Method for analyzing complex system by using surface enhanced Raman spectrum three-dimensional code and application of method

Abstract

The invention discloses a method for analyzing a complex system by using a surface enhanced Raman spectrum three-dimensional code and application of the method. The method specifically comprises the following steps: obtaining SERS sensing channels with different affinity characteristics by using an unmarked or functionally-marked SERS substrate; collecting an SERS spectrum data set by utilizing the constructed SERS sensing channel, and integrating the SERS spectrum data set into a'three-dimensional code 'image through data conversion and integration of three dimensions of Raman vibration band distribution, the SERS sensing channel and SERS intensity, so as to realize imaging of the molecular landscape; the feature information in the SERS three-dimensional code is screened and extracted through a convolutional neural network algorithm, and a mapping relation between the SERS three-dimensional code and a complex system label is established, so that a complex system classification / regression model is established, rapid qualitative or quantitative analysis of a complex system is realized, and the accuracy of the SERS three-dimensional code is improved. More applications can be expanded in the fields of molecular recognition, biological sample classification and identification, disease diagnosis, cell monitoring and the like.

Description

Technical Field

[0001] The present invention relates to the field of molecular detection, and in particular to a method for analyzing a complex system using a surface enhanced Raman spectroscopy three-dimensional code, and also to an application of the method. Background Art

[0002] The analysis of complex systems has always been a research hotspot in the fields of biomedicine, environment and food. Existing complex system analysis technologies, such as fluorescence labeling, mass spectrometry, nuclear magnetic resonance (NMR), etc., have been widely used in some fields, but each has significant limitations. For example, although fluorescence labeling technology has high sensitivity, it has problems such as photobleaching and background signal interference. Especially in complex biological samples, the stability and selectivity of fluorescent probes are poor, and they are easy to induce nonspecific signals. Although mass spectrometry can provide accurate molecular mass information and has advantages in multi-component analysis, its sample pretreatment is complex, the analysis speed is slow, and it is difficult to apply to on-site detection. Although nuclear magnetic resonance is non-destructive and can provide rich molecular structure information, its sensitivity is low and it is difficult to analyze low-concentration components. In contrast, SERS technology shows great potential in the analysis of complex systems. It has high sensitivity and high information throughput, can achieve single-molecule level detection at extremely low concentrations, and can also achieve enhanced output of multi-molecule signals in multi-component complex systems. Summary of the invention

[0003] In view of this, one of the objects of the present invention is to provide a method for analyzing complex systems using surface enhanced Raman spectroscopy three-dimensional codes; a second object of the present invention is to provide application of the analysis method in microbial identification and appraisal.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] 1. Surface enhanced Raman spectroscopy 3D code analysis method for complex systems, the specific method is as follows:

[0006] (1) Using unlabeled or labeled SERS substrates with 11-mercaptoundecanoic acid, 11-mercapto-1-undecanol, 1-dodecanethiol, mercaptoethylamine, vancomycin, 4-mercaptophenylboronic acid, α-lactose, and antimicrobial peptide Cecropin A to obtain SERS sensing channels with different affinity properties;

[0007] (2) The constructed SERS sensing channel is used to collect SERS spectral data sets. Through data conversion, the three dimensions of Raman vibration band distribution, SERS sensing channel and SERS intensity are integrated to integrate the SERS spectral data sets carrying the component information of the molecular landscape of the complex system into a "3D code" to realize the visualization of the molecular landscape.

[0008] (3) The convolutional neural network algorithm is used to screen and extract the characteristic information in the SERS 3D code, and a mapping relationship between the SERS 3D code and the complex system label is established, thereby building a complex system classification / regression model and realizing rapid qualitative / quantitative analysis of the complex system.

[0009] Preferably, the data conversion method is as follows:

[0010] First, the spectrum is smoothed using the Savitsky-Golay method, and the raw SERS spectra are preprocessed using quadratic fitting to subtract the baseline and normalize. Secondly, a cluster analysis is performed on the data sets collected in large quantities from a single sensing channel, the homogenized SERS spectra are integrated, and the spectral data with information-bearing significance in the sensing channel are screened. Then, the second-order derivative of the processed characteristic spectral data set is taken to reduce the influence of background variability and intensity fluctuations, and the spectral data in the form of a series is converted into a barcode form. Finally, multiple SERS barcodes from multiple sensing channels are integrated to construct an imaged SERS 3D code for a complex system.

[0011] Preferably, the present invention uses the SERS 3D code as input, obtains the multi-scale features of the SERS 3D code under different receptive fields through 3 parallel convolution layers, designs a pooling layer to unify the dimensions of the fusion feature vector, and then inputs it into the fusion layer for feature connection to obtain fusion features, and then sets weights through the attention mechanism layer to strengthen more discriminative features, and finally uses it as input for recognition, and the fully connected layer and the Softmax classifier output the recognition accuracy of N classifications, which is expressed as:

[0012]

[0013] Where: z j is the linear probability of the jth category; z N is the sum of the linear probabilities of N categories; A j (z) is the category normalization result.

[0014] Preferably, the method for obtaining the multi-scale features of the SERS three-dimensional code under different receptive fields by the three parallel convolution layers is as follows: using the CNN feature extraction layer to use the three modules of the convolution layer, the maximum pooling layer and the activation function layer in parallel, setting three sizes of convolution kernels to broaden the width of the network, generating multiple receptive fields to extract the features of the SERS three-dimensional code at different scales, using the maximum pooling layer to downsample the output of the convolution layer, achieving dimensionality reduction on the basis of retaining the original main features, and making the network structure less prone to overfitting, and its mathematical expression is:

[0015]

[0016] Where: a h(m，t)is the activation value of the tth neuron in the mth feature map in the hth layer; g is the width of the pooling region; n is the nth pooling kernel; p h(m，n) is the output of the corresponding pooling layer; the features output by the pooling layer are input into the feature union layer for feature fusion to obtain the fused features.

[0017] Preferably, the processing process of the attention mechanism layer is as follows: the fused features output by the CNN feature extraction layer are input into the Softmax function, the weight coefficients of different feature dimensions are calculated, the weight coefficients obtained are multiplied by the corresponding elements of the input features to obtain new weighted features, the context vector extracted by the convolution feature extraction layer is Ht, and the weight vector corresponding to each spatial step of the global attention is a t (s), which is calculated as:

[0018]

[0019] Where: is the hidden state at the tth moment, and the function score calculates the matching score between each feature vector and the entire feature vector as:

[0020]

[0021] Where W T It is a trainable weight matrix. The more important the feature is, the closer the output value of the corresponding attention mechanism module is to 1. The importance of the feature is reflected by the value, and more critical and discriminative feature identification is completed. rReLU is a function, W is a weight matrix, H is a feature vector, U is a union operator, and b represents an intercept term.

[0022] 2. Application of the analytical method in identification and characterization of microorganisms.

[0023] The beneficial effects of the present invention are as follows: the present invention provides a method for analyzing complex systems using surface enhanced Raman spectroscopy three-dimensional codes, obtains SERS sensing channels with different affinity characteristics through two SERS substrate strategies, namely, labeled and unlabeled, and then performs overall signal capture and targeted local molecular portraits on the molecular landscape of the complex system, and obtains a SERS spectral data set of the molecular landscape of the complex system in combination with an automatic spectrum acquisition system. Then, through data conversion, the three dimensions of Raman vibration band distribution, SERS sensing channels and SERS intensity are integrated to integrate the SERS spectral data set carrying the component information of the molecular landscape of the complex system into a "three-dimensional code" to realize the imaging of the molecular landscape. Further, the feature information in the SERS three-dimensional code is screened and extracted through a convolutional neural network algorithm, and a mapping relationship between the SERS three-dimensional code and the complex system label is established, thereby building a complex system classification / regression model to realize rapid qualitative / quantitative analysis of the complex system. The method does not require complex sample pretreatment, can be combined with portable Raman spectroscopy to realize on-site detection and analysis, can capture and characterize the overall component characteristics of the complex system and realize its identification imaging; and can be combined with a convolutional neural network to realize rapid analysis of the complex system. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to make the purpose, technical solution and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration:

[0025] Figure 1 Schematic diagram for constructing functionalized multi-SERS sensing channels and SERS 3D codes;

[0026] Figure 2 Schematic diagram of the process of constructing SERS 3D code;

[0027] Figure 3 Fingerprint information of different bacteria is collected and applied (a: normalized multi-dimensional SERS spectra of 9 bacteria, the nanomaterials represented from left to right are unmodified Ag NPs and Ag NPs modified with 11COOH, 11OH, 12CH3, 2NH2, Van, and MPBA; b: schematic diagram of SERS three-dimensional code technology driven by convolutional neural network for microbial identification; c: confusion matrix of microbial identification based on the optimal dimensional combination results). DETAILED DESCRIPTION

[0028] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0029] Example 1: Rapid identification of pathogenic microorganisms

[0030] Microbial samples contain compounds of different types and concentrations, which are rich in biochemical information. The effective extraction of this information and its use for rapid identification of strains is a research hotspot in the environment and other related fields. The SERS method can provide characteristic molecular fingerprints and has the advantages of high sensitivity, rapidity, and universality.

[0031] (1) Functionalized multi-SERS sensing channels and SERS 3D code construction

[0032] Functionalized SERS sensing channels, 11-mercaptoundecanoic acid (11COOH), 11-mercapto-1-undecyl alcohol (11OH), 1-dodecyl mercaptan (12CH3), mercaptoethylamine (2NH2), vancomycin (Van), 4-mercaptophenylboronic acid (MPBA), α-lactose (α-Lactose), antimicrobial peptide Cecropin A (AMPs) were functionalized into SERS sensing channels ( Figure 1 ). The affinity of the sensing interface is modulated by nanomaterials with different modifications to achieve the selective adsorption of complex system components on it, obtain various bacteria, provide high-dimensional fingerprint information for bacterial samples and apply it to rapid identification. The experiment uses nanomaterials with different modifications to modulate the affinity of the sensing interface to achieve the selective adsorption of complex system components on it, and obtain diverse and refined spectral curves of various bacteria.

[0033] (2) Data collection and conversion

[0034] The functionalized SERS sensing channels were used to collect fingerprint information of 9 bacteria (Staphylococcus aureus, Staphylococcus epidermidis, Klebsiella pneumoniae, Pseudomonas aeruginosa, Acinetobacter baumannii, Enterococcus faecalis, Enterococcus faecium, Escherichia coli, Clostridium tetani). Figure 3 , a), after data enhancement, a total of 3150 spectra were obtained from 9 strains to construct spectral fingerprint datasets of different bacterial types, and the Raman vibration band distribution was integrated through data transformation. The specific methods are as follows: first, the Savitsky-Golay method was used to smooth the spectrum, and the original SERS spectrum was preprocessed by using quadratic fitting to subtract the baseline and normalization; secondly, cluster analysis was performed on the datasets collected in large quantities by a single sensing channel, and the homogenized SERS spectra were integrated and the spectral data with information-bearing significance in the sensing channel were screened; then, the processed characteristic spectral dataset was second-order differentiated to reduce the influence of background variability and intensity fluctuation, and the spectral data in the form of a series was converted into a barcode form; finally, multiple SERS barcodes of multiple sensing channels were integrated to construct an imaged SERS 3D code of a complex system; the SERS 3D code dataset of each bacterium was obtained in three dimensions of each SERS sensing channel and SERS intensity ( Figure 2 ) to visualize the molecular landscape.

[0035] (3) Construct a convolutional neural network (CNN) model consisting of four convolutional layers

[0036] The convolutional neural network (CNN) model is used to extract the SERS 3D code features of bacteria and realize strain identification ( Figure 3 , b), the specific method is as follows:

[0037] The overall structure of the CNN model consists of an input layer, a CNN feature extraction layer, an attention mechanism layer, a fully connected layer, and an output layer, where:

[0038] CNN feature extraction layer: This module consists of 3 convolution layers, a maximum pooling layer, and an activation function layer. In a parallel manner, 3 sizes of convolution kernels are set to widen the width of the network, generating multiple receptive fields to extract the features of SERS 3D codes at different scales. The maximum pooling layer is used to downsample the output of the convolution layer, achieving dimensionality reduction while retaining the original main features, while making the network structure less prone to overfitting. Its mathematical expression is:

[0039]

[0040] Where: a h(m，t) is the activation value of the tth neuron in the mth feature map in the hth layer; g is the width of the pooling region; n is the nth pooling kernel; p h(m，n) is the output of the corresponding pooling layer; the features output by the pooling layer are input into the feature union layer for feature fusion to obtain the fused features.

[0041] Attention mechanism layer: Input the feature matrix output by the CNN feature extraction layer into the Softmax function, calculate the weight coefficients of different feature dimensions, and multiply the corresponding elements of the weight coefficients with the input features to obtain new weighted features. The context vector extracted by LSTM is H t , the weight vector corresponding to each time step of global attention is a t (s), which is calculated as:

[0042]

[0043] Where: is the hidden state at time t. The function score calculates the matching score between each feature vector and the entire feature vector as:

[0044]

[0045] Where W TIt is a trainable weight matrix. The more important the feature is, the closer the output value of the corresponding attention mechanism module is to 1. The importance of the feature is reflected by the value, and more critical and discriminative feature identification is completed. rReLU is a function, W is a weight matrix, H is a feature vector, U is a union operator, and b represents an intercept term.

[0046] The SERS 3D code is used as input, and the multi-scale features of the SERS 3D code under different receptive fields are obtained through three parallel convolution layers (each convolution layer is followed by an activation function layer and a pooling layer). The pooling layer is designed to unify the dimensions of the fusion feature vectors, and then input into the fusion layer for feature connection to obtain fusion features. Then, the weights are set through the attention mechanism layer to strengthen the more discriminative features, and finally used as input for recognition. The fully connected layer and the Softmax classifier output the recognition accuracy of N categories, which is expressed as:

[0047]

[0048] Where: z j is the linear probability of the jth category; z N is the sum of the linear probabilities of N categories; A j (z) is the category normalization result.

[0049] 80% of the data set was divided into a training set and 20% of the data set was used as a test set. After the training set SERS 3D code was input into the CNN model for feature learning training, its classification accuracy reached 99.21%. Then the test set data was input into the model to verify the generalization ability of the model. The confusion matrix of its classification results is shown in ( Figure 3 , c) shows that the accuracy of this model can reach 97.35%, and its performance is better than that of traditional algorithms such as RF, SVM and PLS based on series spectral data. The results show that the construction of SERS 3D codes can improve the recognition accuracy of highly heterogeneous and complex biological systems, but the combination of dimensions needs to be optimized to avoid problems such as data redundancy or overfitting during data processing. Given the versatility and portability of SERS 3D codes, this method will expand more applications in the fields of molecular recognition, biological sample classification / identification, disease diagnosis and cell monitoring.

[0050] The above-described embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or changes made by those skilled in the art based on the present invention are within the protection scope of the present invention. The protection scope of the present invention shall be subject to the claims.

Claims

1. The method for analyzing complex systems using surface enhanced Raman spectroscopy three-dimensional code is characterized by: The specific method is as follows: (1) Using unlabeled or labeled SERS substrates with 11-mercaptoundecanoic acid, 11-mercapto-1-undecanol, 1-dodecanethiol, mercaptoethylamine, vancomycin, 4-mercaptophenylboronic acid, α-lactose, and antimicrobial peptide Cecropin A to obtain SERS sensing channels with different affinity properties; (2) The constructed SERS sensing channel is used to collect SERS spectral data sets. Through data conversion, the three dimensions of Raman vibration band distribution, SERS sensing channel and SERS intensity are integrated to integrate the SERS spectral data sets carrying the component information of the molecular landscape of the complex system into a "three-dimensional code" to realize the visualization of the molecular landscape. (3) The convolutional neural network algorithm is used to screen and extract the characteristic information in the SERS 3D code, and a mapping relationship between the SERS 3D code and the complex system label is established, thereby building a complex system classification / regression model and realizing rapid qualitative / quantitative analysis of the complex system.

2. According to claim 1, the method for analyzing complex systems using surface enhanced Raman spectroscopy three-dimensional codes is characterized by: The method of data conversion is as follows: First, the spectrum is smoothed using the Savitsky-Golay method, and the raw SERS spectra are preprocessed using quadratic fitting to subtract the baseline and normalize. Secondly, a cluster analysis is performed on the data sets collected in large quantities from a single sensing channel, the homogenized SERS spectra are integrated, and the spectral data with information-bearing significance in the sensing channel are screened. Then, the second-order derivative of the processed characteristic spectral data set is taken to reduce the influence of background variability and intensity fluctuations, and the spectral data in the form of a series is converted into a barcode form. Finally, multiple SERS barcodes from multiple sensing channels are integrated to construct an imaged SERS 3D code for a complex system.

3. According to claim 1, the method for analyzing complex systems using surface enhanced Raman spectroscopy three-dimensional codes is characterized by: The SERS 3D code is used as input, and the multi-scale features of the SERS 3D code under different receptive fields are obtained through three parallel convolutional layers. The pooling layer is designed to unify the dimensions of the fusion feature vectors, and then input into the fusion layer for feature connection to obtain fusion features. Then, the weights are set through the attention mechanism layer to strengthen the more discriminative features, and finally used as input for recognition. The fully connected layer and the Softmax classifier output the recognition accuracy of N categories, which is expressed as: Where: z j is the linear probability of the jth category; z N is the sum of the linear probabilities of N categories; A j (z) is the category normalization result.

4. According to claim 1, the method for analyzing complex systems using surface enhanced Raman spectroscopy three-dimensional codes is characterized by: The method for obtaining the multi-scale features of the SERS 3D code under different receptive domains by the three parallel convolution layers is as follows: the CNN feature extraction layer is used to parallelly use the convolution layer, the maximum pooling layer and the activation function layer, and the convolution kernels of three sizes are set to broaden the width of the network, so as to generate multiple receptive domains to extract the features of the SERS 3D code at different scales, and the maximum pooling layer is used to downsample the output of the convolution layer, so as to achieve dimensionality reduction on the basis of retaining the original main features, and at the same time make the network structure less prone to overfitting. The mathematical expression is: Where: a h(m，t) is the activation value of the tth neuron in the mth feature map in the hth layer; g is the width of the pooling region; n is the nth pooling kernel; p h(m，n) is the output of the corresponding pooling layer; the features output by the pooling layer are input into the feature union layer for feature fusion to obtain the fused features.

5. According to claim 1, the method for analyzing complex systems using surface enhanced Raman spectroscopy three-dimensional codes is characterized by: The processing process of the attention mechanism layer is as follows: the fused features output by the CNN feature extraction layer are input into the Softmax function, the weight coefficients of different feature dimensions are calculated, and the corresponding elements of the obtained weight coefficients are multiplied by the input features to obtain new weighted features. The context vector extracted by the convolutional feature extraction layer is Ht, and the weight vector corresponding to each spatial step of the global attention is a t (s), which is calculated as: Where: is the hidden state at the tth moment, and the function score calculates the matching score between each feature vector and the entire feature vector as: Where W T It is a trainable weight matrix. The more important the feature is, the closer the output value of the corresponding attention mechanism module is to 1. The importance of the feature is reflected by the value, and more critical and discriminative feature identification is completed. rReLU is a function, W is a weight matrix, H is a feature vector, U is a union operator, and b represents an intercept term.

6. Use of the analytical method according to any one of claims 1 to 3 in the identification and identification of microorganisms.

Images