阅读说明：本技术 一种利用拉曼光谱快速鉴别细菌和真菌的方法 (Method for rapidly identifying bacteria and fungi by using Raman spectrum ) 是由 徐晓刚 王明贵 衣晓飞 于 2020-06-11 设计创作，主要内容包括：本发明涉及一种利用拉曼光谱快速鉴别细菌和真菌的方法,对待鉴别的样品进行前处理,对处理后的样品进行拉曼图谱采集,对采集到的拉曼图谱进行去背景和归一化处理,利用处理后的图谱进行细菌真菌鉴别。与现有技术相比,本发明方法操作简单,在进行拉曼测试前不需要进行复杂的样品前处理。本方法测试所需菌量少,不需要进行增菌培养,可大大缩短检测时间,快速给出鉴别结果。利用细胞的拉曼图谱可以反应细胞化学成分的特点,通过比较细菌和真菌的拉曼图谱的特异峰位来准确鉴别细菌和真菌,可排除肉眼观察细胞形态大小并主观判断造成的误差。(The invention relates to a method for rapidly identifying bacteria and fungi by utilizing Raman spectrum, which comprises the steps of pretreating a sample to be identified, collecting a Raman spectrum of the treated sample, removing background and normalizing the collected Raman spectrum, and identifying bacteria and fungi by utilizing the treated spectrum. Compared with the prior art, the method is simple to operate, and does not need to carry out complicated sample pretreatment before carrying out Raman test. The method needs less bacteria for testing, does not need bacteria enrichment culture, can greatly shorten the detection time and quickly give an identification result. The characteristic that the Raman spectrum of the cell can reflect the chemical components of the cell is utilized, the bacteria and the fungi are accurately identified by comparing the specific peak positions of the Raman spectra of the bacteria and the fungi, and errors caused by the observation of the shape and the size of the cell by naked eyes and the subjective judgment can be eliminated.)

1. A method for rapidly identifying bacteria and fungi by using Raman spectrum is characterized in that a sample to be identified is pretreated, a Raman spectrum is collected on the treated sample, background removal and normalization treatment are carried out on the collected Raman spectrum, and bacteria and fungi are identified by using the treated spectrum.

2. The method for rapidly identifying bacteria and fungi by using Raman spectroscopy according to claim 1, comprising the steps of:

(1) sample pretreatment

Removing culture solution and blood from a sample to be identified by adopting a centrifugal method, then adding sterile water for washing, carrying out centrifugation to remove supernatant, then carrying out resuspension on the bacteria solution, dropwise adding the resuspended solution onto a low-Raman background glass slide, and carrying out Raman detection after room-temperature air drying;

(2) raman spectrum collection

Collecting a thallus single cell Raman spectrum by using a confocal micro-Raman spectrometer;

(3) raman spectrum pretreatment

Background removal and normalization processing are carried out on the acquired Raman spectrum, and the processed spectrum is used for subsequent data analysis;

(4) bacterial fungus identification

One or two methods are selected and used in combination to distinguish bacteria from fungi: one is to judge the bacteria and fungi by comparing the specific peak positions of the fingerprint areas of the bacteria and the fungi; and the other method is to train the Raman spectrum data of known classification through a machine learning algorithm, establish a bacterial fungi identification database, and substitute the new data into a database for test identification.

3. The method for rapidly identifying bacteria and fungi by using Raman spectroscopy as claimed in claim 2, wherein in step (1), the low Raman background slide comprises a single crystal CaF₂Glass slide, BaF₂A glass slide or an aluminized glass slide.

4. The method for rapidly identifying bacteria and fungi by using Raman spectroscopy as claimed in claim 2, wherein in step (1), the low Raman background glass slide is purchased from Shanghai deuterium peak medical instruments Inc., Cat No 1001.

5. The method for rapidly identifying bacteria and fungi by using Raman spectroscopy as claimed in claim 2, wherein in step (2), the silicon chip is calibrated before the test by using the confocal Raman microscope, so that the Raman spectrum peak shift of the silicon is 520.73cm, and thallus is found in the microscope field.

6. The method for rapidly identifying bacteria and fungi by using Raman spectroscopy according to claim 2, wherein in the step (2), the Raman spectrum is collected under the following collection conditions: a 532nm laser is used, the laser power is set to be 4-15 mW, the integration time is 5-20 s, and the spectrum collection range is set to include 300-1800 cm of the fingerprint area spectrum of the cell^-1。

7. The method for rapidly identifying bacteria and fungi by using Raman spectroscopy according to claim 2, wherein in the step (2), the number of Raman spectra collected per strain in the library construction is more than 200, and the test data is more than 10.

8. The method for rapidly identifying bacteria and fungi by using Raman spectroscopy according to claim 2, wherein in the step (3), the background removal is selected from linear or polynomial algorithms, the order and the number of points are respectively set, the normalization process is selected to be divided by the area of the spectrum or divided by the maximum value, and the same treatment is ensured to be carried out on all the spectra during the background removal and the normalization processes.

9. The method for rapidly identifying bacteria and fungi by using Raman spectroscopy as claimed in claim 2, wherein in the step (4), the specific peak positions of the fingerprint regions of bacteria and fungi are selected, and the fungi are analyzed by testing to contain specific substances such as Cytochrome C which is only distributed in mitochondria and has Raman shift of 750cm^-1、1128cm^-1、1583cm^-1The specific peak position of the bacteria is 810cm^-1And 1240cm^-1Two most typical RNA peaks and a peak at 1480cm^-1The amide II peak of (1) can be accurately determined by using the specific peak positionsBacteria and fungi were identified.

10. The method for rapidly identifying bacteria and fungi by using Raman spectroscopy according to claim 2, wherein in the step (4), the machine learning algorithm comprises PCA-DFA, PCA-SVM, CNN;

before DFA and SVM analysis, PCA analysis is carried out to reduce the dimension of multi-dimensional data, then DFA analysis or SVM analysis is used to carry out linear recombination on complex variables after dimension reduction, complexity is simplified, and therefore the maximum change between two or more groups can be seen, and the changes are transferable, so that the changes can be applied when new data are substituted;

CNN adopts mxnet deep learning frame to construct Densenet neural network for deep network learning and training, each layer in Densenet uses the characteristics of all previous layers as input, and its own characteristics as input of all subsequent layers.

Technical Field

The invention belongs to the technical field of microorganism identification, and particularly relates to a method for rapidly identifying bacteria and fungi by using Raman spectrum.

Background

The identification of bacteria and fungi has very important clinical significance, and the identification of bacteria and fungi infection in the early infection stage can help clinicians to establish a targeted treatment scheme and provide reference for clinically selecting effective antibacterial drugs so as to reduce improper use of broad-spectrum antibacterial drugs and reduce the economic burden of patients.

The commonly used methods for identifying bacteria and fungi mainly include staining methods, biochemical methods, and immunological and molecular biological detection methods. The staining method for distinguishing bacteria and fungi in clinic is mainly a gram staining method, and the fungi after staining is gram positive, dark purple and about 10-100 mu m in size; gram-positive bacteria and gram-negative bacteria respectively show deep purple and red, the size is about 1-10 mu m, bacteria and fungi can be preliminarily identified by observing the shape and the size of thalli under a mirror, but the method is not suitable for identifying the fungi with smaller volume, and errors easily occur depending on manual interpretation. The biochemical method is the 'gold standard' for clinical bacterial and fungal identification, and although the method is accurate, the method is long in time consumption and difficult to meet clinical requirements. The detection of the infected microorganisms by applying immunology and molecular biology techniques provides an important means for identifying microorganisms which are long in culture time and difficult to culture or cannot be cultured, so that the diagnosis of the infection is early, rapid and accurate. However, the two methods have high detection cost and high laboratory requirements, and many hospitals cannot meet the detection conditions.

Disclosure of Invention

The present invention is directed to overcoming the above-mentioned drawbacks of the prior art and providing a method for rapidly identifying bacteria and fungi by using raman spectroscopy.

The purpose of the invention can be realized by the following technical scheme:

the invention provides a method for rapidly identifying bacteria and fungi by utilizing Raman spectrum, which comprises the steps of pretreating a sample to be identified, collecting a Raman spectrum of the treated sample, removing background and normalizing the collected Raman spectrum, and identifying bacteria and fungi by utilizing the treated spectrum.

Further, the method for rapidly identifying bacteria and fungi by using Raman spectroscopy comprises the following steps:

(1) sample pretreatment

Removing culture solution and blood from a sample to be identified by adopting a centrifugal method, then adding sterile water for washing, carrying out centrifugation to remove supernatant, then carrying out resuspension on the bacteria solution, dropwise adding the resuspended solution onto a low-Raman background glass slide, and carrying out Raman detection after room-temperature air drying;

(2) raman spectrum collection

Collecting a thallus single cell Raman spectrum by using a confocal micro-Raman spectrometer;

(3) raman spectrum pretreatment

Background removal and normalization processing are carried out on the acquired Raman spectrum, and the processed spectrum can be used for subsequent data analysis;

(4) bacterial fungus identification

One or two methods are selected and used in combination to distinguish bacteria from fungi: one is to judge the bacteria and fungi by comparing the specific peak positions of the fingerprint areas of the bacteria and the fungi; and the other method is to train the Raman spectrum data of known classification through a machine learning algorithm, establish a bacterial fungi identification database, and substitute the new data into a database for test identification.

In one embodiment of the present invention, in step (1), the sample to be identified comprises a pure cultured bacterial liquid or a clinical infection specimen, and the clinical infection specimen comprises infected urine, blood in positive report, and the like.

In one embodiment of the present invention, the number of times of washing with sterile water in step (1) is recommended to be 3 or more.

In one embodiment of the invention, in step (1), the low raman background slide comprises single crystal CaF₂Glass slide, BaF₂Glass slides or aluminized glass slides, and the like.

In one embodiment of the invention, in step (1), the low raman background slide is purchased from shanghai deuterium peak medical instruments ltd, Cat No 1001.

In one embodiment of the present invention, in step (2), silicon wafer calibration is performed before the confocal raman microscope test so that the raman spectrum peak shift of silicon is 520.73cm and bacterial cells are found in the microscope field.

In one embodiment of the present invention, in step (2), the acquisition conditions set at the time of raman spectrum acquisition are: a 532nm laser is used, the laser power is set to be 4-15 mW, the integration time is 5-20 s, and the spectrum collection range is set to include 300-1800 cm of the fingerprint area spectrum of the cell^-1。

In one embodiment of the invention, in the step (2), in order to increase the identification accuracy, the number of the collected raman spectrums of each strain is more than 200 when the library is built, and the number of the test data is more than 10.

In one embodiment of the present invention, in step (3), the background removal may be performed by selecting linear or polynomial algorithms, and setting the order and the number of points, respectively, and the normalization process may be performed by selecting division by the area of the atlas or division by the maximum value, and it should be noted that during the background removal and normalization processes, it is ensured that the same process is performed on all atlases.

In one embodiment of the invention, in step (4), regarding the specific peak positions of the fingerprint regions of bacteria and fungi, the fungi are analyzed by tests to contain specific substances with Cytochrome C only distributed in mitochondria and Raman shift of 750cm^-1、1128cm^-1、1583cm^-1The specific peak position of the bacteria is 810cm^-1And 1240cm^-1Two most typical RNA peaks and a peak at 1480cm^-1The amide II peak can be used for accurately identifying bacteria and fungi by utilizing the specific peak positions.

In one embodiment of the present invention, in step (4), the machine learning algorithm includes PCA-DFA (principal analysis-differentiation function analysis), PCA-SVM (principal analysis-Support vector machines), cnn (volumetric neural networks).

Before DFA and SVM analysis, PCA analysis is carried out to reduce the dimension of multi-dimensional data, then DFA analysis or SVM analysis is used to carry out linear recombination on complex variables after dimension reduction, complexity is simplified, and therefore the maximum change between two or more groups can be seen, and the changes are transferable and can be applied when new data are substituted. The CNN is one of the algorithms of deep learning (deep learning), has the characteristic learning capacity, adopts an mxnet deep learning framework to construct a DenseNet neural network, and performs deep network learning and training. Each layer in a conventional convolutional network uses only the output features of the previous layer as its input, and each layer in DenseNet uses the features of all previous layers as inputs and its own features as inputs to all subsequent layers. DenseNet has several advantages: the method reduces the problems caused by gradient diffusion, enhances the spread of the characteristics, encourages the reuse of the characteristics and greatly reduces the parameter quantity.

Raman spectroscopy is a vibrational spectroscopy technique defined as: when an incident photon and a molecule have inelastic collision, the laser energy changes, the energy is transferred from the molecule to the photon or vice versa, and the energy of the scattered photon is less or more than that of the incident photon, which is the inelastic raman scattering. The frequency difference between the incident and raman scattered light, called the raman shift, is unique to a single molecule, measured by a machine detector, and expressed as 1/cm. The Raman shift corresponds to the chemical bond vibration frequency, namely the fingerprint information of molecular vibration. The Raman spectrum can reflect the chemical components of the sample, the changes brought by the functional groups and chemical reactions, and the like, thereby reflecting the molecular characteristics of the cells. Therefore, the application obtains the specific information of the cells through the analysis of the molecular map data of the cells.

Bacteria and fungi belong to prokaryotes and eukaryotes respectively, and fungi contain abundant organelles such as mitochondria, endoplasmic reticulum, Golgi bodies and the like in cells compared with the bacteria, and the biomacromolecules forming the cells are also different. Based on the differences of these biological macromolecules, the raman spectrum of fungi also has specific raman spectrum peaks compared with bacteria. Bacteria and fungi can be rapidly identified through the specific spectral peaks.

In addition, the invention can combine Raman spectrum and machine learning, train Raman spectra of two bacteria through machine learning, establish bacteria and fungi Raman spectrum library, substitute test data and can also distinguish bacteria and fungi rapidly and accurately.

Compared with the prior art, the invention has the following advantages and beneficial effects:

(1) the method is simple to operate, and does not need to carry out complicated sample pretreatment before carrying out the Raman test.

(2) The method needs less bacteria for testing, does not need bacteria enrichment culture, can greatly shorten the detection time and quickly give an identification result.

(3) The characteristic that the Raman spectrum of the cell can reflect the chemical components of the cell is utilized, the bacteria and the fungi are accurately identified by comparing the specific peak positions of the Raman spectra of the bacteria and the fungi, and errors caused by the observation of the shape and the size of the cell by naked eyes and the subjective judgment can be eliminated.

(4) The Raman spectrum library of the fungi and bacteria is established through machine learning, newly acquired spectrum data are directly substituted into the library for discrimination, and the accuracy of discrimination results can be better guaranteed through the combination of two discrimination methods of machine learning discrimination and specific peak discrimination.

Drawings

Figure 1 bacterial and fungal fingerprint raman and specific peaks [ fungi (Fungus): (candida albicans (c.albicans), candida krusei (c.krusei), candida parapsilosis (c.parapsilosis),. candida glabrata (c.glabrata), candida tropicalis (c.tropicalis), Bacteria (Bacteria): escherichia coli (e.coli), klebsiella pneumoniae (k.pneumoniae), pseudomonas aeruginosa (p.aeruginosa), acinetobacter baumannii (a.baumannii), staphylococcus aureus (s.aureus), enterococcus faecalis (e.faecis), staphylococcus epidermidis (s.epidermidis) ];

FIG. 2 shows the PCA analysis chart of the Raman spectra of the fingerprint regions of bacteria and fungi;

FIG. 3 Raman spectrum DFA analysis chart of bacterial and fungal fingerprint area.

Detailed Description

The invention provides a method for rapidly identifying bacteria and fungi by using Raman spectrum, which comprises the following steps:

(1) sample pretreatment

Removing culture solution and blood from a sample to be identified by adopting a centrifugal method, then adding sterile water for washing, carrying out centrifugation to remove supernatant, then carrying out resuspension on the bacteria solution, dropwise adding the resuspended solution onto a low-Raman background glass slide, and carrying out Raman detection after room-temperature air drying;

(2) raman spectrum collection

Collecting a thallus single cell Raman spectrum by using a confocal micro-Raman spectrometer;

(3) raman spectrum pretreatment

Background removal and normalization processing are carried out on the acquired Raman spectrum, and the processed spectrum can be used for subsequent data analysis;

(4) bacterial fungus identification

One or two methods are selected and used in combination to distinguish bacteria from fungi: one is to judge the bacteria and fungi by comparing the specific peak positions of the fingerprint areas of the bacteria and the fungi; and the other method is to train the Raman spectrum data of known classification through a machine learning algorithm, establish a bacterial fungi identification database, and substitute the new data into a database for test identification.

In one embodiment, in step (1), the sample to be identified comprises a pure cultured bacterial liquid or a clinical infection specimen, and the clinical infection specimen comprises infected urine, blood from newspaper positive and the like.

In one embodiment, the number of times of washing with sterile water in step (1) is recommended to be 3 or more.

In one embodiment, in step (1), the low raman background slide comprises single crystal CaF₂Glass slide, BaF₂Glass slides or aluminized glass slides, and the like.

In one embodiment, in step (1), the low raman background slide is purchased from shanghai deuterium peak medical instruments ltd, Cat No 1001.

In one embodiment, in step (2), silicon wafer calibration is performed before confocal raman microscopy testing, so that the raman peak shift of the silicon is 520.73cm, and the bacterial cells are found in the microscope field.

In one embodiment thereofIn the step (2), the acquisition conditions are set during the acquisition of the Raman spectrum as follows: a 532nm laser is used, the laser power is set to be 4-15 mW, the integration time is 5-20 s, and the spectrum collection range is set to include 300-1800 cm of the fingerprint area spectrum of the cell^-1。

In one embodiment, in step (2), in order to increase the identification accuracy, the number of raman spectrum collections of each strain in library construction is more than 200, and the test data is more than 10.

In one embodiment, in step (3), the background removal may be performed by selecting linear or polynomial algorithms, respectively setting orders and point numbers, and the normalization process may be performed by selecting division by the area of the atlas or division by the maximum value, and it should be noted that the same process is guaranteed to be performed on all atlases during the background removal and normalization processes.

In one embodiment, in step (4), specific peak positions of fingerprint regions of bacteria and fungi are selected, and the fungi are analyzed by tests to contain specific substances with Cytochrome C distributed only in mitochondria and Raman shift of 750cm^-1、1128cm^-1、1583cm^-1The specific peak position of the bacteria is 810cm^-1And 1240cm^-1Two most typical RNA peaks and a peak at 1480cm^-1The amide II peak can be used for accurately identifying bacteria and fungi by utilizing the specific peak positions.

In one embodiment, in step (4), the machine learning algorithm includes PCA-DFA (principal analysis-differentiation function analysis), PCA-SVM (principal analysis-Support vector machines), cnn (volumetric neural networks).

Before DFA and SVM analysis, PCA analysis is carried out to reduce the dimension of multi-dimensional data, then DFA analysis or SVM analysis is used to carry out linear recombination on complex variables after dimension reduction, complexity is simplified, and therefore the maximum change between two or more groups can be seen, and the changes are transferable and can be applied when new data are substituted. The CNN is one of the algorithms of deep learning (deep learning), has the characteristic learning capacity, adopts an mxnet deep learning framework to construct a DenseNet neural network, and performs deep network learning and training. Each layer in a conventional convolutional network uses only the output features of the previous layer as its input, and each layer in DenseNet uses the features of all previous layers as inputs and its own features as inputs to all subsequent layers. DenseNet has several advantages: the method reduces the problems caused by gradient diffusion, enhances the spread of the characteristics, encourages the reuse of the characteristics and greatly reduces the parameter quantity.

The invention is described in detail below with reference to the figures and specific embodiments.