Method, device and storage medium for identifying conditional pathogens in large cohort metagenomes

Abstract

The present invention discloses a method, device and storage medium for identifying conditional pathogens of a large cohort of metagenomes, and relates to the technical fields of deep learning and bioinformatics. The deep learning network proposed in the present invention adopts a hierarchical (or hierarchical) recognition strategy to construct a method for identifying conditional pathogens from a single species to multiple species. Compared with the prior art, the method for identifying conditional pathogens provided by the present invention can more accurately identify the interactions of conditional pathogens. By analyzing the complex interactive network structure between multiple species, the interaction pattern between conditional pathogens can be identified, which helps to more comprehensively reveal the role of microorganisms in the occurrence and development of diseases, and then provide a more sophisticated interpretation of internal factors for clinical diagnosis and treatment, which helps to formulate clinical diagnosis and treatment strategies.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning and bioinformatics, and in particular to a method, device and storage medium for identifying conditional pathogens of a large cohort of metagenomes. Background Art

[0002] With the rapid development of metagenomic sequencing technology and the continuous reduction in costs, large-scale metagenomic research projects have become increasingly feasible and affordable. This allows researchers to conduct high-throughput metagenomic sequencing analysis on large cohorts of samples, thereby deeply exploring the overall structure and functional characteristics of gene communities in complex environments or populations. Through such technical means, researchers can not only reveal the differences in genetic diversity and population distribution between different samples, but also accurately identify those conditional pathogens that are potentially related to the occurrence and development of diseases in massive data. In the medical field, large-cohort metagenomic sequencing is revolutionary for the detection and discovery of the causes of chronic diseases, infectious diseases, and some unexplained diseases. Through metagenomic analysis of intestinal, respiratory or other body fluid samples from thousands of individuals, scientists can systematically screen for gene types that frequently appear or are significantly enriched under specific health conditions. These genes may become pathogenic bacteria under certain conditions and have adverse effects on host health. This progress provides a more comprehensive perspective for clinical diagnosis, helps the design of personalized treatment plans and the formulation of prevention strategies, and also greatly promotes the basic research progress on the relationship between genome and human health.

[0003] In the human body, the microbiota and the body usually maintain a dynamic equilibrium and steady-state relationship, and this genetic ecosystem plays a vital role in human health. Under normal circumstances, even if there are some potential conditional pathogens, as long as their number, activity and interaction with the host immune system remain within an appropriate range, they will not cause disease. However, under certain conditions, such as decreased immunity, environmental changes or changes in living habits, a certain conditional pathogen that was originally in a symbiotic state may break this steady state due to a sharp increase in number or abnormal function, thereby causing human illness. However, more commonly, the simultaneous changes of multiple conditional pathogens and their interactions together constitute a complex etiological network that causes disease. These conditional pathogens may affect each other's number and activity, as well as the impact on the host's physiological functions through competition, synergy or other interactions. In this case, a single conditional pathogen cannot fully explain the occurrence of the disease, but requires a comprehensive consideration of the changes in the structure of the entire genetic community and its complex interaction mechanism with the host immune system. Therefore, understanding and revealing how multiple conditional pathogens work together to cause human illness has extremely important theoretical value and clinical significance for the prevention and treatment of related diseases.

[0004] Given the importance of the synergistic effects and complex interactions of multiple conditional pathogens in the development of diseases, it is particularly important to develop a software tool that can effectively identify and analyze the interactions between these genes in large cohort metagenomic data. Such software should have strong bioinformatics analysis capabilities and be able to accurately capture the changing dynamics of pathogenic bacterial populations under different conditions and the association patterns between them from massive metagenomic sequencing data.

[0005] Although advanced network analysis methods such as WGCNA (weighted gene co-expression network analysis) have played an important role in metagenomics research and successfully revealed the complex interactions between members within the gene community, they still face some challenges and shortcomings in processing and parsing large cohort metagenomic data. First, when faced with large-scale sample data sets, the computational complexity of methods such as WGCNA will increase significantly, and the demand for storage space and computing resources will also increase, which may lead to a decrease in analysis efficiency. Secondly, due to the high heterogeneity of large cohort metagenomic data and the huge and diverse amount of information contained, existing network construction algorithms may not be able to fully capture the fine interaction patterns between all potential conditional pathogens, especially the dynamic changes in the process of disease occurrence and development.

[0006] With the development of deep learning technology, deep learning network models can be used to identify interactions between conditional pathogens, which will make the identification of interactions between conditional pathogens more accurate.

[0007] In view of this, the present invention is proposed. Summary of the invention

[0008] The purpose of the present invention is to provide a method, device and storage medium for identifying conditional pathogens of a large cohort of metagenomes to solve the above-mentioned technical problems.

[0009] The present invention is achieved in that:

[0010] In a first aspect, the present invention provides a method for identifying conditional pathogens of a large cohort of metagenomes, comprising the following steps:

[0011] S1. Perform species annotation and relative abundance estimation on the metagenomic sequencing data after quality control filtration and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance of all samples;

[0012] S2. Take the data after relative abundance leveling to construct species relative abundance matrix and phenotype matrix; send the species relative abundance matrix and phenotype matrix to the encoder, which is responsible for extracting key features from the original data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; where m represents the number of samples and n represents the number of species;

[0013] S3. The m feature vectors G and n feature vectors Y extracted by S2 are input into the pathogen registration module; the correlation between each feature vector is calculated to construct a correlation matrix P;

[0014] S4. The correlation matrix P is sent to the interaction analysis module, and the interaction between features is deeply analyzed through the deep learning algorithm to build a deep learning model; the species interaction matrix is ​​output;

[0015] S5. Identify conditional pathogens based on the deep learning model and species interaction matrix constructed in S4.

[0016] In a second aspect, the present invention also provides a method for identifying interactions of conditional pathogens in a large cohort of metagenomes, comprising the following steps:

[0017] S1. Perform species annotation and relative abundance estimation on the sequencing data after quality control filtering and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance;

[0018] S2. Take the data after relative abundance leveling to construct species relative abundance matrix and phenotype matrix; send the species relative abundance matrix and phenotype matrix to the encoder, which is responsible for extracting key features from the original data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; where m represents the number of samples and n represents the number of species;

[0019] S3. The m feature vectors G and n feature vectors Y extracted by S2 are input into the pathogen registration module; the correlation between each feature vector is calculated to construct a correlation matrix P;

[0020] S4. The correlation matrix P is sent to the interaction analysis module, and the interaction between features is deeply analyzed through the deep learning algorithm to build a deep learning model;

[0021] S5. Based on the deep learning model constructed in S4 and the species interaction matrix, conditional pathogens are identified, including first-order conditional pathogens, second-order conditional pathogens, and higher-order conditional pathogens.

[0022] First-order conditional pathogen identification is used to: Identify individual species features through deep learning models to determine which species may be conditional pathogens.

[0023] Second-order conditional pathogen identification is used to: analyze the relationship between species through deep learning models and identify species combinations that may work together to cause disease under specific conditions;

[0024] High-order conditional pathogen identification is used to: Analyze the interaction network structure among multiple species to identify the interaction patterns of conditional pathogens among multiple species.

[0025] In a third aspect, the present invention also provides a device for identifying conditional pathogens of a large cohort of metagenomes, comprising: an input module, a control module, and an output module;

[0026] The input module is configured to: input the sequencing data after quality control filtering and host removal or the metagenomic off-machine sequencing data;

[0027] The control module includes: a data preprocessing module, a feature vector extraction module, a pathogen registration module, an interaction analysis module and a conditional pathogen identification module;

[0028] The data preprocessing module is configured to: perform species annotation and relative abundance estimation on the metagenomic sequencing data after quality control filtering and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance; or, perform quality control filtering, host sequence removal, species annotation and relative abundance estimation on the metagenomic sequencing data, then merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance;

[0029] The feature vector extraction module is configured as follows: the species relative abundance matrix and the phenotype matrix are sent to the encoder, which is responsible for extracting key features from the raw data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; where m represents the number of samples and n represents the number of species;

[0030] The pathogen registration module is configured as follows: inputting the feature vector of the feature vector extraction module; calculating the correlation between each feature vector and constructing a correlation matrix P;

[0031] The interaction analysis module is configured as follows: the correlation matrix P of the pathogen registration module is input to conduct in-depth analysis of the interaction between features through a deep learning algorithm to construct a deep learning model; and the species interaction matrix is ​​output;

[0032] The conditional pathogen identification module is configured to: identify conditional pathogens based on the deep learning model and species interaction matrix constructed by the interaction analysis module;

[0033] The output module is configured to output the identification result of the conditional pathogen identification module.

[0034] The total abundance of the sample is very low, which can be defined as less than 25% of the average of the total abundance of all samples. In other embodiments, the percentage of the average of the total abundance of all samples can also be customized.

[0035] In a fourth aspect, the present invention also provides a device for identifying the interaction of conditional pathogens in a large cohort of metagenomes, comprising: an input module, a control module, and an output module;

[0036] The input module is configured to: input the sequencing data after quality control filtering and host removal or the metagenomic off-machine sequencing data;

[0037] The control module includes: a data preprocessing module, a feature vector extraction module, a pathogen registration module, an interaction analysis module and a conditional pathogen identification module;

[0038] The data preprocessing module is configured to: perform species annotation and relative abundance estimation on the metagenomic sequencing data after quality control filtering and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance; or, perform quality control filtering, host sequence removal, species annotation and relative abundance estimation on the metagenomic sequencing data, then merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance;

[0039] The feature vector extraction module is configured as follows: the species relative abundance matrix and the phenotype matrix are sent to the encoder, which is responsible for extracting key features from the raw data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; where m represents the number of samples and n represents the number of species;

[0040] The pathogen registration module is configured as follows: inputting the feature vector of the feature vector extraction module; calculating the correlation between each feature vector and constructing a correlation matrix P;

[0041] The interaction analysis module is configured as follows: the correlation matrix P of the pathogen registration module is input to conduct in-depth analysis of the interaction between features through a deep learning algorithm to construct a deep learning model; and the species interaction matrix is ​​output;

[0042] The conditional pathogen identification module is configured to: identify conditional pathogens based on the deep learning model and species interaction matrix constructed by the interaction analysis module;

[0043] The conditional pathogen identification module includes: a first-order conditional pathogen identification module, a second-order conditional pathogen identification module and a high-order conditional pathogen identification module; wherein the first-order conditional pathogen identification module is configured to: identify the characteristics of a single species through a deep learning model to determine which species may be conditional pathogens; the second-order conditional pathogen identification module is configured to: analyze the relationship between species through a deep learning model to identify the combination of species that may act together to cause disease under specific conditions; the high-order conditional pathogen identification module is configured to: analyze the interaction network structure between multiple species to identify the interaction pattern of conditional pathogens between multiple species;

[0044] The output module is configured to output the identification result of the conditional pathogen identification module.

[0045] In a fifth aspect, the present invention also provides a device for identifying conditional pathogens of a large cohort of metagenomes or identifying the interactions of conditional pathogens of a large cohort of metagenomes. The device includes a processor and a memory. The memory stores at least one instruction, at least one program, a code set or an instruction set. The at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor to implement a method for identifying conditional pathogens of a large cohort of metagenomes or a method for identifying the interactions of conditional pathogens of a large cohort of metagenomes.

[0046] In a sixth aspect, the present invention also provides a computer-readable storage medium, storing at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement a method for identifying conditional pathogens of a large cohort of metagenomes or a method for identifying interactions of conditional pathogens of a large cohort of metagenomes.

[0047] The present invention has the following beneficial effects:

[0048] The present invention provides a deep learning method that can efficiently process and analyze large-scale metagenomic data. This method requires powerful bioinformatics analysis capabilities to accurately capture the changing dynamics of pathogenic bacteria populations and their inter-correlation patterns from massive metagenomic sequencing data. Compared with the prior art, the identification method of conditional pathogens provided by the present invention can more accurately identify the interaction of conditional pathogens. The proposal of the present invention provides a basis for analyzing the role of microorganisms in the occurrence and development of diseases.

[0049] The deep learning network proposed in the present invention adopts a hierarchical (or hierarchical) recognition strategy to construct a method for identifying conditional pathogens from a single species to multiple species. By analyzing the complex interactive network structure among multiple species, the interaction patterns between conditional pathogens are identified. This helps to more comprehensively reveal the role of microorganisms in the occurrence and development of diseases, and further provide a more detailed interpretation of internal factors for clinical diagnosis and treatment, which is helpful for the formulation of clinical diagnosis and treatment strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0051] Figure 1 Flowchart of the method for identification of opportunistic pathogenic interactions in large metagenomic cohorts. DETAILED DESCRIPTION

[0052] References to embodiments of the present invention will now be provided in detail, one or more examples of which are described below. Each example is provided as an explanation rather than a limitation of the present invention. In fact, it will be apparent to those skilled in the art that various modifications and variations may be made to the present invention without departing from the scope or spirit of the present invention. For example, a feature illustrated or described as part of one embodiment may be used in another embodiment to produce a further embodiment.

[0053] In a first aspect, the present invention provides a method for identifying conditional pathogens of a large cohort of metagenomes, comprising the following steps:

[0054] S1. Perform species annotation and relative abundance estimation on the metagenomic sequencing data after quality control filtration and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance of all samples;

[0055] S2. Take the data after relative abundance leveling to construct species relative abundance matrix and phenotype matrix; send the species relative abundance matrix and phenotype matrix to the encoder, which is responsible for extracting key features from the original data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; where m represents the number of samples and n represents the number of species;

[0056] S3. The m feature vectors G and n feature vectors Y extracted by S2 are input into the pathogen registration module; the correlation between each feature vector is calculated to construct a correlation matrix P;

[0057] S4. The correlation matrix P is sent to the interaction analysis module, and the interaction between features is deeply analyzed through the deep learning algorithm to build a deep learning model; the species interaction matrix is ​​output;

[0058] S5. Identify conditional pathogens based on the deep learning model and species interaction matrix constructed in S4.

[0059] The present invention provides a deep learning method that can efficiently process and analyze large-scale metagenomic data. The method requires powerful bioinformatics analysis capabilities in order to accurately capture the changing dynamics of pathogenic bacteria populations and their mutual correlation patterns from massive metagenomic sequencing data. Compared with the prior art, the method for identifying conditional pathogens provided by the present invention can more accurately identify the interactions of conditional pathogens.

[0060] The identification results of conditional pathogens provide detailed information for understanding the role of microorganisms in the occurrence and development of diseases. Through the deep learning network architecture proposed in this invention, we can better understand and predict the role of microorganisms in the occurrence and development of diseases, provide deeper insights for clinical diagnosis and treatment, and help formulate clinical diagnosis and treatment strategies.

[0061] In the above step S1, leveling means placing the abundance of all communities to be analyzed at the same depth, that is, the total species abundance of each community is the same.

[0062] The correlation matrix P in the above step S3 reflects the interaction strength between different features and provides a basis for subsequent interaction analysis.

[0063] In a preferred embodiment of the present invention, step S5 includes: first-order conditional pathogen identification;

[0064] First-order conditional pathogen identification is used to: Identify individual species features through deep learning models to determine which species may be conditional pathogens.

[0065] In an optional embodiment, when the interactive relationship is not of concern, the identification of possible conditional pathogens can also be achieved through first-order conditional pathogen identification.

[0066] In a preferred embodiment of the present invention, step S5 further includes second-order conditional pathogen identification;

[0067] Second-order conditional pathogen identification is used to: analyze the relationship between species through deep learning models and identify species combinations that may work together to cause pathogens under specific conditions.

[0068] By identifying these first- and second-order conditions, deep learning models can reveal subtle interaction patterns between conditional pathogens, especially the dynamic changes during the development of pathogenicity.

[0069] The relationships between species include but are not limited to: symbiotic interactions, persistent mutualism, persistent biotic interactions, mutualism, interspecific reciprocal altruistic behavior, partial mutualism, parasitic symbiosis, obligate interaction, facultative interaction, both species are obligate, one is obligate but the other is facultative, both species are facultative, ectosymbiosis, endosymbiosis, partial ectosymbiosis, mutualistic ectosymbiosis, ectoparasitism, associative symbiosis, segregative symbiosis, antagonistic or antagonistic symbiosis or relationship, necrotrophic interaction, biotrophic interaction, partial harm symbiosis, competitive relationship, antagonistic relationship, and double loss symbiosis.

[0070] In a preferred embodiment of the present invention, the species annotation and relative abundance estimation in step S1 refers to: using species annotation software to obtain the species and abundance in the sample.

[0071] In a preferred embodiment of the present invention, the species annotation software is selected from at least one of Kraken2, Bracken, MetaPhlAn, QIIME and Genometa.

[0072] In a preferred embodiment of the present invention, Kraken2 and Bracken are used to perform species annotation and relative abundance estimation on the metagenome of the sample respectively;

[0073] In a preferred embodiment of the present invention, the sample is selected from at least one of a donor tissue sample, a body fluid sample, an environmental sample and a stool sample.

[0074] Donor tissue samples include but are not limited to diseased tissue and healthy tissue samples; diseased tissue includes but is not limited to tumor tissue samples, inflammatory tissue samples, etc. Tissues include but are not limited to: intestines, respiratory tract, digestive tract, tear duct, arteries, veins, thyroid, skull, bones and liver, etc.

[0075] In a second aspect, the present invention also provides a method for identifying interactions of conditional pathogens in a large cohort of metagenomes, comprising the following steps:

[0076] S1. Perform species annotation and relative abundance estimation on the sequencing data after quality control filtering and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance;

[0077] S2. Take the data after relative abundance leveling to construct species relative abundance matrix and phenotype matrix; send the species relative abundance matrix and phenotype matrix to the encoder, which is responsible for extracting key features from the original data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; where m represents the number of samples and n represents the number of species;

[0078] S3. The m feature vectors G and n feature vectors Y extracted by S2 are input into the pathogen registration module; the correlation between each feature vector is calculated to construct a correlation matrix P;

[0079] S4. The correlation matrix P is sent to the interaction analysis module, and the interaction between features is deeply analyzed through the deep learning algorithm to build a deep learning model;

[0080] S5. Based on the deep learning model constructed in S4 and the species interaction matrix, conditional pathogens are identified, including first-order conditional pathogens, second-order conditional pathogens, and higher-order conditional pathogens.

[0081] First-order conditional pathogen identification is used to: Identify individual species features through deep learning models to determine which species may be conditional pathogens.

[0082] Second-order conditional pathogen identification is used to: analyze the relationship between species through deep learning models and identify species combinations that may work together to cause disease under specific conditions;

[0083] High-order conditional pathogen identification is used to: Analyze the interaction network structure among multiple species to identify the interaction patterns of conditional pathogens among multiple species.

[0084] The above-mentioned high-order conditional pathogen identification enables a more detailed analysis of the complex interaction network structure among multiple species to identify the interaction patterns of conditional pathogens among multiple species, which will help to better understand and predict the role of microorganisms in the occurrence and development of diseases, provide deeper insights for clinical diagnosis and treatment, and facilitate the formulation of clinical diagnosis and treatment strategies.

[0085] In a third aspect, the present invention also provides a device for identifying conditional pathogens of a large cohort of metagenomes, comprising: an input module, a control module, and an output module;

[0086] The input module is configured to: input the sequencing data after quality control filtering and host removal or the metagenomic off-machine sequencing data;

[0087] The control module includes: a data preprocessing module, a feature vector extraction module, a pathogen registration module, an interaction analysis module and a conditional pathogen identification module;

[0088] The data preprocessing module is configured to: perform species annotation and relative abundance estimation on the metagenomic sequencing data after quality control filtering and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance; or, perform quality control filtering, host sequence removal, species annotation and relative abundance estimation on the metagenomic sequencing data, then merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance;

[0089] The feature vector extraction module is configured as follows: the species relative abundance matrix and the phenotype matrix are sent to the encoder, which is responsible for extracting key features from the raw data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; where m represents the number of samples and n represents the number of species;

[0090] The pathogen registration module is configured as follows: inputting the feature vector of the feature vector extraction module; calculating the correlation between each feature vector and constructing a correlation matrix P;

[0091] The interaction analysis module is configured as follows: the correlation matrix P of the pathogen registration module is input to conduct in-depth analysis of the interaction between features through a deep learning algorithm to construct a deep learning model; and the species interaction matrix is ​​output;

[0092] The conditional pathogen identification module is configured to: identify conditional pathogens based on the deep learning model and species interaction matrix constructed by the interaction analysis module;

[0093] The output module is configured to output the identification result of the conditional pathogen identification module.

[0094] The total abundance of the sample is very low, which can be defined as less than 25% of the average of the total abundance of all samples. In other embodiments, the percentage of the average of the total abundance of all samples can also be customized.

[0095] In a preferred embodiment of the present invention, the conditional pathogen identification module includes: a first-order conditional pathogen identification module, the first-order conditional pathogen identification module is configured to: identify single species characteristics through a deep learning model to determine which species may be conditional pathogens;

[0096] In a preferred embodiment of the present invention, the conditional pathogen identification module also includes a second-order conditional pathogen identification module; the second-order conditional pathogen identification module is configured to: analyze the relationship between species through a deep learning model, and identify species combinations that may work together to cause pathogens under specific conditions.

[0097] In a fourth aspect, the present invention also provides a device for identifying the interaction of conditional pathogens in a large cohort of metagenomes, comprising: an input module, a control module, and an output module;

[0098] The input module is configured to: input the sequencing data after quality control filtering and host removal or the metagenomic off-machine sequencing data;

[0099] The control module includes: a data preprocessing module, a feature vector extraction module, a pathogen registration module, an interaction analysis module and a conditional pathogen identification module;

[0100] The data preprocessing module is configured to: perform species annotation and relative abundance estimation on the metagenomic sequencing data after quality control filtering and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance; or, perform quality control filtering, host sequence removal, species annotation and relative abundance estimation on the metagenomic sequencing data, then merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance;

[0101] The feature vector extraction module is configured as follows: the species relative abundance matrix and the phenotype matrix are sent to the encoder, which is responsible for extracting key features from the raw data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; where m represents the number of samples and n represents the number of species;

[0102] The pathogen registration module is configured as follows: inputting the feature vector of the feature vector extraction module; calculating the correlation between each feature vector and constructing a correlation matrix P;

[0103] The interaction analysis module is configured as follows: the correlation matrix P of the pathogen registration module is input to conduct in-depth analysis of the interaction between features through a deep learning algorithm to construct a deep learning model; and the species interaction matrix is ​​output;

[0104] The conditional pathogen identification module is configured to: identify conditional pathogens based on the deep learning model and species interaction matrix constructed by the interaction analysis module;

[0105] The conditional pathogen identification module includes: a first-order conditional pathogen identification module, a second-order conditional pathogen identification module and a high-order conditional pathogen identification module; wherein the first-order conditional pathogen identification module is configured to: identify the characteristics of a single species through a deep learning model to determine which species may be conditional pathogens; the second-order conditional pathogen identification module is configured to: analyze the relationship between species through a deep learning model to identify the combination of species that may act together to cause disease under specific conditions; the high-order conditional pathogen identification module is configured to: analyze the interaction network structure between multiple species to identify the interaction pattern of conditional pathogens between multiple species;

[0106] The output module is configured to output the identification result of the conditional pathogen identification module.

[0107] The total abundance of the sample is very low, which can be defined as less than 25% of the average of the total abundance of all samples. In other embodiments, the percentage of the average of the total abundance of all samples can also be customized.

[0108] In a fifth aspect, the present invention also provides a device for identifying conditional pathogens of a large cohort of metagenomes or a device for identifying the interactions of conditional pathogens of a large cohort of metagenomes, the device comprising a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, at least one program, code set or instruction set being loaded and executed by the processor to implement a method for identifying conditional pathogens of a large cohort of metagenomes or a method for identifying the interactions of conditional pathogens of a large cohort of metagenomes.

[0109] Specifically, the electronic device may include a memory, a processor, a bus, and a communication interface, and the memory, processor, and communication interface are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these elements can be electrically connected to each other via one or more buses or signal lines. The processor can process information and / or data related to target identification to perform one or more functions described in this application.

[0110] The memory can be, but is not limited to, random access memory (RAM), read only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable read-only memory (EEPROM), etc.

[0111] In a sixth aspect, the present invention also provides a computer-readable storage medium, storing at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement a method for identifying conditional pathogens of a large cohort of metagenomes or a method for identifying interactions of conditional pathogens of a large cohort of metagenomes.

[0112] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0113] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme in the embodiments of the present invention will be described clearly and completely below. If the specific conditions are not specified in the embodiments, they are carried out according to conventional conditions or conditions recommended by the manufacturer. If the manufacturer of the reagents or instruments used is not specified, they are all conventional products that can be purchased commercially.

[0114] The features and performance of the present invention are further described in detail below in conjunction with the embodiments.

[0115] Example 1

[0116] This embodiment provides a method for identifying conditional pathogens in a large cohort of metagenomes. The network architecture diagram is shown in Figure 1 As shown, the method for identifying conditional pathogens specifically includes the following steps:

[0117] 1. Data preprocessing

[0118] In the data preprocessing stage, fastp software was first used to perform quality control on the metagenomic sequencing data, including removing adapter sequences and removing low-quality sequences. Then, bwa software was used to remove the host genome sequences, and the unaligned sequences were retained for subsequent analysis. Species annotation and relative abundance estimation were quickly completed using Kraken2 and Bracken software. Finally, the relative abundance data of all samples were merged, samples with low total abundance were deleted, and leveling analysis was performed to prepare data for subsequent deep learning analysis.

[0119] a) Quality control and filtering of offline data. First, use fastp software to perform quality control on the metagenomic offline sequencing data, remove adapter sequences, and remove low-quality sequences;

[0120] b) Removal of the host: The host genome was aligned using bwa software, and the unaligned sequences were retained for subsequent analysis;

[0121] c) Species annotation and relative abundance: Kraken2 and Bracken were used to quickly perform species annotation and relative abundance estimation on metagenomes.

[0122] d) Merge samples. Merge the relative abundance of all samples, delete samples with very low total abundance, perform sampling analysis, and save the data for subsequent analysis.

[0123] 2. Deep Learning Network Architecture

[0124] Deep learning network architectures are designed to process these pre-processed metagenomic data. The network may contain multiple layers, each responsible for extracting a different level of features from the data.

[0125] a) In this deep learning network architecture, the input data includes a species relative abundance matrix and a phenotype matrix. The above phenotype refers to whether the patient is sick. The two matrices are first fed into the encoder, which is responsible for extracting key features from the raw data. For the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y. (m represents the number of samples, and n represents the number of species).

[0126] b) The m feature vectors G and n feature vectors Y extracted above are input into the pathogen registration module; the correlation between each feature vector is calculated to construct a correlation matrix P; this matrix reflects the interaction strength between different features and provides a basis for subsequent interaction analysis.

[0127] c) The correlation matrix P is sent to the interaction analysis module, and the interaction between features is deeply analyzed through the deep learning algorithm to build a deep learning model; the species interaction matrix (Q k );

[0128] The above deep learning model is built based on deep learning, which includes an encoder, an attention mechanism layer, an interaction fusion layer, and an interaction analysis layer.

[0129] Input layer: The input is a feature vector (from the species relative abundance matrix and the phenotype matrix). The input size is [batch_size, num_features], where batch_size is the batch size and num_features is the number of features for each sample.

[0130] Encoder: The type is a multi-layer perceptron (MLP) or a convolutional neural network (CNN). The function is to extract key features from the input data. The output shape is [batch_size, encoded_features], where encoded_features is the feature dimension of the encoder output.

[0131] Correlation calculation layer: The type is a custom operation (for example, dot product or correlation coefficient calculation). The function is to calculate the correlation between feature vectors. The output shape is [batch_size, num_pairs], where num_pairs is the number of feature vector pairs.

[0132] Attention mechanism layer: The type is Self-Attention or Multi-Head Attention. The function is to give different weights to different feature pairs to highlight important interactions. The output shape is [batch_size, num_features, attention_features], where attention_features is the feature dimension of the attention layer output.

[0133] Interaction fusion layer: The type is MLP or recurrent neural network (RNN). The function is to fuse the attention-weighted features to form an interaction representation. The output shape is [batch_size, num_interactions], where num_interactions is the number of interactions.

[0134] Interaction analysis layer: The type is MLP or deep neural network (DNN). The function is to analyze and identify conditional pathogens and their interactions. The output shape is [batch_size, num_organisms], where num_organisms is the number of conditional pathogens.

[0135] Output layer: The type is a fully connected layer. The function is to output the final interaction matrix. Output shape: [batch_size, num_organisms*num_organisms], which indicates the interaction strength between each conditional pathogen and all other conditional pathogens.

[0136] The specific parameter examples are as follows:

[0137] Encoder: Assume a 3-layer MLP with 256 neurons in each layer and ReLU activation function.

[0138] Correlation calculation layer: Assuming there are 1000 feature vector pairs, the dot product is used to calculate the correlation.

[0139] Attention mechanism layer: A multi-head attention layer with 8 attention heads is used, and the dimension of each head is 32.

[0140] Interaction Fusion Layer: Assume that a 2-layer MLP is used, each layer has 128 neurons and ReLU activation function.

[0141] Interaction analysis layer: Assume that a 5-layer DNN is used, each layer has 64 neurons and ReLU activation function.

[0142] Output layer: Assuming there are 100 conditional pathogens, the output layer uses a linear activation function.

[0143] d) Conditional pathogens are identified based on the deep learning model and species interaction matrix constructed above: including first-order conditional pathogens identification and second-order conditional pathogens identification. Figure 1 Zk refers to the final pathogen identification layer, which performs the final pathogen identification based on the interaction matrix output by the deep learning model and the pre-designed identification types (first-order and second-order).

[0144] First-order conditional pathogen identification involves the analysis of individual species characteristics to determine which species may be conditional pathogens (i.e., whether individual species influence disease occurrence);

[0145] Second-order conditions focus on relationships such as symbiosis or antagonism between two species, identifying combinations of species that may work together to cause disease under certain conditions.

[0146] Example 2

[0147] This embodiment provides a method for identifying interactions of conditional pathogens in a large cohort of metagenomes. Compared with Embodiment 1, the only difference is step d) of the deep learning network architecture.

[0148] In this embodiment, step d) includes the following steps:

[0149] Conditional pathogens are identified based on the deep learning model and species interaction matrix constructed above:

[0150] It includes first-order conditional pathogen identification, second-order conditional pathogen identification and higher-order conditional pathogen identification.

[0151] First-order conditional pathogen identification involves analyzing the characteristics of individual species to determine which species may be conditional pathogens; second-order conditions focus on the relationship between two species, such as symbiosis or antagonism, to identify species combinations that may work together to cause disease under certain conditions; high-order conditional pathogen identification: At this level, the network further analyzes the complex interaction network structure between multiple species to identify the interaction patterns of conditional pathogens between multiple species. Through the identification of these conditions, deep learning models can reveal the subtle interaction patterns between conditional pathogens, especially the dynamic changes in the process of pathogenic development.

[0152] These results provide detailed information for understanding the role of microorganisms in the occurrence and development of diseases. Through the deep learning network architecture proposed in this example, people can better understand and predict the role of microorganisms in the occurrence and development of diseases, provide deeper insights for clinical diagnosis and treatment, and help formulate clinical diagnosis and treatment strategies.

[0153] In addition, this embodiment provides a method for identifying the interaction of conditional pathogens in a large cohort of metagenomes, which can also identify conditional pathogens.

[0154] Example 3

[0155] This embodiment provides an identification device for interaction of conditional pathogens of a large cohort of metagenomes, including: an input module, a control module, and an output module;

[0156] The input module is configured to: input the sequencing data after quality control filtering and host removal or the metagenomic off-machine sequencing data;

[0157] The control module includes: a data preprocessing module, a feature vector extraction module, a pathogen registration module, an interaction analysis module and a conditional pathogen identification module;

[0158] The data preprocessing module is configured to: perform species annotation and relative abundance estimation on the metagenomic sequencing data after quality control filtering and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance; or, perform quality control filtering, host sequence removal, species annotation and relative abundance estimation on the metagenomic sequencing data, then merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance;

[0159] The feature vector extraction module is configured as follows: the species relative abundance matrix and the phenotype matrix are sent to the encoder, which is responsible for extracting key features from the raw data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; where m represents the number of samples and n represents the number of species;

[0160] The pathogen registration module is configured as follows: inputting the feature vector of the feature vector extraction module; calculating the correlation between each feature vector and constructing a correlation matrix P;

[0161] The interaction analysis module is configured as follows: the correlation matrix P of the pathogen registration module is input to conduct in-depth analysis of the interaction between features through a deep learning algorithm to construct a deep learning model; and the species interaction matrix is ​​output;

[0162] The conditional pathogen identification module is configured to: identify conditional pathogens based on the deep learning model and species interaction matrix constructed by the interaction analysis module;

[0163] The conditional pathogen identification module includes: a first-order conditional pathogen identification module, a second-order conditional pathogen identification module and a high-order conditional pathogen identification module; wherein the first-order conditional pathogen identification module is configured to: identify the characteristics of a single species through a deep learning model to determine which species may be conditional pathogens; the second-order conditional pathogen identification module is configured to: analyze the relationship between species through a deep learning model to identify the combination of species that may act together to cause disease under specific conditions; the high-order conditional pathogen identification module is configured to: analyze the interaction network structure between multiple species to identify the interaction pattern of conditional pathogens between multiple species;

[0164] The output module is configured to output the identification result of the conditional pathogen identification module.

[0165] Example 4

[0166] This embodiment provides a device for identifying conditional pathogens of a large cohort of metagenomes, including: an input module, a control module, and an output module;

[0167] The input module is configured to: input the sequencing data after quality control filtering and host removal or the metagenomic off-machine sequencing data;

[0168] The control module includes: a data preprocessing module, a feature vector extraction module, a pathogen registration module, an interaction analysis module and a conditional pathogen identification module;

[0169] The data preprocessing module is configured to: perform species annotation and relative abundance estimation on the metagenomic sequencing data after quality control filtering and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance; or, perform quality control filtering, host sequence removal, species annotation and relative abundance estimation on the metagenomic sequencing data, then merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance;

[0170] The feature vector extraction module is configured as follows: the species relative abundance matrix and the phenotype matrix are sent to the encoder, which is responsible for extracting key features from the raw data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; where m represents the number of samples and n represents the number of species;

[0171] The pathogen registration module is configured as follows: inputting the feature vector of the feature vector extraction module; calculating the correlation between each feature vector and constructing a correlation matrix P;

[0172] The interaction analysis module is configured as follows: the correlation matrix P of the pathogen registration module is input to conduct in-depth analysis of the interaction between features through a deep learning algorithm to construct a deep learning model; and the species interaction matrix is ​​output;

[0173] The conditional pathogen identification module is configured to: identify conditional pathogens based on the deep learning model and species interaction matrix constructed by the interaction analysis module;

[0174] The output module is configured to output the identification result of the conditional pathogen identification module.

[0175] The conditional pathogen identification module includes: a first-order conditional pathogen identification module and a second-order conditional pathogen identification module, wherein the first-order conditional pathogen identification module is configured to: identify single species characteristics through a deep learning model to determine which species may be conditional pathogens;

[0176] The second-order conditional pathogen identification module is configured to analyze the relationship between species through a deep learning model and identify species combinations that may work together to cause pathogens under specific conditions.

[0177] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying conditional pathogens in a large cohort of metagenomes, characterized in that: It includes the following steps: S1. Perform species annotation and relative abundance estimation on the metagenomic sequencing data after quality control filtration and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance of all samples; S2. Take the data after relative abundance leveling to construct species relative abundance matrix and phenotype matrix; send the species relative abundance matrix and phenotype matrix to the encoder, which is responsible for extracting key features from the original data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; Where m represents the number of samples and n represents the number of species; S3. The m feature vectors G and n feature vectors Y extracted by S2 are input into the pathogen registration module; the correlation between each feature vector is calculated to construct a correlation matrix P; S4. The correlation matrix P is sent to the interaction analysis module, and the interaction between features is deeply analyzed through the deep learning algorithm to build a deep learning model; Output species interaction matrix; S5. Identify conditional pathogens based on the deep learning model constructed in S4 and the species interaction matrix.

2. The method for identifying conditional pathogens of a large cohort of metagenomes according to claim 1, characterized in that: The step S5 includes: first-order conditional pathogen identification; The first-order conditional pathogen identification is used to: identify individual species characteristics through the deep learning model to determine which species may be conditional pathogens.

3. The method for identifying conditional pathogens of a large cohort of metagenomes according to claim 2, characterized in that: The step S5 also includes second-order conditional pathogen identification; The second-order conditional pathogen identification is used to: analyze the relationship between species through the deep learning model, and identify species combinations that may work together to cause pathogens under specific conditions.

4. The method for identifying conditional pathogens of a large cohort of metagenomes according to claim 1, characterized in that: The species annotation and relative abundance estimation in step S1 refers to: obtaining the species and abundance in the sample using species annotation software.

5. The method for identifying conditional pathogens of a large cohort of metagenomes according to claim 4, characterized in that: The species annotation software is selected from at least one of Kraken2, Bracken, MetaPhlAn, QIIME and Genometa.

6. The method for identifying conditional pathogens of a large cohort of metagenomes according to claim 5, characterized in that: Kraken2 and Bracken were used to perform species annotation and relative abundance estimation on the metagenomes of the samples, respectively.

7. The method for identifying conditional pathogens of a large cohort of metagenomes according to claim 4, characterized in that: The sample is selected from at least one of donor sputum, alveolar lavage fluid and stool samples.

8. A method for identifying interactions of conditional pathogens in a large cohort of metagenomics, characterized in that: It includes the following steps: S1. Perform species annotation and relative abundance estimation on the sequencing data after quality control filtering and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance; S2. Take the data after relative abundance leveling to construct species relative abundance matrix and phenotype matrix; send the species relative abundance matrix and phenotype matrix to the encoder, which is responsible for extracting key features from the original data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; Where m represents the number of samples and n represents the number of species; S3. The m feature vectors G and n feature vectors Y extracted by S2 are input into the pathogen registration module; the correlation between each feature vector is calculated to construct a correlation matrix P; S4. The correlation matrix P is sent to the species interaction analysis module, and the interaction between features is deeply analyzed through the deep learning algorithm to build a deep learning model; Output species interaction matrix; S5. Identify conditional pathogens based on the deep learning model and species interaction matrix constructed in S4, including: first-order conditional pathogens identification, second-order conditional pathogens identification and high-order conditional pathogens identification; The first-order conditional pathogen identification is used to: identify individual species characteristics through the deep learning model to determine which species may be conditional pathogens; The second-order conditional pathogen identification is used to: analyze the relationship between species through the deep learning model to identify species combinations that may work together to cause disease under specific conditions; The high-order conditional pathogen identification is used to: analyze the interaction network structure among multiple species to identify the interaction pattern of conditional pathogens among multiple species.

9. A device for identifying conditional pathogens of a large cohort of metagenomes, characterized in that: include: Input module, control module and output module; The input module is configured to: input the sequencing data after quality control filtering and host removal or the metagenomic off-machine sequencing data; The control module includes: a data preprocessing module, a feature vector extraction module, a pathogen registration module, an interaction analysis module and a conditional pathogen identification module; The data preprocessing module is configured to: perform species annotation and relative abundance estimation on the metagenomic sequencing data after quality control filtering and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance; or, perform quality control filtering, host sequence removal, species annotation and relative abundance estimation on the metagenomic sequencing data, then merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance; The feature vector extraction module is configured as follows: the species relative abundance matrix and the phenotype matrix are sent to an encoder, and the encoder is responsible for extracting key features from the raw data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; wherein m represents the number of samples and n represents the number of species; The pathogen registration module is configured to: input the feature vector of the feature vector extraction module; calculate the correlation between each feature vector and construct a correlation matrix P; The interaction analysis module is configured to: input the correlation matrix P of the pathogen registration module, conduct in-depth analysis of the interaction between features through a deep learning algorithm, and construct a deep learning model; output a species interaction matrix; The conditional pathogen identification module is configured to: identify conditional pathogens based on the deep learning model constructed by the interaction analysis module and the species interaction matrix; The output module is configured to output the identification result of the conditional pathogen identification module.

10. The device for identifying conditional pathogens of a large cohort of metagenomes according to claim 9, characterized in that: The conditional pathogen identification module includes: a first-order conditional pathogen identification module, and the first-order conditional pathogen identification module is configured to: identify single species characteristics through the deep learning model to determine which species may be conditional pathogens.

11. The device for identifying conditional pathogens of a large cohort of metagenomes according to claim 10, characterized in that: The conditional pathogen identification module also includes a second-order conditional pathogen identification module; the second-order conditional pathogen identification module is configured to: analyze the relationship between species through the deep learning model, and identify species combinations that may work together to cause pathogens under specific conditions.

12. A device for identifying the interaction of conditional pathogens in a large cohort of metagenomes, characterized in that: include: Input module, control module and output module; The input module is configured to: input the sequencing data after quality control filtering and host removal or the metagenomic off-machine sequencing data; The control module includes: a data preprocessing module, a feature vector extraction module, a pathogen registration module, an interaction analysis module and a conditional pathogen identification module; The data preprocessing module is configured to: perform species annotation and relative abundance estimation on the metagenomic sequencing data after quality control filtering and host removal, merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance; or, perform quality control filtering, host sequence removal, species annotation and relative abundance estimation on the metagenomic sequencing data, then merge the relative abundance of all samples, delete samples with very low total abundance, and level the relative abundance; The feature vector extraction module is configured as follows: the species relative abundance matrix and the phenotype matrix are sent to an encoder, and the encoder is responsible for extracting key features from the raw data; for the species relative abundance matrix, the encoder extracts m feature vectors G; for the phenotype matrix, the encoder extracts n feature vectors Y; wherein m represents the number of samples and n represents the number of species; The pathogen registration module is configured to: input the feature vector of the feature vector extraction module; calculate the correlation between each feature vector and construct a correlation matrix P; The interaction analysis module is configured to: input the correlation matrix P of the pathogen registration module, conduct in-depth analysis of the interaction between features through a deep learning algorithm, and construct a deep learning model; output a species interaction matrix; The conditional pathogen identification module is configured to: identify conditional pathogens based on the deep learning model constructed by the interaction analysis module and the species interaction matrix; The conditional pathogen identification module includes: a first-order conditional pathogen identification module, a second-order conditional pathogen identification module and a high-order conditional pathogen identification module; wherein the first-order conditional pathogen identification module is configured to: identify the characteristics of a single species through the deep learning model to determine which species may be conditional pathogens; the second-order conditional pathogen identification module is configured to: analyze the relationship between species through the deep learning model to identify the combination of species that may work together to cause disease under specific conditions; the high-order conditional pathogen identification module is configured to: analyze the interaction network structure between multiple species to identify the interaction pattern of conditional pathogens between multiple species; The output module is configured to output the identification result of the conditional pathogen identification module.

13. A device for identifying conditional pathogens of a large cohort of metagenomes or a device for identifying interactions of conditional pathogens of a large cohort of metagenomes, characterized in that: The device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the method for identifying conditional pathogens of a large cohort of metagenomes as described in any one of claims 1 to 7 or the method for identifying interactions of conditional pathogens of a large cohort of metagenomes as described in claim 8.

14. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the method for identifying conditional pathogens of a large cohort of metagenomes as described in any one of claims 1 to 7 or the method for identifying the interaction of conditional pathogens of a large cohort of metagenomes as described in claim 8.

Images