Pathogen recognition system and method based on multi-dimensional information and machine learning

Abstract

The invention discloses a pathogen recognition system and method based on multi-dimensional information and machine learning, and relates to the technical field of pathogen recognition, and the system comprises a pathogen sample collection module which is used for obtaining known background sample sequencing data and corresponding multi-environment index parameters; the gold standard type determination module is used for acquiring pathogen types of known background samples; the data set construction module is used for constructing a multi-dimensional information data set with labels; the pathogen recognition model module is used for constructing a machine learning classification model based on improved ensemble learning and training to obtain a pathogen recognition model; and the pathogen sample acquisition module also acquires to-be-identified sample sequencing data and corresponding multi-environment index parameters, and inputs the data into the pathogen identification model to obtain a pathogen identification result. According to the method, the multi-dimensional information data set with the labels and the machine learning classification model based on improved ensemble learning are constructed, the pathogen recognition model is obtained through training, and the accuracy and generalization ability of pathogen recognition are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pathogen identification, in particular to a pathogen identification system and method based on multi-dimensional information and machine learning. BACKGROUND

[0002] Pathogen identification is the core link of guaranteeing public health safety, food safety and clinical precise diagnosis and treatment. The contradiction between current actual demand and traditional technology limitation is highlighted, and the research in this field is imminent. Pathogen survival and transmission are affected by multiple environmental factors such as temperature and humidity. Traditional technology only relies on sequencing data, and the accuracy of pathogen identification in complex environment is low. Therefore, it is necessary to study the pathogen identification technology.

[0003] In the prior art, Chinese patent CN118072831A discloses a pathogen identification method, device, medium and product based on metagenome, which classifies the metagenome sample sequencing sequence first, determines the target pathogen according to the classification result, determines the type of housekeeping gene sequence according to the bacterial pathogen, metagenome sample sequencing sequence, metagenome alignment database and bacterial typing database when the target pathogen belongs to bacterial pathogen, and determines the type of bacterial pathogen based on this; when the target pathogen belongs to viral pathogen, all first genome sequences are obtained according to the viral pathogen, metagenome sample sequencing sequence, metagenome alignment database and viral typing database, and the type of viral pathogen is determined based on this.

[0004] However, the above-mentioned prior art only relies on metagenome sequencing sequence and specific typing database for pathogen identification and typing, does not consider the influence of multiple environmental factors, is easy to have information deviation, and has poor feature capturing ability, so that the accuracy and generalization ability of pathogen identification are poor. SUMMARY

[0005] The present application provides a pathogen identification system and method based on multi-dimensional information and machine learning, to solve the problem that the existing pathogen identification technology does not consider the influence of multiple environmental factors, is easy to have information deviation, has poor feature capturing ability, and has poor accuracy and generalization ability of pathogen identification.

[0006] In one aspect, the present application provides a pathogen identification system based on multi-dimensional information and machine learning, comprising: a pathogen sample collection module, a gold standard species determination module, a data set construction module, and a pathogen identification model module.

[0007] The pathogen sample collection module is configured to collect known background pathogen samples, and obtain corresponding known background sample sequencing data and corresponding multi-environment index parameters.

[0008] The gold standard species determination module is configured to obtain the pathogen species of the known background pathogen sample by using a gold standard method, i.e., a known background sample pathogen species.

[0009] The data set construction module is configured to construct a labeled multi-dimensional information data set by taking the known background sample pathogen species as a label of the known background sample sequencing data and combining the corresponding multi-environment index parameters.

[0010] The pathogen identification model module is configured to construct a machine learning classification model based on improved ensemble learning, train the machine learning classification model by using the labeled multi-dimensional information data set, and obtain a pathogen identification model.

[0011] The pathogen sample collection module is further configured to collect a to-be-identified pathogen sample, and obtain corresponding to-be-identified sample sequencing data and corresponding multi-environment index parameters.

[0012] The pathogen identification model module is further configured to input the to-be-identified sample sequencing data and the corresponding multi-environment index parameters into the pathogen identification model, and obtain a pathogen identification result.

[0013] In a possible implementation, the directional collection of the known background pathogen sample, the obtaining of the corresponding known background sample sequencing data, and the obtaining of the corresponding multi-environment index parameters include: The known background pathogen sample is collected in different pathogen hosts and environment types, and each group of the known background pathogen sample is divided into a sequencing part and an environment part.

[0014] The known background sample sequencing data is obtained by sequencing the sequencing part of the known background pathogen sample.

[0015] The corresponding multi-environment index parameters are obtained by sequencing the environment part of the known background pathogen sample.

[0016] In a possible implementation, the gold standard method includes pathogen isolation culture, specific nucleic acid detection, and serological identification.

[0017] In a possible implementation, the construction of the labeled multi-dimensional information data set by taking the known background sample pathogen species as the label of the known background sample sequencing data and combining the corresponding multi-environment index parameters includes: The known background sample sequencing data and the corresponding multi-environment index parameters are subjected to feature extraction and feature fusion to obtain a fusion feature matrix.

[0018] The known background sample pathogen species is taken as a label of the fusion feature matrix to obtain a labeled feature matrix, and a set of the labeled feature matrix is the labeled multi-dimensional information data set.

[0019] In a possible implementation, the machine learning classification model based on improved ensemble learning includes: a random forest base model, an XGBoost base model, a one-dimensional convolution base model, and a logistic regression meta-model.

[0020] The random forest base model, the XGBoost base model, and the one-dimensional convolution base model output a prediction probability vector as an input of the logistic regression meta-model.

[0021] When the machine learning classification model is trained by using the labeled multi-dimensional information dataset, the random forest base model, the XGBoost base model, and the one-dimensional convolution base model are trained respectively, and then the prediction probability vectors output by the trained random forest base model, XGBoost base model, and one-dimensional convolution base model are used as an input of the logistic regression meta-model to obtain a pathogen identification model.

[0022] In a possible implementation, the logistic regression meta-model uses a weighted cross-entropy loss as a loss function.

[0023] In a possible implementation, the pathogen identification model is input with the to-be-identified sample sequencing data and the corresponding multi-environment index parameters to obtain a pathogen identification result, including: The to-be-identified sample sequencing data and the corresponding multi-environment index parameters are subjected to feature extraction and feature fusion to obtain a to-be-identified fusion feature matrix.

[0024] The to-be-identified fusion feature matrix is input into the pathogen identification model to obtain a pathogen identification result.

[0025] In a possible implementation, the pathogen identification model module is further configured to verify the pathogen identification result by using a Bootstrap confidence.

[0026] In another aspect, the present application provides a pathogen identification method based on multi-dimensional information and machine learning, which uses the pathogen identification system based on multi-dimensional information and machine learning as described above, and includes the following steps: Step one, directional collection of known background pathogen samples, acquisition of corresponding known background sample sequencing data and corresponding multi-environment index parameters.

[0027] Step two, use of a gold standard method to obtain the pathogen species of the known background pathogen samples, i.e., known background sample pathogen species.

[0028] Step three, use of the known background sample pathogen species as a label of the known background sample sequencing data, in combination with the corresponding multi-environment index parameters, to construct a labeled multi-dimensional information dataset.

[0029] Step four, constructing a machine learning classification model based on improved ensemble learning, training the machine learning classification model with the labeled multi-dimensional information dataset to obtain a pathogen identification model.

[0030] Step five, collecting a to-be-identified pathogen sample, obtaining corresponding to-be-identified sample sequencing data and corresponding multi-environment index parameters.

[0031] Step six, inputting the to-be-identified sample sequencing data and the corresponding multi-environment index parameters into the pathogen identification model to obtain a pathogen identification result.

[0032] The pathogen identification system and method based on multi-dimensional information and machine learning in the present application have the following advantages: By constructing a labeled multi-dimensional information dataset and a machine learning classification model based on improved ensemble learning, a pathogen identification model is trained, wherein the labeled multi-dimensional information dataset is constructed based on known background sample sequencing data and corresponding multi-environment index parameters, which expands the feature dimension and reduces information bias; the machine learning classification model based on improved ensemble learning can capture the differentiated features through the integration of different classification models, further improving the reliability. The combination of the two technologies improves the accuracy and generalization ability of pathogen identification.

[0033] The proposed machine learning classification model based on improved ensemble learning includes a random forest base model, an XGBoost base model, a one-dimensional convolution base model, and a logistic regression meta-model; the prediction probability vectors output by the random forest base model, the XGBoost base model, and the one-dimensional convolution base model are used as the input of the logistic regression meta-model, wherein the random forest base model and the XGBoost base model process structured features, the one-dimensional convolution base model processes sequence features, and the logistic regression meta-model fuses the prediction probability vectors to improve the accuracy of pathogen identification.

[0034] The proposed logistic regression meta-model uses weighted cross-entropy loss as the loss function, balances the sample weight, and improves the identification accuracy of rare pathogens.

[0035] The proposed Bootstrap confidence is used to verify the pathogen identification result, reducing the misjudgment rate. DETAILED DESCRIPTION

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0037] Figure 1 A structural schematic diagram of a pathogen identification system based on multi-dimensional information and machine learning provided by an embodiment of the present application is provided. Figure 2 A flowchart of a pathogen identification method based on multi-dimensional information and machine learning provided by an embodiment of the present application is provided. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0039] As shown in Figure 1 The present application provides a pathogen identification system based on multi-dimensional information and machine learning, which comprises a pathogen sample collection module, a gold standard species determination module, a data set construction module, and a pathogen identification model module.

[0040] The pathogen sample collection module is configured to collect known background pathogen samples in a targeted manner, and obtain corresponding known background sample sequencing data and corresponding multi-environment index parameters.

[0041] The gold standard species determination module is configured to obtain the pathogen species of the known background pathogen samples by using a gold standard method, i.e., known background sample pathogen species.

[0042] The data set construction module is configured to use the known background sample pathogen species as the label of the known background sample sequencing data, and construct a labeled multi-dimensional information data set in combination with the corresponding multi-environment index parameters.

[0043] The pathogen identification model module is configured to construct a machine learning classification model based on improved ensemble learning, train the machine learning classification model using the labeled multi-dimensional information data set, and obtain a pathogen identification model.

[0044] The pathogen sample collection module is further configured to collect a to-be-identified pathogen sample, and obtain corresponding to-be-identified sample sequencing data and corresponding multi-environment index parameters.

[0045] The pathogen identification model module is further configured to input the to-be-identified sample sequencing data and the corresponding multi-environment index parameters into the pathogen identification model, and obtain a pathogen identification result.

[0046] Exemplarily, the targeted collection of known background pathogen samples, and the obtaining of corresponding known background sample sequencing data and corresponding multi-environment index parameters comprise: The known background pathogen samples are collected according to different pathogen hosts and environment types, and each group of known background pathogen samples is divided into sequencing portions and environment portions.

[0047] The sequencing portions of the known background pathogen samples are subjected to corresponding known background sample sequencing data.

[0048] The environment portions of the known background pathogen samples are subjected to corresponding multi-environment index parameters.

[0049] Specifically, in the present embodiment, the pathogen hosts include humans, birds, and plants; and the environment types include hospitals, food processing plants, freshwater lakes, transportation sites, and cold chains. The known background sample sequencing data is obtained by using existing sequencing techniques, which will not be described herein.

[0050] In the present embodiment, the multi-environment index parameters include temperature, humidity, light intensity, osmotic pressure, pH value, dissolved oxygen, soluble organic carbon concentration, total nitrogen concentration, disinfectant residue concentration, pollutant concentration, host cell density, and environmental microbial community abundance, a total of 12 indexes. Each index is collected three times, and the mean value is taken after removing the abnormal values. In other possible embodiments, the types of multi-environment index parameters can also be increased or decreased according to actual needs.

[0051] Exemplarily, the gold standard method includes pathogen isolation culture, specific nucleic acid detection, and serological identification.

[0052] Specifically, pathogen isolation culture is suitable for bacteria, fungi, and part of viruses that can be cultured in vitro. The known background pathogen samples are inoculated into selective culture medium, cultured under specific conditions, and the culture time is set according to the growth characteristics of the pathogen. The colony morphology, staining characteristics (such as Gram staining), and biochemical reactions (such as oxidase test and sugar fermentation test) are observed, and the pathogen species is determined by combining the colony characteristics database.

[0053] Specific nucleic acid detection uses real-time fluorescent quantitative PCR technology. Specific primers and probes are designed for the conserved gene sequences (such as bacterial 16S rRNA gene and viral ORF gene) of the target pathogen. The Ct value of the amplification curve is detected to determine whether the pathogen nucleic acid exists in the sample. The Ct value determination threshold is usually set between 30-38. The specific value can be adjusted according to the type of pathogen, the instruction manual of the detection kit, and the laboratory verification results. Positive determination is required to determine the pathogen species in combination with primer specificity.

[0054] Serological identification is based on the principle of antigen-antibody specific binding, and the enzyme-linked immunosorbent assay (ELISA) is used to detect pathogen-specific antigens (such as virus capsid protein) or antibodies (such as IgM / IgG) produced by the host against the pathogen in the sample. The result is determined by comparing the color intensity (OD value) with the critical value, and the specific correspondence of the antigen / antibody determines the pathogen species.

[0055] In other possible embodiments, other existing gold standard methods can also be used to obtain the pathogen species of the known background pathogen sample.

[0056] Exemplarily, the pathogen species of the known background sample is taken as a label of the known background sample sequencing data, and a corresponding multi-environment index parameter is used to construct a labeled multi-dimensional information data set. Feature extraction and feature fusion are performed on the known background sample sequencing data and the corresponding multi-environment index parameter to obtain a fusion feature matrix.

[0057] The pathogen species of the known background sample is taken as a label of the fusion feature matrix to obtain a labeled feature matrix, and a set of the labeled feature matrix is the labeled multi-dimensional information data set.

[0058] Specifically, in the present embodiment, GC content, gene density, ORF average length, repeat sequence proportion, low complexity sequence proportion, sequencing depth, genome coverage, virulence gene number, drug resistance gene number, and characteristic k-mer frequency are extracted from the known background sample sequencing data to form a 10-dimensional sequencing feature matrix Finit. seq In other possible embodiments, the types of sequencing features can also be increased or decreased according to actual needs.

[0059] The multi-environment index parameter is subjected to Z-score standardization processing to eliminate dimensional differences to obtain a multi-environment index matrix Einit. known =[e ij ] n×12 , where e ij represents the i-th group of j-th type of index parameter after standardization, n represents the number of samples, and the dimension is 12.

[0060] The fusion feature matrix Finit=[E known |F seq ] n×22 , and the dimension is 22.

[0061] In the present embodiment, the set of labeled feature matrices, i.e., the labeled multi-dimensional information data set, is divided into a training set, a test set, and a validation set in a ratio of 8:1:1, which are respectively used for subsequent training, testing, and validation of the machine learning classification model based on the improved ensemble learning.

[0062] Exemplarily, the machine learning classification model based on improved ensemble learning comprises a random forest base model, an XGBoost base model, and a one-dimensional convolution base model.

[0063] The random forest base model, the XGBoost base model, and the one-dimensional convolution base model output a prediction probability vector as an input of the logistic regression meta-model.

[0064] When the machine learning classification model is trained by using the labeled multi-dimensional information data set, the random forest base model, the XGBoost base model, and the one-dimensional convolution base model are trained respectively, and then the prediction probability vectors output by the trained random forest base model, XGBoost base model, and one-dimensional convolution base model are used as an input of the logistic regression meta-model to obtain a pathogen identification model.

[0065] Specifically, in the embodiment, the random forest base model processes structured features, 100-500 decision trees are set, and the maximum depth is adaptively adjusted; the XGBoost base model strengthens feature interaction (such as the synergistic effect of “temperature + GC content”), the learning rate is 0.01-0.1, and the number of trees is 50-200; the one-dimensional convolution base model processes sequence features, the input is one-hot encoding of reads, and the network structure is: one-dimensional convolution layer (64 filters, kernel size 3) → maximum pooling layer → one-dimensional convolution layer (128 filters, kernel size 3) → full connection layer (64 neurons). The adaptive PSO algorithm is used to optimize the hyperparameters of the three base models, and the target is the accuracy of the validation set: .

[0066] wherein, indicates the inertia weight in the t+1th iteration of the adaptive PSO algorithm, and respectively indicate the maximum inertia weight and the minimum inertia weight (in the embodiment, 0.9 and 0.4 are set respectively), t indicates the current iteration number, and T indicates the maximum iteration number. Through the adaptive PSO algorithm, when the initial iteration, t is small, close to 1, close to the maximum inertia weight, at this time, more attention is paid to global search, and help to explore more possible hyperparameter combinations of the three base models; when the iteration number is close to the maximum iteration number, close to 0, Close to the minimum inertia weight, more emphasis is placed on local search, fine adjustment within the found better hyperparameter region, and avoiding missing the optimal solution. By dynamically adjusting the inertia weight, the adaptive PSO algorithm can efficiently search for the optimal hyperparameters of the random forest base model, the XGBoost base model, and the one-dimensional convolution base model, and provide high-quality prediction probability vectors for the subsequent logistic regression meta-model.

[0067] Illustratively, the logistic regression meta-model adopts a weighted cross-entropy loss as a loss function.

[0068] Specifically, the weighted cross-entropy loss is as follows: .

[0069] wherein Loss represents the weighted cross-entropy loss, N represents the total number of samples participating in training, represents the weight coefficient of the kth pathogen, represents the true label of the ith sample corresponding to the kth pathogen (if the ith sample belongs to the kth pathogen, = 1, otherwise = 0), represents the probability that the ith sample is predicted to be the kth pathogen, N k represents the number of samples of the kth pathogen, and K represents the total number of pathogen categories. The weighted cross-entropy loss differentiates the errors of samples of different pathogen categories by differential weighting, which can solve the model bias problem caused by the small sample size of rare pathogens and improve the recognition accuracy of rare pathogens.

[0070] Illustratively, the sequencing data of the sample to be identified and the corresponding multi-environment index parameters are input into the pathogen identification model to obtain a pathogen identification result, which includes: The sequencing data of the sample to be identified and the corresponding multi-environment index parameters are subjected to feature extraction and feature fusion to obtain a fusion feature matrix to be identified.

[0071] The fusion feature matrix to be identified is input into the pathogen identification model to obtain a pathogen identification result.

[0072] Specifically, in this embodiment, the process of feature extraction and feature fusion on the sequencing data of the sample to be identified and the corresponding multi-environment index parameters is the same as the process of feature extraction and feature fusion on the sequencing data of the known background sample and the corresponding multi-environment index parameters.

[0073] Illustratively, the pathogen identification model module is further configured to verify the pathogen identification result using Bootstrap confidence.

[0074] Specifically, in the present embodiment, the pathogen identification result output by the pathogen identification model is (i.e., the predicted pathogen category of the i-th sample), and Bootstrap confidence is used for verification: the confidence is calculated by 100 times of resampling with replacement as follows: .

[0075] wherein B represents the total number of resampling (100 in the present embodiment), represents the predicted pathogen category of the i-th sample in the b-th resampling, represents an indicator function (1 if the prediction is consistent, otherwise 0). When the confidence is higher than a preset confidence threshold (0.85 in the present embodiment), it is indicated that the pathogen identification result is reliable.

[0076] As shown in Figure 2 , the present embodiment further provides a pathogen identification method based on multi-dimensional information and machine learning, which uses the pathogen identification system based on multi-dimensional information and machine learning described above, and includes the following steps: Step one, directional collection of known background pathogen samples, acquisition of corresponding known background sample sequencing data and corresponding multi-environment index parameters.

[0077] Step two, acquisition of the pathogen species of the known background pathogen sample by using a gold standard method, i.e., the known background sample pathogen species.

[0078] Step three, using the known background sample pathogen species as the label of the known background sample sequencing data, combining the corresponding multi-environment index parameters, and constructing a labeled multi-dimensional information dataset.

[0079] Step four, construction of a machine learning classification model based on improved ensemble learning, training of the machine learning classification model by using the labeled multi-dimensional information dataset, and obtaining a pathogen identification model.

[0080] Step five, collection of a to-be-identified pathogen sample, acquisition of corresponding to-be-identified sample sequencing data and corresponding multi-environment index parameters.

[0081] Step six, input of the to-be-identified sample sequencing data and the corresponding multi-environment index parameters into the pathogen identification model, and obtaining a pathogen identification result.

[0082] The embodiment of the application trains a pathogen identification model by constructing a labeled multi-dimensional information dataset and an improved ensemble learning-based machine learning classification model, wherein the labeled multi-dimensional information dataset is constructed based on known background sample sequencing data and corresponding multi-environment index parameters, which expands the feature dimension and reduces information bias; the improved ensemble learning-based machine learning classification model can capture the differentiated feature capture ability of different classification models, thereby further improving reliability. The combination of the two technologies improves the accuracy and generalization ability of pathogen identification.

[0083] The improved ensemble learning-based machine learning classification model includes a random forest base model, an XGBoost base model, a one-dimensional convolution base model, and a logistic regression meta-model; the prediction probability vectors output by the random forest base model, the XGBoost base model, and the one-dimensional convolution base model are used as the input of the logistic regression meta-model, wherein the random forest base model and the XGBoost base model process structured features, the one-dimensional convolution base model processes sequence features, and the logistic regression meta-model fuses the prediction probability vectors, thereby improving the accuracy of pathogen identification.

[0084] The logistic regression meta-model uses a weighted cross-entropy loss as the loss function, balances the sample weights, and improves the identification accuracy of rare pathogens.

[0085] The Bootstrap confidence is used to verify the pathogen identification result, thereby reducing the misjudgment rate.

[0086] Although the preferred embodiments of the application have been described, those skilled in the art who, once aware of the basic inventive concept, can make further changes and modifications to the embodiments. Therefore, the appended claims are intended to include the preferred embodiments and all changes and modifications falling within the scope of the application.

[0087] Obviously, those skilled in the art can make various modifications and variations to the application without departing from the spirit and scope of the application. Thus, if these modifications and variations of the application fall within the scope of the claims of the application and their equivalent technologies, the application also intends to include these modifications and variations.

Claims

1. A pathogen identification system based on multidimensional information and machine learning, characterized in that, include: Pathogen sample collection module, gold standard type determination module, dataset construction module, and pathogen identification model module; The pathogen sample collection module is configured to: collect known background pathogen samples in a directional manner, and obtain the corresponding known background sample sequencing data and corresponding multi-environmental index parameters; The gold standard type determination module is configured to: use the gold standard method to obtain the pathogen type of the known background pathogen sample, that is, the pathogen type of the known background sample; The dataset construction module is configured to: use the known background sample pathogen types as labels for the known background sample sequencing data, and combine them with corresponding multi-environmental index parameters to construct a labeled multidimensional information dataset; The pathogen identification model module is configured to: construct a machine learning classification model based on improved ensemble learning, and train the machine learning classification model using the labeled multidimensional information dataset to obtain the pathogen identification model; The pathogen sample collection module is also configured to: collect pathogen samples to be identified, and obtain corresponding sequencing data of the samples to be identified and corresponding multi-environmental index parameters. The pathogen identification model module is further configured to input the sequencing data of the sample to be identified and the corresponding multi-environmental index parameters into the pathogen identification model to obtain the pathogen identification result.

2. The pathogen identification system based on multi-dimensional information and machine learning according to claim 1, characterized in that, The targeted collection of known background pathogen samples, and the acquisition of corresponding known background sample sequencing data and corresponding multiple environmental index parameters, include: Collect known background pathogen samples according to different pathogen hosts and environmental types, and divide each group of known background pathogen samples into sequencing samples and environmental samples; Sequencing data of the known background samples were obtained from the sequencing copies of the known background pathogen samples. Obtain the corresponding multiple environmental index parameters for the environmental components of the known background pathogen samples. 3.The pathogen identification system based on multi-dimensional information and machine learning according to claim 1, wherein, The gold standard method includes: pathogen isolation and culture, specific nucleic acid detection, and serological identification. 4.The pathogen identification system based on multi-dimensional information and machine learning according to claim 1, wherein, Using the known background sample pathogen types as labels for the known background sample sequencing data, and combining them with corresponding multi-environmental index parameters, a labeled multidimensional information dataset is constructed, including: Feature extraction and feature fusion are performed on the known background sample sequencing data and corresponding multi-environment index parameters to obtain a fused feature matrix; The known background sample pathogen types are used as labels for the fused feature matrix to obtain labeled feature matrices. The set of labeled feature matrices is the labeled multidimensional information dataset. 5.The pathogen identification system based on multi-dimensional information and machine learning according to claim 1, wherein, The machine learning classification models based on improved ensemble learning include: random forest base model, XGBoost base model, one-dimensional convolutional base model, and logistic regression meta-model; The predicted probability vectors output by the random forest base model, the XGBoost base model, and the one-dimensional convolutional base model are used as the input to the logistic regression meta-model. When the labeled multi-dimensional information dataset is used to train the machine learning classification model, the random forest base model, the XGBoost base model and the one-dimensional convolution base model are trained respectively, and the prediction probability vectors output by the trained random forest base model, XGBoost base model and one-dimensional convolution base model are used as the input of the logistic regression meta-model to obtain a pathogen identification model.

6. The pathogen identification system based on multi-dimensional information and machine learning according to claim 5, characterized in that, The logistic regression meta-model uses a weighted cross-entropy loss as a loss function. 7.The pathogen identification system based on multi-dimensional information and machine learning according to claim 1, wherein, The pathogen identification model is input with the to-be-identified sample sequencing data and corresponding multi-environment index parameters to obtain a pathogen identification result. The to-be-identified sample sequencing data and corresponding multi-environment index parameters are subjected to feature extraction and feature fusion to obtain a to-be-identified fusion feature matrix. The to-be-identified fusion feature matrix is input into the pathogen identification model to obtain a pathogen identification result. 8.The pathogen identification system based on multi-dimensional information and machine learning according to claim 1, wherein, The pathogen identification model module is further configured to verify the pathogen identification result by using Bootstrap confidence.

9. A pathogen identification method based on multi-dimensional information and machine learning, using the pathogen identification system based on multi-dimensional information and machine learning according to any one of claims 1 to 8, characterized in that, The method comprises the following steps: Step one, directional collection of known background pathogen samples, acquisition of corresponding known background sample sequencing data and corresponding multi-environment index parameters; Step two, acquisition of the pathogen species of the known background pathogen samples by using a gold standard method, i.e., known background sample pathogen species; Step three, using the known background sample pathogen species as the label of the known background sample sequencing data, and combining the corresponding multi-environment index parameters to construct a labeled multi-dimensional information dataset; Step four, construction of a machine learning classification model based on improved ensemble learning, training of the machine learning classification model by using the labeled multi-dimensional information dataset, and obtaining a pathogen identification model; Step five, collection of to-be-identified pathogen samples, acquisition of corresponding to-be-identified sample sequencing data and corresponding multi-environment index parameters; Step six, input of the to-be-identified sample sequencing data and corresponding multi-environment index parameters into the pathogen identification model to obtain a pathogen identification result.

Images