A method to identify key clusters in microbial co-occurrence networks

Abstract

The present invention discloses a method for identifying key clusters in a microbial co-occurrence network, which relates to the field of microbial technology, collects and processes microbial flora species data, identifies differential flora species, and constructs a microbial co-occurrence network; identifies the number of all connected networks in the microbial co-occurrence network, and divides them into recognition modules and isomorphic modules; uses a community discovery algorithm to cluster the connected networks in the recognition module, obtains bacterial species clusters with similar structural characteristics, summarizes bacterial species characteristics of the same cluster, and discovers key clusters; for the connected networks in the isomorphic modules, uses a graph matching technology to match subgraphs with similar structures from the connected networks of the recognition module, and further discovers key clusters. The present invention identifies microbial clusters with similar structures or functions to the maximum extent, matches similar networks for small microbial networks, and analyzes the potential association mechanisms and common characteristics of these clusters in combination with flora biological characteristics, so as to assist in targeted intervention based on similar clusters in microbial networks.

Description

Technical Field

[0001] The present invention relates to the field of microbial technology, in particular to a method for identifying key clusters in a microbial co-occurrence network. Background Art

[0002] Microorganisms such as intestinal flora play an important role in the occurrence, development, prevention and control of chronic non-communicable diseases. In addition to the biological functions of microorganisms themselves and their metabolites, the mutually beneficial symbiotic relationship and ecological succession within the microbial flora make it easy for microorganisms with similar structures or functions or close associations to aggregate into clusters with common characteristics. Finding clusters with similar characteristics based on microbial networks is of great significance for assisting targeted intervention of microbial clusters, preventing the occurrence and development of chronic diseases, or effectively controlling chronic diseases.

[0003] In the existing technology, there is a method of cluster identification of microbial co-occurrence networks through community discovery technology. However, since microbial networks are relatively large and complex, especially when there are dozens or hundreds of microbial networks showing non-connectivity characteristics, relying solely on community discovery for cluster identification may lead to difficulties in operation or low efficiency in identifying key clusters in small cluster networks. Summary of the invention

[0004] The purpose of the present invention is to provide a method for identifying key clusters in a microbial co-occurrence network, so as to solve the problem that cluster identification in the prior art that only relies on community discovery may lead to difficult operation or low efficiency of key cluster identification in a small cluster network.

[0005] The present invention solves the above problems through the following technical solutions:

[0006] A method for identifying key clusters in a microbial co-occurrence network, comprising:

[0007] Step S100, collecting and processing microbial flora data, and identifying differential bacterial species;

[0008] Step S200: constructing a microbial co-occurrence network for the identified differential bacterial species;

[0009] Step S300, identifying the number N of all connected networks in the microbial co-occurrence network, dividing the connected networks whose number of nodes of the bacterial community is greater than or equal to a preset value into recognition modules, and defining the remaining connected networks as isomorphic modules; and sorting the connected networks in the recognition modules and isomorphic modules from large to small according to the size of the bacterial community;

[0010] Step S400: using a community discovery algorithm to perform cluster discovery on the connected network in the recognition module, obtaining bacterial species clusters with similar structural characteristics, summarizing bacterial species characteristics of the same cluster, and discovering key clusters;

[0011] Step S500: For the connected networks in the isomorphic modules, a graph matching technique is used to match key clusters with similar structures from the connected networks of the recognition modules.

[0012] Furthermore, the step S100 specifically includes:

[0013] Step S110, collecting a sample containing microbial flora species, extracting genomic deoxynucleotide DNA from the sample, and measuring the genomic DNA content;

[0014] Step S120, randomly breaking the DNA, performing end repair, adding an A tail (adding a single adenine (A) nucleotide), adding a sequencing adapter, purification, and polymerase chain reaction (PCR) amplification to complete library preparation and quantification;

[0015] Step S130, mixing different libraries according to effective concentrations and performing double-end sequencing;

[0016] Step S140, performing species taxonomy annotation to obtain microbial flora species;

[0017] Step S150: perform basic feature analysis on the relative abundance data at the species level of the microbial flora, including species proportion and diversity analysis, and use linear discriminant analysis (LDA) to identify differential species.

[0018] Furthermore, the step S200 is specifically as follows: for the identified differential bacterial species, under the condition that the correlation is greater than a set correlation strength threshold and the correlation is statistically significant, constructing a microbial co-occurrence network based on the correlation adjacency matrix;

[0019]

[0020] Among them, i and j are two bacterial colonies, A ij is the association between bacterial colony i and bacterial colony j. When the value range is greater than the predetermined association strength threshold, A ij =1, otherwise A ij =0, the association strength threshold r≥0.3 can be predetermined.

[0021] Furthermore, the step S300 specifically includes:

[0022] Step S310, using the adjacency matrix, identifying the number N of all connected networks in the microbial co-occurrence network;

[0023] Step S320, sort the N connected networks from large to small according to the size of the bacterial colony, and form the M1 connected networks whose number of nodes is greater than or equal to the preset value into an identification module G (Vg, Eg), where Vg represents the vertex of the bacterial colony in the identification module, and Eg represents the edge of the bacterial colony connection relationship in the identification module; form the M2 connected networks whose number of nodes is less than the preset value into an isomorphic module Q (Vq, Eq), where Vq represents the vertex of the bacterial colony in the isomorphic module, and Eq represents the edge of the bacterial colony connection relationship in the isomorphic module;

[0024] Step S330, sort the M1 connected networks in the recognition module and the M2 connected networks in the isomorphic module from large to small according to the number of bacterial species.

[0025] Furthermore, the step S310 specifically includes:

[0026] Step A: For bacterial colony i, if there is a bacterial colony j that is not connected to bacterial colony i, and the bacterial colony connected to bacterial colony j in order n is not connected to bacterial colony i, then the adjacency matrix is ​​divided into two at bacterial colony i; where n is the longest path distance;

[0027] Step B: Repeat step A for the regenerated adjacency matrix until it cannot be divided any further, exclude the adjacency matrix with the number of rows and columns less than the set value, and identify N connected networks.

[0028] Furthermore, the step S400 is specifically as follows: using the modularity optimization method Louvain or the physical model-based method Spinglass to perform cluster discovery on the connected network in the identification module, dividing the key clusters that affect the network structure, obtaining bacterial species clusters with similar structural characteristics, summarizing the bacterial species characteristics of the same cluster, and discovering key clusters in combination with centrality index analysis.

[0029] Furthermore, the step S500 specifically includes:

[0030] Step S510: Based on the topological structure information in the isomorphic modules, similar nodes or connection modes in the isomorphic modules are mapped to the identification module;

[0031] Step S520, determining whether there is a subgraph match between the connected network of the isomorphic module and the connected network in the identification module, and performing subgraph isomorphism analysis. If there is an isomorphic graph, then they have the same isomorphism class;

[0032] Step S530: Map each isomorphic module to the cluster divided by the identification module, and obtain all matched key clusters based on the result of isomorphic class identification.

[0033] Furthermore, the step S510 specifically includes:

[0034] Step S511: For the x-th connected network Qx in the isomorphic module, compare it with the connected network Gy in the identification module in order, where x∈(1,2,…,M1) and y∈(1,2,…,M2);

[0035] Step S512: For each node in Qx, there must be a corresponding node in Gy; for each edge in Qx, there must be an edge in Gy, and the nodes at both ends of them correspond one to one; if there is an edge between any two nodes in Qx, there is also an edge between the two nodes in Qx;

[0036] Step S513 traverses the connected networks in the identification module and searches for all connected networks in Gy that are isomorphic to it;

[0037] Step S514: traverse all connected networks in the isomorphic modules to expand the number of all key clusters in the bacterial community network.

[0038] Furthermore, the step S520 specifically includes: if MQm*m and MGn*n are respectively an adjacency matrix composed of a connected network Q in the isomorphic module and G of all connected networks in the identification module (such as an association matrix composed of the connection relationship of microorganisms), where M is the adjacency matrix, Q contains m bacterial species, and G contains n bacterial species. If a graph matching technology based on subgraph isomorphism is adopted, a connected network with a similar structure to MQm*m can be matched from MGn*n, that is, there is a subgraph matching of the connected network Q->connected network G.

[0039] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0040] (1) The present invention divides the microbial network into recognition modules and isomorphic modules. For medium and large (cluster) networks with a large number of nodes, community discovery technology is used to perform further cluster detection to maximize the identification of microbial clusters with similar structures or functions, and further matches small (cluster) networks with similar microbial networks. For the discovered clusters, the potential association mechanisms and common characteristics of these clusters are analyzed in combination with the biological characteristics of the microbial community to assist in targeted intervention based on similar clusters in the microbial network.

[0041] (2) The present invention effectively realizes the detection of similar structures and features of small cluster network structures in large cluster network structures, thereby improving the effectiveness and accuracy of key cluster identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flow chart of the present invention;

[0043] Figure 2 A microbial co-occurrence network composed of exemplary bacterial species of the present invention;

[0044] Figure 3 Schematic diagram of the identification of key clusters in the microbial co-occurrence network based on community discovery and subgraph isomorphism, where (A) is the community discovery in the recognition module; (B) is the exploration of whether isomorphic modules can be mapped to the recognition module. DETAILED DESCRIPTION

[0045] The present invention is further described in detail below in conjunction with examples, but the embodiments of the present invention are not limited thereto.

[0046] Embodiment 1:

[0047] Combined with Figure 1 As shown in the figure, a method for identifying key clusters in a microbial co-occurrence network is used. Taking intestinal metagenomic data as an example, human fecal samples are collected, and the genomic deoxyribonucleic acid DNA of the samples is extracted by CTAB (hexadecyl trimethylammonium bromide) extraction method. The DNA purity and integrity are analyzed by 1% agarose gel electrophoresis, and the DNA content is further determined. The library is constructed using the kit NEBNext UltraTM DNA LibraryPrep Kit for Illumina. The ultrasonic crusher Covaris M220 randomly interrupts the genomic DNA, and the library is prepared and quantified after end repair, A-tailing, sequencing adapter addition, purification, polymerase chain reaction PCR amplification and other steps. After the library inspection is qualified, different libraries are mixed according to the effective concentration and double-end sequencing is performed using the Illumina PE150 sequencing platform.

[0048] The open source software kneadData was used for quality control. Species taxonomy annotation was performed based on the standard database of Kraken 2 (v.2.1.2). MEGAHIT (v.1.2.9) and MetaBAT2 (v.2.2.15) were used to assemble and bin metagenomic reads to obtain the detected intestinal flora species.

[0049] R software was used to perform basic feature analysis on the relative abundance data of intestinal flora at the species level, including species proportion and diversity analysis, and LEfSe was used to identify differential bacterial species using linear discriminant analysis (LDA).

[0050] For the identified differential bacterial species, under the conditions of a certain correlation strength threshold and statistical significance, the igraph package was used to construct a co-occurrence network based on the correlation adjacency matrix;

[0051]

[0052] The adjacency matrix is ​​used to identify the number of all connected networks in the microbial network; that is, the dichotomy method is adopted, that is, for bacterial colony i, if there is a bacterial colony j that is not connected to it, and the bacterial colony that is connected to bacterial colony j in order n (n is the longest path distance) is not connected to it, then the adjacency matrix is ​​divided into two at bacterial colony i.

[0053] Repeat the above process for the regenerated adjacency matrix until it can no longer be divided. After excluding the adjacency matrix with less than 5 rows and columns, N connected networks are identified.

[0054] Furthermore, M1 connected networks among the N connected networks are determined to form a recognition module G (Vg, Eg), where Vg represents the vertex representing the bacterial community in the recognition module, and Eg represents the edge representing the connection relationship between the bacterial community in the recognition module; the remaining M2 connected network sets form an isomorphic module Q (Vq, Eq), where Vq represents the vertex representing the bacterial community in the isomorphic module, and Eq represents the edge representing the connection relationship between the bacterial community in the isomorphic module.

[0055] From an operational perspective, N connected networks are sorted from large to small according to the size of the bacterial community in the network, and M1 (larger, such as the number of nodes ≥ 5) connected networks are defined as identification modules, and M2 (smaller, such as the number of nodes < 5) networks are defined as isomorphic modules.

[0056] Sort the M1 connected networks in the recognition module and the M2 connected networks in the isomorphic module from large to small according to the number of bacterial species in the network;

[0057] Since the identification module is a large microbial cluster network, combined with the topological structure information of the microbial network, the Louvain, Spinglass, Louvain and other algorithms are used to discover communities (clusters) in the connected network. The community discovery algorithm is used to divide the key clusters that affect the network structure.

[0058] Based on the results of community mining, bacterial species clusters with similar structural characteristics were obtained, and the characteristics of bacterial species in the same cluster were summarized. The key clusters were discovered by combining centrality indicators such as degree centrality.

[0059] For key clusters, we can combine the previous level (genus) to determine whether it is across bacterial genera, indicating the effect of this bacterial genus on other bacterial genera.

[0060] Furthermore, for the M2 isomorphic modules, a graph matching technique based on subgraph isomorphism is used to match the networks with similar structures from the M1 recognition modules. Specifically:

[0061] Based on the topological structure information in the isomorphic modules, similar nodes or connections in the isomorphic modules are mapped to the recognition modules. The module division process is as follows:

[0062] For the x-th connected network Qx in the isomorphic module, it is sequentially compared with the connected network Gy in the identification module;

[0063] For each node in Qx, there must be a corresponding node in Gy; for each edge in Qx, there must be an edge in Gy, and the nodes at both ends of them correspond one to one; for any two nodes in Qx, if there is an edge between their corresponding nodes in Gy, then there is also an edge between these two nodes in Qx.

[0064] Traverse the connected networks in the identification module and search for all connected networks in Gy that are isomorphic to it;

[0065] Traverse all connected networks in the isomorphic module and expand the number of all key clusters C in the bacterial network;

[0066] Ullmann and VF2 algorithms are commonly used in graph matching techniques, which can find subgraphs in the recognition module that match the large cluster network one by one. If MQm*m and MGn*n are the association matrices composed of a connected network Q in the isomorphic module and G of all connected networks in the recognition module, respectively, where M is the adjacency matrix, Q contains m bacterial species, and G contains n bacterial species. If a graph matching technique based on subgraph isomorphism is used, a connected network with a similar structure to MQm*m can be matched from MGn*n, that is, there is a subgraph match of connected network Q—>connected network G.

[0067] In R language programs, functions such as subgraph_isomorphic and subgraph_isomorphisms in the Topology package can be used to perform subgraph isomorphism analysis, and isomorphic graphs always have the same isomorphism class.

[0068] Map each isomorphic module to the cluster divided by the recognition module as much as possible, and obtain all matched key clusters based on the results of isomorphic class recognition.

[0069] The key clusters that can be matched with the clusters in the recognition module and the key clusters captured in the recognition module are further analyzed for biological characteristics. Similar clusters can be assigned a theme, and the unmatched networks do not define the theme type. Similar key clusters can provide reference targets for microbiome intervention.

[0070] Taking bacterial communities as an example, this paper proposes a new method for integrating community discovery and graph matching technology to identify key bacterial community clusters in microbial co-occurrence networks. It has the following advantages:

[0071] 1. The co-occurrence network constructed based on the correlation between microorganisms can reflect the fine associations between microorganisms. Combining complex network community discovery and graph matching technology to capture microbial network clusters with similar characteristics is conducive to the implementation of targeted intervention.

[0072] 2. Through training, the microbial network is divided into recognition modules and isomorphic modules, taking into account the non-connectivity of the microbial network and the accuracy of cluster division, making it possible to use the microbial network comprehensively and efficiently.

[0073] 3. Complex network centrality indicators such as degree centrality can further discover the core nodes in the microbial co-occurrence network, which is conducive to providing targets for intervention implementation and assisting in targeted intervention of microorganisms.

[0074] 4. Combined with the biological characteristics of microorganisms, the characteristics of microbial clusters are described to discover the potential mechanism of action between microorganisms in microbial clusters.

[0075] The following is the verification process of this method:

[0076] Based on the intestinal metagenomic data of obese and healthy people in a natural population cohort in the southwest region, the present invention identified 57 differential bacterial species between obese and healthy people in an R language program (version 4.1.3).

[0077] The Spearman correlation coefficients between different strains were calculated, and a correlation matrix was constructed after screening according to a certain correlation strength threshold (r=0.5) and statistically significant correlation (P<0.05).

[0078] The igraph package was used to visualize the network graph and construct the co-occurrence network of the microbiome composed of bacterial species (Co-occurrence Network), see Figure 2 .

[0079] The adjacency matrix is ​​used to identify the number of all connected networks in the strain network; that is, the dichotomy method is used, that is, for strain i, if there is a strain j that is not connected to it, and the strain that is connected to strain j in order n (n is the longest path distance) is also not connected to it, then the adjacency matrix is ​​divided into two at strain i.

[0080] Repeat the above process for the regenerated adjacency matrix until it can no longer be divided, discard the adjacency matrix with less than 5 rows and columns, and find all the connected networks in the network.

[0081] The Louvain algorithm is used to discover communities in connected networks. The degree centrality and other metrics are used to analyze the key clusters in the bacterial clusters, and to determine the characteristics of the key clusters and their connections with other clusters.

[0082] Furthermore, the isomorphic modules (smaller networks, with the number of nodes <5) are mapped in the identification modules (larger networks, with the number of nodes ≥5). The Topology package is used in the R language program to perform subgraph isomorphism analysis, and clusters with the same connection mode in the isomorphic modules are mapped to the identification modules, such as Figure 3 Middle (A) and Figure 3 As shown in (B).

[0083] Four communities were revealed in the recognition module (with a full-graph modularity of 0.55), two of which overlapped. Without considering the weight of the network weighted adjacency matrix, the four isomorphic modules can all find corresponding mapping parts in the recognition module. When conducting subgraph isomorphism analysis on a bacterial community network based on a multi-source, large-scale database, more accurate subgraph matching can be achieved by combining the network edge weights.

[0084] Although the present invention is described herein with reference to the illustrative embodiments of the present invention, the above embodiment is only one of the preferred embodiments of the present invention, and the embodiments of the present invention are not limited to the above embodiment. It should be understood that those skilled in the art can design many other modifications and embodiments, which will fall within the scope and spirit of the principles disclosed in this application.

Claims

1. A method for identifying key clusters in a microbial co-occurrence network, characterized in that: include: Step S100, collecting and processing microbial flora data, and identifying differential bacterial species; Step S200: constructing a microbial co-occurrence network for the identified differential bacterial species; Step S300, identifying the number N of all connected networks in the microbial co-occurrence network, dividing the connected networks whose number of bacterial communities is greater than or equal to a preset value into identification modules, and defining the remaining connected networks as isomorphic modules; Sort the connected networks in the recognition module and the isomorphic module from large to small according to the size of the bacterial colony; Step S400: using a community discovery algorithm to perform cluster discovery on the connected network in the recognition module, obtaining bacterial species clusters with similar structural characteristics, summarizing bacterial species characteristics of the same cluster, and discovering key clusters; Step S500: for the connected networks in the isomorphic modules, a graph matching technique is used to match the connected networks of the identification modules to obtain key clusters with similar structures; The step S200 specifically includes: for the identified differential bacterial species, under the condition that the correlation is greater than a set correlation strength threshold and the correlation is statistically significant, constructing a microbial co-occurrence network based on the correlation adjacency matrix; Among them, i and j are two bacterial colonies, A ij is the association between bacterial colony i and bacterial colony j. When the value range is greater than the predetermined association strength threshold, A ij =1, otherwise A ij =0.

2. A method for identifying key clusters in a microbial co-occurrence network according to claim 1, characterized in that: The step S100 specifically includes: Step S110, collecting a sample containing microbial flora species, extracting genomic deoxynucleotide DNA from the sample, and measuring the genomic DNA content; Step S120, randomly breaking the DNA, performing end repair, adding A tails, adding sequencing adapters, purification, and polymerase chain reaction (PCR) amplification to complete library preparation and quantification; Step S130, mixing different libraries according to effective concentrations and performing double-end sequencing; Step S140, performing species taxonomy annotation to obtain microbial flora species; Step S150: perform basic feature analysis on the relative abundance data at the species level of the microbial flora, including species proportion and diversity analysis, and use linear discriminant analysis (LDA) to identify differential species.

3. A method for identifying key clusters in a microbial co-occurrence network according to claim 1, characterized in that: The step S300 specifically includes: Step S310, using the adjacency matrix, identifying the number N of all connected networks in the microbial co-occurrence network; Step S320, sort the N connected networks from large to small according to the size of the bacterial colony, and form the M1 connected networks whose number of nodes is greater than or equal to the preset value into an identification module G (Vg, Eg), where Vg represents the vertex of the bacterial colony in the identification module, and Eg represents the edge of the bacterial colony connection relationship in the identification module; form the M2 connected networks whose number of nodes is less than the preset value into an isomorphic module Q (Vq, Eq), where Vq represents the vertex of the bacterial colony in the isomorphic module, and Eq represents the edge of the bacterial colony connection relationship in the isomorphic module; Step S330, sort the M1 connected networks in the recognition module and the M2 connected networks in the isomorphic module from large to small according to the number of bacterial species.

4. A method for identifying key clusters in a microbial co-occurrence network according to claim 3, characterized in that: The step S310 specifically includes: Step A: For bacterial colony i, if there is a bacterial colony j that is not connected to bacterial colony i, and the bacterial colony connected to bacterial colony j in order n is not connected to bacterial colony i, then the adjacency matrix is ​​divided into two at bacterial colony i; where n is the longest path distance; Step B: Repeat step A for the regenerated adjacency matrix until it cannot be divided any further, exclude the adjacency matrix with the number of rows and columns less than the set value, and identify N connected networks.

5. A method for identifying key clusters in a microbial co-occurrence network according to claim 1, characterized in that: The step S400 specifically includes: using the modularity optimization method Louvain or the physical model-based method Spinglass to cluster the connected network in the identification module, dividing the key clusters that affect the network structure, obtaining bacterial species clusters with similar structural characteristics, summarizing the bacterial species characteristics of the same cluster, and combining centrality index analysis to discover key clusters.

6. A method for identifying key clusters in a microbial co-occurrence network according to claim 3, characterized in that: The step S500 specifically includes: Step S510: Based on the topological structure information in the isomorphic modules, similar nodes or connection modes in the isomorphic modules are mapped to the identification module; Step S520, determining whether there is a subgraph match between the connected network of the isomorphic module and the connected network in the identification module, and performing subgraph isomorphism analysis. If there is an isomorphic graph, then they have the same isomorphism class; Step S530: Map each isomorphic module to the cluster divided by the identification module, and obtain all matched key clusters based on the result of isomorphic class identification.

7. A method for identifying key clusters in a microbial co-occurrence network according to claim 6, characterized in that: The step S510 specifically includes: Step S511: For the x-th connected network Qx in the isomorphic module, compare it with the connected network Gy in the identification module in order, where x∈(1,2,…,M1) and y∈(1,2,…,M2); Step S512: For each node in Qx, there must be a corresponding node in Gy; for each edge in Qx, there must be an edge in Gy, and the nodes at both ends of them correspond one to one; if there is an edge between any two nodes in Qx, there is also an edge between the two nodes in Qx; Step S513 traverses the connected networks in the identification module and searches for all connected networks in Gy that are isomorphic to it; Step S514: traverse all connected networks in the isomorphic modules to expand the number of all key clusters in the bacterial community network.

8. A method for identifying key clusters in a microbial co-occurrence network according to claim 6, characterized in that: The step S520 specifically includes: if MQm*m and MGn*n are respectively adjacency matrices composed of a connected network Q in the isomorphic module and G of all connected networks in the identification module, where M is the adjacency matrix, Q contains m bacterial species, and G contains n bacterial species; using a graph matching technology based on subgraph isomorphism, a network with a structure similar to MQm*m is matched from MGn*n, that is, a subgraph matching of the connected network Q—>connected network G.

Images