Method for rapidly identifying streptococcus agalactiae and highly virulent strains thereof

Abstract

A method for rapidly identifying highly virulent strains of Streptococcus agalactiae, comprising: using a laser confocal Raman spectrometer to collect single-cell Raman spectra of clinical strains, and establishing a database; classifying data of III / ST17-type highly virulent GBS strains into a first group, named as GBS-HIGH, data of non-III / ST17-type GBS strains into a second group, named as GBS-OTHERS, and data of Escherichia coli into a third group, named as ECO; randomly dividing the three groups of data into a training set, a validation set and a test set in a ratio of 7:1:2; and on the basis of a ResNet18 residual neural network architecture, changing 2D convolution in a network to one-dimensional (1D) convolution suitable for a 1D Raman spectrum, determining optimal parameters by means of simulation experiments, training the weight of a model in the training set, and then selecting optimal hyperparameters by means of a result of the model on the verification set, so as to determine a trained model.

Description

Method for rapidly identifying streptococcus agalactiae and high virulence strains thereof TECHNICAL FIELD

[0001] The present application belongs to the field of bacterial identification methods, deep learning and Raman spectroscopy technology, and specifically relates to a method for rapidly identifying streptococcus agalactiae and high virulence strains thereof. BACKGROUND

[0002] Bacterial meningitis is an acute central nervous system infectious disease caused by bacteria, with high morbidity and mortality, affecting all populations, especially neonates. The neonatal period is the most vulnerable period of life with the highest risk of serious infection. Streptococcus agalactiae, also known as group B streptococcus (GBS), is a gram-positive bacterium and the main pathogen causing bacterial meningitis in neonates. Among the isolated clinical strains, most cases are serotypes III, Ia, V, Ib and II by serotyping of the capsular polysaccharide, with the most important being serotype III. GBS can also be subjected to multilocus sequence typing (MLST), a single cystic serotype III GBS clone belonging to the high virulence clone complex 17, which is almost only associated with meningitis and late-onset neonatal GBS infection cases, and is usually referred to as a high virulence clone, also known as III / ST17 GBS high virulence strain.

[0003] Meningitis caused by III / ST17 GBS high virulence strain is an important cause of neurological disability worldwide, having considerable emotional and economic impact on individuals, families and society. Delayed diagnosis and treatment have serious and long-term consequences for individuals, and the pathogenicity of neonatal GBS overlaps with the critical period of neurodevelopment, so timely diagnosis and treatment are particularly important.

[0004] Raman spectroscopy is a powerful and reliable technique for bacterial detection and characterization. Raman spectra originating from molecular group vibrations are considered as the "fingerprint" of the structure and composition of biological molecules, and have been widely used in various fields such as biomedical science, chemical detection and mineral analysis. Deep learning and machine learning methods can mine the knowledge hidden in data in a data-driven manner, and have been widely applied in image processing, speech recognition, classification and identification of different substances, etc. due to their strong characterization ability. The present application proposes the combination of Raman spectroscopy technology and deep learning, the collection of Raman spectroscopy data of clinical strains, and the rapid detection of GBS and III / ST17 high virulence strains. TECHNICAL PROBLEM

[0005] Traditional pathogen identification methods rely on phenotypic identification, mainly using staining method, culture method and simple biochemical test. Macro-genome next-generation sequencing technology can detect III / ST17 type GBS high virulence strain, which has the key advantage of complete bias compared with traditional PCR-based method. This method is not limited by the prior selection of suspicious pathogens, and can identify potential infectious factors in one test, which is especially important for limited samples obtained from newborns, but has the problems of high price and long time-consuming. Technical solutions

[0006] In order to overcome the deficiencies in the prior art, the present application provides a detection method for Streptococcus agalactiae and high virulence strain based on deep learning combined with Raman spectrum.

[0007] Raman spectrum can reflect the energy of chemical bond and judge the structure of compound. The single cell Raman spectrum of bacteria can reflect the composition and concentration of biological macromolecules and metabolites in intracellular, cell membrane and cell wall. The cell wall structure of high virulence and low virulence Streptococcus agalactiae has significant difference, so their Raman spectra also have corresponding characteristics. However, the difference of such Raman spectrum is difficult to observe by naked eye, so the present application uses the method of deep learning to train the Raman spectrum of bacteria, so as to establish an algorithm model which can correctly identify and classify. The bacteria are divided into three categories: GBS-HIGH (Streptococcus agalactiae high virulence strain), GBS-OTHERS (Streptococcus agalactiae non-high virulence strain) and ECO (Escherichia coli). The training group data is analyzed by deep learning, and then a deep learning analysis model is established based on the analysis result. Finally, the validation group data is used for verification, the model parameters are adjusted, and finally the Raman spectrum of the bacteria to be tested is tested.

[0008] In order to achieve the above application purposes and solve the technical problems, the technical solutions adopted are as follows:

[0009] A detection method for Streptococcus agalactiae and high virulence strain based on deep learning combined with Raman spectrum, comprising the following steps:

[0010] Step 1: Establishing a Raman spectrum database;

[0011] Step 2: Establishing a GBS and III / ST17 high virulence strain typing model;

[0012] Step 3: Evaluating the performance of GBS and III / ST17 high virulence strain typing model.

[0013] Step 1 includes the following contents:

[0014] Step 1.1 Strain identification: the stored clinical isolates were recovered to Columbia blood agar medium, and after overnight culture, a suitable amount of single colony was picked and evenly smeared on the target points of the MALDI target plate to form a thin layer. 1 μl of HCCA standard solvent was overlaid on the sample points and naturally air-dried at room temperature. The target plate was placed in the matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF-MS), and bacterial identification was performed according to the automatic identification process. When the identification result score was ≥2.0 and the consistency classification was A, the strain was confirmed as GBS;

[0015] Step 1.2 Determination of GBS III / ST17 high virulence strain: DNA of the strain after step 1.1 review was extracted, and the purity and concentration of the extracted DNA were detected by BioDrop ultraviolet spectrophotometer. The absorbance A260 / A280 was 1.8-2.0, and the DNA concentration was >50 ng / μl. 150 bp double-end sequencing strategy was used for whole genome sequencing, and Unicycler was used for splicing sequencing of original short read sequence data. According to the typing information, GBS was divided into III / ST17 high virulence strain and non-III / ST17 strain two groups;

[0016] Step 1.3 Strain treatment process: GBS III / ST17 high virulence strain group, GBS non-III / ST17 strain and control group strains were treated. An appropriate amount of colony was washed and suspended in a 1.5 ml centrifuge tube with 500 μl of sterile deionized water. Centrifugation was performed at 12000 rpm / min for 3 min, and the washing was repeated for 3 times. The turbidity reached 0.5 Macleod, and finally 200 μl was reserved for standby.

[0017] Step 1.4 Raman spectrum database establishment: laser confocal Raman spectrometer was used to detect GBS III / ST17 high virulence strain group, GBS non-III / ST17 strain group and Escherichia coli group to excite different Raman spectra, and collect and save.

[0018] Sample pretreatment before collecting Raman spectrum: 2 μl of sample liquid was taken and dropped on an aluminum sheet to form a droplet with a diameter of about 2 mm, and then naturally dried.

[0019] The parameters for collecting Raman spectrum were as follows: laser wavelength: 532 nm; microscope objective magnification: 100 times lens (numerical aperture N.A. = 0.9); laser power: 1%; grating line density: 1200 gr / mm; signal collection time: 50 seconds, single; pinhole aperture size: 300 μm; slit width: 200 μm; signal collection frequency range: 500 cm -1 -2000 cm -1 .

[0020] Step 2 includes the following:

[0021] Step 2.1 The following grouping is performed on the Raman spectrum data set collected in step 1: GBS III / ST17 high virulent strains, GBS non-III / ST17 strains, and Escherichia coli group. Each group is randomly divided into a training set, a validation set, and a test set in a ratio of 7:1:2;

[0022] Step 2.2 Data set processing: self-normalization is used as pre-processing for all Raman spectra collected in step 1;

[0023] Step 2.3 Model structure design: based on the ResNet18 residual neural network architecture, the 2D convolution in the network is changed to 1D convolution suitable for 1D Raman spectrum, the neural network structure model is improved, and the best parameters are determined through simulation experiments;

[0024] Step 2.4 Model training: cross-entropy is used as the loss function, ADAM is used as the training optimizer, the weights of the model are trained in the training set, and then the optimal hyperparameters are selected by the results of the model on the validation set to determine the trained model;

[0025] Step 2.5 Model testing: for the trained convolutional neural network model, the test set is used to test the classification effect of the model.

[0026] Step 3 includes the following:

[0027] The constructed typing model is evaluated for laboratory specificity and accuracy using subsequently collected GBS III / ST17 high virulent strains, GBS non-III / ST17 strains, Escherichia coli, and a blank control group. The application of the constructed typing model is evaluated using 8 clinical strains to evaluate the sensitivity, specificity, and accuracy of the typing model. The cerebrospinal fluid culture results are used as the standard for evaluation, and the discordant results are verified by metagenomics. Advantages

[0028] Compared with the prior art, the present application has the following advantages and positive effects:

[0029] 1. The present application overcomes the traditional pathogenic bacteria phenotype identification method, and innovatively uses a laser confocal Raman spectrometer to construct a GBS and III / ST17 high virulent strain typing model, realizing rapid identification of GBS and III / ST17 high virulent strains, and having the advantages of simplicity, rapidness, and high sensitivity.

[0030] 2. The application overcomes the shortcomings of traditional analysis methods requiring complex pretreatment of Raman spectra, and uses a deep learning method of convolutional neural network to innovatively establish a Raman fingerprint database of GBS and III / ST17 high virulent strains, thereby providing a high-efficiency and convenient rapid diagnostic tool for III / ST17 high virulent strain GBS. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed to be used in the embodiment description will be briefly introduced as follows:

[0032] Fig. 1 is a Raman spectrum of GBS III / ST17 high virulent strain, non-III / ST17 strain and Escherichia coli;

[0033] Fig. 2 is a PCA analysis diagram of GBS III / ST17 high virulent strain, non-III / ST17 strain and Escherichia coli in the embodiment 1 of the application.

[0034] Fig. 3 is a discrimination result of GBS III / ST17 high virulent strain, non-III / ST17 strain and Escherichia coli in the embodiment 1 of the application. BEST MODE FOR CARRYING OUT THE INVENTION

[0035] In order to make the above-mentioned objects, features and advantages of the application more apparent and easy to understand, the specific embodiments of the application will be described in detail below. EMBODIMENT

[0036] As shown in Fig. 1, the embodiment discloses a detection method of Streptococcus agalactiae and high virulent strain based on deep learning combined with Raman spectrum, which comprises the following steps:

[0037] Step 1: Establishing a Raman spectrum database;

[0038] Step 1 comprises the following contents:

[0039] Step 1.1 Strain identification: resuscitate the clinical isolates stored in Columbia blood agar medium, pick up an appropriate amount of single colony after overnight culture, evenly smear on the target points of MALDI target plate to form a thin layer. Cover 1 μl HCCA solution on the sample points and dry naturally at room temperature, and put the target plate into MALDI-TOF-MS to perform bacterial identification according to the automatic identification process. When the identification result score is ≥2.0 and the consistency classification is A, it is confirmed that the strain is GBS;

[0040] Step 1.2 Determination of GBS III / ST17 high virulent strain: DNA of the strains compounded in step 1.1 was extracted, and the purity and concentration of the extracted DNA were detected by BioDrop ultraviolet spectrophotometer, the absorbance A260 / A280 was 1.8-2.0, and the DNA concentration was >50 ng / μl. Whole genome sequencing was performed by using 150 bp double-end sequencing strategy, and the original short read sequence data was spliced by Unicycler. According to the typing information, GBS was divided into two groups of III / ST17 high virulent strain and non-III / ST17 strain, and the corresponding database was compared and searched.

[0041] Step 1.3 Strain processing flow: GBS III / ST17 high virulent strain group, GBS non-III / ST17 strain and Escherichia coli group were treated, and an appropriate amount of colonies were suspended in a 1.5 ml centrifuge tube with 500 μl of sterile deionized water, centrifuged at 12000 rpm / min for 3 min, and washed repeatedly for 3 times to make the turbidity reach 0.5 Mache, and finally 20 μl was reserved.

[0042] Step 1.4 Raman spectrum database establishment: laser confocal Raman spectrometer was used to detect GBS III / ST17 high virulent strain group, GBS non-III / ST17 strain group and Escherichia coli group, including bacterial suspension in step 1.3, to excite different Raman spectra, and collect and save.

[0043] Sample pretreatment before collecting Raman spectrum: 2 μl of sample liquid was taken and dropped on an aluminum sheet to form a liquid drop with a diameter of about 2 mm, and then naturally dried.

[0044] The parameters for collecting Raman spectrum are as follows: laser wavelength: 532 nm; microscope objective magnification: 100 times lens (numerical aperture N.A. = 0.9); laser power: 1%; grating line density: 1200 gr / mm; signal acquisition time: 30 seconds, single; pinhole aperture size: 300 μm; slit width: 200 μm; signal collection frequency range: 500 cm -1 -2000 cm -1 .

[0045] Step 2: Establishment of GBS and III / ST17 high virulent strain typing model;

[0046] Step 2 includes the following contents:

[0047] Step 2.1 The Raman spectrum data set collected in step 1 was grouped as follows: GBS III / ST17 high virulent strain, GBS non-III / ST17 strain, and Escherichia coli. Each group was randomly divided into training set, validation set and test set in the ratio of 7:1:2;

[0048] Step 2.2 Dataset Processing: The Raman spectra acquired in Step 1 were preprocessed using self-normalization;

[0049] Step 2.3 Model Structure Design: Based on the ResNet18 residual neural network architecture, the 2D convolution in the network is changed to 1D convolution suitable for 1-dimensional Raman spectra, and the optimal parameters are determined through simulation experiments;

[0050] Step 2.4 Model Training: Using cross-entropy as the loss function and ADAM as the training optimizer, train the model weights on the training set, and then select the optimal hyperparameters based on the model's results on the validation set to determine the trained model.

[0051] Step 2.5 Model Testing: For the trained convolutional neural network model, the classification performance of the model is tested using a test set. Specifically, the PCA function of the sklearn package in the Python software is used to perform PCA analysis on the last layer features extracted by the deep neural network, and deep learning analysis is used on the test group data to identify the highly virulent Streptococcus agalactiae type III / ST17.

[0052] Step 3: Evaluate the performance of the GBS and III / ST17 highly virulent strain typing models;

[0053] Step 3 includes the following:

[0054] The laboratory specificity and accuracy of the constructed typing model were evaluated using subsequently collected GBS III / ST17 highly virulent strains, GBS non-III / ST17 strains, and Escherichia coli groups. The typing model was then applied to eight clinical strains to evaluate its sensitivity, specificity, and accuracy. Cerebrospinal fluid culture results were used as the standard for evaluation, and the divergence results were validated using metagenomic sequencing.

[0055] Figure 1 shows the Raman spectral differences between the highly virulent GBS III / ST17 strain, non-GBSIII / ST17 strains, and Escherichia coli, where the horizontal axis represents Raman shift (Raman shift / cm). ‑1 The vertical axis represents Raman intensity, with black representing Escherichia coli, red representing GBSIII / ST17 highly virulent strains, and blue representing the Raman spectra of non-III / ST17 strains.

[0056] Figure 2 is a PCA analysis diagram of the three types of data in Example 1 of the present invention: GBS III / ST17 highly virulent strain, non-III / ST17 strain, and Escherichia coli.

[0057] The data is input into the established model, and the test group data is analyzed by deep learning to identify whether it is a III / ST17 high virulence strain. When the test data is tested, the model outputs the percentage of the test data identified as a high virulence strain, and when a strain is greater than 80% identified as a III / ST17 type GBS high virulence strain, it indicates that the strain is a III / ST17 type GBS high virulence strain. The identification accuracy is 93.18%, as shown in Figure 3.

[0058] It should be noted that the above examples are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, which should be covered by the scope of the claims of the present application.

Claims

The application relates to a method for rapidly identifying streptococcus agalactiae and high-toxicity strains of the same. (1) sample pretreatment: the bacterial samples are divided into three groups of high-toxicity streptococcus agalactiae, non-high-toxicity streptococcus agalactiae and escherichia coli, the bacterial samples are dropped on the surface of aluminum sheets, naturally dried, and Raman spectra are collected; (2) Raman spectrum database establishment: laser confocal Raman spectrometers are used to detect the three groups of bacterial samples of high-toxicity streptococcus agalactiae, non-high-toxicity streptococcus agalactiae and escherichia coli, different Raman spectra are excited to form, and the Raman spectra are collected and saved; (3) establishing a deep learning analysis model: the three groups of bacterial samples of high-toxicity streptococcus agalactiae, non-high-toxicity streptococcus agalactiae and escherichia coli are randomly divided into a training set, a verification set and a test set at a ratio of 7:1:2; the Raman spectra collected in step (2) are pretreated by using self-normalization; based on a ResNet18 residual neural network architecture, 2D convolution in the network is changed into 1D convolution suitable for 1D Raman spectrum, and the best parameters are determined through simulation experiments; the weights of the model are trained in the training set, and then the optimal hyperparameters are selected through the results of the model on the verification set to determine the trained model; for the trained convolutional neural network model, the classification effect of the model is tested by using the test set, and the performance of the high-toxicity streptococcus agalactiae, non-high-toxicity streptococcus agalactiae and escherichia coli typing model is evaluated; (4) the sample to be detected is pretreated according to step (1), the Raman spectrum is collected according to (2), and the deep learning analysis model in step (3) is used to identify the sample to be detected. The method for rapid identification of Streptococcus agalactiae and its high virulence strain according to claim 1, characterized in that: In step (1), the bacterial sample is suspended with sterile water, the turbidity is 0.5 Mie, 2 microliters are dropped on the surface of an aluminum sheet, and the aluminum sheet is naturally dried. The method for rapidly identifying Streptococcus agalactiae and its high virulence strain according to claim 1 or 2, characterized in that: In step (2), a laser confocal Raman spectrometer is selected, a 532 nm laser is used, the laser power is set to 1%, the grating of the spectrometer is selected to be 1200 g / mm, a single cell of the bacterial sample to be detected is found under a 100 times objective lens, the single spectrum collection time is set to 50s, and 100 spectra are collected for each strain of bacteria in the training group. The method for rapidly identifying Streptococcus agalactiae and its high virulence strain according to claim 1 or 2, characterized in that: In step (2), the spectral range 500-2000 cm -1 Pinhole aperture size: 300 μm; slit width: 200 μm. The method for rapidly identifying Streptococcus agalactiae and its high virulence strain according to claim 1 or 2, characterized in that: In step (3), the last layer of features extracted by the deep neural network is analyzed by using the PCA function of the sklearn package in the pytyhon software, and the test group data is analyzed by using the deep learning analysis to identify high-toxicity streptococcus agalactiae. The method for rapidly identifying Streptococcus agalactiae and its high virulence strain according to claim 1 or 2, characterized in that: In step (3), the weights of the model are trained in the training set, and the weights of the model are trained in the training set. The method for rapidly identifying Streptococcus agalactiae and its high virulence strain according to claim 1 or 2, characterized in that: In step (4), the sample data to be detected is input into the trained convolutional neural network model, the model outputs the percentage of the sample to be detected identified as a high-toxicity strain, and when the percentage is greater than 80%, the sample to be detected is identified as high-toxicity streptococcus agalactiae. The method for rapidly identifying Streptococcus agalactiae and its high virulence strain according to claim 1 or 2, characterized in that: The high-toxicity streptococcus agalactiae is serotype III and sequence type 17 (III / ST17) high-toxicity streptococcus agalactiae.

Images