A method for detecting a protein aggregation state and a detection device thereof

Abstract

The present application relates to a kind of protein aggregation state detection method and its detection device.The protein aggregation state detection method described in the present application includes: the solution to be detected is mixed with optical microcavity, obtains the second original WGM spectrum data after the solution to be detected and optical microcavity effect;The optical bar code of the second original WGM spectrum data is obtained;The optical bar code is input into detection classification model, and the protein aggregate class is obtained;Wherein, the detection classification model includes deep learning classification model.The detection method of the present application can automatically, efficiently extract deep features from complex optical bar code, and can learn the nonlinear discrimination boundary between different protein aggregation states, overcome the limitations of traditional methods relying on artificial feature extraction and simple linear judgment, greatly improve the accuracy and robustness of protein aggregation state classification.

Description

Technical Field

[0001] This invention relates to the field of protein detection technology, and in particular to a method and apparatus for detecting protein aggregation state. Background Technology

[0002] Abnormal protein aggregation is closely related to the development and progression of various neurodegenerative diseases, such as Alzheimer's and Parkinson's. Accurate and rapid detection and differentiation of protein aggregation states (monomers, oligomers, fibers, etc.) are of great significance for understanding disease mechanisms, developing early diagnostic methods, and evaluating drug efficacy.

[0003] Currently, commonly used techniques for detecting protein aggregation include size exclusion chromatography, mass spectrometry, dynamic light scattering, atomic force microscopy, and transmission electron microscopy. However, these methods often have some limitations, such as expensive equipment, complex operation, cumbersome sample preparation, time-consuming detection, difficulty in achieving real-time in-situ monitoring, or insufficient sensitivity for detecting early low-concentration oligomers.

[0004] Optical whispering gallery mode (WGM) microcavities, due to their extremely high quality factor and tiny mode volume, can interact strongly with light and are highly sensitive to changes in refractive index in micro- and nano-scale environments, making them a powerful and highly sensitive biosensing platform. Currently, traditional WGM sensors primarily detect the presence or concentration changes of biomolecules by monitoring single or multiple resonant wavelength shifts caused by target molecule binding. However, when the sensing target shifts from "detecting presence" to "distinguishing complex states" (such as differentiating the same protein in different conformations or aggregate states), single frequency shift information often lacks sufficient specificity. Furthermore, the interaction between proteins in different aggregate states and the microsphere surface can produce similar but not identical global spectral responses. These subtle differences are difficult to effectively resolve and distinguish using simple thresholds or linear models, limiting their application in complex biomedical detection. Summary of the Invention

[0005] Based on this, the purpose of the present invention is to provide a method and device for detecting protein aggregation states, which achieves high accuracy and rapid detection and identification of multiple protein aggregation states.

[0006] A method for detecting protein aggregation state includes the following steps:

[0007] The solution to be tested is mixed with the optical microcavity to obtain the second original WGM spectral data after the interaction between the solution to be tested and the optical microcavity.

[0008] Obtain the optical barcode of the second raw WGM spectral data;

[0009] The optical barcode is input into a detection and classification model to obtain the protein aggregate category; wherein, the detection and classification model includes a deep learning classification model.

[0010] This invention treats the entire WGM spectrum obtained after the test solution interacts with the optical microcavity as a high-dimensional "optical barcode" and uses a deep learning model to detect and classify protein aggregation states. It can automatically and efficiently extract deep features from the complex optical barcode and learn the nonlinear discrimination boundary between different protein aggregation states. This method overcomes the limitations of traditional methods that rely on manual feature extraction and simple linear judgment, and greatly improves the accuracy and robustness of protein aggregation state classification.

[0011] Furthermore, obtaining the optical barcode from the second original WGM spectral data includes: assigning the center wavelength of each resonance peak in the second original WGM spectral data to the position of the barcode, the full width at half maximum (FWHM) of the resonance peak to the width of the barcode, and the ratio of the resonance peak to the fluorescence background intensity generated by the fluorescent dye excited under the same laser power to the color depth of the barcode.

[0012] Furthermore, the deep learning classification model includes convolutional neural networks, recurrent neural networks, Transformers, or fully connected neural networks.

[0013] Furthermore, the deep learning classification model is a four-class convolutional neural network model, which includes an input layer, two convolutional layers, two max pooling layers, one flattening layer, two fully connected layers, and an output layer.

[0014] Furthermore, the optical microcavity is an optical microcavity doped with fluorescent dye, and the detection method further includes: acquiring first raw WGM spectral data after the fluorescent dye doped in the optical microcavity interacts with the optical microcavity, and simultaneously acquiring an optical barcode of second raw WGM spectral data based on the first raw WGM spectral data.

[0015] Furthermore, the fluorescent dye is a coumarin, rhodamine, or fluorescein dye.

[0016] Furthermore, the first original WGM spectral data is subtracted from the second original WGM spectral data, and then the optical barcode is obtained from the processed spectral data.

[0017] Furthermore, after subtracting the first original WGM spectral data from the second original WGM spectral data, smoothing filtering and normalization processing are performed sequentially, and then the optical barcode is obtained from the normalized spectral data.

[0018] Furthermore, the present invention also provides a protein aggregation state detection device, comprising an optical sensing unit, a stage, an optical excitation unit, an optical acquisition unit, and a data processing unit; wherein, the optical sensing unit is an optical microcavity; the optical excitation unit includes an excitation light source and a first objective lens connected in sequence by optical paths; the optical acquisition unit includes a second objective lens, a filter, and a spectrometer connected in sequence by optical paths; the data processing unit is connected to the spectrometer and includes an optical barcode generation module and a detection classification module. The optical barcode generation module is used to convert the second raw WGM spectral data generated by the interaction of the test solution acquired by the spectrometer with the optical microcavity into an optical barcode. The detection classification module includes a deep learning classification model for detecting the protein aggregation state based on the optical barcode.

[0019] Furthermore, the optical microcavity is an optical microcavity doped with fluorescent dye, and the data processing unit further includes a preprocessing module for subtracting the first original WGM spectral data generated by the interaction of the fluorescent dye and the optical microcavity from the second original WGM spectral data. The optical barcode generation module is used to convert the WGM spectral data processed by the preprocessing module into the optical barcode.

[0020] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the protein aggregation state detection device of the present invention;

[0022] Figure 2 This is a schematic flowchart of the protein aggregation state detection method of the present invention;

[0023] Figure 3 The above is the spectral data of the second original WGM spectral data corresponding to the test solution in a specific embodiment of the present invention after data processing. Curve A represents the spectral data of the original WGM spectral data of the test solution obtained in step S2 after removing the fluorescence background generated by the fluorescent dye, and curve B represents the spectral data of the original WGM spectral data of the test solution obtained in step S2 after removing the fluorescence background generated by the fluorescent dye, smoothing with Savitzky-Golay filter, and normalizing to the [0, 1] interval.

[0024] Figure 4 The following is a partial spectral data of different WGM spectra generated after the interaction of protein monomer solution, protein dimer solution, protein polymer solution, and protein-free pure PBS solution with polystyrene microspheres in a specific embodiment of the present invention. The wavelength unit in the figure is nm.

[0025] Figure 5This is a schematic diagram illustrating the conversion of WGM spectrum into an optical barcode in a specific embodiment of the present invention;

[0026] Figure 6 This is a schematic block diagram of the detection and classification model structure of the present invention;

[0027] Figure 7 This is a schematic diagram of the four-class confusion matrix of the detection and classification model of the present invention on the test set. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention more readily understood by those skilled in the art, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] refer to Figure 1 , Figure 1 A schematic diagram of the protein aggregation state detection device of the present invention is shown. The protein aggregation state detection device of the present invention includes: an optical sensing unit, a stage, an optical excitation unit, an optical acquisition unit, a data processing unit 6, and a display unit 7.

[0030] The optical sensing unit is preferably an optical microsphere doped with a fluorescent dye, which is dispersed and fixed on the stage. The optical microcavity is selected from polystyrene microspheres, silica microspheres, and polymethyl methacrylate microspheres. The fluorescent dye is a coumarin-based, rhodamine-based, or fluorescein-based dye. In this invention, under laser pumping, the fluorescent dye is excited, and its spontaneous emission photons are confined within the optical microcavity, forming stimulated emission or random laser light. This achieves active light emission directly from inside the optical microcavity, eliminating the need for precision components such as optical fibers to introduce the laser into the optical microcavity.

[0031] The optical excitation unit includes an excitation source 1 and a first objective lens 2 connected in sequence by optical paths. The excitation source 1 is a Ti:sapphire femtosecond laser, which loads a pump source with a wavelength of 532nm onto Ti:sapphire to excite a pulsed laser with a wavelength of 800nm.

[0032] When the optical sensing unit is an optical microsphere without fluorescent dye, the optical excitation unit includes an excitation source 1, a first objective lens 2, and an optical coupler (not shown) connected in sequence by optical paths, so that the laser is introduced into the optical microcavity through the optical coupler, wherein the optical coupler can be an optical fiber.

[0033] The optical acquisition unit includes a second objective lens 3, a filter 4, and a spectrometer 5 connected in sequence by optical paths. The spectrometer 5 has a spectral resolution of 0.2 nm and can be a fiber optic spectrometer or a micro spectrometer.

[0034] In this invention, the laser emitted by the excitation source 1 is focused by the first objective lens 2 and can irradiate the optical microcavity or be coupled by an optical coupler to excite its whispering gala mode. The fluorescence signal containing WGM emitted by the optical microcavity is collected by the second objective lens 3, and after the excitation light is filtered out by the filter 4, it is collected and received by the spectrometer 5.

[0035] The data processing unit is connected to the spectrometer and includes a WGM spectrum acquisition module, a preprocessing module, an optical barcode generation module, and a detection and classification module. The WGM spectrum acquisition module is configured to receive raw WGM spectral data acquired by the spectrometer 5. This raw WGM spectral data includes first raw WGM spectral data after the interaction of fluorescent dyes doped in the optical microcavity with the optical microcavity, and second raw WGM spectral data after the interaction of the solution to be detected with the optical microcavity. The preprocessing module is used to sequentially perform fluorescence background removal, smoothing filtering, and normalization processing on the second raw WGM spectral data. The optical barcode generation module is used to convert the WGM spectral data processed by the preprocessing module into optical barcodes. The detection and classification module includes a deep learning classification model, which can be selected from one of convolutional neural networks, recurrent neural networks, Transformers, and fully connected neural networks, for detecting protein aggregation states based on the optical barcodes and outputting the detection results.

[0036] The display unit is configured to display the detection results of the detection and classification module.

[0037] refer to Figure 2 , Figure 2 A schematic flowchart of the protein aggregation state detection method of the present invention is shown. Specifically, the protein aggregation state detection method of the present invention includes the following steps:

[0038] S1. Obtain the first raw WGM spectral data after the fluorescent dye doped in the optical microcavity interacts with the optical microcavity.

[0039] In one specific embodiment, the optical microcavity employs optical microspheres doped with fluorescent dye, which are dispersed and fixed on a stage on a glass substrate. A Ti:sapphire femtosecond laser with a center wavelength of 800 nm, a pulse width of 100 fs, and a repetition frequency of 80 MHz is used as the excitation source 1. The laser emitted by the femtosecond laser is focused by the first objective lens 2 and irradiates an optical microcavity that has not been mixed with the solution to be detected. Coumarin 153 doped in the polystyrene microspheres is excited by the laser and forms stimulated emission in the optical microcavity, generating a whispering gallery mode. The fluorescence signal containing WGM emitted by the optical microcavity is collected by the second objective lens 3, and after the excitation light is filtered out by the filter 4, it is introduced into the fiber receiver of the fiber optic spectrometer 5. After fiber coupling, it enters the fiber optic spectrometer 5, thereby obtaining the WGM spectral data generated after the interaction between the fluorescent dye and the optical microcavity, which is the first original WGM spectral data. The optical microcavity is made of polystyrene microspheres with a diameter of 5 μm. During the preparation of the polystyrene microspheres, 0.1% coumarin 153 fluorescent dye is doped.

[0040] S2. Mix the solution to be tested with the optical microcavity to obtain the second original WGM spectral data after the interaction between the solution to be tested and the optical microcavity.

[0041] In this step, the solution to be tested contains proteins. In one specific embodiment, the solution to be tested is transported to a stage on a glass substrate with dispersed polystyrene microspheres and mixed with the polystyrene microspheres for a certain period of time (e.g., 15 min) to ensure that the protein molecules contained in the solution to be tested are stably attached to the surface of the optical microcavity. Then, in the same manner as obtaining the first original WGM spectral signal corresponding to the fluorescent dye, the laser emitted by the aforementioned Ti:sapphire femtosecond laser is focused by the first objective lens 2 and irradiated into an optical microcavity that has been mixed with the solution to be tested. The fluorescence signal containing the WGM mode emitted by the optical microcavity is collected by the same second objective lens 3, and after the excitation light is filtered out by the filter 4, it is collected by the aforementioned fiber optic spectrometer 5, thereby obtaining the WGM spectral data generated after the interaction between the solution to be tested and the optical microcavity, which is the second original WGM spectral data.

[0042] In this invention, there is no requirement for the order of steps S1 and S2, but step S1 is performed first and then step S2.

[0043] S3. Based on the first original WGM spectral data, obtain the optical barcode corresponding to the second original WGM spectral data.

[0044] In this invention, when using optical microspheres doped with fluorescent dyes, this step involves preprocessing the WGM spectral data generated after the interaction between the test solution and the optical microcavity, and then obtaining an optical barcode from the preprocessed data. Specifically:

[0045] The preprocessing module in the data processing unit first subtracts the baseline, i.e., the first original WGM spectral data obtained in step S1, from the second original WGM spectral data to remove the fluorescence background generated by the fluorescent dye; then, the Savitzky-Golay filter can be used for smoothing filtering, and then the spectral intensity is normalized to the [0, 1] interval.

[0046] refer to Figure 3 ,exist Figure 3 In the specific embodiment shown, curve A represents the original WGM spectral data of the test solution obtained in step S2 after removing the fluorescence background generated by the fluorescent dye, and curve B represents the original WGM spectral data of the test solution obtained in step S2 after removing the fluorescence background generated by the fluorescent dye, smoothing with the Savitzky-Golay filter, and normalizing to the [0, 1] interval. In this specific embodiment, after the test solution in PBS solution is mixed with the optical microcavity, the optical microcavity is placed in the PBS buffer environment. Coumarin 153 doped in the polystyrene microspheres is excited by a femtosecond laser with a center wavelength of 800 nm, generating stimulated emission in the optical microcavity. The emission mode of this emission is sensitive to the change in refractive index of the optical microsphere surface. Therefore, the protein molecules attached to the surface of the optical microspheres cause a change in its emission mode, and finally produce fluorescence in the wavelength range of 425-700 nm, with the fluorescence showing the highest intensity in the range of 500-520 nm. Furthermore, after the fluorescence emitted by coumarin 153 resonates inside the polystyrene microspheres, the light in the wavelength range that meets the resonance condition is amplified, forming the sawtooth peak in the figure, which is the WGM mode peak.

[0047] Figure 4 The figure illustrates WGM spectral data generated after the interaction of protein monomer solution, protein dimer solution, protein polymer solution, and protein-free pure PBS solution with polystyrene microspheres in a specific embodiment. These WGM spectral data are obtained by removing fluorescence background, smoothing, filtering, and normalizing the original WGM spectral data. The protein-free pure PBS solution serves as a control group, and its corresponding WGM spectral data is shown as curve C. Curve D represents the WGM spectral data corresponding to the protein monomer solution, curve E represents the WGM spectral data corresponding to the protein dimer solution, and curve F represents the WGM spectral data corresponding to the protein polymer solution. The figure indicates the wavelengths corresponding to the peaks. Figure 4 As can be seen, the WGM spectral data corresponding to the four different test solutions are different in terms of the distance between two adjacent peaks and the peak intensity.

[0048] Of course, if optical microspheres doped with fluorescent dyes are not used, step S1 is not required. In addition, the aforementioned fluorescence background removal process can be omitted in step S3. Preferably, the second raw WGM spectral data generated after the test solution is mixed with the optical microcavity and optically interacts is smoothed by a Savitzky-Golay filter, and then the spectral intensity is normalized to the [0, 1] interval for preprocessing. After that, an optical barcode is obtained from the preprocessed spectral data.

[0049] Further, refer to Figure 5 For the second original WGM spectral data or the preprocessed WGM spectral data, the optical barcode generation module in the data processing unit uses Python to assign the center wavelength of each WGM mode resonance peak to the barcode position, the full width at half maximum (FWHM) of the resonance peak to the barcode width, and the ratio of the resonance peak to the fluorescence background intensity generated by the fluorescent dye excited under the same laser power to the barcode color depth. This ratio is defined as the extinction ratio. The larger the peak value of the WGM spectrum resonance peak, the higher the extinction ratio, and the darker the barcode stripe color. Finally, each resonance peak is converted into a barcode, ultimately forming an "optical barcode". In this invention, the optical barcode completely retains multi-dimensional information such as the peak position, intensity, peak shape, and relative mode intensity ratio of the WGM spectrum, i.e., the extinction ratio.

[0050] S4. Input the optical barcode into the detection and classification model to obtain the protein aggregate category.

[0051] In this step, the detection and classification module in the data processing unit 6 detects and classifies the protein aggregation state category of the optical barcode, wherein the detection and classification module includes a deep learning classification model.

[0052] refer to Figure 6In one specific embodiment, the deep learning classification model is a four-class CNN (Convolutional Neural Network) model, which sequentially includes an input layer, two convolutional layers, two max-pooling layers, one flattening layer, two fully connected layers, and an output layer. The input layer receives data at a length equal to the number of barcodes, N. The first convolutional layer uses a filter with 32 3×1 convolutional kernels and the ReLU activation function, while the second convolutional layer uses a filter with 64 3×1 convolutional kernels and the ReLU activation function. The first fully connected layer has 128 neurons and the ReLU activation function, while the second fully connected layer has 64 neurons and the ReLU activation function. The output layer has 4 neurons and uses the Softmax activation function to output a four-dimensional probability vector, corresponding to the probabilities of "blank," "monomer," "dimer," and "multimer," respectively. The category corresponding to the maximum value is taken as the final detection result. In this embodiment, two small-sized convolutional kernels (3×1) can effectively extract local features of optical barcode sequences, avoiding excessive feature abstraction and better adapting to the weak and localized feature distribution in biomolecule detection. Simultaneously, the increasing number of convolutional channels, from 32 to 64 layers, gradually enhances feature representation capabilities, improving the model's feature discrimination against different molecular aggregates (monomers, dimers, and multimers) without significantly increasing the number of parameters. The max-pooling layer retains key feature information while reducing dimensionality, lowering computational complexity, improving the model's anti-interference ability, and reducing the impact of noise on classification results. Two fully connected layers achieve nonlinear fusion and high-order abstraction of features, ensuring model fitting ability while avoiding overfitting through a moderate number of neurons. The four-class Softmax output directly corresponds to four classification results: blank (i.e., no protein), monomer, dimer, and multimer, with clear and intuitive output probabilities and stable and reliable classification decisions. Therefore, the detection and classification model in this embodiment has a lightweight structure, a moderate number of parameters, and strong feature extraction targeting. It is highly efficient and robust while ensuring classification accuracy, making it very suitable for molecular aggregation state detection.

[0053] This invention treats the entire WGM spectrum as a high-dimensional "optical barcode." This optical barcode fully preserves multi-dimensional information of the WGM spectrum, including peak position, intensity, peak shape, and relative mode intensity ratio (i.e., extinction ratio), unlike existing technologies that only extract the frequency shift of one or two resonance peaks. This multi-dimensional information can more comprehensively reflect the subtle differences in spectral characteristics caused by the interaction between different protein aggregation states and the microsphere surface. Therefore, when using the optical barcode containing this multi-dimensional information for detection and classification, the accuracy of distinguishing different protein aggregation states can be improved. This invention utilizes high-dimensional spectral information with strong specificity. Furthermore, this invention employs a deep learning model for protein aggregation state detection and classification, which can automatically and efficiently extract deep features from the complex optical barcode and learn the nonlinear discrimination boundary between different protein aggregation states. This overcomes the limitations of traditional methods that rely on manual feature extraction and simple linear judgment, greatly improving the accuracy and robustness of protein aggregation state classification.

[0054] To improve the accuracy of the detection and classification model in detecting protein aggregate categories, it is preferable to train the detection and classification model. Specifically:

[0055] First, solutions of protein monomers, protein dimers, protein polymers, and a protein-free PBS solution at pH 7.4 were prepared, all with a concentration of 0.1 mg / mL. Specifically, β-lactoglobulin was dissolved in PBS buffer at pH 7.4 to obtain a β-lactoglobulin monomer solution, which was stored at low temperature to prevent aggregation. β-lactoglobulin was then dissolved in PBS buffer at pH 7.4, and a solution predominantly composed of dimers was prepared by controlling protein concentration and ionic strength, or by using a low concentration of a chemical cross-linking agent (such as glutaraldehyde), and verified by size exclusion chromatography. Further, the aforementioned β-lactoglobulin solutions (i.e., the β-lactoglobulin monomer solutions or the predominantly dimer protein solutions) were heated at 65°C for a period of time to induce the formation of large aggregates, and dynamic light scattering confirmed a significant increase in their hydrodynamic radius, thus obtaining a polymer solution.

[0056] In one specific embodiment, according to step S1, the raw WGM spectral data of the fluorescent dye is obtained; according to step S2, multiple independent raw WGM spectral data samples corresponding to the protein monomer solution, protein dimer solution, protein polymer solution, and PBS buffer are collected respectively, and corresponding optical barcodes are generated according to step S3.

[0057] The obtained datasets of multiple optical barcodes were randomly divided into training, validation, and test sets in a 7:2:1 ratio. The detection and classification model was trained using the training set with the Adam optimizer and cross-entropy loss. During training, the model performance was monitored using the validation set to prevent overfitting. After training, the detection and classification model was evaluated on an independent test set, referencing... Figure 7 The test set includes four datasets, each containing 120 spectral data points. The confusion matrix's x-axis represents the true label classification of the test set data, and the y-axis represents the label predicted by the detection classification model. Figure 7 It can be seen that the overall accuracy of the detection and classification model can reach 98.12%.

[0058] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0059] This invention treats the entire WGM spectrum as a high-dimensional "optical barcode." This optical barcode fully preserves multi-dimensional information of the WGM spectrum, including peak position, intensity, peak shape, and relative mode intensity ratio (i.e., extinction ratio), unlike existing technologies that only extract the frequency shift of one or two resonance peaks. This multi-dimensional information can more comprehensively reflect the subtle differences in spectral characteristics caused by the interaction between different protein aggregation states and the microsphere surface. Therefore, when using this optical barcode, which embodies this multi-dimensional information, for detection and classification, the accuracy of distinguishing different protein aggregation states can be improved. This invention utilizes high-dimensional spectral information with strong specificity. Furthermore, this invention employs a deep learning model for protein aggregation state detection and classification, thereby automatically and efficiently extracting deep features from the complex optical barcode and learning the nonlinear discrimination boundary between different protein aggregation states. This approach overcomes the limitations of traditional methods that rely on manual feature extraction and simple linear judgment, greatly improving the accuracy and robustness of protein aggregation state classification. Furthermore, based on the inherent high Q-value of the WGM microcavity, this invention is extremely sensitive to minute refractive index perturbations caused by nanoscale biomolecules, which is beneficial for detecting low-concentration early protein aggregates. It can also obtain classification results within seconds, making it rapid and ideal for continuous, real-time monitoring of the dynamic aggregation process of proteins. Simultaneously, this invention requires no chemical modification or fluorescent labeling of the target protein, preserving its native conformation and activity, and simplifying the detection process. This invention can be applied to research on protein conformational diseases, assessment of protein stability in biopharmaceuticals, and detection of early clinical biomarkers.

[0060] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and the present invention also intends to include these modifications and variations.

Claims

1. A method for detecting protein aggregation state, characterized in that, Includes the following steps: Includes an optical microcavity, which is a polystyrene optical microcavity doped with fluorescent dye; In the absence of mixing between the optical microcavity and the solution to be detected, the first raw WGM spectral data after the interaction between the fluorescent dye doped in the optical microcavity and the optical microcavity is obtained. The solution to be tested is mixed with the optical microcavity; The second raw WGM spectral data is obtained after the test solution interacts with the optical microcavity; wherein the test solution is a protein monomer solution, a protein dimer solution, a protein polymer solution, or a protein-free pure PBS solution; Based on the first original WGM spectral data, the second original WGM spectral data is preprocessed, and the optical barcode of the preprocessed spectral data is obtained; wherein, the preprocessing includes: subtracting the first original WGM spectral data from the second original WGM spectral data, and then performing smoothing filtering and normalization processing in sequence; The optical barcode is input into a detection and classification model to obtain the protein aggregate category; wherein, the detection and classification model includes a deep learning classification model, which outputs the corresponding classification results of monomers, dimers, multimers, and blanks; The process of obtaining an optical barcode from the preprocessed spectral data includes: assigning the center wavelength of each resonance peak in the preprocessed spectral data to the position of the barcode, assigning the full width at half maximum (FWHM) of the resonance peak to the width of the barcode, and assigning the ratio of the resonance peak to the fluorescence background intensity generated by the fluorescent dye excited under the same laser power to the color depth of the barcode.

2. The method for detecting protein aggregation state according to claim 1, characterized in that, The deep learning classification model includes convolutional neural networks, recurrent neural networks, Transformers, or fully connected neural networks.

3. The method for detecting protein aggregation state according to claim 2, characterized in that, The deep learning classification model is a four-class convolutional neural network model, which includes an input layer, two convolutional layers, two max pooling layers, one flattening layer, two fully connected layers, and an output layer.

4. The method for detecting protein aggregation state according to claim 1, characterized in that, The fluorescent dye is a coumarin-based, rhodamine-based, or fluorescein-based dye.

5. A detection device based on the protein aggregation state detection method according to any one of claims 1-4, characterized in that, The system includes an optical sensing unit, a stage, an optical excitation unit, an optical acquisition unit, and a data processing unit. The optical sensing unit is a polystyrene optical microcavity doped with fluorescent dye. The optical excitation unit includes an excitation source and a first objective lens connected in sequence via optical paths. The optical acquisition unit includes a second objective lens, a filter, and a spectrometer connected in sequence via optical paths. The data processing unit is connected to the spectrometer and includes a preprocessing module, an optical barcode generation module, and a detection and classification module. The preprocessing module is used to process the first raw WGM spectral data generated after the interaction between the fluorescent dye and the optical microcavity, acquired by the spectrometer, and the data of the solution to be detected acquired by the spectrometer. The second raw WGM spectral data generated after the optical microcavity action is preprocessed; the optical barcode generation module is used to convert the spectral data obtained after the preprocessing module into optical barcodes; the detection and classification module includes a deep learning classification model for detecting protein aggregation state based on the optical barcodes; wherein, when converting the spectral data obtained after the preprocessing module into optical barcodes, the center wavelength of each resonance peak in the spectral data corresponds to the position of the barcode, the full width at half maximum (FWHM) of the resonance peak corresponds to the width of the barcode, and the ratio of the resonance peak to the fluorescence background intensity generated by the fluorescent dye excited under the same laser power corresponds to the color depth of the barcode.

Images