of node v _i in k≠p is obtained by formula (8):

W(v _i ,v _j )＝ECC(v _i ,v _j )×(PCC(v _i ,v _j )+GO_sim(v _i ,v _j )) formula (9)

In formula (8), a and b are coefficients, satisfying a+b=1, Nei(v _i ) means the set of neighbor nodes of node v _i , DIC(v _i ) means node v _i is inside the protein complex degree.

Further, in step (7), δ=Cn-Tn.

Compared with existing methods, the present invention has the following advantages:

1. The present invention selects some known key proteins as prior knowledge, and according to the fact that the key proteins are more likely to connect with each other, the prediction of the key proteins is completed by searching the neighbor nodes that make up the protein of the artificial fish through the foraging behavior of the artificial fish. The topological properties of key proteins in the network are fully considered.

2. In the present invention, when artificial fish perform clustering behavior to score protein nodes, edge clustering coefficient (ECC), Pearson correlation coefficient (PCC), and GO functional similarity (GO_sim) are used, and two The tightness of the connection between interacting proteins, the similarity of gene expression, and the functional correlation of proteins; and the degree of participation of proteins in the complex (DIC), considering the relationship between protein criticality and complexes, and various characteristics The fusion makes the identification of key proteins more precise.

3. The present invention simulates the process of foraging or finding companions in artificial fish schools to identify key proteins, construct a reliable protein interaction network, and comprehensively consider the topological characteristics of protein interaction networks, protein gene expression values, and GO semantic similarity properties, protein complex information and prior knowledge, and adding the optimization mechanism of artificial fish swarms, the use of various features makes the accuracy of the key proteins identified by the present invention better than that of other key protein identification methods currently used high.

4. The method of the present invention can accurately identify key nodes; the simulation experiment results show that the performance of indicators such as sensitivity, specificity, positive predictive value, and negative predictive value is better; compared with other key protein identification methods, artificial fish school The optimization characteristics of the network are combined with the topological characteristics of the node interaction network to realize the identification process of key nodes and improve the identification accuracy of key nodes.

5. Using the present invention can effectively identify key proteins from the protein interaction network, which can not only help understand the growth regulation process of cells and the operation mechanism of life activities, but also have an extremely important theory on how to accurately develop drugs and diagnose and treat diseases value.

Description of drawings

Fig. 1 is the process flow chart of embodiment 1 of the present invention.

Fig. 2 is a partial schematic diagram of key proteins in the whole protein interaction network obtained by using Example 1.

Figure 3 is the situation of key proteins in the standard library corresponding to Figure 2.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

As shown in Figure 1, the method for identifying key proteins based on the artificial fish swarm optimization algorithm of the present invention comprises the following steps:

(1) Transform the protein interaction network into an undirected graph

Transform the protein interaction network into an undirected graph G=(V, E), where V={v _i ,i=1,2,...,n} is the set of nodes v _i , E is the edge e Set, the node v _i represents the protein, and the edge e represents the interaction between proteins;

(2) Construction of purified protein interaction network

At the time point t, if the gene expression value Ep _it of the node v _i is greater than the gene expression activity threshold Active_Th(i), the node v _i is considered to be active at the time point t; otherwise, the node is considered not to be active at the time point t is active; if any two different nodes v and u in V are active at the time point t at the same time, it is considered that the nodes v and u are co-expressed at the time point t; The edges corresponding to the protein interactions without co-expression are deleted, and a new protein interaction network is constructed, that is, the purified protein network;

Ep _it is the gene expression value of node v _i at time point t;

Gene expression activity threshold Active_Th(i) is obtained from formula (1):

Active_Th(i)=μ(i)+3σ(i)(1-F(i)) Formula (1)

In formula (1), μ(i) is the average gene expression value of node v _i , σ(i) is the standard deviation of gene expression value; F(i)=1/(1+σ ² ) is the weight function;

(3) Process the edges and nodes of the purified protein interaction network

Calculate the edge clustering coefficient according to formula (2):

In the formula, N _i , N _j represent the neighbor node sets of nodes v _i , v _j respectively;

Calculate the Pearson correlation coefficient of the side according to formula (3):

In the formula, Ep _it and Ep _jt respectively represent the gene expression values of nodes v _i and v _j at time point t, μ(i) and μ(j) are the average gene expression values of nodes v _i and v _j , T is the maximum value at the time point;

Calculate the GO functional similarity of edges according to formula (4):

In the formula, GO _i and GO _j represent the GO terms of annotation node v _i and node v _j respectively.

Pretreatment of node v _i : Calculate the degree of node v _i inside the protein complex according to formula (5):

In the formula, V(|C|) represents the set of protein nodes contained in the protein complex, C _vi represents the protein complex containing node v _i , D _in (v _i , C _vi ) represents the node v _i in degree in the protein complex C _{vi ,} v _j is the neighbor node of vi;

(4) Select known key proteins as the initial artificial fish

Let N be the population size of the artificial fish, and m be the number of known key proteins contained in each artificial fish; randomly select m known key proteins in the standard library (currently known key proteins) to form a prior knowledge Artificial fish; Fish(k) represents the set of known key proteins contained in the k initial artificial fish, k=1,2...N; Cn is the number of candidate key proteins;

(5) Foraging behavior

The artificial fish looks for food within the visible range, that is, it looks for the protein that directly interacts with the protein of the artificial fish, finds out all the neighbor proteins of the protein in each artificial fish, and forms the neighbor protein node set Neighbor(k), and the set Neighbor( k) is different from the nodes in the set Neighbor(l) (k=1, 2…N, l=1, 2…N, k≠l), for each node v _i in Neighbor(k) According to the formula score1(i)=fitness1(v _i , Fish(k)) to determine the possibility of merging into the artificial fish Fish(k), the nodes in the neighbor protein node set Neighbor(k) are calculated according to their score1 scores Sort in descending order, add the node with the highest value of score1 to Fish(k), and add it to the set Add(k);

Random behavior: If no suitable protein is added to the artificial fish during the execution of foraging behavior, random behavior is performed, and a protein node is randomly selected to be added to the set Neighbor(k);

The foraging behavior is repeated Tn times, that is, Tn protein nodes are added to the initial artificial fish;

(6) Rear-end behavior

After the foraging behavior is executed, determine the artificial fish in the optimal state according to the formula Score2(k)=fitness2(Add(k)) for each artificial fish, sort all artificial fish in descending order according to their Score2 scores, and the value of Score2 is the highest The artificial fish is the optimal artificial fish Fish(p), p∈[1,N], add the protein node in the set Add(p) corresponding to the optimal artificial fish Fish(p) to the set Candidate;

(7) Crowd behavior

Except for the set Add(p) corresponding to the optimal artificial fish Fish(p), the nodes v _i in the set Add(k)(k≠p) corresponding to the other artificial fish Fish(k)(k≠p) according to The formula Score3(i)=fitness3(v _i ) calculates the score, and sorts all v _i in descending order according to their Score3 scores, let δ be the crowding factor, δ=Cn-Tn, select the first δ protein nodes to add to the collection Candidate;

(8) Produce key proteins

Export the proteins in the set Candidate as key proteins.

In the step (5) of the present invention, the possibility fiitness1 of the collection Neighbor (k) node v _i added to the artificial fish Fish (k) is obtained by formula (6):

In the formula, v _j is the node in the artificial fish Fish(k); ECC is the aggregation coefficient of the edge between node v _i and node v _j , which is obtained by formula (2); PCC is the relationship between node v _i and node v j The Pearson correlation coefficient of the edge between points v _j is obtained by formula (3); GO_sim is the functional similarity between node v _i and node v _j , which is obtained by formula (4).

In the step (6) of the present invention, the possibility fitness2 of determining that the artificial fish is in the optimum state is obtained by formula (7):

In the formula, Add(k) represents the set of protein nodes added by the kth artificial fish after Tn foraging behaviors, and fitness1(v _i , Fish(k)) is shown in formula (6).

In the step (7) of the present invention, the score fitness3 of the protein node in the set Add(k) (k≠p) is determined to be obtained by formula (8):

W(v _i ,v _j )＝ECC(v _i ,v _j )×(PCC(v _i ,v _j )+GO_sim(v _i ,v _j )) formula (9)

In formula (8), a and b are coefficients, satisfying a+b=1, Nei(v _i ) represents the set of neighbor nodes of node v _i , DIC(v _i ) represents the position of node v _i inside the complex The degree is obtained by formula (5).

The present invention is described in further detail below by specific embodiment:

Example 1

Taking protein network as an example, the steps of a method for identifying key proteins based on artificial fish swarm optimization algorithm are as follows:

In this example, the yeast data set (DIP 20160114 version) collected from the DIP database was used as the simulation data set, and the DIP data contained 5028 proteins and 22303 interaction relationships. The gene expression dataset is collected from the yeast metabolic expression dataset GSE3431 in the GEO database, which includes 9336 genes, gene values at 36 time points in 3 cycles, covering 95% of the proteins in DIP. GO data includes annotation spectrum and SGD. Known protein complex information is from CYC2008, including 408 protein complexes, covering 1492 proteins. Key protein data are obtained by integrating data from four databases: MIPS, SGD, DEG and SGDP , contains a total of 1285 key proteins, corresponding to 1152 of the 5028 proteins are key proteins, and the rest are regarded as non-key proteins. The experimental platform is Windows 10 operating system, Intel Core i5-6600 dual-core 3.31GHz processor, 8GB physical memory, and uses Matlab R2014a software to realize the method of the present invention.

1. Transform the protein interaction network into an undirected graph

Transform the protein interaction network containing 5028 proteins and 22303 interactions into an undirected graph G=(V, E), where V={v _i ,i=1,2,…,5028} is the knot The set of points v _i , E is the set of edge e, the node v _i represents the protein, and the edge e represents the interaction between proteins.

2. Construction of purified protein interaction network

At the time point t, if the gene expression value Ep _it of the node v _i is greater than the gene expression activity threshold Active_Th(i), the node v _i is considered to be active at the time point t; otherwise, the node is considered not to be active at the time point t is active; if any two different nodes v and u in V are active at the time point t at the same time, it is considered that v and u are co-expressed at the time point t; the gene expression activity threshold Active_Th(i) is given by formula (1) get:

Active_Th(i)=μ(i)+3σ(i)(1-F(i)) Formula (1)

In formula (1), μ(i) is the average gene expression value of node v _i , σ(i) is the standard deviation of gene expression value; F(i)=1/(1+σ ² ) is the weight function. Through the above processing, correspond to the original protein interaction network, delete the protein interaction that is not co-expressed at all time points, and form a new protein interaction network with 5028 protein nodes and 9576 edges, that is, the purified protein interaction network.

3. Process the edges and nodes of the purified protein interaction network

Calculate the edge clustering coefficient according to formula (2):

In the formula, N _i and N _j represent the number of neighbor nodes of points v _i and v _j respectively, and d _i and d _j are the degrees of points v _i and v _j respectively; the Pearson correlation of edges is calculated according to formula (3) coefficient:

In the formula, EP _it , EP _jt represent the gene expression value of node v _i , v _j at time point t, μ(i), μ(j) are the average gene expression value of node v _i , v _j , T is the maximum value at the time point; calculate the GO function similarity of edges according to formula (4):

In the formula, GO _i and GO _j represent the GO terms annotating protein v _i and protein v _j , respectively.

Pretreatment of node v _i : i=1, 2,...,5028, given a certain i, the degree of participation of node v _i in the protein complex can be calculated, and the node can be calculated according to formula (5) The degree of v _i inside the protein complex:

In the formula, V(|C|) represents the set of protein nodes contained in the protein complex, Cv _i represents the protein complex containing protein v _i , D _in (v _i ,Cv _i ) represents the protein v _i in the protein complex The degree of object Cv _i , v _j is the neighbor node of v _i .

4. Select known key proteins to form the initial artificial fish

Let N be the size of the artificial fish population, m be the number of known key proteins contained in each artificial fish; for each artificial fish, randomly select 100 known key proteins from the 1152 key proteins in the standard library to form a Artificial fish of empirical knowledge, Fish(k) represents the protein set contained in the kth initial artificial fish; in this example, N=100, m=100; Cn is the number of candidate key proteins.

5. Foraging behavior

The artificial fish looks for food within the visible range, that is, it looks for proteins that directly interact with the artificial fish protein, finds out all the neighbor proteins Neighbor(k) of the protein in each artificial fish, and sets Neighbor(k) and set Neighbor(l ) are different from each other (k=1,2...100, l=1,2...100, k≠l), for each protein v _i in Neighbor(k) according to score1(i)=fitness1(v _i , Fish(k)) determines the possibility of merging into the artificial fish Fish(k), sorts the nodes in the protein node set Neighbor(k) in descending order according to their score1 scores, and ranks the node with the highest score1 value Add to Fish(k) and add to the set Add(k) at the same time, where score1(i) is the intimacy between protein v _i and all proteins in the artificial fish, and the intimacy is obtained by formula (6):

In the formula, v _j is the protein node in the artificial fish Fish(k), ECC is the aggregation coefficient of the edge between node v _i and node v _j obtained from the formula (2), PCC is the connection between node v _i and node v j The Pearson correlation coefficient of the edge between points v _j is obtained by formula (3), and GO_sim is the functional similarity between node v _i and node v _j obtained by formula (4).

If no suitable protein is added to the artificial fish during the execution of the foraging behavior, a random behavior is performed, and a protein node is randomly selected to be added to the set Neighbor(k). The foraging behavior (or random behavior) is repeated Tn times, that is, Tn protein nodes are added to the initial artificial fish.

6. Rear-end behavior

After the foraging behavior (or random behavior) is executed, determine the artificial fish in the optimal state according to the formula Score2(k)=fitness2(Add(k)) for each artificial fish, and sort all artificial fish in descending order according to their Score2 scores , the artificial fish with the highest value of Score2 is the optimal artificial fish Fish(p)(p∈[1,100]), and the protein node in the set Add(p) corresponding to the optimal artificial fish Fish(p) is added to In the set Candidate, fitness2 represents the fitness of each artificial fish after adding protein, which can be obtained from formula (7):

In the formula, Add(k) represents the set of protein nodes added by the kth artificial fish after Tn foraging behaviors, and fitness1(v _i , Fish(k)) is shown in formula (6).

7. Crowd behavior

Except for the set Add(p) corresponding to the optimal artificial fish Fish(p), the protein node v _i in the set Add(k) (k≠p) corresponding to the other artificial fish Fish(k) (k≠p) Calculate the score according to the formula Score3(i)=fitness3(v _i ), sort all v _i in descending order according to their Score3 scores, let δ(δ=Cn-Tn) be the crowding factor, and select the top δ protein knots Points are added to the set Candidate, and fitness3 represents the score of the protein node in the set Add(k) (k≠p), which is obtained by formula (8):

W(v _i ,v _j )＝ECC(v _i ,v _j )×(PCC(v _i ,v _j )+GO_sim(v _i ,v _j )) formula (9)

In the formula, a, b are coefficients, a=0.8, b=0.2, Nei(v _i ) represents the set of neighbor nodes of node v _i , DIC(v _i ) represents the degree of node v _i in the complex by Formula (5) is obtained.

9. Produce key proteins

Export the proteins in the set Candidate as key proteins.

In order to verify the effectiveness of the present invention, the inventor adopts the method for identifying key proteins by the artificial fish swarm optimization algorithm in Example 1 of the present invention to identify key proteins in the protein network in the DIP database, and the number of candidate key proteins (Cn) is 100, 200, 300, 400, 500 and 600, the correctly identified key proteins were analyzed. In this experiment, we took 100 known key proteins for each artificial fish as prior knowledge. In view of the fact that during the experiment as a priori The known key proteins are randomly selected, so the experiment is carried out 50 times, and the average value of the 50 experimental results is taken as the final result. The results are shown in Table 1, Figure 2 and Figure 3. Compare the recognition accuracy of the results identified by the above methods. Figure 2 shows the distribution of some of the key proteins identified by the present invention in the network, and Figure 3 shows the corresponding part of the standard library in Figure 2 .

Table 1 The comparison between the present invention and other methods for identifying key proteins in accuracy

Table 1 shows that the present invention compares the identified 100, 200, 300, 400, 500, 600 proteins as candidate key proteins with the key proteins in the standard library, and the identification results with other current methods for identifying key proteins contrast. When identifying the first 600 key proteins, compared with the remaining 8 key protein identification methods, it shows that the present invention has higher prediction accuracy. It can be seen from Table 2 that the present invention can effectively identify key proteins, and the number of candidate key proteins ranges from 100 to 600, and the present invention has the highest recognition accuracy. Fig. 2 shows the positions of some key proteins identified by the present invention in the protein interaction network. In Fig. 2, the key proteins correctly identified by the present invention are those with a dark background, the key proteins incorrectly identified with a light background, and the non-key proteins are those with a white background. Figure 3 is the situation of key proteins in the standard library corresponding to Figure 2. Through the comparison of Figure 2 and Figure 3, it can be found that the wrong proteins identified by the present invention include "YDR283C" and "YPL246C", and the key proteins that were not identified include "YBR152W". If some known key proteins are used as prior knowledge, the method of the present invention can correctly identify most of the key proteins around the prior knowledge.

The method of the present invention is based on the artificial fish swarm optimization algorithm to identify key proteins, transforms the protein interaction network into an undirected graph, constructs a purified protein interaction network, obtains the ribonucleic acid gene expression value corresponding to the protein, GO annotation information, and The degrees in the complex are known, and the purified protein interaction network edges and nodes are processed, and the known key proteins are selected as the initial artificial fish. key protein. The method of the present invention can accurately identify key proteins; the simulation experiment results show that the performance of indicators such as sensitivity, specificity, positive predictive value, and negative predictive value is better; compared with other key protein identification methods, the optimized characteristics of artificial fish swarm Combined with the topological features of the node interaction network to realize the identification process of key proteins, and improve the identification accuracy of key proteins.

The above is a preferred embodiment of the present invention. Through the above description, relevant workers in the technical field can make various improvements and replacements without departing from the technical principles of the present invention. These improvements and replacements should also be regarded as protection scope of the present invention.

Claims (8)

1. the method based on Artificial Fish Swarm Optimization Algorithm identification key protein matter, it is characterised in that：Comprise the following steps：

(1) protein-protein interaction network is converted into non-directed graph：

Protein-protein interaction network is changed into a non-directed graph G=(V, E), wherein, V={ v_i, i=1,2 ..., n } it is knot Point v_iSet, E be side e set, node v_iProtein is represented, side e represents the interaction between protein；

(2) protein-protein interaction network of structure purification：

In time point t, node v_iGene expression values Ep_itIf being more than activity of gene expression threshold value A ctive_Th (i), recognize For node v_iIt is active in time point t, otherwise it is assumed that the node does not have activity in time point t；If any two is different in V Node v, u time point t simultaneously it is active, then it is assumed that the node v under time point t, u co-express；By in non-directed graph in institute All leave out under having time point without the side corresponding to the protein interaction of coexpression, the protein for building a purification is mutual Act on network；

(3) side and node of the protein-protein interaction network of purification are handled：Calculate while convergence factor ECC, while The degree of Pearson correlation coefficients PCC, the GO functional similarity and node on side inside protein complex；

(4) key protein matter composition original manual fish known to choosing：

It is artificial fingerling group scale to make N, and m is the quantity of the known key protein matter included in every Artificial Fish；It is being currently known Key protein matter in randomly select the Artificial Fish that the known key protein matter of m form a priori；Fish (k) represents the The known key protein matter set included in k bar original manual fishes, k=1,2 ... N；Cn is the number of candidate key protein；

(5) foraging behavior：

All neighbours' protein of protein in every Artificial Fish are found out, form neighbours protein node set Neighbor (k), And set Neighbor (k) and the protein in set Neighbor (l) are different, k=1,2 ... N, l=1,2 ... N, k ≠l；For each node v in Neighbor (k)_iAccording to formula score1 (i)=fitness1 (v_i, Fish (k)) determine to close And to the possibility in Artificial Fish Fish (k), by the node in neighbours protein node set Neighbor (k) according to it Score1 scores carry out descending sort, and score1 value highest protein node is added in Fish (k), is added to simultaneously In set Add (k)；Foraging behavior is repeated Tn times, and Tn protein node is added into original manual fish；

(6) knock into the back behavior：

After foraging behavior performs, every Artificial Fish is determined to be according to formula S core2 (k)=fitness2 (Add (k)) The Artificial Fish of optimum state, descending sort, Score2 value highest people are carried out according to its Score2 score to all Artificial Fishs Work fish is optimal Artificial Fish Fish (p), p ∈ [1, N], corresponding in the set Add (p) of optimal Artificial Fish Fish (p) Protein node is added in set Candidate；

(7) bunch behavior：

In addition to set Add (p) corresponding to optimal Artificial Fish Fish (p), by set Add (k) corresponding to remaining Artificial Fish Fish (k) In node v_iAccording to formula S core3 (i)=fitness3 (v_i) calculate score, wherein k ≠ p；To all v_iAccording to it Score3 scores carry out descending sort, and it is the crowding factor to make δ, and δ protein node for selecting to come above is added to set In Candidate；

(8) key protein matter is produced：

Exported the protein node in the set Candidate obtained by step (7) as key protein matter.

2. the method as claimed in claim 1 based on Artificial Fish Swarm Optimization Algorithm identification key protein matter, it is characterised in that：

Gene expression threshold value A ctive_Th (i) is obtained by formula (1)：

Active_Th (the i)=σ of μ (i)+3 (i) (1-F (i)) formula (1)

μ (i) is node v in formula (1)_iAverage gene expression value, σ (i) are the standard deviations of gene expression values；F (i)=1/ (1+ σ²) It is weight function.

3. the method as claimed in claim 1 based on Artificial Fish Swarm Optimization Algorithm identification key protein matter, it is characterised in that：Step Suddenly in (3), the convergence factor on side is calculated by formula (2)：

In formula, N_i,N_jNode v is represented respectively_i,v_jNeighbor node collection；

The Pearson correlation coefficients on side are calculated by formula (3)：

In formula, Ep_itAnd Ep_jtNode v is represented respectively_i,And v_jGene expression values in time point t, μ (i) and μ (j) are node v_i And v_jAverage gene expression value, T be time point maximum；

The GO functional similarity on side is calculated by formula (4)：

In formula, GO_i,GO_jAnnotation node v is represented respectively_iWith node v_jGO terms；

Node v is calculated by formula (5)_iDegree inside protein complex：

In formula, V (| C |) represent the node set included in protein complex, C_viExpression includes node v_iProtein answer Compound, D_in(v_i, C_vi) represent node v_iIn protein complex C_viIn degree, v_jIt is v_iNeighbor node.

4. the method as claimed in claim 1 based on Artificial Fish Swarm Optimization Algorithm identification key protein matter, it is characterised in that：Step Suddenly node v in (5) middle set Neighbor (k)_iThe possibility fiitness1 being added in Artificial Fish Fish (k) is obtained by formula (6) Arrive：

V in formula_jIt is the protein node inside Artificial Fish Fish (k), ECC is node v_iWith node v_jBetween side aggregation system Number, PCC is node v_iWith node v_jBetween side Pearson correlation coefficients, GO_sim is node v_iWith node v_jBetween work( Can similitude.

5. the method as claimed in claim 1 based on Artificial Fish Swarm Optimization Algorithm identification key protein matter, it is characterised in that：Step Suddenly in (5), if there is no suitable protein node to be added in Artificial Fish in foraging behavior implementation procedure, perform random Behavior, one protein node of random selection are added in neighbours protein node set Neighbor (k).

6. the method as claimed in claim 1 based on Artificial Fish Swarm Optimization Algorithm identification key protein matter, it is characterised in that：Step Suddenly the possibility fitness2 that (6) middle determination Artificial Fish is in optimum state is obtained by formula (7)：

In formula, Add (k) represents that kth bar Artificial Fish passes through the protein node set that Tn foraging behavior is added.

7. the method as claimed in claim 1 based on Artificial Fish Swarm Optimization Algorithm identification key protein matter, it is characterised in that：Step Suddenly node v in determination set Add (k) in (7), k ≠ p_iScore fitness3 obtained by formula (8)：

W(v_i,v_j)=ECC (v_i,v_j)×(PCC(v_i,v_j)+GO_sim(v_i,v_j)) formula (9)

In formula (8), a, b are coefficients, meet a+b=1, Nei (v_i) represent node v_iNeighbor node set, DIC (v_i) represent knot Point v_iDegree inside protein complex.

8. the method as claimed in claim 1 based on Artificial Fish Swarm Optimization Algorithm identification key protein matter, it is characterised in that：Step Suddenly δ=Cn-Tn in (7).