Method and system for identifying Shigella and / or diarrhea-causing Escherichia coli

Abstract

The present invention relates to a method and system for identifying Shigella and / or diarrhea-causing Escherichia coli, comprising: collecting mass spectrum data of known strains through paper spray ionization; generating training samples based on the mass spectrum data of known strains; constructing a strain prediction model, and training the strain prediction model using the training samples; collecting mass spectrum data of the strain to be tested through paper spray ionization; predicting the strain to be tested using the trained strain prediction model based on the mass spectrum data of the strain to be tested, and determining the type of the strain to be tested. The method and system can directly identify strain samples without sample pretreatment, have low analysis cost, are easy to operate, can identify strain categories in high throughput and real time, and have high accuracy and reliability in identification results and fast identification speed.

Description

Technical Field

[0001] The invention belongs to the technical field of microbial testing, and in particular relates to a method and a system for identifying Shigella and / or diarrhea-causing Escherichia coli. Background Art

[0002] Shigella ( Shigella , SH), also known as Shigella dysenteriae, a type of Gram-negative bacillus, is the most common pathogenic microorganism that causes bacterial dysentery. According to the difference in bacterial antigens, Shigella is divided into four groups, namely: Group A: Shigella dysenteriae ( S.dysenteriae , SDA), Group B: Shigella flexneri ( S. fexneri , SFB), Group C: Shigella boydii ( S.boydii , SBC), Group D: Shigella sonnei ( S. sonnei , SSD).

[0003] Diarrhea-causing Escherichia coli ( Diarrheagenic Escherichia coli DEC) is a type of Escherichia coli that can cause diarrhea in humans. It is also a type of Gram-negative bacillus that can cause human disease by contaminating food. Common DEC mainly include enteropathogenic Escherichia coli ( Enteropathogenic Escherichia coli , EPEC), enteroinvasive Escherichia coli ( Enteroinvasive Escherichia coli , EIEC), enterotoxigenic Escherichia coli ( Enterotoxigenic Escherichia coli , ETEC), Shiga toxin-producing Escherichia coli ( Shiga toxin-producing Escherichia coli , STEC) (also known as enterohemorrhagic Escherichia coli ( Enterohemorrhagic Escherichia coli , EHEC)) and enteroaggregative Escherichia coli ( Enteroaggregative Escherichia coli , EAEC) 5 categories.

[0004] All strains of SH and DEC are Gram-negative bacteria and belong to the family Enterobacteriaceae. They can cause serious intestinal diseases. They all have the same diarrheal symptoms in the early stage of infection, but their pathogenic mechanisms are different. The clinical symptoms and severity of the late infection are different. For example, ETEC infection is often self-limiting diarrhea, which usually heals itself in 2 to 3 days; EPEC infection is highly contagious and can cause death in severe cases; SDA and STEC can secrete highly toxic Shiga toxin protein, causing symptoms such as hemolytic uremic syndrome and causing death. Compared with SDA, some specific subtypes of STEC may have higher toxicity and severity. Therefore, accurate identification of Shigella and diarrhea-causing Escherichia coli and their strains is very important for clinical diagnosis, prevention and control of Shigella and Escherichia coli infections, and formulation of effective treatment plans.

[0005] However, since the strains of SH and DEC are closely related in phenotype and highly similar in genotype, they are highly similar and closely related to each other, and are closely related microorganisms (Zhao, AL, et al. Journal of Disease Surveillance, 2011, 26(1), 65-67). It is difficult to directly achieve accurate differentiation and identification of SH and DEC strains through traditional detection methods including biochemical detection, serotype identification, nucleic acid amplification, whole genome sequencing, etc., and the identification steps are complicated, time-consuming, and require the combination of multiple methods. There are also problems such as false positives in the identification results. Therefore, it is necessary to develop methods for accurately and quickly distinguishing and identifying SH and / or DEC. Summary of the invention

[0006] The present invention provides a method for identifying Shigella and / or diarrhea-causing Escherichia coli, comprising:

[0007] Mass spectrometric data of known strains were collected by paper spray ionization;

[0008] Generate training samples based on mass spectrometry data of known strains;

[0009] Construct a strain prediction model and use training samples to train the strain prediction model;

[0010] The mass spectrometry data of the strains to be tested were collected by paper spray ionization;

[0011] Based on the mass spectrometry data of the strain to be tested, the strain to be tested is predicted using the trained strain prediction model to determine the type of the strain to be tested, where:

[0012] The known strains include at least two strains selected from the group consisting of Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli, and enteroaggregative Escherichia coli;

[0013] The mass spectral data of the known strains include mass spectral data of at least two strains selected from Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli and enteroaggregative Escherichia coli;

[0014] The strain prediction model is the LightGBM model, the Bagging model or the Extra Trees model.

[0015] In certain embodiments, the method of identifying Shigella and / or diarrheagenic Escherichia coli is a method for non-diagnostic purposes.

[0016] In certain embodiments, after collecting mass spectrometry data of known strains, the method further comprises:

[0017] Mass spectrometry data of known strains were normalized to eliminate bias and noise in the data.

[0018] In certain embodiments, collecting mass spectrometry data of a known strain or a test strain by paper spray ionization comprises:

[0019] Preparing the paper base, including: soaking the paper base in methanol for 3 h to 12 h, and drying;

[0020] Place the known strain or the strain to be tested on a dry paper base;

[0021] Add 3 μL to 10 μL DMF to the known strain or the strain to be tested on the paper substrate and dry;

[0022] Collect mass spectrometry data of known strains or strains to be tested.

[0023] In certain embodiments, the paper base includes a sharp corner having an angle of 18° to 22°.

[0024] In certain embodiments, the paper base includes a sharp corner having an angle of 20°.

[0025] In certain embodiments, collecting mass spectrometry data of a known strain or a strain to be tested includes:

[0026] Methanol was selected as the elution solvent, and the elution volume was 100 μL ~180 μL.

[0027] In certain embodiments, collecting mass spectrometry data of a known strain or a strain to be tested includes:

[0028] Methanol was selected as the elution solvent, and the elution volume was 110 μL ~150 μL.

[0029] In certain embodiments, after collecting the mass spectrometry data of the strain to be tested, the method further comprises: normalizing the mass spectrometry data of the strain to be tested to eliminate deviations and noises in the data.

[0030] The present invention also provides a system for identifying Shigella and / or diarrhea-causing Escherichia coli, comprising:

[0031] a paper spray ionization mass spectrometer configured to collect mass spectral data of known strains or strains to be tested;

[0032] A training sample generation module is configured to generate training samples based on mass spectrometry data of known strains;

[0033] A model training module is configured to construct a strain prediction model and train the strain prediction model using training samples;

[0034] The prediction module is configured to predict the strain to be tested based on the mass spectrometry data of the strain to be tested using the trained strain prediction model to determine the type of the strain to be tested, wherein:

[0035] The known strains include at least two strains selected from the group consisting of Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli, and enteroaggregative Escherichia coli;

[0036] The mass spectral data of the known strains include mass spectral data of at least two strains selected from Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli and enteroaggregative Escherichia coli;

[0037] The strain prediction model is the LightGBM model, the Bagging model or the Extra Trees model.

[0038] In certain embodiments, the system further comprises:

[0039] The preprocessing module is configured to perform normalization processing on the mass spectrometry data of the known strain or the mass spectrometry data of the strain to be tested to eliminate the deviation and noise in the data.

[0040] The present invention also provides a device for identifying Shigella and / or diarrhea-causing Escherichia coli, comprising:

[0041] A memory; and a processor coupled to the memory, the processor being configured to execute the method according to any one of the embodiments of the present invention based on instructions stored in the memory.

[0042] The present invention also provides a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the steps of the method described in any embodiment of the present invention.

[0043] Beneficial effects of the present invention:

[0044] The method for identifying Shigella and / or diarrhea-causing Escherichia coli provided by the present invention successfully solves the problem that traditional technologies are difficult to accurately distinguish and identify Shigella and / or diarrhea-causing Escherichia coli species, and has one or more of the following advantages:

[0045] 1) Strains can be directly identified without sample pretreatment.

[0046] 2) Using paper spray ionization mass spectrometry (PSI-MS) technology, the strain sample analysis can be completed on a piece of paper, with low analysis cost.

[0047] 3) Easy to operate,

[0048] 4) Ability to identify strain types in high throughput and real time,

[0049] 5) The identification results are highly accurate and reliable, with an accuracy rate of up to 94.44%.

[0050] 6) Fast identification speed,

[0051] 7) It can be applied to strain selection, poisoning identification, food safety, etc., and can be used to accurately identify the cause of disease and trace the source of infection in social public safety incidents such as food safety and poisoning control, and has a wide range of application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A schematic diagram showing an embodiment of the method for identifying SH and / or DEC provided by the present invention;

[0053] Figure 2 A schematic diagram of the paper base shape is shown;

[0054] Figure 3 A schematic diagram of the placement of the paper base is shown;

[0055] Figure 4 A schematic diagram showing a component carrying a paper base is placed in a connecting component and connected to a sample injection component of a mass spectrometer;

[0056] Figure 5 and Figure 6 PSI-MS spectra of nine strains, including SDA, SFB, SBC, SSD, EPEC, EIEC, EAEC, ETEC, and STEC, are shown;

[0057] Figure 7 The performance evaluation results of the trained LightGBM model provided in Example 1 of the present invention as a prediction model are shown, wherein A shows the learning curve evaluation results; B shows the ROC curve and AUC value evaluation results;

[0058] Figure 8 The performance evaluation results of the trained Bagging model provided in Example 2 of the present invention as a prediction model are shown, wherein A shows the learning curve evaluation results; B shows the ROC curve and AUC value evaluation results;

[0059] Fig. 9 The performance evaluation results of the trained ExtraTrees model provided in Example 3 of the present invention as a prediction model are shown, wherein A shows the learning curve evaluation results; B shows the ROC curve and AUC value evaluation results;

[0060] Fig.10 The results of external evaluation and verification of the prediction model provided by the embodiment of the present invention using new strain mass spectrometry data are shown, wherein A shows the prediction results of the new strain type using the trained LightGBM model as the prediction model; B shows the prediction results of the new strain type using the trained Bagging model as the prediction model; C shows the prediction results of the new strain type using the trained ExtraTrees model as the prediction model;

[0061] Fig.11 PSI-MS spectra obtained by culturing STEC in different LB culture dishes are shown;

[0062] Fig.12 The PSI-MS spectra obtained by culturing STEC for different time periods on the same LB culture dish are shown;

[0063] Fig.13 The PSI-MS spectrum obtained by directly applying STEC onto a paper substrate that has not been treated with methanol soaking is shown;

[0064] Fig.14 The PSI-MS spectrum obtained by applying STEC onto a paper substrate treated with acetone is shown;

[0065] Fig.15 The results of discriminatory identification of SH and / or DEC by applying the trained PCA model as a prediction model are shown;

[0066] Fig.16 The results of discriminatory identification of SH and / or DEC by applying the trained OPLS-DA model as a prediction model are shown;

[0067] Fig.17 The performance evaluation results of the trained SVM model provided in Comparative Example 4 of the present invention as a prediction model are shown, wherein A shows the learning curve evaluation results; B shows the ROC curve and AUC value evaluation results;

[0068] Fig.18 The performance evaluation results of the trained naive Bayes model provided in Comparative Example 4 of the present invention as a prediction model are shown, wherein A shows the learning curve evaluation results; B shows the ROC curve and AUC value evaluation results;

[0069] Fig.19A schematic diagram showing a system for identifying SH and / or DEC according to one embodiment of the present invention;

[0070] Fig. 20 A schematic diagram showing a system for identifying SH and / or DEC according to another embodiment of the present invention;

[0071] Fig.21 A schematic diagram showing the structure of an apparatus for identifying SH and / or DEC according to an embodiment of the present invention. DETAILED DESCRIPTION

[0072] The substantive content of the present invention is further described below in conjunction with specific embodiments of the present invention. It should be understood that the following examples are only used to illustrate the present invention, but are not intended to limit the scope of protection of the present invention. In the following examples, if specific conditions are not specified, they are carried out according to conventional conditions or manufacturer recommendations. The raw materials used are not specified by the manufacturer, and are all conventional products that can be obtained commercially.

[0073] Although many materials and operating methods used in the following examples are well known in the art, the present invention is still described in as much detail as possible. It is clear to those skilled in the art that the materials and operating methods used in the following examples are well known in the art unless otherwise specified.

[0074] The method for identifying Shigella and / or diarrhea-causing Escherichia coli provided by the present invention is described in detail below through specific examples.

[0075] Figure 1 The figure is a flow chart of an embodiment of the method for identifying Shigella and / or diarrhea-causing Escherichia coli according to the present invention.

[0076] like Figure 1 As shown:

[0077] Step 101, collecting mass spectrometry data of known strains by paper spray ionization.

[0078] The known strains include at least two strains selected from Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli and enteroaggregative Escherichia coli.

[0079] The mass spectral data of the known strains include mass spectral data of at least two strains selected from Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli and enteroaggregative Escherichia coli.

[0080] In one embodiment, the known strains include the above 9 strains, and the mass spectrum data of the known strains include the mass spectrum data of the above 9 strains.

[0081] Step 102: Generate training samples based on mass spectrometry data of known strains.

[0082] Step 103, constructing a strain prediction model, and using training samples to train the strain prediction model, wherein the strain prediction model is a LightGBM model, a Bagging model, or an Extra Trees model.

[0083] Step 104, collecting mass spectrum data of the strain to be tested by paper spray ionization.

[0084] Step 105 , predicting the strain to be tested using the trained strain prediction model based on the mass spectrum data of the strain to be tested, and determining the type of the strain to be tested.

[0085] In one embodiment, the method for identifying Shigella and / or diarrheagenic Escherichia coli is a method for non-diagnostic purposes.

[0086] In one embodiment, the method further comprises, after collecting mass spectrometry data of known strains:

[0087] Mass spectrometry data of known strains were normalized to eliminate bias and noise in the data.

[0088] In one embodiment, collecting mass spectrometry data of a known strain or a strain to be tested by paper spray ionization includes:

[0089] Preparing the paper base, including: soaking the paper base in methanol for 3 h to 12 h, and drying;

[0090] Place the known strain or the strain to be tested on a dry paper base;

[0091] Add 3 μL to 10 μL DMF to the known strain or the strain to be tested on the paper substrate and dry;

[0092] Collect mass spectrometry data of known strains or strains to be tested.

[0093] In one embodiment, the paper base includes a sharp corner with an angle of 18° to 22°.

[0094] In one embodiment, the paper base includes a sharp corner having an angle of 20°.

[0095] In one embodiment, 3 μL to 8 μL of DMF is added to the known strain or the strain to be tested carried on the paper substrate. In one embodiment, 4 μL to 6 μL of DMF, such as 5 μL of DMF, is added to the known strain or the strain to be tested carried on the paper substrate.

[0096] In one embodiment, collecting mass spectrometry data of a known strain or a strain to be tested includes:

[0097] Methanol was selected as the elution solvent, and the elution volume was 100 μL ~180 μL.

[0098] In one embodiment, collecting mass spectrometry data of a known strain or a strain to be tested includes:

[0099] Methanol was selected as the elution solvent, and the elution volume was 110 μL ~150 μL.

[0100] In one embodiment, collecting mass spectrometry data of a known strain or a strain to be tested includes:

[0101] Methanol was selected as the elution solvent, and the elution volume was 120 μL.

[0102] In one embodiment, PSI-MS is used to collect mass spectrometry data of known strains or strains to be tested, the mass range is set to 500-2000 Da, methanol is used as the elution solvent, 60 μL is added each time, and 120 μL is added twice for a total of 120 μL. In the positive ion mode, the spray voltage is set to 2.8 Kv, and the MS capillary temperature is set to 320°C.

[0103] In one embodiment, after collecting the mass spectrum data of the strain to be tested, the method further comprises: normalizing the mass spectrum data of the strain to be tested to eliminate deviation and noise in the data.

[0104] In one embodiment, the strain prediction model is a LightGBM model.

[0105] In one embodiment, the parameters of the LightGBM model trained include the maximum depth of the decision tree (max_depth), the number of decision trees (n_estimators), the learning rate (learning_rate), and the regularization coefficients (lambda, alpha).

[0106] In one embodiment, after training, the parameter values ​​of the LightGBM model are set to max_depth=3, n_estimators=100, learning_rate=0.1, lambda=0.1, and alpha=0.1. After evaluation, the trained LightGBM model predicts the strain to be tested with an accuracy of 93.52%, a precision of 94.46%, a recall of 95.17%, and an F1 score of 94.45%.

[0107] In one embodiment, the strain prediction model is a Bagging model.

[0108] In one embodiment, the parameters of the bagging model to be trained include the maximum number of samples (max_samples), the maximum number of features (max_features), and the number of decision trees (n_estimators).

[0109] In one embodiment, after training, the parameter values ​​of the bagging model are set to max_samples=1.0, max_features=0.5, and n_estimators=200. After evaluation, the trained bagging model predicts the strain to be tested with an accuracy of 91.67%, a precision of 92.04%, a recall of 91.5%, and an F1 score of 91.15%.

[0110] In one embodiment, the strain prediction model is an Extra Trees model.

[0111] In one embodiment, the parameters of the ExtraTrees model being trained include the maximum depth of the decision tree (max_depth), the minimum number of samples required to split an internal node (min_samples_split), the minimum number of samples required for a leaf node (min_samples_leaf), and the number of decision trees (n_estimators).

[0112] In one embodiment, after training, the parameter values ​​of the ExtraTrees model are set to max_depth = 20, min_samples_split = 2, min_samples_leaf = 1, and n_estimators = 100. After evaluation, the trained ExtraTrees model predicts the strain to be tested with an accuracy of 95.37%, a precision of 95.06%, a recall of 95.34%, and an F1 score of 94.99%.

[0113] The above embodiment realizes a method for identifying Shigella and / or diarrhea-causing Escherichia coli, successfully solving the problem that traditional technologies are difficult to accurately distinguish and identify SH and / or DEC strains, and the method can directly identify strain samples, does not require sample pretreatment, has low analysis cost, high throughput, simplicity, and real-time, with high accuracy of results, and fast and effective identification. Even the slowest prediction model only takes tens of seconds to complete.

[0114] In particular, the above-mentioned embodiment adopts PSI-MS technology to collect mass spectrometry data of samples, and the analysis of strain samples can be completed on a piece of paper. The analysis cost is low, and the cost of paper-based consumables is much lower than the consumables used in other mass spectrometry technologies, such as chromatographic columns, expensive solvents, etc., and the process is very simple, with ultra-low instrument dependence, convenient and fast, and the samples can be directly analyzed in an open atmospheric pressure environment.

[0115] The above embodiment of the present invention combines PSI-MS technology with machine learning to achieve the identification of closely related microorganisms SH and / or DEC strains. This method can be applied to strain selection, poisoning identification, food safety, etc., and provides the possibility for accurate identification of other closely related microorganisms. It can be used to accurately identify the cause of disease, trace the source of infection, etc. in social public safety events such as food safety and poisoning control, and has a wide range of application scenarios.

[0116] In one embodiment, Fig.19 As shown, the present invention also provides a system for identifying Shigella and / or diarrhea-causing Escherichia coli, comprising:

[0117] A paper spray ionization mass spectrometer 21 is configured to collect mass spectrum data of known strains or strains to be tested;

[0118] A training sample generating module 22 is configured to generate training samples based on mass spectrometry data of known strains;

[0119] The model training module 23 is configured to construct a strain prediction model and train the strain prediction model using training samples;

[0120] The prediction module 24 is configured to predict the strain to be tested based on the mass spectrometry data of the strain to be tested using the trained strain prediction model to determine the type of the strain to be tested, wherein:

[0121] The known strains include at least two strains selected from the group consisting of Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli, and enteroaggregative Escherichia coli;

[0122] The mass spectral data of the known strains include mass spectral data of at least two strains selected from Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli and enteroaggregative Escherichia coli;

[0123] The strain prediction model is the LightGBM model, the Bagging model or the Extra Trees model.

[0124] like Fig. 20 As shown, in one embodiment, the system further includes: a preprocessing module 25, configured to perform normalization processing on the mass spectrometry data of the known strain or the mass spectrometry data of the strain to be tested to eliminate deviations and noises in the data.

[0125] In one embodiment, Fig.21 As shown, the present invention also provides a device for identifying Shigella and / or diarrhea-causing Escherichia coli, comprising:

[0126] A memory 31; and a processor 32 coupled to the memory 31, wherein the processor 32 is configured to execute the method for identifying Shigella and / or diarrhea-causing Escherichia coli according to any embodiment of the present invention based on instructions stored in the memory 31.

[0127] In one embodiment, the apparatus may include a memory 31, a processor 32, a communication interface 33, and a bus 34. The memory 31 is used to store instructions, the processor 32 is coupled to the memory 31, and the processor 32 is configured to implement the above method for identifying Shigella and / or diarrhea-causing Escherichia coli based on the instructions stored in the memory 31.

[0128] The memory 31 may be a high-speed RAM memory, a non-volatile memory, etc., or a memory array. The memory 31 may also be divided into blocks, and the blocks may be combined into virtual volumes according to certain rules. The processor 32 may be a central processing unit CPU, or an application specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the method for identifying Shigella and / or diarrhea-causing Escherichia coli of the present invention.

[0129] The present invention also provides a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the steps of the method for identifying Shigella and / or diarrhea-causing Escherichia coli described in any embodiment of the present invention.

[0130] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, systems and apparatuses according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0131] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0132] The following are specific examples of the practical application of the method for identifying Shigella and / or diarrhea-causing Escherichia coli of the present invention.

[0133] The consumables and instruments used in the embodiment of the present invention include: filter paper Whatman 1 (460×570 mm, thickness 180 μm, purchased from Cytiva) and a Thermo Fisher LTQ Obitrap ETD linear ion trap electrostatic field orbital trap mass spectrometer.

[0134] Embodiment 1:

[0135] The present embodiment provides a prediction model for identifying Shigella and / or diarrhea-causing Escherichia coli based on PSI-MS. The strain is first subjected to resuscitation induction culture, and then a paper base is prepared and the strain is smeared on the paper base to prepare a sample. PSI-MS is used to collect mass spectrometry data, and a prediction model is established based on the collected mass spectrometry data. The model can be used to identify the type of unknown Shigella and / or diarrhea-causing Escherichia coli samples, thereby achieving accurate differentiation and identification between Shigella and / or diarrhea-causing Escherichia coli species.

[0136] 1) Culture medium preparation and strain culture

[0137] Non-selective LB solid medium: weigh 20 g LB broth and 15 g agar powder, dissolve in 1000 mL distilled water, sterilize at 121°C for 20 minutes, cool to about 40-50°C, pour into a 100 mm × 20 mm culture dish, spread evenly and solidify to form LB solid medium for later use.

[0138] Strain culture: Take 2 strains of frozen SDA, SFB, SBC, SSD, EPEC, EIEC, ETEC, STEC, and EAEC, respectively, and perform resuscitation and activation culture. First, streak the different strains on LB solid medium, and then culture them at 37°C for 16 h to obtain multiple single colonies for mass spectrometry analysis of strain samples.

[0139] 2) Paper base cutting and processing

[0140] The carrier Whatman1 paper was cut into a triangular shape with an angle of 20°, such as Figure 2 Then the cut paper base was placed in methanol for immersion for 6 h to remove impurities. After the immersion was completed, the paper base was placed in an oven and dried at 60°C for 1 h.

[0141] 3) Sample preparation

[0142] Use a sterile pipette tip to pick up multiple single colonies in the LB solid medium in step 1) and directly apply them to the treated paper base to form a thin layer. Dry it naturally at room temperature. After drying, add 5 μL of DMF solvent to the strain sample thin layer. After drying naturally, put it into the part corresponding to the paper base, such as Figure 3 shown.

[0143] 4) Collect mass spectrometry data of strains using PSI-MS

[0144] Place the component with the paper base into the connecting component and connect it to the injection component of the mass spectrometer. The paper tip of the paper base is aligned axially with the injection port of the mass spectrometer. Figure 4 PSI-MS was used for mass spectrometry data acquisition, with the mass range set to 500-2000 Da, methanol was used as the elution solvent, 60 μL was added each time, and 120 μL was added twice in total. The spray voltage was set to 2.8 Kv in the positive ion mode, and the MS capillary temperature was set to 320°C. Each strain was measured 30 times in parallel, and a total of 540 mass spectrometry data were obtained.

[0145] Figure 5 and Figure 6 The PSI-MS spectra of 9 strains are shown as examples. It can be seen that the spectra of different strains are relatively similar and cannot be directly distinguished and identified by PSI-MS mass spectrometry data.

[0146] 5) Build a prediction model

[0147] The mass spectrometry data of the strains were normalized using the minimum-maximum normalization method to scale the mass spectrometry data that might have been distributed in different ranges to the interval of [0,1], eliminating the bias and noise in the data. After normalization, the 540 mass spectrometry data were divided into a training set and a test set. In this example, the classification model—LightGBM model was selected to establish a prediction model, the prediction model was trained using samples from the training set, and the prediction model was verified using samples from the test set, which was specifically implemented using the LightGBM library (v3.3.2) of Python (v3.7.3).

[0148] The parameters of the LightGBM model training include the maximum depth of the decision tree (max_depth), the number of decision trees (n_estimators), the learning rate (learning_rate), and the regularization coefficients (lambda, alpha).

[0149] The hyperparameters of the LightGBM model, max_depth, n_estimators, learning_rate, lambda, and alpha, were adjusted and optimized by grid search and five-fold cross validation, and the final hyperparameter values ​​of the LightGBM model were determined to be max_depth=3, n_estimators=100, learning_rate=0.1, lambda=0.1, and alpha=0.1. Furthermore, the data set was divided into a training set and a test set in a ratio of 8:2, and the model was trained and predicted. The classification performance of the model was preliminarily evaluated by accuracy, precision, recall, and F1 score, with an accuracy of 93.52%, a precision of 94.46%, a recall of 95.17%, and an F1 score of 94.45%. This result shows that the LightGBM model can effectively solve complex multi-classification problems after training, and can better achieve the classification and identification of SH and / or DEC strains.

[0150] 6) Performance evaluation of prediction models — learning curve

[0151] Use Python (v3.7.3) to draw the learning curve of the prediction model, observe the changes in the training score and verification score of the learning curve as the number of training samples increases, evaluate the learning effect of the model, and help identify whether the model is overfitting or underfitting. Ideally, as the amount of training data increases, the training score and verification score will increase and the gap will gradually decrease, and eventually stabilize. The learning curve results are as follows Figure 7 As shown in A, the results show that after training, the LightGBM model predicts the accuracy of the test set samples to 93.52%.

[0152] 7) Performance evaluation of the prediction model—ROC curve and AUC value

[0153] Use Python (v3.7.3) to draw the ROC curve and AUC value of the prediction model. The ROC curve measures the classification ability of the model by showing the relationship between the true positive rate and the false positive rate of the model under different thresholds. The ROC curve should be as close to the upper left corner as possible. The closer the curve is to the upper left corner, the better the performance of the model. If the curve is close to the diagonal, it indicates that the classification effect of the model is close to random guessing. The AUC value can be interpreted as randomly selecting a positive example and a negative example. The probability that the model correctly judges that the positive example is higher than the probability of the negative example. The AUC value ranges from 0 to 1. The larger the value, the better the performance of the model. The ROC curve and AUC value results are shown in the figure. Figure 7 As shown in Figure B, the results show that after training, the LightGBM model is used to predict the types of SH and / or DEC, with good classification effect and high accuracy.

[0154] 8) Performance evaluation of the prediction model - Difference in accuracy between the training set and the test set

[0155] Use Python (v3.7.3) to calculate the accuracy difference between the training set and the test set of the prediction model, evaluate the generalization ability of the model, that is, the ability of the model to predict or classify new data that has not been seen, and evaluate the practicality and reliability of the model. The prediction accuracy difference (small difference) range of the training set and test set samples is set to 10%. When the prediction accuracy of the training set and test set samples is high, the difference between the two is within 10%. It is considered that the model has good performance, no over / underfitting, and good generalization ability. The model predicts the accuracy of the training set samples to be 100%, and the accuracy of the test set samples to be 93.52, with an accuracy difference of 6.48%.

[0156] 9) Performance evaluation of prediction models — cross validation

[0157] Python (v3.7.3) was used to cross-validate the prediction model and evaluate the generalization ability of the model. In the cross-validation, K-fold cross-validation was used to randomly divide the data set into K equal subsets, and one of the subsets was selected as the test set each time, and the remaining K-1 subsets were used as the training set. This process was repeated K times, and each subset had a chance to be used as the test set. In this way, it was ensured that each data point was used for training and testing, thereby reducing the variance of the model evaluation and providing a more stable and reliable estimate of the model's generalization ability. After cross-validation, the prediction accuracy of the model for 9 bacterial species was SDA-99%, SFB-97%, SBC-99%, SSD-97%, STEC-97%, EAEC-96%, EIEC-95%, EPEC-95%, and ETEC-96%, respectively.

[0158] 10) Performance evaluation of prediction models - calculation speed

[0159] Computation speed is one of the key indicators to measure the performance of the prediction model, which directly affects the feasibility and efficiency of the model in practical applications. By evaluating and optimizing the computation speed, the practicality of the model can be improved and the cost can be reduced. The training time of the LightGBM model is 3.54 s, the prediction time of the model for a single test sample is <0.01 s, and the prediction time of the model for the entire test set is 0.08 s.

[0160] Embodiment 2:

[0161] This embodiment provides another prediction model for identifying Shigella and / or diarrhea-causing Escherichia coli based on PSI-MS, which is different from Embodiment 1 only in that a classification model, namely a Bagging model, is selected to establish the prediction model. The prediction model is trained using samples of the training set, and the prediction model is verified using samples of the test set, which is specifically implemented by using the Scikit-learn library (v1.0.2) of Python (v3.7.3).

[0162] The parameters of the Bagging model training include the maximum number of samples (max_samples), the maximum number of features (max_features), and the number of decision trees (n_estimators).

[0163] The hyperparameters of the bagging model, max_samples, max_features, and n_estimators, were adjusted and optimized by grid search and five-fold cross validation, and the final values ​​of the hyperparameters of the bagging model were max_samples=1.0, max_features=0.5, and n_estimators=200. Furthermore, the data set was divided into a test set and a training set in a ratio of 8:2, and the model was trained and predicted. The classification performance of the model was preliminarily evaluated by accuracy, precision, recall, and F1 score, and then the model performance was further evaluated by learning curves, ROC curves and AUC values, the accuracy difference between the training set and the test set, cross validation, and calculation speed.

[0164] The evaluation results show that after training, the prediction accuracy of the Bagging model is 91.67%, the precision is 92.04%, the recall is 91.5%, and the F1 score is 91.15%. The learning curve results are as follows: Figure 8 As shown in A, the accuracy of the Bagging model after training is 91.67%. The ROC curve and AUC value are shown in Figure 8As shown in Figure B, after training, the accuracy of the bagging model for predicting the training set samples is 100%, the accuracy of predicting the test set samples is 100%, and the accuracy difference is 0%. Cross-validation shows that the prediction accuracy of the model for 9 bacterial species is SDA-96%, SFB-95%, SBC-96%, SSD-95%, STEC-94%, EAEC-95%, EIEC-94%, EPEC-91%, and ETEC-94%, respectively. The training time of the model is 30.43 s, the prediction time of the model for a single test sample is 0.07 s, and the prediction time of the model for the entire test set is 0.87 s. The results show that the bagging model can also effectively solve complex multi-classification problems and better realize the classification and identification of SH and / or DEC species.

[0165] Embodiment 3:

[0166] This embodiment provides another prediction model for identifying Shigella and / or diarrhea-causing Escherichia coli based on PSI-MS, which is different from Embodiment 1 only in that the classification model, ExtraTrees model, is selected to establish the prediction model. The prediction model is trained using samples of the training set, and the prediction model is verified using samples of the test set, which is specifically implemented by using the Scikit-learn library (v1.0.2) of Python (v3.7.3).

[0167] The parameters for which the ExtraTrees model is trained include the maximum depth of the decision tree (max_depth), the minimum number of samples required to split an internal node (min_samples_split), the minimum number of samples required for a leaf node (min_samples_leaf), and the number of decision trees (n_estimators).

[0168] The hyperparameters of the ExtraTrees classification model, max_depth, min_samples_split, min_samples_leaf, and n_estimators, were adjusted and optimized by grid search and five-fold cross validation, and the final hyperparameter values ​​of the ExtraTrees model were determined to be max_depth=20, min_samples_split=2, min_samples_leaf=1, and n_estimators=100. Furthermore, the data set was divided into a test set and a training set in a ratio of 8:2, and the model was trained and predicted. The accuracy, precision, recall, and F1 scores were used to preliminarily evaluate the model effect, and then the learning curve, ROC curve, AUC value, accuracy difference between the training set and the test set, cross validation, and calculation speed were used to further evaluate the model.

[0169] The evaluation results show that after training, the ExtraTrees model has a prediction accuracy of 95.37%, a precision of 95.06%, a recall of 95.34%, and an F1 score of 94.99%. The learning curve results are as follows: Fig. 9 As shown in A, the ExtraTrees model has been trained and the accuracy of predicting the test set samples is 100%. The ROC curve and AUC value are shown in Fig. 9 As shown in Figure B, after training, the ExtraTrees model predicts 100% accuracy for training set samples and 100% accuracy for test set samples, with an accuracy difference of 0%. Cross-validation shows that the prediction accuracy of the model for 9 bacterial species is SDA-98%, SFB-97%, SBC-98%, SSD-98%, STEC-97%, EAEC-95%, EIEC-97%, EPEC-98%, and ETEC-95%, respectively. The model training time is 0.85s, the model prediction time for a single test sample is 0.01s, and the model prediction time for the entire test set is 0.63s. The results show that the ExtraTrees model can also effectively solve complex multi-classification problems and achieve good classification and identification among SH and / or DEC species.

[0170] Embodiment 4:

[0171] Referring to steps 1) to 4) in Example 1, mass spectrometry data of 36 new strains were collected (two strains were cultured for SDA, SFB, SBC, SSD, STEC, EAEC, EIEC, EPEC, and ETEC, and two mass spectrometry data were collected for each strain), and the data were normalized. Then, the prediction model constructed in Examples 1-3 was externally validated and evaluated by four indicators: accuracy, precision, recall, and F1 score, and the results were visualized by a confusion matrix.

[0172] For the trained LightGBM model constructed in Example 1, the results of its external verification of the overall performance indicators showed that the prediction accuracy was 94.44%, the precision was 96.3%, the recall was 94.44%, and the F1 score was 94.6%. The prediction effect of each specific category of strains is shown in Table 1, where the macro average is the average of the precision, recall, and F1 scores of all strain categories, without considering the number of samples in each category, providing an average performance indicator that does not consider the category distribution. The weighted average is the average of the precision, recall, and F1 scores of all strain categories, but takes into account the number of samples in each category, providing an average performance indicator weighted according to the category distribution. Support is the number of samples. The performance details of the model in each specific category help to identify which categories the model performs well and which categories it performs poorly. The model parameters can be adjusted more finely to improve the prediction performance of specific categories. The confusion matrix and its visualization results are shown in the figure. Fig.10 As shown in A.

[0173] Table 1 LightGBM model-specific category prediction effect

[0174]

[0175] For the trained Bagging model constructed in Example 2, the external verification overall performance index results show that the prediction accuracy is 94.44%, the precision is 95.56%, the recall is 94.44%, and the F1 score is 94.36%. The prediction effect of each specific strain category is shown in Table 2, and the confusion matrix results are shown in Fig.10 As shown in B.

[0176] Table 2 Bagging model-specific category prediction effect

[0177]

[0178] For the trained ExtraTrees model constructed in Example 3, the external validation overall performance index results showed that the prediction accuracy was 94.44%, the precision was 95.56%, the recall was 94.44%, and the F1 score was 94.36%. The prediction effect of each specific strain category is shown in Table 3, and the confusion matrix results are shown in Fig.10 As shown in C.

[0179] Table 3 ExtraTrees model-specific category prediction effect

[0180]

[0181] The above verification results show that the trained LightGBM, Bagging, and ExtraTrees classification models have good classification effects on mass spectrometry data of different bacterial species, and can better achieve accurate differentiation and identification between Shigella and DEC species.

[0182] Example 5 Stability evaluation of mass spectrometry data collected using PSI-MS:

[0183] The same strain was inoculated into different LB solid media and cultured at 37°C for 16 h, or the same LB solid media was used for culture at different times under the same conditions, and then PSI-MS spectra were collected according to steps 2) to 4) of Example 1 to evaluate the stability of the mass spectrometry data of the strain. Taking STEC as an example, the spectra obtained by culturing with different LB solid media are as follows: Fig.11 As shown; the same LB culture dish was used for different culture times, and the obtained spectrum was as follows Fig.12 The results showed that the mass spectra of the same strain cultured in different LB culture dishes or cultured in the same LB culture dish for different time periods were highly similar, indicating that the mass spectrometry data of SH and DEC collected by PSI-MS were stable and reliable.

[0184] Comparative Example 1:

[0185] PSI-MS spectra were collected by referring to steps 2) to 4) of Example 1, except that in step 2), the cut paper base was not soaked in methanol, or was soaked in acetone for 6 h. Taking STEC as an example, the cut paper base was not soaked in methanol, and the mass spectrum obtained was as follows: Fig.13 As shown; the cut paper base was soaked in acetone for 6 hours, and the obtained mass spectrum was as shown Fig.14 The above results show that the correct mass spectrum of the strain cannot be obtained without methanol pretreatment of the paper substrate.

[0186] Comparative Example 2:

[0187] The difference between Comparative Example 2 and Example 1 is that the PCA model is selected to establish the prediction model. The prediction model is trained using samples of the training set, and the prediction model is verified using samples of the test set, which is specifically achieved by using the PCA algorithm of SIMCA.

[0188] PCA (Principal Component Analysis) is a method for analyzing data without label information or predefined classifications. It discovers patterns and structures in data by reducing dimensionality. It is an unsupervised discriminant analysis.

[0189] The trained PCA model can distinguish and identify the SH and DEC strains. Fig.15 As shown, the PCA model cannot achieve accurate differentiation and identification among SH and / or DEC strains.

[0190] Comparative Example 3:

[0191] The difference between Comparative Example 3 and Example 1 is that the orthogonal partial least squares discriminant analysis (OPLS-DA) model is selected to establish the prediction model. The prediction model is trained using samples of the training set, and the prediction model is verified using samples of the test set, which is specifically implemented by using the OPLS-DA model of the SIMCA software.

[0192] The orthogonal partial least squares discriminant analysis (OPLS-DA) model is a supervised discriminant analysis statistical method. The trained OPLS-DA model can distinguish and identify the SH and DEC strains. Fig.16 As shown, the OPLS-DA model cannot achieve accurate differentiation and identification among SH and / or DEC strains.

[0193] Comparative Example 4:

[0194] The difference between Comparative Example 4 and Example 1 is that a support vector machine (SVM) model is selected to establish a prediction model. The prediction model is trained using samples of the training set, and the prediction model is verified using samples of the test set, which is specifically implemented by using the Scikit-learn library (v1.0.2) of Python (v3.7.3).

[0195] The parameters of the SVM model that are trained include the penalty factor (C), kernel function parameter (gamma), and kernel function (kernel).

[0196] The hyperparameters C, gamma, and kernel of the SVM model were adjusted and optimized by grid search and five-fold cross validation, and the hyperparameter values ​​of the SVM model were finally determined to be C=1, gamma=scale, and kernel=rbf. Furthermore, the data set was divided into a test set and a training set in a ratio of 8:2, and the model was trained and predicted. The effect of the model was preliminarily evaluated by accuracy, precision, recall, and F1 score, and then the model was further evaluated by learning curves, ROC curves and AUC values, the accuracy difference between the training set and the test set, cross validation, and calculation speed.

[0197] The evaluation results show that after training, the SVM model has a prediction accuracy of 62.04%, a precision of 78.74%, a recall of 65.4%, and an F1 score of 64.8%. The learning curve results are as follows: Fig.17 As shown in A, after training, the SVM model can predict the accuracy of the test set samples to 62.04%. The ROC curve and AUC value are shown in Fig.17As shown in Figure B, after training, the SVM model has a prediction accuracy of 73.28% for the training set samples and 64.81% for the test set samples, with an accuracy difference of 8.47%. Cross-validation showed that the prediction accuracy of the model for 9 bacterial species was SDA-94%, SFB-90%, SBC-91%, SSD-91%, STEC-81%, EAEC-92%, EIEC-94%, EPEC-94%, and ETEC-89%, respectively. The model training time was 1.32 s, the model prediction time for a single test sample was 0.12 s, and the model prediction time for the entire test set was 0.16 s. The results show that the SVM model cannot achieve good classification and identification among SH and / or DEC species.

[0198] Comparative Example 5:

[0199] The difference between Comparative Example 5 and Example 1 is that the classification model, naive Bayes model, is selected to establish the prediction model. The prediction model is trained using samples of the training set, and the prediction model is verified using samples of the test set, which is specifically implemented by using the Scikit-learn library (v1.0.2) of Python (v3.7.3).

[0200] The parameters of the Naive Bayes model are trained as prior probabilities.

[0201] The hyperparameters priors of the naive Bayes classification model were adjusted and optimized by grid search and five-fold cross validation, and the hyperparameter value of the SVM model was finally determined as priors = none. Furthermore, the data set was divided into a test set and a training set in a ratio of 8:2, and the model was trained and predicted. The effect of the model was preliminarily evaluated by accuracy, precision, recall, and F1 score, and then the model was further evaluated by learning curves, ROC curves and AUC values, accuracy differences between training sets and test sets, cross validation, and calculation speed.

[0202] The evaluation results show that after training, the Naive Bayes model has a prediction accuracy of 50.93%, a precision of 54.93%, a recall of 48.18%, and an F1 score of 48.8%. The learning curve results are as follows: Fig.18 As shown in A, the Naive Bayes model has been trained and the accuracy of predicting the test set samples is 50.93%. The ROC curve and AUC value are shown in Fig.18As shown in Figure B, after training, the naive Bayes model has a prediction accuracy of 62.59% for the training set samples and 56.3% for the test set samples, with an accuracy difference of 6.29%. Cross-validation showed that the prediction accuracy of the model for 9 bacterial species was SDA-91%, SFB-71%, SBC-91%, SSD-92%, STEC-86%, EAEC-89%, EIEC-92%, EPEC-89%, and ETEC-87%, respectively. The model training time was 0.01 s, the model prediction time for a single test sample was 0.03 s, and the model prediction time for the entire test set was 0.14 s. The results show that the naive Bayes model cannot achieve accurate differentiation and identification between SH and / or DEC species.

[0203] In summary, the method for identifying SH and / or DEC provided in the embodiment of the present invention is suitable for the accurate differentiation and identification between SH and / or DEC strains. It only needs to collect PSI-MS mass spectrometry data of the strain, and then apply the prediction model provided in the embodiment of the present invention for analysis and calculation, so as to quickly and efficiently realize the accurate differentiation and identification between closely related microbial species. No sample pretreatment is required, the analysis cost is low, simple, real-time and fast, and it can be applied to strain detection, poisoning identification, food safety and other aspects, and has extremely high application value.

[0204] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying Shigella and / or diarrhea-causing Escherichia coli, comprising: Mass spectrometric data of known strains were collected by paper spray ionization; Generate training samples based on mass spectrometry data of known strains; Construct a strain prediction model and use training samples to train the strain prediction model; The mass spectrometry data of the strains to be tested were collected by paper spray ionization; Based on the mass spectrometry data of the strain to be tested, the strain to be tested is predicted using the trained strain prediction model to determine the type of the strain to be tested, where: The known strains include at least two strains selected from the group consisting of Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli, and enteroaggregative Escherichia coli; The mass spectral data of the known strains include mass spectral data of at least two strains selected from Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli and enteroaggregative Escherichia coli; The strain prediction model is the LightGBM model, the Bagging model, or the Extra Trees model; Mass spectrometry data of known or tested strains collected by paper spray ionization include: Preparing the paper base, including: soaking the paper base in methanol for 3 h to 12 h, and drying; Place the known strain or the strain to be tested on a dry paper base; Add 3 μL to 10 μL DMF to the known strain or the strain to be tested on the paper substrate and dry; Collect mass spectrometry data of known strains or strains to be tested.

2. The method of claim 1, wherein after collecting mass spectrometry data of the known strain or the strain to be tested, the method further comprises: The mass spectrometry data of known strains or strains to be tested are normalized to eliminate bias and noise in the data.

3. The method of claim 1, wherein the paper base comprises a sharp angle of 18 to 22 degrees.

4. The method of claim 1, wherein collecting mass spectrometry data of a known strain or a strain to be tested comprises: Methanol was selected as the elution solvent, and the elution volume was 100 μL ~180 μL.

5. A system for identifying Shigella and / or diarrhea-causing Escherichia coli, comprising: a paper spray ionization mass spectrometer configured to collect mass spectral data of known strains or strains to be tested; A training sample generation module is configured to generate training samples based on mass spectrometry data of known strains; A model training module is configured to construct a strain prediction model and train the strain prediction model using training samples; The prediction module is configured to predict the strain to be tested based on the mass spectrometry data of the strain to be tested using the trained strain prediction model to determine the type of the strain to be tested, wherein: The known strains include at least two strains selected from the group consisting of Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli, and enteroaggregative Escherichia coli; The mass spectral data of the known strains include mass spectral data of at least two strains selected from Shigella dysenteriae, Shigella flexneri, Shigella boydii, Shigella sonnei, enteropathogenic Escherichia coli, enteroinvasive Escherichia coli, enterotoxigenic Escherichia coli, Shiga toxin-producing Escherichia coli and enteroaggregative Escherichia coli; The strain prediction model is the LightGBM model, the Bagging model or the Extra Trees model.

6. The system of claim 5, further comprising: The preprocessing module is configured to normalize the mass spectrometry data of the known strain or the strain to be tested to eliminate the bias and noise in the data.

7. A device for identifying Shigella and / or diarrhea-causing Escherichia coli, comprising: Memory; and a processor coupled to the memory, wherein the processor is configured to execute the method according to any one of claims 1 to 4 based on instructions stored in the memory.

8. A computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 4.

Images