A semi-quantitative immunoreaction detection method based on serum SERS using lithium niobate thin film nanostructures and machine learning.

Abstract

This invention belongs to the field of immunoreaction detection technology, specifically relating to a semi-quantitative immunoreaction detection method based on serum SERS using a lithium niobate thin film nanostructure as a substrate and machine learning. The method involves designing mouse animal experiments to obtain mouse serum with four levels of immune response: negative, borderline positive, weakly positive, and strongly positive, establishing a mouse semi-quantitative immunoreaction system. A two-dimensional periodic lithium niobate nanostructure is prepared on the surface of a femtosecond laser-induced lithium niobate material as the SERS substrate. Trace amounts of serum samples at each level are added to the SERS substrate, and SERS spectra are collected to obtain four sets of mouse serum SERS spectra with varying intensities of immune response. Machine learning-assisted spectral analysis is used to establish a SERS spectrum-immune response intensity system. The SERS spectra of another set of test serum samples with known response grades are input into this system for verification. The results show that this system can rapidly, intelligently, and accurately output the immune response grade, achieving rapid semi-quantitative detection of immune responses from trace serum samples.

Description

Technical Field

[0001] This invention belongs to the field of immunoreaction detection technology, specifically relating to a semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate. Background Technology

[0002] Immunological response testing is a crucial aspect of assessing the safety of biomedical materials / devices and biological agents. Currently, immunological response testing primarily involves obtaining serum from experimental and negative control groups through animal experiments. Enzyme-linked immunosorbent assay (ELISA) is then used to detect the levels of various soluble immunomarkers in the serum, such as immunoglobulins (IgE, IgG, IgM), complement (C3), and cytokines (TNF-α, IL-6). Data from the experimental group and the negative control group are then statistically analyzed. The presence or absence of a statistically significant difference in each indicator is used to determine whether an immune response has occurred [GB / T 16886.20-2015 Biological Evaluation of Medical Devices Part 20: Principles and Methods of Immunotoxicological Testing of Medical Devices]. However, traditional ELISA requires the individual detection of known marker molecules, which is cumbersome and time-consuming. After obtaining the levels, complex statistical analysis is required to draw conclusions, resulting in low overall efficiency, high sample consumption, and the ability to provide only qualitative (negative or positive) results.

[0003] For example, Chinese patent document CN106526201A discloses a method for qualitative and semi-quantitative detection of antigens based on the distance of an immunoreaction using a paper chip. This method is based on the fact that after a competitive immunoreaction is complete, the remaining horseradish peroxidase-labeled antibody reacts with TMB on Whatman filter paper, resulting in different migration distances due to varying capillary forces. This establishes a method for qualitative and semi-quantitative detection of antigens based on the distance of different concentrations of analytes. The properties of the substance can be determined simply by measuring the distance generated by different concentrations of the analyte with a ruler. While this method simplifies the detection process, it still has the following limitations: it requires the individual detection of known marker molecules, making the operation cumbersome; and the results obtained are semi-quantitative results for a single protein, with lower precision than the quantitative results of traditional ELISA methods, making statistical analysis with control groups inconvenient, and unable to comprehensively determine the intensity of the immunoreaction by considering the content of various known and unknown immunoreaction markers.

[0004] Surface-enhanced Raman spectroscopy (SERS) utilizes the localized surface plasmon resonance effect excited by noble metal nanostructures and charge transfer in the substrate-molecule system to elevate the Raman signal intensity to the single-molecule detection level [Nat. Rev. Methods Primers 1, 87 (2021)]. It boasts advantages such as high sensitivity and high efficiency, and has great application potential in fields such as environmental trace pollutant detection and biomedical diagnostics. Traditional SERS detection substrates generally employ colloidal sols or chemically synthesized nanoparticles, which are limited by the high randomness, low stability, and non-reproducibility of the substrates, failing to meet the requirements of immunoassay.

[0005] Serum, as a complex bodily fluid sample, contains more than 2,000 biomolecules (such as proteins, lipids, metabolites, etc.). Its SERS spectrum contains complex fingerprint information formed by the superposition of all known and unknown immune molecules. Traditional manual analysis methods cannot effectively distinguish the characteristic peaks of highly overlapping biomarkers and cannot reliably identify the weak signals of early immune responses.

[0006] Therefore, how to utilize the high sensitivity and information enrichment advantages of SERS technology, while overcoming the shortcomings of traditional manual analysis methods in effectively processing complex SERS spectral fingerprint information, and establish an intelligent correlation model between serum SERS spectra and immune response levels to achieve rapid, label-free, and semi-quantitative grading detection of immune responses in trace serum samples, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a semi-quantitative immunoreaction detection method based on serum SERS using a lithium niobate thin-film nanostructure as a substrate and machine learning. This invention designs a mouse experiment, using ELISA to obtain mouse serum with four levels of immune response: negative, borderline positive, weakly positive, and strongly positive, establishing a mouse semi-quantitative immunoreaction system. A two-dimensional periodic lithium niobate nanostructure is prepared on the surface of a femtosecond laser-induced lithium niobate material as the SERS substrate. Trace amounts of serum samples at each level are added to the SERS substrate, and SERS spectra are collected, obtaining four sets of mouse serum SERS spectra with varying intensities of immune response. Machine learning-assisted spectral analysis is used to explore the linear and nonlinear correlations between response intensity and spectra, establishing a SERS spectrum-immune response intensity system. The SERS spectra of another set of test serum samples with known response grades are input into this system for verification. The results show that this system can rapidly, intelligently, and accurately output the immune response grade (consistent with ELISA results), achieving rapid semi-quantitative detection of immune responses from trace serum samples. This invention provides a more advanced solution for immune response assessment. Compared with traditional methods, this invention has advantages such as small sample consumption, high sensitivity, simultaneous identification of all indicators, high detection efficiency, and semi-quantitative analysis.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate includes the following steps: S1. Through mouse animal experiments, obtain mouse serum with four levels of immune response, including negative, borderline positive, weak positive and strong positive, establish a semi-quantitative immune response system with gradient characteristics, and verify the established semi-quantitative immune response system with gradient characteristics. S2. Using single-crystal thin films as the matrix material for SERS substrates, SERS substrates with two-dimensional periodic nanostructures are formed by processing with femtosecond laser technology. S3. The four levels of immune response mouse serum obtained in step S1 are dropped onto the SERS substrate with a two-dimensional periodic nanostructure prepared in step S2, and their SERS spectra are collected to construct a spectral dataset. The dataset is used to train and test the machine learning model to achieve semi-quantitative detection of the immune response.

[0009] Preferably, step S1 specifically includes: S11. Experimental mice whose body weight difference does not exceed ±10% of the average value are randomly divided into 4 groups: negative control group, borderline positive group, weak positive group, and strong positive group. Negative control group: Mice were not treated in any way and were fed normally; Borderline positive group: Bovine serum albumin solution with a concentration of 0.01C1 was emulsified with an equal volume of Freund's complete adjuvant to obtain a suspension. The suspension was injected into mice intraperitoneally three times a week at a dose of 0.12 ml / mouse. Weak positive group: Bovine serum albumin solution with a concentration of 0.1C1 was emulsified with an equal volume of Freund's complete adjuvant to obtain a suspension. The suspension was injected into mice intraperitoneally three times a week at a dose of 0.12 ml / mouse. Strong positive group: Bovine serum albumin solution with a concentration of C1 was fully emulsified with an equal volume of Freund's complete adjuvant to obtain a suspension. The suspension was injected into mice intraperitoneally three times a week at a dose of 0.12 ml / mouse. Four weeks after the first injection, blood samples were collected to prepare serum samples. The concentrations of immunoglobulins IgE, IgG, IgM, complement C3, and inflammatory factors TNF-α and IL-6 in each group of serum samples were measured to establish a semi-quantitative immune response system with gradient characteristics. S12. Validate the established semi-quantitative immunoreaction system with gradient characteristics: The test results of each indicator in the borderline positive group, weak positive group, strong positive group, and negative control group were compared pairwise. The statistical P value was used to determine whether there was a significant difference between the indicator groups (P < 0.05). At the same time, the number of indicators that showed significant differences compared with the negative control group and the increase of the indicators were used to verify whether a semi-quantitative immune response system with gradient characteristics was successfully established.

[0010] More preferably, C1 in step S11 is 0.2-0.4 mg / mL; more preferably, C1 is 0.33 mg / mL; the bovine serum albumin solution is prepared by dissolving bovine serum albumin in a phosphate buffer solution, and the pH of the phosphate buffer solution is 7.2-7.5.

[0011] More preferably, the serum sample in step S12 is the supernatant collected by centrifuging whole blood from mice.

[0012] In a further preferred embodiment, step S12 specifically includes: Analysis of variance and Dunnett's test were used to analyze the significant differences. Borderline positive group: In serum samples, the mean values ​​of at least four indicators increased by more than 10% compared with the negative control group, but none of the indicators reached statistical significance (P value ≥ 0.05). Weakly positive group: In serum samples, compared with the negative control group, the mean values ​​of at least 5 indicators (IgE, IgG, IgM, complement C3, and inflammatory factors TNF-α and IL-6) increased by more than 10%, at least 3 indicators increased by more than 50%, and at least 2 indicators showed statistically significant increases (P value < 0.05); compared with the borderline positive group, the mean values ​​of at least 3 indicators increased by more than 10%. Strong positive group: Compared with the negative control group, the mean values ​​of at least 4 indicators increased by more than 50%, at least 2 indicators increased by more than 500%, and the increase of 6 indicators reached statistical significance (P value less than 0.05); Compared with the weak positive group, the mean values ​​of at least 4 indicators increased by more than 10%, at least 2 indicators increased by more than 200%, and at least 3 indicators increased significantly (P value less than 0.05).

[0013] Preferably, step S2 specifically includes: S21. Using single-crystal thin films as the matrix material for SERS substrates, the substrates are surface polished, cut, and cleaned. S22. Orthogonal bidirectional processing is performed on the surface of a single-crystal thin film using a femtosecond laser to form a two-dimensional periodic nanostructure on the surface of the single-crystal thin film. S23. A noble metal layer is deposited on the surface of the nanostructure by thermal evaporation, and then annealed to form a uniform array of silver nanoparticles, thus forming a SERS substrate with a two-dimensional periodic nanostructure.

[0014] More preferably, the single-crystal thin film in step S21 is a lithium niobate single-crystal thin film or a lithium niobate single-crystal thin film.

[0015] Further preferably, the parameters for femtosecond laser processing in step S22 are as follows: orthogonal bidirectional processing is performed on the thin film surface using a femtosecond laser, with a wavelength of 800 nm or 1030 nm, a pulse repetition frequency of 1 kHz to 1 MHz, a pulse width of 65 to 150 femtoseconds, a processing speed of 1 μm / s to 100 μm / s, and the laser polarization is linear polarization.

[0016] More preferably, the precious metal layer in step S23 is a gold layer or a silver layer.

[0017] Preferably, step S3 specifically includes: S31. Dilute the mouse serum obtained in step S1 with deionized water and add it dropwise to the SERS substrate with two-dimensional periodic nanostructure prepared in step S2, and collect the serum SERS spectrum. S32. Expand the data volume using the random averaging method of the same group of spectra to obtain the dataset; S33. Construct a machine learning model, and use a dataset to train and test the machine learning model in order to achieve accurate identification of serum spectra.

[0018] Preferably, the parameters of the spectra acquired by the confocal micro Raman spectrometer in step S31 are: excitation wavelength of 488, 532 or 785 nm, power of 0.1-10 mW, integration time of 0.5-10 s, and integration times of 1-10.

[0019] Preferably, in step S31, the volume ratio of mouse serum to deionized water is 1:8 to 1:10.

[0020] Preferably, the machine learning model in step S33 is a CNN model or a ReNet model.

[0021] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention integrates animal experiments, femtosecond laser precision processing, highly stable SERS substrates, and machine learning spectral diagnostic technology to construct an end-to-end rapid immune detection system, overcoming the technical contradiction of the incompatibility between "sensitivity, throughput, and comprehensive index detection" in immune detection. It provides an efficient and reliable integrated tool for the safety evaluation of biomaterials and the diagnosis of clinical immune diseases. Compared with traditional ELISA methods, the machine learning-assisted SERS technology used in this invention has significant advantages, as shown in Table 1: Table 1 Comparison of Machine Learning-Assisted SERS Technology with Traditional Methods

[0022] (2) This invention successfully established a four-level semi-quantitative immune response system for mice by scientifically designing the BSA concentration and fold relationship between each positive group, and obtained serum from mice with four levels of immune response. This response system breaks through the limitations of the traditional purely qualitative immune response. Through gradient dose design, it successfully constructed an immune response system covering four levels: "negative—borderline positive—weak positive—strong positive". Among them, the establishment of the borderline positive group is particularly crucial. Although its immune indicators did not reach the statistical significance threshold of the traditional ELISA method, it represented the early activation stage of the immune response. This invention, combined with machine learning-assisted SERS technology, can accurately identify the weak spectral features of this stage, achieve sensitive capture of the initial stage of the immune response, and fill the blind spot of traditional methods in early immune recognition.

[0023] (3) This invention employs femtosecond laser orthogonal bidirectional processing technology to construct a highly uniform and stable two-dimensional nanostructure on the surface of a lithium niobate single-crystal thin film, and then thermally vapor-deposits silver nanoparticles on its surface to form a SERS substrate. This substrate exhibits high detection sensitivity, high preparation efficiency, good structural uniformity, and high stability, making it suitable for trace detection of various types and components of biomolecules. The lithium niobate thin film (LNOI) retains the acousto-optic, electro-optic, ferroelectric, and nonlinear optical properties of lithium niobate (LN), while also improving the refractive index contrast, which is beneficial for enhancing the Raman signal. The two-dimensional periodic nanostructure demonstrates good stability and repeatability in Raman signal detection, and its excellent SERS performance gives it a significant advantage in detecting low-concentration molecules or biomolecules in complex environments. Traditional SERS detection substrates generally use colloidal sols or chemical methods to synthesize nanoparticles, which are limited by the high randomness, low stability, and non-repeatability of the substrate, and cannot meet the needs of immunoassay. SERS substrates prepared by traditional chemical synthesis methods generally suffer from problems such as strong structural randomness, uneven hot spot distribution, and poor batch-to-batch reproducibility, making it difficult to meet the high requirements of immunoassay for signal stability and reproducibility.

[0024] (4) This invention uses SERS technology to detect immune response, realizes simultaneous detection of all indicators, greatly shortens the detection time and reduces sample consumption. It has high throughput, high sensitivity and high accuracy, and provides a more advanced solution for high-throughput immune response detection of trace serum samples. It is suitable for rapid semi-quantitative detection of immune response through trace serum samples.

[0025] (5) This invention innovatively uses the random averaging method of the same group spectrum to expand the data volume, and combines the machine learning model to automatically identify overlapping spectral features and suppress individual variation interference, breaking through the limitation of high error rate of manual interpretation, and realizing high-precision intelligent identification of immune status. Brief Description of the Drawings

[0026] Figure 1 This is a process flow chart of the semi - quantitative immunoassay detection method based on serum SERS and machine learning with a lithium niobate thin - film nanostructure as the substrate for the present invention; Figure 2 This is the end - face view (a) and top - face view (b) of the lithium niobate single - crystal thin film used in Example 1 of the present invention; Figure 3 This is a flow chart of femtosecond laser orthogonal two - direction processing on the thin - film surface in Example 1 of the present invention. Among them, in Figure (a), the femtosecond laser is focused on the thin - film surface for processing. In Figure (b), a series of processing lines are generated by the first laser processing, and each line contains a periodic nanostructure schematic diagram. In Figure (c), a schematic diagram of the nanostructures generated at the intersection after the second orthogonal processing; Figure 4 This is the CNN model structure diagram in Example 1 of the present invention; Figure 5 This is the heat - map of the confusion matrix of the 20% test set of 4 groups of sera in the present invention; Figure 6 This is the test result of the verification example in the present invention. In Figure (a) is the result of the negative group, in Figure (b) is the result of the low - dose group, in Figure (c) is the result of the medium - dose group, in Figure (d) is the result of the high - dose group, and in Figure (e) is the result of the positive group; Figure 7 This is the Raman spectrum diagram of the comparative experiment on SERS substrates prepared by the single - direction processing method and the two - direction processing method in Comparative Example 3 of the present invention; The reference signs are as follows: 1. Lithium niobate single - crystal thin - film layer; 2. Silicon dioxide layer; 3. Lithium niobate crystal layer; 4. Focusing objective lens; 5. Femtosecond laser; 6. Periodic nanowire structure; 7. Nanostructures generated at the intersection after the second orthogonal processing. Detailed Embodiments

[0027] The mouse strain used in the examples and comparative examples of the present invention is BALB / C, purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd., license number: SCXK (Beijing) 20210006; Bovine serum albumin was purchased from Sigma, product number: SRE0096, purity ≥ 98.0%; Freund's complete adjuvant was purchased from Sigma, product number: F5881; 1×PBS buffer was purchased from Biosharp, product number: BL302A; The Mouse IgG ELISA Kit was purchased from Alpha Diagnostic International (ADI) in the United States, catalog number 6320, lot number 250627K2; The Mouse IgM ELISA Kit was purchased from Alpha Diagnostic International (ADI) in the United States, catalog number 6380, lot number 250627K4; The Mouse IgE ELISA Kit was purchased from Alpha Diagnostic International (ADI) in the United States, catalog number 6370, lot number 250801K1; The Mouse Complement C3 ELISA Kit was purchased from AlphaDiagnostic International (ADI) in the United States, catalog number 6270, lot number MsC3-0461A1; The Mouse TNF-α ELISA Kit was purchased from Invitrogen, catalog number BMS607-3, lot number 386423-007; The Mouse IL-6 ELISA Kit was purchased from Invitrogen, catalog number KMC0061, lot number 398400-008.

[0028] Example 1 This embodiment provides a semi-quantitative immunoreaction detection method based on serum SERS and machine learning using a lithium niobate thin film nanostructure as a substrate, such as... Figure 1 As shown, it includes the following steps: S1. Through mouse animal experiments, obtain mouse serum with four levels of immune response, including negative, borderline positive, weak positive and strong positive, establish a semi-quantitative immune response system with gradient characteristics, and verify the established semi-quantitative immune response system with gradient characteristics. S11. Experimental mice (with a weight difference not exceeding ±10% of the average) were randomly divided into 4 groups (10 mice in each group), which were used as negative control group, borderline positive group, weak positive group and strong positive group, respectively. Negative control group: Mice were not treated in any way and were fed normally; Prepare a mixed solution by mixing 3 mg bovine serum albumin with 9 mL PBS (pH=7.4). After thorough mixing, dilute 10-fold to create three dose groups: a strong positive group (0.33 mg / mL), a weak positive group (0.033 mg / mL), and a borderline positive group (0.0033 mg / mL). Take 3 mL of each dose group's mixed solution and mix it with an equal volume of 3 mL Freund's complete adjuvant. Inject 0.12 mL of the suspension into the corresponding group of mice via intraperitoneal injection. Administer the injections to mice intraperitoneally three times a week, with each group of mice receiving the injection at the same time.

[0029] Four weeks after the first injection, blood was collected to prepare serum samples. ELISA kits were purchased to measure the concentrations of immunoglobulins IgE, IgG, IgM, complement C3, and inflammatory factors TNF-α and IL-6 in each group of serum samples, establishing a semi-quantitative immune response system with gradient characteristics. The kits include: mouse IgG ELISA kit, mouse IgM ELISA kit, mouse IgE ELISA kit, mouse complement C3 ELISA kit, mouse TNF-α ELISA kit, and mouse-derived IL-6 kit.

[0030] S12. Validate the established semi-quantitative immunoreaction system with gradient characteristics: The test results of each indicator in the borderline positive group, weak positive group, strong positive group, and negative control group were compared pairwise. The statistical P value was used to determine whether there was a significant difference between the indicator groups (P < 0.05). At the same time, the number of indicators that showed significant differences compared with the negative control group and the increase of the indicators were used to verify whether a semi-quantitative immune response system with gradient characteristics was successfully established.

[0031] The serum sample was the supernatant collected by centrifuging whole blood from mice.

[0032] Analysis of variance and Dunnett's test were used to analyze the significant differences.

[0033] The results show: Compared with the negative control group, the mean values ​​of each indicator in the borderline positive group were as follows: IgG increased by 1.9% (P=0.858, P>0.05), IgM increased by 31.4% (P=0.156, P>0.05), IgE increased by 6.7% (P=0.296, P>0.05), C3 increased by 16.6% (P=0.222, P>0.05), TNF-α increased by 65.7% (P=0.36, P>0.05), and IL-6 increased by 66.8% (P=0.257, P>0.05).

[0034] Compared with the negative control group, the mean values ​​of each indicator in the weakly positive group were as follows: IgG increased by 15.9%, P=0.099 (P>0.05); IgM increased by 68.7%, P=0.003 (P<0.05); IgE increased by 8.8%, P=0.171 (P>0.05); C3 increased by 31.8%, P=0.019 (P<0.05); TNF-α increased by 143.8%, P=0.048 (P<0.05); and IL-6 increased by 87.0%, P=0.142 (P>0.05). Compared with the borderline positive group, the mean values ​​of each indicator in the weakly positive group were as follows: IgG increased by 13.8%, P=0.140 (P>0.05); IgM increased by 28.4%, P=0.094 (P>0.05); IgE increased by 2.0%, P=0.742 (P>0.05); C3 increased by 13.1%, P=0.241 (P>0.05); TNF-α increased by 47.1%, P=0.276 (P>0.05); and IL-6 increased by 12.1%, P=0.731 (P>0.05).

[0035] Compared with the negative control group, the mean values ​​of all indicators in the strongly positive group were as follows: IgG increased by 57.4% (P < 0.001, P < 0.05), IgM increased by 73.4% (P < 0.001, P < 0.05), IgE increased by 22.4% (P < 0.001, P < 0.05), C3 increased by 34.4% (P < 0.001, P < 0.05), TNF-α increased by 5139.5% (P < 0.001, P < 0.05), and IL-6 increased by 595.4% (P < 0.001, P < 0.05). 5) Compared with the weakly positive group, the mean values ​​of each indicator in the strongly positive group were as follows: IgG increased by 35.8% (P<0.001, P<0.05), IgM increased by 2.8% (P=0.832, P>0.05), IgE increased by 12.6% (P=0.035, P<0.05), C3 increased by 2.0% (P=0.843, P>0.05), TNF-α increased by 2049% (P<0.001, P<0.05), and IL-6 increased by 271.9% (P<0.001, P<0.05).

[0036] The results above indicate that the mean values ​​of each indicator in the borderline positive group increased to varying degrees compared with the negative control group, but the differences did not reach statistical significance (P>0.05); some indicators (IgG, IgM, and C3) in the weakly positive group showed statistically significant differences compared with the negative control group (P<0.05); all six indicators in the strongly positive group showed statistically significant differences compared with the negative control group (P<0.05); the levels of some indicators (IgE, IgG, TNF-α, and IL-6) also showed statistically significant differences in gradient among the positive groups (P<0.05).

[0037] The above results fully demonstrate that a semi-quantitative immune response system was successfully established, and serum from mice with a grade 4 immune response was obtained.

[0038] S2. Using single-crystal thin films as the matrix material for SERS substrates, SERS substrates with two-dimensional periodic nanostructures are formed by processing with femtosecond laser technology. The SERS substrate was prepared using lithium niobate single-crystal thin film as the matrix material, and its SERS performance was verified. The process included the following steps: S21. Using lithium niobate single-crystal thin film as the matrix material for SERS substrate, the substrate is subjected to surface polishing, cutting, and surface cleaning, as detailed below: The SERS substrate uses an X-shaped tangential lithium niobate single-crystal thin film as the matrix material, such as... Figure 2 As shown, the substrate has a three-layer structure, consisting of a lithium niobate crystal layer with a thickness of about 500 micrometers, a SiO2 layer with a thickness of about 2 micrometers, and a lithium niobate single crystal thin film layer with a thickness of about 600 nm from bottom to top. After polishing, it is cut to a surface size of 20(y)×20(z) square millimeters, cleaned, and then fixed on the processing platform. S22. Orthogonal bidirectional processing is performed on the surface of a lithium niobate single-crystal thin film using a femtosecond laser to form a two-dimensional periodic nanostructure on the surface of the single-crystal thin film, as detailed below: Orthogonal bidirectional processing was performed on the thin film surface using a femtosecond laser. The femtosecond laser wavelength was 800 nm, the pulse repetition frequency was 1 kHz, the pulse width was 65 femtoseconds, the processing speed was 90 μm / s, and the laser polarization was linear. A 20× objective lens (numerical aperture 0.4) was used to focus the laser onto the thin film surface. Figure 3 As shown in (a), the processing platform is moved to scan the thin film surface, with the scanning direction parallel to the laser polarization direction. The initial processing laser single pulse energy is 15 nanojoules, the length of a single processing line is 100 micrometers, the center-to-center distance between adjacent processing lines is 1 micrometer, and 100 lines are processed. Figure 3 As shown in (b). The sample was rotated 90 degrees and then subjected to secondary processing. The processing energy was 9 nanojoules, the length of a single processing line was 100 micrometers, the center-to-center distance between adjacent processing lines was 1 micrometer, and 100 lines were processed, as shown. Figure 3 As shown in (c), uniform large-area nanoisland structures were prepared using the above method. Scanning electron microscopy verified that the average length of the nanoislands was 125 nm, the average width was 150 nm, the average size was <150 nm, and the average period was 163 nm, <170 nm.

[0039] S23. A noble metal layer is deposited on the surface of the nanostructure by thermal evaporation, followed by annealing to form a uniform array of silver nanoparticles, thus forming a SERS substrate with a two-dimensional periodic nanostructure, as detailed below: At a vacuum degree of 5×10-4 A 5 nm thick silver film was thermally deposited on the surface of a nanostructure under Pa conditions at a deposition rate of 0.1 Å / s. After thermal deposition, the same 5 × 10⁻⁶ film was maintained. -4 Silver nanoparticles were formed by annealing at 500 degrees Celsius for 30 minutes under a vacuum atmosphere. The average diameter of the silver nanoparticles was verified to be 25 nanometers and the average gap was 5 nanometers using scanning electron microscopy.

[0040] Select 10 -5 Rhodamine 6G (R6G) at a concentration of mol / L was used as the analyte. Five microliters of the analyte solution were dropped onto the surface of a SERS substrate with a two-dimensional periodic nanostructure. The SERS values ​​of R6G (500-1800 mC / cm) were acquired using a confocal Raman spectrometer (excitation wavelength 488 nm, power 2.55 mW, integration time 10 s). Using the characteristic peak of R6G at 613 mC / cm as a reference, the substrate enhancement factor was verified to be >10. 6 Twenty different points were randomly selected to test the characteristic peak of 613 per centimeter. The relative standard deviation was verified to be 4.61% < 5%, indicating that the two-dimensional periodic nanostructure SERS substrate exhibits good stability and repeatability in Raman signal detection.

[0041] S3. The four levels of immune response mouse serum obtained in step S1 are dropped onto the SERS substrate with a two-dimensional periodic nanostructure prepared in step S2, and their SERS spectra are collected to construct a spectral dataset. The dataset is used to train and test the machine learning model to achieve semi-quantitative detection of the immune response.

[0042] S31. Dilute the mouse serum obtained in step S1 with a volume ratio of serum:deionized water = 1:9, and add 2 μL to the SERS substrate with two-dimensional periodic nanostructure prepared in step S2. After drying at room temperature, collect the serum SERS spectrum using a confocal micro Raman spectrometer. S32. The data volume is expanded by the random averaging method of the same group of spectra to obtain a dataset; the random averaging method of the same group of spectra is prior art and is not the technical solution to be protected in this application. S33. Construct a machine learning model, using 80% of the SERS spectra from the negative group, borderline positive group, weak positive group, and strong positive group as the training set, and the remaining 20% ​​as the test set. Train and test the machine learning model to achieve accurate identification of serum spectra (100% identification rate).

[0043] Building a CNN model: This model is a serum spectral intelligent classification model based on a one-dimensional convolutional neural network (1D-CNN), specifically designed for the automatic identification and classification of Raman spectral data of four different concentration categories, such as... Figure 4As shown. First, during the data loading stage, spectral data was read to filter the 400-1800 cm⁻¹ range. - ¹ Wavenumber range and spectral length of all samples were standardized; In the data visualization stage, functions were called to plot the distribution of sample numbers for each category and randomly selected spectral curves, helping to intuitively understand the balance of the dataset and differences in spectral features; then data preprocessing was performed, and the spectral data was standardized; after preprocessing, the dataset was split into 80% training set and 20% test set to ensure consistent distribution across categories; in the model building stage, a deep learning model was constructed, consisting of four convolutional layers (the number of filters increased from 64 to 512), a batch normalization layer, a max pooling layer, and a Dropout regularization layer. A one-dimensional convolutional neural network is used, which outputs a Softmax classification result after global average pooling and two fully connected layers. The model is then compiled using the Adam optimizer and classification cross-entropy loss function, with accuracy as the evaluation metric. During model training, the results are visualized, plotting accuracy and loss curves for training and validation, as well as a confusion matrix heatmap. In the model evaluation phase, loss and accuracy are calculated on the validation set, and a detailed classification report including precision, recall, and F1 score is output. Finally, the trained model is saved for subsequent direct prediction of unknown serum samples.

[0044] Specifically, training ends when the model converges.

[0045] Serum SERS (400-1800 per centimeter) were collected using a confocal Raman spectrometer (excitation wavelength 488 nm, power 2.55 mW, integration time 10 s), with 100 spectra collected for each serum group. The trained model was tested using a test set, and the results are as follows: Figure 5 As shown, the horizontal axis represents the predicted category after model training, and the vertical axis represents the original actual category of the serum. The values ​​in the matrix represent the percentage of the predicted result relative to the total number of samples of that true category. In the figure, the accuracy of the predicted category is 100% in the negative group, strong positive group, borderline positive group, and weak positive group. The diagonal elements of the confusion matrix (the predicted category is consistent with the true category).

[0046] The accurate identification of borderline positive groups indicates that this detection system is more sensitive than traditional methods.

[0047] Verification Example Experimental mice (with a weight difference not exceeding ±10% of the mean) were randomly divided into 5 groups (n=10 per group), serving as the negative control group, positive control group, low-dose sample group, medium-dose sample group, and high-dose sample group, respectively. In the low-dose sample group, each mouse underwent a subcutaneous implantation of a 0.16 cm... 2The sample (absorbable biomembrane: composed of porcine small intestinal submucosa and extracellular matrix, classified as a medical device for tissue repair) was subcutaneously implanted at a depth of 0.82 cm in each mouse in the medium-dose sample group. 2 In the high-dose sample group, each mouse underwent subcutaneous implantation of 4.10 cm. 2 The negative control group underwent sham surgery, while the positive control group underwent the same procedure as the strongly positive group in S11 of Example 1. Serum samples were prepared 28 days after implantation. Using an ELISA kit, the concentrations of immunoglobulins IgE, IgG, IgM, complement C3, and inflammatory factors TNF-α and IL-6 in the serum samples of each group were measured. The mean values ​​of each indicator in the sample groups and positive groups were statistically compared with those in the negative control group. Results showed that compared with the negative control group, there were no statistically significant differences in the mean values ​​of any indicator in the three dose sample groups, while the mean values ​​of all six indicators in the positive control group were significantly higher than those in the negative control group. Serum from each group was collected, and SERS spectra were acquired using the SERS substrate with a two-dimensional periodic nanostructure prepared in step S2. The smoothed SERS spectra of the five groups were input into a tested machine learning model, and the output results were as follows: Figure 6 As shown: negative group (100% negative), positive group (100% positive), low-dose sample group (92.2% negative), medium-dose sample group (92.8% negative), high-dose sample group (99.8% negative), consistent with the results of the traditional ELISA method.

[0048] Comparative Example 1 The steps are the same as in Example 1, except that in step S1, during the construction of the four-level immune response, the fold ratio between the borderline positive group, the weakly positive group, and the positive group is 2. 3 mg of bovine serum albumin and 9 mL of PBS (pH=7.4) are mixed to prepare a solution. After thorough mixing, it is diluted 2-fold to form three dose groups: 0.33 mg / mL for the strong positive group, 0.165 mg / mL for the weakly positive group, and 0.0825 mg / mL for the borderline positive group. 3 mL of each dose group mixture is mixed with an equal volume of 3 mL of Freund's complete adjuvant, and 0.12 mL of the suspension is injected intraperitoneally into the corresponding group of mice. Mice are injected intraperitoneally three times a week.

[0049] Four weeks later, blood samples were collected to prepare serum samples. ELISA kits were purchased to measure the concentrations of immunoglobulins IgE, IgG, IgM, complement C3, and inflammatory factors TNF-α and IL-6 in the serum samples of each group. The results of the borderline positive group, weakly positive group, and strongly positive group were statistically compared with the negative control group. Based on the number of indicators showing significant differences in each group and the P-value, the success of establishing a semi-quantitative immune response system with gradient characteristics was verified.

[0050] The results showed that the mean values ​​of all indicators in the borderline positive group increased to varying degrees compared with the negative control group. The increases in mean values ​​of IgG and IgM were statistically significant (IgG P=0.017, IgM P=0.010), indicating an overall weak positive reaction, failing to achieve the ideal borderline positive reaction. In the weakly positive group, except for IL-6, all other indicators showed statistically significant increases compared with the negative control group (P<0.05), indicating a relatively strong positive reaction. In the strongly positive group, all six indicators showed statistically significant differences compared with the negative control group (P<0.05), with no significant gradient in difference between the group and the weakly positive group. These results indicate that an ideal semi-quantitative immune response system could not be established.

[0051] Comparative Example 2 The rest is the same as in Example 1, except that in step S1, during the construction of the four levels of immune response, the ratio of critical positive group, weak positive group and positive group is 15, and the concentration of critical positive group is 0.0015 mg / mL. Excessive dilution cannot stimulate the body to produce an immune response and critical positive serum cannot be obtained.

[0052] Comparative Example 3 Step S2 uses a unidirectional processing method to process the SERS substrate, and the processing parameters are the same as in Example 1: A comparative experiment was conducted on SERS substrates prepared by unidirectional and bidirectional processing methods. A concentration of 10 was used in the experiment. -5 Using R6G as a probe molecule, 5 μL of R6G analyte solution was dropped onto the surfaces of two different SERS substrates. Subsequently, surface-enhanced Raman scattering (SERS) signals of R6G were obtained using Raman incident light at a wavelength of 488 nm. The results are shown below. Figure 7 As shown. Specifically, R6G at 613 cm - Taking the characteristic Raman peak at position ¹ as an example, the signal intensity of this characteristic peak measured on a unidirectional processed substrate is lower than the signal intensity measured on a bidirectional processed substrate.

[0053] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate, characterized in that, Includes the following steps: S1. Through mouse animal experiments, obtain mouse serum with four levels of immune response, including negative, borderline positive, weak positive and strong positive, establish a semi-quantitative immune response system with gradient characteristics, and verify the established semi-quantitative immune response system with gradient characteristics. S2. Using single-crystal thin films as the matrix material for SERS substrates, SERS substrates with two-dimensional periodic nanostructures are formed by processing with femtosecond laser technology. S3. The four levels of immune response mouse serum obtained in step S1 are dropped onto the SERS substrate with a two-dimensional periodic nanostructure prepared in step S2, and their SERS spectra are collected to construct a spectral dataset. The dataset is used to train and test the machine learning model to achieve semi-quantitative detection of the immune response.

2. The semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate, as described in claim 1, is characterized in that... Step S1 is as follows: S11. Experimental mice whose body weight difference does not exceed ±10% of the average value are randomly divided into 4 groups: negative control group, borderline positive group, weak positive group, and strong positive group. Negative control group: Mice were not treated in any way and were fed normally; Borderline positive group: Bovine serum albumin solution with a concentration of 0.01C1 was emulsified with an equal volume of Freund's complete adjuvant to obtain a suspension. The suspension was injected into mice intraperitoneally three times a week at a dose of 0.12 ml / mouse. Weak positive group: Bovine serum albumin solution with a concentration of 0.1C1 was emulsified with an equal volume of Freund's complete adjuvant to obtain a suspension. The suspension was injected into mice intraperitoneally three times a week at a dose of 0.12 ml / mouse. Strong positive group: Bovine serum albumin solution with a concentration of C1 was fully emulsified with an equal volume of Freund's complete adjuvant to obtain a suspension. The suspension was injected into mice intraperitoneally three times a week at a dose of 0.12 ml / mouse. Four weeks after the first injection, blood samples were collected to prepare serum samples. The concentrations of immunoglobulins IgE, IgG, IgM, complement C3, and inflammatory factors TNF-α and IL-6 in each group of serum samples were measured to establish a semi-quantitative immune response system with gradient characteristics. S12. Validate the established semi-quantitative immunoreaction system with gradient characteristics: The test results of each indicator in the borderline positive group, weak positive group, strong positive group, and negative control group were compared pairwise. The statistical P value was used to determine whether there was a significant difference between the indicator groups (P < 0.05). At the same time, the number of indicators that showed significant differences compared with the negative control group and the increase of the indicators were used to verify whether a semi-quantitative immune response system with gradient characteristics was successfully established.

3. The semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate, as described in claim 2, is characterized in that... In step S11, C1 is 0.2-0.4 mg / mL; more preferably, C1 is 0.33 mg / mL; the bovine serum albumin solution is prepared by dissolving bovine serum albumin in a phosphate buffer solution, and the pH of the phosphate buffer solution is 7.2-7.

5.

4. The semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate, as described in claim 3, is characterized in that... Step S12 is as follows: Analysis of variance and Dunnett's test were used to analyze the significant differences. Borderline positive group: In serum samples, the mean values ​​of at least four indicators increased by more than 10% compared with the negative control group, but none of the indicators reached statistical significance. Weakly positive group: In serum samples, compared with the negative control group, the mean values ​​of at least 5 indicators (IgE, IgG, IgM, complement C3, and inflammatory factors TNF-α and IL-6) increased by more than 10%, at least 3 indicators increased by more than 50%, and at least 2 indicators showed statistically significant increases; compared with the borderline positive group, the mean values ​​of at least 3 indicators increased by more than 10%. Strong positive group: Compared with the negative control group, the mean values ​​of at least 4 indicators increased by more than 50%, at least 2 indicators increased by more than 500%, and the increase of 6 indicators reached statistical significance; compared with the weak positive group, the mean values ​​of at least 4 indicators increased by more than 10%, at least 2 indicators increased by more than 200%, and the increase of 3 indicators was statistically significant.

5. The semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate, as described in claim 1, is characterized in that... Step S2 is as follows: S21. Using single-crystal thin films as the matrix material for SERS substrates, the substrates are surface polished, cut, and cleaned. S22. Orthogonal bidirectional processing is performed on the surface of a single-crystal thin film using a femtosecond laser to form a two-dimensional periodic nanostructure on the surface of the single-crystal thin film. S23. A noble metal layer is deposited on the surface of the nanostructure by thermal evaporation, and then annealed to form a uniform array of silver nanoparticles, thus forming a SERS substrate with a two-dimensional periodic nanostructure.

6. The semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate, as described in claim 5, is characterized in that... The single-crystal thin film mentioned in step S21 is a lithium niobate single-crystal thin film or a lithium niobate single-crystal thin film.

7. The semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate, as described in claim 5, is characterized in that... The parameters for femtosecond laser processing in step S22 are as follows: orthogonal bidirectional processing is performed on the thin film surface using a femtosecond laser, with a wavelength of 800 nm or 1030 nm, a pulse repetition frequency of 1 kHz to 1 MHz, a pulse width of 65-150 femtoseconds, a processing speed of 1 μm / s to 100 μm / s, and the laser polarization is linear polarization.

8. The semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate, as described in claim 5, is characterized in that... The precious metal layer mentioned in step S23 is a gold layer or a silver layer.

9. The semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate, as described in claim 1, is characterized in that... Step S3 is as follows: S31. Dilute the mouse serum obtained in step S1 with deionized water and add it dropwise to the SERS substrate with two-dimensional periodic nanostructure prepared in step S2, and collect the serum SERS spectrum. S32. Expand the data volume using the random averaging method of the same group of spectra to obtain the dataset; S33. Construct a machine learning model, and use a dataset to train and test the machine learning model in order to achieve accurate identification of serum spectra.

10. The semi-quantitative immunoreaction detection method based on serum SERS and machine learning using lithium niobate thin film nanostructures as a substrate, as described in claim 1, is characterized in that... In step S31, the volume ratio of mouse serum to deionized water is 1:8 to 1:10; the machine learning model mentioned in step S33 is a CNN model or a ReNet model.

Images