A method of normalizing and comparing horizontal gene transfer networks

Abstract

The application discloses a method for normalizing and comparing horizontal gene transfer (HGT) network, comprising the following steps: S1, using Bayesian probability model to normalize HGT according to sequencing amount; S2, in the HGT network, each genome is represented as a vertex, and if HGT occurs between two genomes, there is an edge between the two vertices; S3, using different classification levels to annotate genome sequence, and obtaining HGT network of different classification levels; S4, analyzing HGT network through several topological properties of graph density, transitivity, homophily and algebraic connectivity. The graph density, transitivity, homophily and algebraic connectivity are used to describe the network, and the HGT network is compared among samples in different groups, so that a new analysis direction is provided for the correlation between metagenome and human phenotype, and the metagenome can be more systematically modeled and analyzed by calculating various topological properties to analyze the HGT network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gene transfer network, and particularly relates to a method for normalizing and comparing horizontal gene transfer network. BACKGROUND

[0002] Human microbiota and human have complex interactions. Studies have found that human microbiota and a variety of diseases are related. Human microbiota is a complex ecosystem. For example, Ferretti et al. used strain-level metagenomic profiling to track bacterial transmission between mother and infant; they found that mother and infant shared strains; in recent studies on gut microbiota, many researchers established the connection between metagenome and human phenotype, but these works failed to model the gut microbiota from a systematic perspective. In recent years, researchers applied HGT network to analyze metagenome, extract microbiome features, and explore the correlation between microbiome and phenotype.

[0003] Disadvantages or deficiencies exist:

[0004] Although researchers have established the connection between metagenome and human phenotype, these works failed to model the gut microbiota from a systematic perspective. HGT network is a tool for systematic analysis of metagenome, but due to the influence of sequencing data volume on HGT network, some false positive correlations between metagenome and phenotype may occur, so it is necessary to normalize and compare horizontal gene transfer network. The present application relates to a method for normalizing HGT network and calculating HGT network topology attributes. The method can standardize the comparison of metagenome between test group and control group, thereby modeling and analyzing metagenome systematically. SUMMARY

[0005] The present application aims to solve the defects in the prior art and provides a method for normalizing and comparing horizontal gene transfer network.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0007] A method for normalizing and comparing horizontal gene transfer network comprises the following steps:

[0008] S1: normalizing HGT according to sequencing volume by using Bayesian probability model;

[0009] S2: in the HGT network, each genome is represented as a vertex, and if HGT occurs between two genomes, there is an edge between the two vertices;

[0010] S3: annotating genome sequences by using different classification levels to obtain HGT networks of different classification levels;

[0011] S4: Analyzing the HGT network by several topological properties of the graph, including density, transitivity, assortativity and algebraic connectivity.

[0012] Further, in step S1: using LocalHGT to detect HGT in each sample, and obtaining the number of reads supporting each HGT breakpoint pair, wherein when the average number of reads in all samples is Q, the number of reads supporting each HGT breakpoint pair is inferred according to the model;

[0013] When the total number of reads is Q, then the probability distribution of the number of reads supporting each HGT breakpoint pair is obtained, and the HGT breakpoint is filtered according to the probability distribution;

[0014] The filtering criterion is that when the total number of reads is Q, the probability that the expected number of reads is greater than x (default is 2) is higher than 90%.

[0015] Further, in step S4: the graph density represents the number of edges existing in the graph divided by the maximum number of edges the graph can have;

[0016] Transitivity is calculated by observing the ratio of the number of triangles to the maximum number of triangles that the graph can have, and the transitivity measures the overall probability that the neighbors of a vertex are connected to each other;

[0017] Assortativity is the tendency of nodes in the graph to connect to nodes with similar degrees, and the degree is quantified by the Pearson correlation coefficient of the degree and the degree of association;

[0018] Algebraic connectivity is the second smallest eigenvalue of the Laplacian matrix of the graph;

[0019] The topological properties are calculated using the Python module NetworkX.

[0020] Compared with the prior art, the beneficial effects of the present application are:

[0021] Since the HGT event is affected by the amount of sequencing data, the Bayesian model is used in all samples to normalize the HGT event according to the sequencing amount, which greatly reduces the occurrence of some false positive macrogenomic and phenotype association caused by the influence of sequencing data amount on the HGT network, and makes the comparison of macrogenomes between the experimental group and the control group more standardized.

[0022] The genome is annotated at different classification levels, and different levels of HGT network are obtained. The graph density, transitivity, assortativity and algebraic connectivity are used to describe the network, and the HGT network is compared between samples in different groups to provide a new analysis direction for the association between macrogenome and human phenotype. By calculating various topological properties to analyze the HGT network, the macrogenome can also be modeled and analyzed more systematically. Detailed Implementation

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0024] A method for normalizing and comparing horizontal gene transfer networks, comprising the following steps:

[0025] S1: The HGT was normalized based on the sequencing volume using a Bayesian probability model;

[0026] S2: In the HGT network, each genome is represented as a vertex, where if the HGT occurs between two genomes, there is an edge between the two vertices;

[0027] S3: Different taxonomic levels were used to annotate the genome sequences, resulting in HGT networks of different taxonomic levels;

[0028] S4: Analyze HGT networks using several topological properties, including graph density, transitivity, isomatch, and algebraic connectivity.

[0029] In a specific embodiment of this application, in step S1: LocalHGT is used to detect HGT in each sample and the number of reads supporting each HGT breakpoint pair is obtained, wherein when the average number of reads in all samples is Q, it is inferred from the model;

[0030] When the total number of reads is Q, the probability distribution of the number of reads for each HGT breakpoint pair is supported, and HGT breakpoints are filtered according to this probability distribution.

[0031] The filtering criterion is: given a total number of reads of Q, the probability that the expected number of reads is greater than x (default is 2) is higher than 90%.

[0032] In step S4: Graph density represents the number of edges present in the graph divided by the maximum number of edges the graph can have;

[0033] Transitivity is calculated by the ratio of the observed number of triangles to the maximum possible number of triangles in the graph. Transitivity measures the overall probability that adjacent vertices of a vertex are connected to each other.

[0034] Isomatch is the tendency for nodes in a graph to connect with nodes that have similarity. The degree of association is quantified by the correlation degree and the Pearson correlation coefficient.

[0035] Algebraic connectivity is the second smallest eigenvalue of the Laplacian matrix of a graph;

[0036] Topology properties are computed using the Python module NetworkX.

[0037] In order to better understand the technical solutions of the present application, the following will be further described in conjunction with examples.

[0038] One: using Bayesian probability model to normalize HGT according to sequencing amount, sequencing amount affects the number of detected HGT breakpoint pairs, wherein the HGT of low abundance species may be missed in the case of low sequencing amount, therefore we use Bayesian probability model to normalize HGT according to sequencing amount to reduce the influence of sequencing amount on subsequent analysis.

[0039] For each breakpoint H, it is assumed that the probability of read alignment to H is θ, then the number of reads aligned to breakpoint H conforms to binomial distribution.

[0040] It is assumed that the prior distribution of θ is Beta distribution θ~Beta(a, b), wherein a and b values are self-defined:

[0041] Beta(a, b) is

[0042] B(a, b)=Γ(a+b) / [Γ(a)+Γ(b)].

[0043] In a given sample, it is assumed that the sample has g reads, wherein m reads are aligned to breakpoint H, according to Bayesian inference, the posterior distribution of θ is

[0044]

[0045] It is assumed that given θ, the probability that m reads in g reads are aligned to H can be calculated by

[0046]

[0047] Since θ~Beta(a, b), we obtain

[0048]

[0049] Therefore, the posterior distribution of θ is θ~Beta(a+m, b+g-m).

[0050] It is assumed that α=a+m, β=b+g-m, and

[0051]

[0052] It is assumed that in the case of n reads in the sample, the probability that σ reads are aligned to breakpoint H can be obtained by

[0053]

[0054] There are

[0055]

[0056] The integral can be computed using the Beta function

[0057]

[0058] The final result is the function

[0059]

[0060] In this model, HGT can be filtered according to sequencing reads.

[0061] Suppose the average read count in all samples is denoted as Q, the HGT detection software LocalHGT provides the number of reads supporting each HGT breakpoint pair, for each sample, according to the observed number of reads supporting each HGT breakpoint pair, we can estimate the probability distribution of supporting the breakpoint pair if the total number of reads is Q, we set the HGT filtering standard is, under the condition of total read count Q, the probability of expected number of reads greater than x (default 2) is greater than 90%.

[0062] Two: analysis of HGT network by several topological properties of graph density, transitivity, homophily and algebraic connectivity.

[0063] HGT events are the linkage of two different genome sequences, in the HGT network, each genome is represented as a vertex, if HGT occurs between two genomes, there is an edge between the two vertices;

[0064] Then, adopt different classification levels to annotate the genome sequence, such as genus, species, etc., we get the HGT network of different classification levels, and:

[0065] The graph density represents the number of edges in the graph divided by the maximum number of edges the graph could have.

[0066] Given a graph G (V, E), the formula to calculate the density is:

[0067]

[0068] Transitivity is calculated by observing the ratio of the number of triangles to the maximum number of triangles the graph could have. Transitivity measures the overall probability of adjacent vertices of a vertex being connected to each other.

[0069] Homophily refers to the tendency of nodes in the graph to connect to nodes with similar degrees. The degree is quantified by the Pearson correlation coefficient of the degree with the degree.

[0070] Algebraic connectivity is the second smallest eigenvalue of the Laplacian matrix of a graph. To compute the algebraic connectivity, the Laplacian matrix of the graph can be computed by L = D - A, where D is the degree matrix of the graph and A is the adjacency matrix of the graph. Then the Laplacian matrix L is spectrally decomposed to obtain the second smallest eigenvalue of L;

[0071] Topological properties are computed using the Python module NetworkX.

[0072] The above description is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art, according to the technical solution and the inventive concept of the present application, should be covered within the protection scope of the present application.

Claims

1. A method of normalizing and comparing horizontal gene transfer networks, characterized by, The method comprises the following steps: S1: using Bayesian probability model to normalize HGT according to sequencing amount; S2: in HGT network, each genome is represented as a vertex, and if HGT occurs between two genomes, there is an edge between the two vertices; S3: using different classification levels to annotate genome sequences, and obtaining HGT networks of different classification levels; S4: analyzing HGT network through several topological properties of graph, such as density, transitivity, assortativity and algebraic connectivity.

2. The method of normalizing and comparing horizontal gene transfer networks of claim 1, wherein, In step S1: using LocalHGT to detect HGT in each sample, and obtaining the number of reads supporting each HGT breakpoint pair, wherein when the average number of reads in all samples is Q, the number of reads supporting each HGT breakpoint pair is inferred according to the model; When the total number of reads is Q, the probability distribution of the number of reads supporting each HGT breakpoint pair is obtained, and according to the probability distribution, the HGT breakpoint is filtered.

3. The method of normalizing and comparing horizontal gene transfer networks of claim 2, wherein, The filtering standard for filtering HGT breakpoint is that when the total number of reads is Q, the probability that the expected number of reads is greater than x is higher than 90%.

4. The method of normalizing and comparing horizontal gene transfer networks of claim 1, wherein, In step S4: the graph density represents the number of edges existing in the graph divided by the maximum number of edges that the graph can have; Transitivity is obtained by calculating the ratio of the number of observed triangles to the maximum number of triangles that the graph can have, and the transitivity measures the overall probability that the adjacent vertices of a vertex are connected to each other; Assortativity is the tendency of nodes in the graph to be connected to nodes with similar degrees, which is quantified by the Pearson correlation coefficient of the correlation degree and the correlation degree; Algebraic connectivity is the second smallest eigenvalue of the Laplacian matrix of the graph; The topological properties are calculated using the Python module NetworkX.

Images