Analysis and identification system for Raman spectrum signal data of biomolecular crown

Abstract

The invention discloses an analysis and identification system for Raman spectrum signal data of a biomolecular crown, which comprises a memory, a processor and a computer program stored in the memory, and a Raman spectrum identification model is stored in the memory; inputting the surface enhanced Raman spectrum data of the biomolecular crown to be analyzed into the Raman spectrum model, and executing a computer program by the processor to obtain a prediction score of a physiological state category corresponding to the input surface enhanced Raman spectrum data; the construction method of the Raman spectrum recognition model comprises the following steps: (1) collecting a surface enhanced Raman spectrum of a biomolecular crown obtained from blood plasma of a volunteer, and labeling according to the physiological state of the volunteer to obtain a characteristic Raman spectrum fingerprint map library; and (2) constructing a weighted average machine learning model, and training the weighted average machine learning model by using the characteristic Raman spectrum fingerprint library to obtain a Raman spectrum recognition model. The system provided by the invention can efficiently and accurately analyze and process Raman spectrum signal data.

Description

Technical Field

[0001] The invention relates to the technical field of bioinformatics, and in particular to an analysis and recognition system for biological molecule Raman spectroscopy signal data. Background Art

[0002] Cancer is one of the leading causes of death worldwide. Early screening is of vital importance to improving treatment outcomes and patient survival rates. Currently, the most commonly used cancer screening methods in clinical practice include imaging examinations (such as CT and MRI), biochemical tests (such as tumor marker tests), and histological analysis (such as biopsy and pathological examinations). Although these methods have achieved certain results in diagnosis, they still have many shortcomings, such as high invasiveness, diagnostic results limited by doctor experience, poor diagnostic accuracy, and difficulty in early detection.

[0003] With the rapid development of molecular biology, spectroscopy technology and artificial intelligence, new non-invasive, rapid and accurate cancer screening methods have gradually become a research hotspot. Raman spectroscopy is a scattering spectrum that analyzes scattering spectra with different frequencies from the incident light to obtain information on molecular vibration and rotation, and is applied to the study of molecular structure. As an important molecular vibration spectroscopy technology, Raman spectroscopy has broad application prospects in molecular structure characterization, chemical composition analysis and biological sample detection. Compared with traditional spectral analysis technology, Raman spectroscopy can not only provide molecular-level component information, but also realize real-time, label-free detection, which is especially suitable for the characterization of biological tissues and cells. With the development of Raman spectroscopy technology, from the study of tissue specimens to the detection of single cells, from cancer tissue detection to the identification of body fluid specimens, Raman spectroscopy technology has been widely used in the field of medical research and has great application potential. For example, a Chinese patent document with publication number CN 113116302 A discloses an endoscopic Raman spectroscopy detection system for early cancer screening, comprising: an endoscope; a light source module for providing illumination light and Raman laser; a channel module comprising a biopsy channel for the endoscope body to pass through and an optical path channel provided with a light source optical fiber and a collection optical fiber; a spectroscopic module, a white light camera, a Raman spectrometer and a computer.

[0004] The biomolecular crown (BC) is a multi-molecular aggregate that can be formed on the surface of a specific modified material. It has a high degree of specificity and affinity and can selectively bind to biomolecules related to the physiological state of the human body through its crown structure. Combining the biomolecular crown and Raman spectroscopy technology can assist in the identification of biomarkers related to the physiological state of the human body at the molecular level, thereby achieving early screening of diseases.

[0005] However, the original Raman spectrum contains noise such as fluorescence and various interference sources. In addition, there are many interference sources in the Raman spectrum during actual measurement, which will bring inaccuracy and instability to the subsequent analysis results. When screening patients for diseases, due to the complexity and high noise of Raman spectral signals, relying solely on manual analysis and identification of Raman spectral signals not only requires a large workload, but also has the problem of low recognition accuracy. Summary of the invention

[0006] The present invention provides an analysis and recognition system for biomolecule Raman spectroscopy signal data, which can efficiently and accurately analyze and process Raman spectroscopy signal data.

[0007] The technical solution of the present invention is as follows: A system for analyzing and identifying biomolecule Raman spectroscopy signal data, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory stores a Raman spectroscopy identification model; The surface enhanced Raman spectroscopy data of the biomolecular corona to be analyzed is input into the Raman spectroscopy model, and the processor executes the computer program to obtain a prediction score of the physiological state category corresponding to the input surface enhanced Raman spectroscopy data.

[0008] The method for constructing the Raman spectrum recognition model includes: (1) Collect surface-enhanced Raman spectra of biomolecular coronas obtained from volunteers’ plasma and annotate them according to the volunteers’ physiological states to obtain a characteristic Raman spectral fingerprint library; (2) Construct a weighted average machine learning model and use the characteristic Raman spectrum fingerprint library to train the weighted average learning model to obtain the Raman spectrum recognition model.

[0009] Machine learning can extract potential feature information from a large amount of Raman spectral data and establish an accurate Raman spectral recognition model, which can analyze and process Raman spectral data efficiently and accurately.

[0010] Preferably, the weighted average machine learning model includes a multi-layer convolutional neural network model, a random forest model and a support vector machine model; after the surface enhanced Raman spectroscopy data is input into the weighted average machine learning model, the prediction category scores are obtained by the three models respectively, and the prediction category scores obtained by the three models are weighted averaged to obtain the final prediction category score.

[0011] The multi-layer convolutional neural network model includes four convolutional layer modules connected in sequence, a flattening module, four connection modules and a fully connected layer; the activation function adopts the ReLU function.

[0012] In the random forest model, the optimal value of the number of trees is 100; in the support vector machine model, the regularization parameter is set to 100.

[0013] Preferably, the weighted ratios of the multi-layer convolutional neural network model, the random forest model, and the support vector machine model are 52%, 36%, and 12%, respectively.

[0014] The weighted average learning model is trained using the characteristic Raman spectrum fingerprint library, including: using the characteristic Raman spectrum fingerprint library to train the multi-layer convolutional neural network model, the random forest model and the support vector machine model; when training the multi-layer convolutional neural network model, the initial learning rate is 10 -4 , which decreases exponentially with the increase of the number of iterations.

[0015] Preferably, the training data is enhanced by replicating the Raman spectrum and introducing 1% random noise into the replicated Raman spectrum, and the enhanced training data is used to train a layer convolutional neural network model, a random forest model and a support vector machine model respectively.

[0016] Step (1) includes: (1-1) Preparation of surface iodine-modified gold nanoparticles; (1-2) using surface iodine-modified gold nanoparticles to co-incubate with the plasma of volunteers in different physiological states, and forming a biomolecular corona on the surface of the surface iodine-modified gold nanoparticles; (1-3) Surface-enhanced Raman spectra of biomolecular coronas obtained from the plasma of volunteers in different physiological states were collected to obtain a characteristic Raman spectral fingerprint library.

[0017] Step (1-1) comprises: preparing citric acid-coated nano-gold particles by using a seed growth method, and modifying the surface of the nano-gold particles by using potassium iodide to obtain surface iodine-modified nano-gold particles.

[0018] The surface activity of the nano-gold particles synthesized by traditional sodium citrate reduction will be affected by citrate. The surface of the nano-gold particles modified by potassium iodide in the present invention can reduce the interference of the outer matrix of the nano-gold particles and optimize the interface properties of the nano-gold particles, change the interaction mode between the adsorbed biological molecules and the nano-gold particles with surface enhanced Raman effect activity, so as to retain the natural structure of biological molecules such as proteins and nucleic acids and improve the reliability of the Raman enhancement effect signal.

[0019] Preferably, the average particle size of the citric acid-coated gold nanoparticles is 65-75 nm.

[0020] Preferably, the surface of the nano-gold particles is modified with potassium iodide, comprising: adding potassium iodide solution to the nano-gold particles, mixing evenly, and incubating at 30-40° C.; repeating 1-3 times, centrifuging, and obtaining nano-gold particles with surface iodine modified.

[0021] The surface iodine-modified gold nanoparticles were co-incubated with the plasma of volunteers in different physiological states to form a biomolecular corona on the surface of the surface iodine-modified gold nanoparticles, including: The surface iodine-modified gold nanoparticles were added to a phosphate buffer solution containing plasma and incubated at 37°C for 1-3 hours to form a biomolecular corona on the surface of the surface iodine-modified gold nanoparticles.

[0022] In step (1-3), collecting the surface enhanced Raman spectrum of the biomolecular corona includes: dropping a dispersed liquid of gold nanoparticles coated with the biomolecular corona on a glass slide wrapped with aluminum foil, and after drying, randomly selecting multiple points of each sample to obtain the surface enhanced Raman spectrum using a laser confocal Raman spectrometer.

[0023] The scanning range of the laser confocal Raman spectrometer is 400-2000 cm -1 , the excitation wavelength was 785 nm, the laser intensity was 1, and the integration time was 10 s.

[0024] Preferably, the surface enhanced Raman spectroscopy data of the biomolecule corona to be analyzed is 400-500 cm -1 、560-620cm -1 、650-700 cm -1 , 1250-1300 cm -1 、1350-1480 cm -1 、1540-1650 cm -1 , 1920-2000 cm -1 , 1000-1100 cm -1 , 1700-1750 cm -1 , 1850-1900 cm -1 Data at the Raman shift.

[0025] Compared with the prior art, the present invention has the following beneficial effects: The present invention constructs, trains and optimizes an integrated weighted average machine learning model based on data obtained from a laser confocal Raman spectrometer, and evaluates the model performance using the receiver operating curve. The results show that the integrated machine learning model constructed by the present invention can better extract feature information from different Raman spectra and improve the accuracy of analysis and recognition of Raman spectral signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1Schematic diagram of the process for preparing potassium iodide-modified gold nanoparticles and biomolecule crown-coated gold nanoparticles from citric acid-coated gold nanoparticles.

[0027] Figure 2 Transmission electron microscope photographs (A), particle size distribution intensity diagram (B), hydrated particle size diagram (C), surface Zeta potential analysis diagram (D), UV-visible spectrum analysis result diagram (E), and dynamic light scattering particle size analysis diagram (F) of the gold nanoparticles, potassium iodide-modified gold nanoparticles, and biomolecule corona-coated nanoparticles of Example 1.

[0028] Figure 3 Schematic diagram of the architecture of the weighted average machine learning model.

[0029] Figure 4 The relationship between the loss (A), accuracy (B) and training rounds of the CNN model. DETAILED DESCRIPTION

[0030] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0031] 1. Preparation of biomolecule corona-coated gold nanoparticles.

[0032] like Figure 1 As shown, gold nanoparticles, potassium iodide-modified gold nanoparticles, and biomolecule crown-coated gold nanoparticles are prepared according to the following steps: (1) Prepare citric acid-coated gold nanoparticles with a diameter of 70 nm using the sodium citrate reduction method and the seed growth method. First, prepare gold nanoparticles with a diameter of 15 nm using the sodium citrate reduction method. Dissolve 100 mg of chloroauric acid trihydrate in 300 mL of water and heat the solution to boiling before adding 30 mL of sodium citrate solution. Continue stirring at boiling temperature for 10 min. After the solution is cooled to room temperature, a gold nanoparticle solution with a diameter of 15 nm is obtained. Mix 30 mL of the gold nanoparticle solution with a diameter of 15 nm with 118 mL of water, add 2 mL of sodium citrate solution (60 mM), stir vigorously and heat the mixture to 90 °C. Add 1 mL of chloroauric acid trihydrate solution (25 mM) 4 times and every 30 min to obtain a gold nanoparticle solution with a diameter of 30 nm. Repeat this step more than twice to obtain a gold nanoparticle solution coated with citric acid with a diameter of about 70 nm. (2) Modify the citric acid-coated gold nanoparticles using potassium iodide. Take 200 μL of the concentrated gold nanoparticle solution, add 100 μL of potassium iodide (1 mM), mix and sonicate for 5 min, and incubate in a 35 ℃ water bath for 30 min. After centrifugation at 9000 rpm for 10 min, discard the supernatant, add 100 μL of potassium iodide (1 mM) again and repeat the above operation. After centrifugation again and discarding the supernatant, resuspend in 50 μL of water for later use.

[0033] (3) Potassium iodide-modified gold nanoparticles were incubated with human serum to form a biomolecular corona: The prepared 70 nm potassium iodide-modified gold nanoparticles were added to 500 μL of phosphate buffer solution containing 10% plasma from healthy people, lung cancer patients, or breast cancer patients, and incubated in a 37 °C water bath for 2 h to form a plasma biomolecular corona. Then, the nanoparticles were centrifuged at 9000 rpm for 15 min, 500 μL of phosphate buffer solution was added to disperse the nanomaterials and mix them evenly, and the nanoparticles were centrifuged at 9000 rpm for 15 min and washed twice, and finally dissolved in 50 μL of phosphate buffer solution to form a gold nanoparticle complex coated with a biomolecular corona.

[0034] 2. Detect and analyze the physical and chemical properties of gold nanoparticles, potassium iodide-modified gold nanoparticles, and biomolecule crown-coated nanoparticles.

[0035] (1) Morphology and surface properties of gold nanoparticles (AuNP), potassium iodide-modified gold nanoparticles (AuNP-KI), and biomolecule corona-coated nanoparticles (BC-Coated AuNP) Citric acid-coated gold nanoparticles, potassium iodide-modified gold nanoparticles and biomolecule crown-coated nanoparticles were negatively stained with phosphotungstic acid and then analyzed by TEM, and the hydrated particle size and nanoparticle size tracking analysis were performed.

[0036] The morphologies of gold nanoparticles (AuNP), potassium iodide-modified gold nanoparticles (AuNP-KI), and biomolecule corona-coated nanoparticles (BC-Coated AuNP) were observed in turn using TEM. Figure 2 In (A), the results showed that AuNP and AuNP-KI were nearly spherical, with uniform and similar particle sizes and uniform dispersion. According to the particle size statistics of TEM photos, the average particle sizes of AuNP and AuNP-KI were 72.63 and 72.32 nm, respectively. After the formation of BC-Coated AuNP, it was observed through negative staining with phosphotungstic acid that the surface of AuNP was covered with a layer of organic matter with a thickness of about 9 nm, which proved the successful formation of the biomolecular corona after incubation with plasma. The TEM particle size distribution of AuNP, AuNP-KI and BC-Coated AuNP is shown in Figure 2. Figure 2 As shown in (B).

[0037] (2) Dynamic light scattering particle size analysis of AuNP, AuNP-KI and BC-coated AuNP The hydrated particle sizes of the three materials were determined by dynamic light scattering ( Figure 2 In (C), the hydrated particle sizes of AuNP, AuNP-KI and BC-Coated AuNP were 78±0.36, 74.8±0.56 and 127.1±1.78 nm, respectively, proving that the biomolecular corona formation increased the particle size of AuNP by 50 nm, which is consistent with the particle size intensity distribution diagram ( Figure 2 The results of (B) are consistent. The polydispersity index (PDI) of the three materials is all below 0.3, indicating that the materials have good stability and dispersibility in aqueous solution before and after the formation of the biomolecular corona (BC). The DLS measurement results show a larger particle size, which may be due to the optical scattering phenomenon of the material in the solution environment. The measurement results include the solvent layer or other molecules bound to the surface. In addition, the nanoparticle tracking analysis technology (NTA) was used to record the concentration and size of the three materials, and similar results were obtained with TEM and DLS. The size of the material before the formation of the biomolecular corona was close to 70 nm, and the particle size increased by 35 nm after formation ( Figure 2 Middle (F).

[0038] (3) Surface Zeta potential analysis of AuNP, AuNP-KI and BC-coated AuNP The Zeta potential of the three materials was determined by DLS. Figure 2 In (D), the citrate groups on the surface of AuNPs make them negatively charged in aqueous solution, with an average potential of -49 ± 4.29 mV. The potential of iodine-modified AuNPs increased slightly after iodine ion substitution. After BC was formed on the AuNP surface, the potential of biocorona-AuNPs increased to -32 ± 0.6 mV, which may be due to the adsorption of proteins with different charges in plasma.

[0039] (4) UV-visible spectroscopic analysis of AuNP, AuNP-KI and BC-coated AuNP The UV absorption spectra of the three materials between 400-800 nm were measured and normalized ( Figure 2 Middle (E). AuNP and AuNP-KI show the strongest UV-visible absorption peaks at 546 nm and 543 nm, respectively. After incubation with plasma to form BC, the UV absorption peak of the material slightly blue-shifts to 537 nm, which may be caused by the adsorption of biomolecules on the surface of the material and indicates that the material has not undergone obvious aggregation.

[0040] 3. Composition analysis of biocorona nanoparticles based on surface enhanced Raman effect (SESR).

[0041] (1) Blood sample collection from subjects with different health conditions With the approval of the Ethics Review Committee, blood samples from 92 patients with lung cancer and 65 patients with breast cancer who were clinically diagnosed in the Second Affiliated Hospital of Shandong University of Chinese Medicine were collected, and blood samples from 105 healthy volunteers were included as controls. The blood was placed in a collection tube containing an appropriate amount of anticoagulant (heparin sodium) and placed on ice. The supernatant was centrifuged at 2000 rpm for 10 min to obtain plasma, which was stored in a -80 ℃ refrigerator for long-term storage. Before use, it was thawed in a 4 ℃ refrigerator and centrifuged at 1000 rpm for 5 min to remove flocs in the plasma.

[0042] (2) Surface enhanced Raman spectroscopy information collection The potassium iodide-modified gold nanoparticles were added to 500 μL of PBS containing 10% plasma from healthy people, lung cancer patients, and breast cancer patients, and incubated in a 37 °C water bath for 2 h to form a biomolecular corona. Then centrifuged at 9000 rpm for 15 min, discarded the supernatant, and added 10 μL of PBS to disperse and mix the nanomaterials. Take 10 μL and drop it on a tin foil-wrapped slide to air dry. Ten points were randomly selected on each droplet formed, and SERS measurements were performed using a laser confocal Raman microscope. All ten points showed strong Raman characteristic peaks, and the position of each characteristic peak was consistent, but the intensity was slightly different, proving that this acquisition method has excellent SERS reproducibility.

[0043] In order to better analyze SERS data and reduce the deviation caused by intensity differences, baseline subtraction, smoothing and normalization data processing are uniformly performed after spectrum acquisition to obtain reliable SERS data that can be used for comparison and calculation.

[0044] The present invention uses plasma from 105 healthy people, 92 lung cancer (LC) and 65 breast cancer (BCa) patients to form BC-Coated AuNPs. Based on the above processing, 1050, 920 and 650 Raman spectra of disease-specific bio-crown-gold nanoparticles are collected respectively and the Raman intensity at each wavelength is collected. Finally, 1547700, 1356080 and 958100 Raman signal values ​​are collected respectively for subsequent machine learning modeling.

[0045] (3) Analysis of SESR spectra differences of biomolecular coronas formed by blood samples from subjects with different health conditions The average SERS spectra of the Raman data of the plasma biomolecule corona samples of healthy subjects, lung cancer patients, and breast cancer patients before and after normalization were compared together. The peaks at 447, 550, 570, 870, 942, 888, 1002, 1142, 1264, 1374, 1447, and 1489 cm were consistently observed in the three groups. −1 After comparing the average spectra of lung cancer and breast cancer with those of healthy controls, it was found that the Raman peaks at 447, 550, 570, 870, 942, 1002, 1142, 1264, 1447, and 1489 cm −1 The peak intensity at has changed significantly.

[0046] 4. Experimental methods were used to analyze and quantify the differential characteristic proteins of the biomolecular crowns of healthy people, lung cancer patients, and breast cancer patients.

[0047] AuNP-KI was incubated with plasma from three different healthy individuals, lung cancer patients, and breast cancer patients to form BCs. The proteins in the BCs were separated after ultrasound, centrifugation, and high-temperature denaturation, and the adsorbed protein coronas were analyzed by SDS-PAGE.

[0048] After silver staining, the gels showed obvious bands, indicating that a considerable amount of protein in the plasma was adsorbed. ImageJ was used to perform quantitative grayscale analysis of the images, which more intuitively demonstrated the difference in the average concentration of protein bands between the two groups. The results showed that compared with healthy people, the plasma PC of lung cancer had obvious differential proteins between 15-25 kD and 35-50 kD, while the protein bands between 25-50 kD of breast cancer plasma PC were significantly different from those of healthy people, indicating that there were certain differences in the composition of plasma PC in different disease states.

[0049] 5. Establish a machine learning model to classify Raman spectral information.

[0050] To further evaluate SERS spectra and their ability to classify healthy and cancer groups, 2620 SERS spectral data sets (1050 healthy individuals, 920 lung cancer, and 650 breast cancer spectra) were used as training and test sets (the training data SERS spectra were randomly divided into 70% and 30% validation data sets), and a weighted average machine learning model was trained to classify Raman spectral information and effectively distinguish Raman spectral information of healthy individuals and cancer.

[0051] Weighted average machine learning models include convolutional neural network models (CNN), random forest models (RF) and support vector machine models (SVM) ( Figure 3). For the CNN model, features in unidirectional spectral data are extracted through the convolution layer, and then connected with the connection layer for classification. The numbers below the basic block represent the length and number of channels at output, respectively. The CNN model is organized in the order of 4 convolution layer modules, a flattening layer (Flatten), 4 connection modules, and 1 fully connected layer. Each convolution layer module includes a convolution layer, a linear rectification layer, a normalization layer, a convolution layer, a linear rectification layer, and a maximum pooling layer connected in sequence. Each connection module includes a dense layer, a linear rectification layer, and a dropout layer. A dropout layer is used in the connection module to prevent overfitting. The linear rectification function is used as the activation function to enhance nonlinearity without affecting the accuracy.

[0052] The learning rate of the CNN model is increased from 10 -4 At the beginning, it decreases exponentially with the increase of the number of iterations (lr=0.0001×e -0.02×epoch The validation set loss for each learning iteration ( Figure 4 (A)) and accuracy ( Figure 4 There is no significant difference between the training set and the validation set in (B), and the accuracy of the training set and validation set are close to 1.0 and 0.9, respectively, indicating that the CNN model is not overfitting.

[0053] For the random forest (RF) model, the number of trees was optimized to 100. In the support vector machine (SVM), the regularization parameter was set to 100. For the weighted average machine learning model, an optimizer was used to evaluate the weight factor of each sub-model by fitting the model output to the training dataset. The optimized weight factors of the CNN model, RF model, and SVM model obtained through training were 0.52, 0.36, and 0.12, respectively.

[0054] 6. A semi-empirical method is used to determine the model response when each Raman value changes.

[0055] Due to the complexity and nonlinearity of the machine learning model, the weighted machine learning average model can accurately distinguish samples from healthy people and cancer patients. However, the huge number of model parameters in the model (48,361,795 parameters in total) makes it impossible to discern and distinguish the contribution of each model parameter. This embodiment develops a semi-empirical method to diagnose the model response at each Raman shift. The Raman spectrum matrix is ​​processed to contain unique signals under specific Raman shifts, and then input into the weighted average machine learning model for classification. The model predicts the category score for each class and selects the one with the highest score as the category output. All predicted category scores were obtained to see the relative relationship between the Raman shift and the model output. The results show that the lung cancer group has a strong response to 1000-1100 cm -1 , 1700-1750 cm -1 and 1850-1900 cm-1 The breast cancer group has a more relevant Raman shift of 400-500 cm -1 、560-620 cm -1 、650-700 cm -1 , 1250-1300 cm -1 、1350-1480 cm -1 、1540-1650 cm -1 , 1920-2000 cm -1 The above results strongly confirm that the constructed model has excellent classification performance in distinguishing three different disease states.

[0056] The surface activity of the nano-gold particles synthesized by traditional sodium citrate reduction will be affected by citrate. The surface of the nano-gold particles modified by potassium iodide in the present invention can reduce the interference of the nano-gold outer layer matrix and optimize the interface properties of the nano-particles, change the interaction mode between the adsorbed biological molecules and the nano-gold particles with surface enhanced Raman effect activity, so as to retain the natural structure of biological molecules such as proteins and nucleic acids and improve the reliability of the Raman enhancement effect signal.

[0057] In view of the problems of complex composite substrate preparation in the traditional method of using surface-enhanced Raman spectroscopy to detect human serum in different disease states for cancer diagnosis, the present invention selected the simplest aluminum foil detection substrate and explored the effects of two pre-treatment methods, namely, directly mixing nanogold and serum and co-incubating nanogold and serum, on the specificity and sensitivity of Raman enhanced spectroscopy signals. At the same time, to ensure the reliability of the results, the present invention uses SDS-Page, proteomics and other means to analyze the differences in the biomolecular coronas formed by incubating human serum in different disease states with gold nanoparticles, explore the peak position of the surface-enhanced Raman spectrum of the plasma protein corona and the possible molecular attribution, and provide a biological explanation for the differences observed in the spectrum.

[0058] The matters not described in detail in the present invention are all known technologies to those skilled in the art.

[0059] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A system for analyzing and identifying biomolecule Raman spectroscopy signal data, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The memory stores a Raman spectrum recognition model; Inputting the surface enhanced Raman spectroscopy data of the biomolecular corona to be analyzed into the Raman spectroscopy model, the processor executing the computer program to obtain a prediction score of the physiological state category corresponding to the input surface enhanced Raman spectroscopy data; The method for constructing the Raman spectrum recognition model includes: (1) Collect surface-enhanced Raman spectra of biomolecular coronas obtained from volunteers’ plasma and annotate them according to the volunteers’ physiological states to obtain a characteristic Raman spectral fingerprint library; (2) Construct a weighted average machine learning model and use the characteristic Raman spectrum fingerprint library to train the weighted average learning model to obtain the Raman spectrum recognition model.

2. The system for analyzing and identifying biomolecule Raman spectroscopy signal data according to claim 1, characterized in that: The weighted average machine learning model includes a multi-layer convolutional neural network model, a random forest model and a support vector machine model; after the surface enhanced Raman spectroscopy data is input into the weighted average machine learning model, the prediction category scores are obtained by the three models respectively, and the prediction category scores obtained by the three models are weighted averaged to obtain the final prediction category score.

3. The system for analyzing and identifying biomolecule Raman spectroscopy signal data according to claim 2, characterized in that: The multi-layer convolutional neural network model includes four convolutional layer modules connected in sequence, a flattening module, four connection modules and a fully connected layer; the activation function adopts the ReLU function.

4. The system for analyzing and identifying biomolecule Raman spectroscopy signal data according to claim 2, characterized in that: In the random forest model, the optimal value of the number of trees is 100; in the support vector machine model, the regularization parameter is set to 100.

5. The system for analyzing and identifying biomolecule Raman spectroscopy signal data according to claim 2, characterized in that: The weighted proportions of the multi-layer convolutional neural network model, random forest model, and support vector machine model are 52%, 36%, and 12%, respectively.

6. The system for analyzing and identifying biomolecule Raman spectroscopy signal data according to claim 2, characterized in that: The weighted average learning model is trained using the characteristic Raman spectrum fingerprint library, including: using the characteristic Raman spectrum fingerprint library to train the multi-layer convolutional neural network model, the random forest model and the support vector machine model; when training the multi-layer convolutional neural network model, the initial learning rate is 10 -4 , which decreases exponentially with the increase of the number of iterations.

7. The system for analyzing and identifying biomolecule Raman spectroscopy signal data according to claim 6, characterized in that: The training data was enhanced by replicating the Raman spectrum and introducing 1% random noise on the replicated Raman spectrum. The enhanced training data was used to train the layer-convolutional neural network model, random forest model and support vector machine model respectively.

8. The system for analyzing and identifying biomolecule Raman spectroscopy signal data according to claim 1, characterized in that: Step (1) includes: (1-1) Preparation of surface iodine-modified gold nanoparticles; (1-2) using surface iodine-modified gold nanoparticles to co-incubate with the plasma of volunteers in different physiological states, and forming a biomolecular corona on the surface of the surface iodine-modified gold nanoparticles; (1-3) Surface-enhanced Raman spectra of biomolecular coronas obtained from the plasma of volunteers in different physiological states were collected to obtain a characteristic Raman spectral fingerprint library.

9. The system for analyzing and identifying biomolecule Raman spectroscopy signal data according to claim 8, characterized in that: Step (1-1) comprises: preparing citric acid-coated nano-gold particles by using a seed growth method, and modifying the surface of the nano-gold particles by using potassium iodide to obtain surface iodine-modified nano-gold particles.

10. The biological molecule Raman spectroscopy signal data analysis and identification system according to claim 1, characterized in that: The surface enhanced Raman spectroscopy data of the biomolecule corona to be analyzed is 400-500 cm -1 、560-620 cm -1 、650-700cm -1 , 1250-1300 cm -1 、1350-1480 cm -1 、1540-1650 cm -1 , 1920-2000 cm -1 , 1000-1100 cm -1 , 1700-1750 cm -1 , 1850-1900 cm -1 Data at the Raman shift.

Images