Biosensor-based method for detecting platelet-rich plasma

Abstract

This invention relates to the field of biomedical detection technology, specifically a method for detecting platelet-rich plasma based on biosensors. The method includes: preparing a biosensor interface capable of recognizing platelet surface-specific antigens; applying the sample to be tested to the interface for a specific binding reaction; monitoring signal changes during the reaction process in real time and acquiring raw data; preprocessing the data to obtain a purification response curve; extracting characteristic parameters such as binding rate, equilibrium response value, and dissociation rate constant from the curve; and finally, analyzing the above parameters based on a pre-established quantitative relationship model to simultaneously calculate the platelet concentration and platelet activation index of the sample. This method achieves simultaneous, rapid, and integrated quantitative assessment of platelet quantity and functional activity in platelet-rich plasma.

Description

Technical Field

[0001] This invention relates to the field of biomedical detection technology, and in particular to a method for detecting platelet-rich plasma based on biosensors. Background Technology

[0002] The key to quality assessment of platelet-rich plasma (PRP) formulations lies in the accurate quantification of their core components: platelet concentration and activation status. Currently, the detection of these two parameters is technically separate. Platelet concentration is primarily obtained through blood cell counting, while the assessment of platelet activation levels relies on the detection of surface-specific markers or released active substances. These two detections belong to different technical categories and require independent sample processing and instrument platforms.

[0003] Existing detection methods have shortcomings. Concentration detection only provides quantitative information and does not involve the functional activity of platelets at all. Furthermore, detection of the activation state is mostly an endpoint method, which only obtains the final signal after the reaction is complete and cannot reflect the dynamic process of platelet-target molecule interaction. This results in a static and one-sided assessment of platelet bioactivity. Concentration and activation state measurements cannot be completed simultaneously in a single test, leading to fragmented data, cumbersome procedures, and difficulty in forming a unified and timely judgment of the formulation's bioefficacy.

[0004] Current technology cannot achieve simultaneous, real-time, integrated quantitative analysis of platelet concentration and functional status in a single sample. There is an urgent need to acquire not only platelet quantity information in a single testing event but also to simultaneously capture dynamic characteristic parameters reflecting platelet activity, in order to achieve a more fundamental and comprehensive characterization of platelet-rich plasma quality. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a method for detecting platelet-rich plasma based on biosensors.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for detecting platelet-rich plasma based on biosensors, comprising:

[0007] Prepare a biosensing interface that can recognize platelet surface specific antigens;

[0008] A platelet-rich plasma sample to be tested is obtained, and the platelet-rich plasma sample to be tested is applied to the biosensor interface to induce a specific binding reaction between platelets and the biosensor interface.

[0009] Real-time monitoring of the physicochemical signal changes at the biosensing interface during the specific binding reaction process to obtain raw sensing response data;

[0010] The original sensor response data is preprocessed to eliminate baseline drift and environmental noise interference, resulting in a purified sensor response curve.

[0011] Characteristic response parameters are extracted from the purified sensor response curve, including reaction binding rate, equilibrium response signal value, and dissociation rate constant.

[0012] Based on a pre-established quantitative relationship model between characteristic response parameters and platelet concentration and platelet activation state, the extracted characteristic response parameters are analyzed to calculate the platelet concentration and platelet activation index of the platelet-rich plasma sample to be tested.

[0013] As a further aspect of the present invention, the preparation of the biosensing interface having the function of recognizing platelet surface specific antigens includes:

[0014] A substrate sensor chip is provided, and the surface of the substrate sensor chip is cleaned and activated.

[0015] Recognition molecules that can specifically bind to platelet surface target antigens are immobilized on the surface of the activated substrate sensor chip to form a recognition molecule layer.

[0016] The surface areas of the substrate sensor chip not covered by the recognition molecules are sealed with a sealing agent to reduce non-specific adsorption.

[0017] The biosensing interface is obtained by stabilizing the substrate sensor chip on which the recognition molecule is fixed and after sealing treatment.

[0018] As a further aspect of the present invention, the step of immobilizing the recognition molecule capable of specifically binding to platelet surface target antigens on the surface of the activated substrate sensor chip includes:

[0019] Select specific antibodies or aptamers targeting platelet membrane glycoproteins as the recognition molecules;

[0020] The specific functional groups of the recognition molecule are covalently coupled to the active groups on the surface of the activated substrate sensor chip using a chemical crosslinking agent.

[0021] The conditions for controlling the coupling reaction, including the pH of the reaction solution, ionic strength, reaction temperature, and reaction time, are used to optimize the fixation density and orientation of the recognition molecules on the surface of the substrate sensor chip.

[0022] As a further aspect of the present invention, the real-time monitoring of the physicochemical signal changes at the biosensing interface during the specific binding reaction includes:

[0023] The biosensing interface is signal acquired using one of the following techniques: surface plasmon resonance, quartz crystal microbalance, or electrochemical impedance spectroscopy.

[0024] Throughout the entire process of the specific binding reaction, signal values ​​reflecting changes in interface quality, optical properties, or electrical properties are continuously recorded at set time intervals.

[0025] The signal values ​​constitute the original sensing response data that varies over time.

[0026] As a further aspect of the present invention, the raw sensing response data is preprocessed, including:

[0027] The original sensor response data is smoothed using a digital filtering algorithm to suppress high-frequency random noise;

[0028] A baseline correction algorithm is used to identify and subtract background signals caused by instrument drift or nonspecific binding.

[0029] The data, after smoothing and baseline correction, is normalized to a uniform dimensional range to obtain the purified sensor response curve.

[0030] As a further aspect of the present invention, characteristic response parameters are extracted from the purified sensing response curve, including:

[0031] Identify the starting point of the binding phase, the ending point of the binding phase, the starting point of the dissociation phase, and the signal equilibrium plateau region on the purified sensor response curve;

[0032] Within the binding phase range, the reaction binding rate is calculated by curve fitting;

[0033] In the signal balance plateau region, the average response signal value is calculated as the balance response signal value;

[0034] Within the dissociation phase interval, the dissociation rate constant is calculated through curve fitting;

[0035] The calculation of the reaction binding rate by curve fitting includes:

[0036] Identify data points in the binding phase interval of the purified sensor response curve, the binding phase interval being the stage from the start of the reaction to the signal reaching the plateau period;

[0037] A first-order combined dynamic model is selected as the fitting function, which describes the mathematical relationship between the response signal and time.

[0038] The fitting function is fitted to the data points in the binding phase interval using the nonlinear least squares method, and the fitting parameters are optimized.

[0039] The binding rate constant is extracted from the optimized fitting parameters, and the binding rate constant is the reaction binding rate.

[0040] As a further aspect of the present invention, the process of establishing the pre-established quantitative relationship model between the characteristic response parameters and platelet concentration and platelet activation state includes:

[0041] Prepare a series of standard platelet-rich plasma samples with known platelet concentrations and known activation states;

[0042] For each of the standard platelet-rich plasma samples, the specific binding reaction, real-time monitoring, data preprocessing, and feature response parameter extraction steps are performed to obtain the feature response parameter dataset corresponding to each standard sample.

[0043] By applying multiple regression analysis or machine learning algorithms, a mathematical mapping relationship is established between the feature response parameter dataset and the corresponding known platelet concentration values ​​and known platelet activation state indicators;

[0044] The accuracy and robustness of the mathematical mapping relationship are verified and optimized to form the quantitative relationship model.

[0045] As a further aspect of the present invention, the application of multiple regression analysis or machine learning algorithms to establish a mathematical mapping relationship between the feature response parameter dataset and the corresponding known platelet concentration values ​​and known platelet activation state indicators includes:

[0046] The feature response parameter dataset is used as the input variable, and the known platelet concentration value and the known platelet activation state index are used as the output variables.

[0047] Use one of the following algorithms: partial least squares regression, support vector regression, or artificial neural network to train and learn the input and output variables.

[0048] Determine the model coefficients or network weights that connect the input and output variables to construct the mathematical mapping relationship.

[0049] As a further aspect of the present invention, the analysis of the extracted characteristic response parameters based on a pre-established quantitative relationship model between characteristic response parameters and platelet concentration and platelet activation state includes:

[0050] The extracted feature response parameters are input into the quantitative relationship model;

[0051] Using the algorithm built into the quantitative relationship model, preliminary platelet concentration estimates and platelet activation state estimates are calculated based on the input feature response parameters.

[0052] Based on the collection information or preprocessing conditions of the platelet-rich plasma sample to be tested, necessary corrections are made to the preliminary platelet concentration estimate and platelet activation state estimate;

[0053] Output the corrected platelet concentration and platelet activation index of the platelet-rich plasma sample to be tested.

[0054] As a further aspect of the present invention, after calculating the platelet concentration and platelet activation index of the platelet-rich plasma sample to be tested, the method further includes:

[0055] The platelet concentration value, the platelet activation index, and related characteristic response parameters are stored in a database.

[0056] The quantitative relationship model is periodically updated and retrained using newly measured standard sample data to maintain the model's predictive accuracy.

[0057] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0058] The signal changes throughout the entire process of platelet binding to the sensor interface are monitored in real time, and continuous response curves are recorded. Multiple parameters, including binding rate, equilibrium response value, and dissociation rate constant, are extracted from these kinetic curves. These parameters directly reflect the speed and intensity characteristics of the interaction. The endpoint equilibrium signal only reflects the total amount bound, while the binding and dissociation rates are sensitively dependent on the conformation, accessibility, and distribution of platelet surface antigens. The level of platelet functional activation alters these membrane surface properties, thus affecting the kinetic characteristics of binding and dissociation. Therefore, analyzing the rate parameters can obtain dynamic information reflecting the state of platelet biological activity, achieving a shift from static total quantity measurement to dynamic functional characteristic assessment.

[0059] Multiple extracted feature parameters, including equilibrium signal, binding rate, and dissociation rate, are input into a pre-defined quantitative relationship model for integrated analysis. This model processes these multidimensional input data using an algorithm, simultaneously outputting two independent results: platelet concentration and platelet activation index. The equilibrium response signal is primarily related to the number of platelets in the sample, while the binding and dissociation kinetics are closely related to the degree of platelet activation. Utilizing a multi-parameter model effectively distinguishes and isolates the contributions of concentration and activation factors to the integrated sensor signal. This makes it possible to extract two distinct biological indicators from a single dataset obtained from the same experiment. This method changes the traditional approach that requires two independent detections to obtain concentration and functional information separately, achieving simultaneous, integrated quantitative analysis of concentration and activation status. Attached Figure Description

[0060] Figure 1This is a flowchart of the platelet-rich plasma detection method based on biosensors described in this invention;

[0061] Figure 2 A flowchart for fixing and recognizing molecules;

[0062] Figure 3 A flowchart for real-time monitoring of signal changes;

[0063] Figure 4 To standardize the distribution of feature parameters;

[0064] Figure 5 Violin plots showing the corrected distribution of platelet activation index at different storage times. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0066] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0067] See Figure 1A biosensor interface capable of recognizing platelet-specific antigens was prepared, enabling specific capture of platelets. Platelet-rich plasma samples were acquired and applied to the prepared biosensor interface, allowing platelets in the sample to specifically bind to the recognition molecules on the interface. During this reaction, the physicochemical signal changes of the biosensor interface were monitored in real-time and continuously to obtain raw sensor response data. These raw sensor response data were preprocessed to eliminate baseline drift and environmental noise interference, resulting in purified sensor response curves. Characteristic response parameters reflecting reaction kinetics, including the reaction binding rate, equilibrium response signal value, and dissociation rate constant, were extracted from these purified curves. The extracted characteristic response parameters were input into a pre-established quantitative relationship model, which describes the intrinsic mathematical relationship between the characteristic response parameters and platelet concentration and platelet activation state. Through model analysis and calculation, the platelet concentration and platelet activation index of the platelet-rich plasma sample were obtained.

[0068] See Figure 2In one embodiment of the present invention, a biosensor interface capable of recognizing platelet surface-specific antigens is prepared. The steps include providing a substrate sensor chip, the surface material of which can be gold-plated glass, quartz, or a metal electrode; cleaning and activating the surface of the substrate sensor chip; repeatedly rinsing the surface with piranha solution and deionized water for cleaning; and activating the surface by oxygen plasma bombardment or strong acid etching. The purpose of the activation is to introduce active groups such as hydroxyl, carboxyl, or amino groups. A recognition molecule capable of specifically binding to platelet surface target antigens is immobilized on the activated substrate sensor chip surface to form a recognition molecule layer. The recognition molecule is a monoclonal antibody targeting platelet membrane glycoproteins GPIIb / IIIa. The immobilization process involves modifying the activated substrate sensor chip surface with a mixed solution containing N-hydroxysuccinimide and 1-ethyl-(3-dimethylaminopropyl)carbodiimide hydrochloride to form an active ester intermediate. Subsequently, phosphate buffer containing the monoclonal antibody is added to the modified surface, and incubation is performed under set conditions to allow the amino group of the monoclonal antibody to undergo an amidation reaction with the active ester, thereby achieving covalent coupling. In practice, controlling the coupling reaction conditions involves maintaining the pH of the reaction solution at 7.4, using a sodium chloride solution with an ionic strength of 0.15 mol / L, controlling the reaction temperature at 25°C, and maintaining the reaction time for 2 hours. These conditions are used to optimize the immobilization density and orientation of the monoclonal antibody on the substrate sensor chip surface. The immobilization density is evaluated by measuring the change in response value using surface plasmon resonance (SPR) technology. A blocking agent, a 2% (w / w) bovine serum albumin solution, is used to block the surface areas of the substrate sensor chip that are not covered by the recognized molecules, to reduce the non-specific adsorption of other protein components in the plasma on the substrate sensor chip surface. The substrate sensor chip, immobilized with the recognized molecules and after blocking, undergoes a stabilization treatment. This stabilization treatment involves immersing the prepared substrate sensor chip in a phosphate buffer solution containing preservatives and storing it at 4°C to obtain a biosensing interface that can be directly used for detection.

[0069] In some embodiments, the immobilization method for the recognition molecule is not limited to the covalent coupling described above. Indirect immobilization using a streptavidin-biotin system can be employed. A biotinylated antibody specific to platelet-specific P-selectin (CD62P) serves as the recognition molecule. Streptavidin is pre-immobilized onto the activated substrate sensor chip surface using the aforementioned chemical cross-linking agent. Subsequently, a biotinylated antibody solution is flowed through the substrate sensor chip surface, achieving immobilization of the recognition molecule through the high affinity between streptavidin and biotin. The conditions controlling the coupling reaction include a pH of 7.2 and an ionic strength of 0.01 mol / L phosphate buffer for the streptavidin immobilization stage, with a reaction time of 1 hour. The subsequent antibody binding solution has a pH of 7.4 and an ionic strength of 0.15 mol / L phosphate buffer, with a reaction time of 30 minutes.

[0070] It is understandable that the recognition molecule is not limited to antibodies; it can also be a nucleic acid aptamer targeting specific antigens on the platelet surface, such as a DNA aptamer targeting GPVI. Immobilization is achieved by modifying the ends of the nucleic acid sequences with thiol groups, allowing the thiol-modified DNA aptamer to directly self-assemble with the gold-based sensor chip via gold-sulfur bonds. Controlling the coupling reaction conditions involves using a 1 μmol / L solution of thiolized DNA aptamer, a phosphate buffer solution containing 1 mol / L potassium chloride, a pH of 7.4, a reaction temperature of 37°C, and a reaction time of 16 hours to ensure the formation of a dense, ordered monolayer.

[0071] In practical implementation, the surface coverage density of the immobilized recognition molecules can be theoretically estimated using the following formula:

[0072]

[0073] Where: symbol Represents the surface coverage density of the identifying molecules, measured in nanograms per square centimeter; symbol This represents the frequency change measured by a quartz crystal microbalance, expressed in Hertz; symbol This represents the mass sensitivity constant of a quartz crystal microbalance, expressed in Hertz square centimeters per nanogram; symbol This represents the effective area on the sensor chip where the immobilization reaction occurs, measured in square centimeters. This formula provides a calculation method for quantitatively characterizing the immobilization process.

[0074] It is understandable that the activated substrate sensor chip surface can also be treated with a silane coupling agent containing epoxy groups, making the substrate sensor chip surface rich in epoxy groups. Subsequently, an antibody-containing solution is reacted with the epoxy group surface under alkaline conditions, and the amino groups on the antibody molecules undergo ring-opening reactions to achieve fixation. Blocking can be performed using ethanolamine hydrochloride solution or casein solution, and stabilization treatment also includes encapsulation and storage under an inert gas environment. The cleaning and activation processes of the substrate sensor chip surface can be performed using ultraviolet ozone cleaning, and activation can be achieved by using electrochemical cyclic voltammetry to generate active functional groups on the electrode surface.

[0075] See Figure 3 In one embodiment of the present invention, real-time monitoring of the physicochemical signal changes at the biosensing interface during the specific binding reaction includes acquiring signals from the biosensing interface using surface plasmon resonance (SPR) technology. The SPR instrument is equipped with an optical prism, a laser source, and a photodetector. The laser source emits a beam with a wavelength of 670 nanometers, which is incident on the biosensing interface at a set angle. The photodetector continuously records changes in the reflected light intensity. Throughout the specific binding reaction, the control software of the SPR instrument continuously records the signal value of the reflected light intensity at a set time interval of 0.1 seconds. The reflected light intensity values ​​constitute the raw sensing response data changing over time, and the raw sensing response data is recorded in units of light intensity change. A digital filtering algorithm is applied to smooth the raw sensing response data. The digital filtering algorithm selected is the Savitzky-Golay filter, with a filter window width set to 15 data points and a polynomial order set to 3, to suppress high-frequency random noise in the raw sensing response data. A baseline correction algorithm is employed to identify and subtract background signals caused by instrument drift or nonspecific binding. This algorithm identifies the steady-state signal range before the reaction begins, calculates the average signal value within that range as the baseline, and uniformly subtracts this baseline value from the entire original sensor response data sequence. The data, after being smoothed by a Savitzky-Golay filter and baseline corrected, is normalized to a uniform dimension. Normalization involves dividing all data points by the average signal value of the smoothed reaction-end plateau region, resulting in a purified sensor response curve ranging from 0 to 1.

[0076] In some embodiments, real-time monitoring employs quartz crystal microbalance technology to acquire signals from the biosensing interface. The quartz crystal microbalance sensor operates at a frequency of 10 MHz and is connected to a frequency counter. Throughout the specific binding reaction, the frequency counter continuously records the frequency change values ​​of the quartz crystal microbalance sensor at set time intervals of 1 second. These frequency change values ​​constitute the raw sensing response data over time, recorded in units of frequency change (Hertz). A digital filtering algorithm is applied to smooth the raw sensing response data. A low-pass Butterworth filter is selected, with a cutoff frequency set to 0.1 Hz to suppress high-frequency random noise. A baseline correction algorithm is used to identify and subtract background signals. This algorithm fits a linear trend of frequency change over time for a period before the reaction begins and subtracts this linear trend line from the entire raw sensing response data sequence.

[0077] It is understandable that real-time monitoring can also employ electrochemical impedance spectroscopy (EIS) to acquire signals from the biosensor interface. The electrochemical workstation is configured with a three-electrode system, including a working electrode, a counter electrode, and a reference electrode, with the biosensor interface serving as the working electrode. Throughout the specific binding reaction, the EIS continuously records the phase angle or magnitude changes of the electrochemical impedance at set time intervals and frequencies. The time interval is set to 2 seconds, and the frequency to 10 Hz. The phase angle changes constitute the raw sensor response data over time. A digital filtering algorithm is applied to smooth the raw sensor response data. A moving average filter is selected, and the filter window width is set to 5 data points. A baseline correction algorithm is used to identify and subtract background signals. This baseline correction algorithm extracts the initial value of the phase angle at each measurement time point before the reaction begins, forming a baseline sequence, which is then subtracted from the entire raw sensor response data sequence.

[0078] In practical implementation, when preprocessing the raw sensor response data, the Savitzky-Golay filter used for smoothing can be described by convolution operations. Its core is to perform a weighted average of the data within the window, with the weights determined by the polynomial fitting coefficients. The normalization process after baseline correction can be represented by a formula that outlines the data processing flow:

[0079]

[0080] Where: symbol Represents the normalized signal value, i.e., a point on the purified sensor response curve; symbol Represents a point in time The signal value after being smoothed by a digital filtering algorithm; symbol Represents the baseline signal value calculated using the baseline correction algorithm; symbol This represents the average signal value during the plateau period after smoothing. The formula describes the normalization transformation relationship from raw data to cleaned data.

[0081] It's understandable that digital filtering algorithms can also be median filters. A median filter replaces each data point with the median of all data points within its neighborhood window, with the window width set to 7 points. Baseline correction algorithms can employ exponentially decaying background fitting, suitable for subtracting background signals with slow, nonlinear drift. The normalization dimension range can also be set from 0 to 100, achieved by multiplying by a scaling factor. For electrochemical impedance spectroscopy, the raw sensor response data can simultaneously contain variations in both the real and imaginary parts of the impedance; the preprocessing process requires smoothing, baseline correction, and normalization of the real and imaginary parts, respectively.

[0082] In one embodiment of the present invention, characteristic response parameters are extracted from the purified sensor response curve, including identifying the binding phase start point, binding phase end point, dissociation phase start point, and signal equilibrium plateau region on the purified sensor response curve. The binding phase start point is defined as the fluid switching moment when the platelet-rich plasma sample to be tested is introduced into the biosensor interface. The binding phase end point is defined as the time point when the absolute value of the derivative of the purified sensor response curve first falls below a set threshold and remains below a preset duration. The dissociation phase start point is defined as the fluid switching moment when a platelet-free buffer solution is introduced into the biosensor interface to initiate the dissociation reaction. The signal equilibrium plateau region is a continuous data interval of the purified sensor response curve after the binding phase end point and before the dissociation phase start point, where the signal fluctuation is less than a preset fluctuation range. Within the binding phase interval, the reaction binding rate is calculated through curve fitting. In specific implementation, the calculation of the reaction binding rate through curve fitting includes identifying data points in the binding phase interval of the purified sensor response curve. The data points in the binding phase interval correspond to all sampling time points from the starting point to the ending point of the binding phase and their corresponding normalized response signal values. A first-order binding kinetic model is selected as the fitting function. The fitting function of the first-order binding kinetic model describes the mathematical relationship between the response signal and time. The nonlinear least squares method is used to fit the fitting function of the first-order binding kinetic model to the data points in the binding phase interval. The optimized fitting parameters include the reaction binding rate, the theoretically calculated value of the equilibrium response signal value, and the fitting residual. The binding rate constant is extracted from the optimized fitting parameters, and the binding rate constant is the reaction binding rate. In the signal equilibrium plateau region, the arithmetic mean of the normalized response signal values ​​of all data points in the signal equilibrium plateau region is calculated, and this arithmetic mean is used as the equilibrium response signal value. Within the dissociation phase interval, the dissociation rate constant is calculated through curve fitting. The data points in the dissociation phase interval correspond to all data from the start point of the dissociation phase to the signal drop to near the initial baseline or the end point of the preset duration. The first-order dissociation dynamics model is selected as the fitting function, and the dissociation rate constant is obtained by fitting using the nonlinear least squares method.

[0083] In some embodiments, the dynamic threshold method is used to identify the binding phase endpoint. The sliding window derivative of the purified sensor response curve is calculated. The sliding window derivative is defined as the ratio of the signal difference between adjacent data points to the time difference. When the absolute value of the sliding window derivative of 10 consecutive data points is less than 0.0001 normalized signal units per second, the first point is determined to be the binding phase endpoint. The analysis of variance method is used to identify the signal equilibrium plateau region. The variance of the data points within the signal equilibrium plateau region is calculated. A region with a variance less than the square of 0.00001 normalized signal units is considered a qualified signal equilibrium plateau region. The equilibrium response signal value can also be the median of the data points within the signal equilibrium plateau region.

[0084] It is understandable that the curve fitting process can employ the Levenberg-Marquardt algorithm to achieve nonlinear least squares fitting. Optimizing the fitting parameters involves iteratively adjusting the parameters in the fitting function of the first-order binding kinetic model to minimize the sum of squared residuals between the fitted function curve and the data points in the binding phase interval. The dissociation rate constant can also be calculated using a logarithmic linearization method. After taking the natural logarithm of the relationship between the normalized response signal value and time within the dissociation phase interval, a linear fit is performed, and the absolute value of the slope of the fitted line is the dissociation rate constant.

[0085] In practical implementation, the fitting function form of the first-order combined dynamic model is as follows:

[0086]

[0087] Where: symbol Represents a point in time The fitted response signal value at the location, sign The symbol represents the theoretical equilibrium response signal value obtained through fitting calculation. Represents the observed apparent binding rate constant, with the sign... This represents the base of the natural logarithm. The formula describes the ideal trajectory of the binding phase response signal under assumed conditions. The reaction binding rate is derived from the fitted parameters. Obtained from.

[0088] It is understandable that the characteristic response parameters can also include the reciprocal of the time required for the signal to rise to 63% of the equilibrium response signal value within the binding phase interval. This reciprocal is defined as another parameter characterizing the reaction binding rate. The equilibrium response signal value can be calculated by averaging the remaining data points in the signal equilibrium plateau region after removing the first and last 10% of the data points. The dissociation rate constant can be calculated by using the reciprocal of the time required for the signal to drop from the plateau value to 37% of the plateau value within the dissociation phase interval. The process of extracting the characteristic response parameters can be completely automated by a computer program. The program reads the purified sensor response curve data, automatically identifies key points, divides intervals, and performs curve fitting calculations according to preset algorithm logic, ultimately outputting three values: the reaction binding rate, the equilibrium response signal value, and the dissociation rate constant.

[0089] In one embodiment of the present invention, the process of establishing a pre-established quantitative relationship model between characteristic response parameters and platelet concentration and platelet activation state includes preparing a series of standard platelet-rich plasma samples with known platelet concentrations and known activation states. The known platelet concentrations are calibrated using a complete blood cell counter, and the known activation states are calibrated by flow cytometry detection of the positive expression percentage of P-selectin (CD62P) on the platelet membrane surface. The standard platelet-rich plasma samples cover a platelet concentration gradient of 100,000 to 2,000,000 per microliter and an activation state gradient of 5% to 60% CD62P positivity. For each standard platelet-rich plasma sample, specific binding reaction, real-time monitoring, data preprocessing, and characteristic response parameter extraction steps are performed. The specific binding reaction is performed on a surface plasmon resonance biosensor, and the optical response signals of the binding and dissociation processes are monitored and recorded in real time. Data preprocessing includes smoothing, baseline correction, and normalization. Characteristic response parameter extraction includes calculating the reaction binding rate, equilibrium response signal value, and dissociation rate constant to obtain a characteristic response parameter dataset containing three characteristic response parameters for each standard sample. By applying multiple regression analysis or machine learning algorithms, a mathematical mapping relationship is established between the feature response parameter dataset and the corresponding known platelet concentration and known platelet activation state indices. The multiple regression analysis method uses the feature response parameter dataset as input variables and the known platelet concentration and activation state indices as output variables. Partial least squares regression (PLR) is used to train the input and output variables, projecting them into a new latent variable space. A linear regression model is then established in this space, determining the model coefficients connecting the input and output variables, thus constructing the mathematical mapping relationship. The accuracy and robustness of the mathematical mapping relationship are verified and optimized. Verification employs leave-one-out cross-validation to calculate the coefficient of determination between predicted and known platelet concentrations, and between predicted and known platelet activation state indices. Optimization is achieved by adjusting the number of latent variables retained by the PLS algorithm, ultimately forming a quantitative relationship model.

[0090] In some embodiments, the mathematical mapping relationship established using machine learning algorithms is achieved through Support Vector Regression (SVR). SVR uses the feature response parameter dataset as input variables and the known platelet concentration value as the output variable for separate modeling. Alternatively, it uses the feature response parameter dataset as input variables and the known platelet activation state index as the output variable for separate modeling. SVR employs radial basis functions as kernel functions, and the penalty parameters and kernel function parameters are determined through grid search. During training, the optimal separating hyperplane for the input variables in the high-dimensional feature space is found, and the support vectors and network weights are determined, thereby constructing mathematical mapping relationships for predicting platelet concentration and platelet activation state, respectively. Validation employs five-fold cross-validation, and optimization is achieved by adjusting the parameter combinations of the SVR algorithm.

[0091] It is understandable that machine learning algorithms can also utilize artificial neural network (ANN) algorithms. An ANN constructs a multilayer perceptron model with one hidden layer. The input layer has three neurons corresponding to three feature response parameters, and the output layer has two neurons corresponding to platelet concentration and platelet activation index, respectively. The hidden layer has five neurons. The ANN uses backpropagation and gradient descent to train the input and output variables. The training process determines the network weights and biases connecting the input and hidden layers, and between the hidden and output layers, thus establishing a mathematical mapping relationship. Validation uses an independent validation set method, and optimization is achieved by adjusting the learning rate and number of iterations of the ANN algorithm.

[0092] In practice, the characteristic response parameter dataset and known values ​​of standard platelet-rich plasma samples can be organized into the following table for model training and validation, see Table 1.

[0093] Table 1: Correspondence between characteristic response parameters and known values ​​of standard platelet-rich plasma samples

[0094] Sample No. Reaction Association Rate Equilibrium Response Signal Value Dissociation Rate Constant Known Platelet Concentration Known Platelet Activation Index S01 0.015 125.4 0.0021 1.0 5.2 S02 0.032 251.8 0.0018 2.0 12.8 S03 0.047 380.5 0.0030 3.0 30.5 S04 0.062 498.7 0.0045 4.0 45.1 S05 0.078 625.9 0.0038 5.0 22.3 S06 0.095 752.3 0.0052 6.0 58.7

[0095] The mathematical mapping established by partial least squares regression can be expressed as a system of linear equations, used to calculate predicted values ​​from feature response parameters. For a model with three latent variables, the prediction formula for platelet concentration is:

[0096]

[0097] Where: symbol Represents the predicted platelet concentration value, symbol The intercept term of the model, symbol These represent the three latent variables calculated by the partial least squares regression model. The regression coefficient, sign The scores represent the three latent variable scores obtained after projecting the original characteristic response parameters (reaction binding rate, equilibrium response signal value, and dissociation rate constant) using a partial least squares regression algorithm. The formula illustrates the linear transformation relationship from the latent variable space to the concentration prediction value.

[0098] It is understandable that the process of establishing a quantitative relationship model can also involve standardizing the feature response parameter dataset to ensure that each feature parameter has a mean of zero and a variance of one. The validation process can calculate the root mean square error of prediction as a model performance indicator. The optimization process may include removing parameters in the feature response parameter dataset that are highly collinear with other parameters. For known platelet activation state indicators, in addition to the CD62P positivity rate, platelet mitochondrial membrane potential or calcium ion fluorescence intensity can also be used as calibration values. Standard platelet-rich plasma samples can be prepared by centrifuging whole blood from healthy volunteers and adjusting platelet activation state by adding different concentrations of adenosine diphosphate. The acquisition of the feature response parameter dataset needs to be completed under the same biosensor chip, fluid conditions, and ambient temperature.

[0099] See Figure 4 In the data preprocessing stage of the biosensor-based platelet-rich plasma (PRP) detection method, the three core feature response parameters (reaction binding rate, equilibrium response signal value, and dissociation rate constant) were standardized to eliminate dimensional differences and unify the data distribution. As shown in the figure, all three feature parameters were standardized to a distribution range with a mean of 0 and a variance of 1. This operation was performed to meet the distribution requirements of subsequent quantitative modeling methods such as partial least squares regression, support vector regression, or artificial neural networks. The box plot clearly shows the distribution characteristics of each parameter: the boxes (interquartile range) are symmetrically distributed around the mean of 0, indicating that the standardization process effectively eliminated the bias of the original data. The whiskers (extreme range) cover an interval of approximately ±1.5, while also containing several discrete original data points (red scatter dots). These points represent the measured standardized values ​​of different standard PRP samples on this feature, and their distribution reflects the biological heterogeneity among samples. The similar box shapes of the three feature parameters indicate that they have similar distribution scales after standardization, avoiding model weight skew caused by dimensional differences. This standardized preprocessing is a key preliminary step in constructing a quantitative model of the relationship between feature response parameters and platelet concentration and platelet activation state. It ensures that feature parameters with different physical meanings can be learned and utilized fairly by the model on the same scale, thereby improving the model's prediction accuracy and robustness.

[0100] In one embodiment of the present invention, based on a pre-established quantitative relationship model between characteristic response parameters and platelet concentration and platelet activation state, the extracted characteristic response parameters are analyzed. This includes inputting three characteristic response parameters—reaction binding rate, equilibrium response signal value, and dissociation rate constant—into the quantitative relationship model, which is a mathematical mapping relationship constructed using a support vector regression algorithm. Using the algorithm built into the quantitative relationship model, preliminary estimates of platelet concentration and platelet activation state are calculated based on the three input characteristic response parameters. The kernel function of the support vector regression algorithm is a radial basis function. The calculation process involves mapping the three input characteristic response parameters to a high-dimensional feature space and performing a linear combination based on the trained support vectors, Lagrange multipliers, and bias terms to output the estimated platelet concentration and platelet activation state. Based on the collection information or preprocessing conditions of the platelet-rich plasma sample to be tested, necessary corrections are made to the preliminary platelet concentration and platelet activation state estimates. The collection information includes the time interval from blood collection to the preparation of platelet-rich plasma. For samples with a time interval exceeding four hours, the preliminary platelet activation state estimate needs to be multiplied by a time decay correction factor of 0.95. The corrected platelet concentration and platelet activation index of the platelet-rich plasma sample are output. The platelet activation index is expressed as the equivalent value of the percentage of P-selectin positive expression on the platelet membrane surface. After calculating the platelet concentration and platelet activation index of the platelet-rich plasma sample, the platelet concentration, platelet activation index, and related characteristic response parameters are stored in a relational database. The stored fields include sample number, detection timestamp, reaction binding rate, equilibrium response signal value, dissociation rate constant, platelet concentration, platelet activation index, and collection time information. The quantitative relationship model is periodically updated and retrained using newly measured standard sample data to maintain the model's prediction accuracy. The periodic update is set to be performed once every fifty newly added standard sample data. The newly measured standard sample data is a new feature response parameter dataset obtained from platelet-rich plasma samples with known concentrations and activation states under the same detection conditions. The retraining process involves merging the newly added feature response parameter dataset with the historical training data and re-executing the training process of the support vector regression algorithm to optimize the model parameters.

[0101] In some embodiments, the quantitative relationship model is a mathematical mapping relationship constructed using a partial least squares regression algorithm. After the extracted feature response parameters are input into the quantitative relationship model, the model calculation process involves projecting the three input feature response parameters onto the latent variable space and calculating the output value using pre-stored regression coefficients. Correction of the preliminary estimate considers preprocessing conditions. If the platelet-rich plasma sample to be tested is stored at 4 degrees Celsius for more than six hours before testing, the preliminary platelet concentration estimate needs to be multiplied by a temperature storage correction factor of 1.05. The corrected platelet concentration value and platelet activation index are output, with the platelet activation index expressed as the equivalent value of the platelet mitochondrial membrane potential relative to fluorescence intensity. A time-series database is used to facilitate tracking the evolution of model performance over time. Regular updates are set to be performed quarterly, retaining the most recent 300 standard sample data during training and discarding the oldest historical data to maintain the adaptability of the quantitative relationship model to current testing conditions.

[0102] It is understandable that quantitative relationship models can also be mathematical mappings constructed using artificial neural network algorithms. The process of calculating preliminary estimates is the result of the forward propagation of feature response parameters within the neural network. The correction process can consider sample dilution factors. If the platelet-rich plasma sample to be tested is over-diluted with buffer before use, the preliminary platelet concentration estimate needs to be divided by the dilution factor to obtain the final platelet concentration value. The platelet activation index can be directly output as a dimensionless exponential form. The database can record the version number, training data size, and key performance indicators for each model update. Periodic updates can also be manually triggered by operators. During training, incremental learning algorithms can be used to fine-tune the quantitative relationship model using only newly added standard sample data, based on historical model parameters, to reduce computational resource consumption.

[0103] In practice, the formula for correcting the preliminary estimate of platelet activation status by combining the collection time information is as follows:

[0104]

[0105] Where: symbol Represents the corrected platelet activation index, symbol This represents a preliminary estimate of platelet activation status calculated by a quantitative relationship model, with the symbol... An empirical constant representing the decay of platelet activation state over time, with the symbol... Represents the actual storage time of the platelet-rich plasma sample from collection to testing, symbolized by [symbol]. The symbol represents the average storage time of the standard samples used in establishing the quantitative relationship model. This represents the base of the natural logarithm. The formula describes a method for calculating the exponential decay correction of the activation index estimate based on differences in storage time.

[0106] Understandably, when periodically updating the model, the new standard sample data should cover the common concentration and activation state ranges of the current test samples. After retraining, an independent validation set should be used to evaluate the predictive performance of the updated quantitative relationship model. Evaluation metrics include the correlation coefficient and mean absolute percentage error between predicted and known values. If performance deteriorates, a regression to the previous version of the quantitative relationship model should be implemented. Feature response parameters stored in the database may include intermediate parameters calculated during the extraction process, such as goodness of fit. Correction of the initial estimates can also be based on a multiple linear regression correction model built around multiple factors, with correction factors integrating multiple variables such as temperature, time, and anticoagulant type. The output can simultaneously include platelet concentration values, platelet activation index, and their corresponding confidence intervals.

[0107] See Figure 5In assessing the impact of storage conditions on the platelet activation index, the distribution characteristics of the corrected activation index under different storage times can be visually presented through a violin plot. Specifically, the horizontal axis represents the sample storage time, divided into two groups: "<4 hours" and "≥4 hours". The vertical axis represents the corrected platelet activation index (%), which is expressed as the equivalent value of the percentage of P-selectin positive expression on the platelet membrane surface. Its value has been corrected using a time decay correction factor; that is, for samples stored for more than four hours, the initial estimated activation state needs to be multiplied by a time decay correction factor of 0.95. From a distributional perspective, the activation index distribution in the "<4 hours" group was relatively concentrated, with data mainly concentrated in the 15%–30% range, a median of approximately 22%, and extreme values ​​ranging from 11%–34%. In contrast, the distribution in the "≥4 hours" group exhibited a clear right skew and dispersion, with the data distribution range widening to 10%–41%, while the median remained at approximately 22%. However, the proportion of samples with high values ​​(>30%) increased significantly, suggesting that with prolonged storage time, individual differences in platelet activation index increased, and the proportion of highly activated samples rose. The distributional differences between the two groups can be further explained by statistical characteristics: the interquartile range (IQR) of the "<4 hours" group was approximately 10%, while the IQR of the "≥4 hours" group expanded to 15%, indicating that the dispersion of the activation index significantly increased after storage time ≥4 hours. This change in distributional characteristics is consistent with the biological mechanism of increased spontaneous activation and heterogeneity of platelets during long-term storage due to changes in the microenvironment and metabolic stress. At the data visualization level, the kernel density estimation curve of the violin plot clearly shows the probability density distribution of the two sets of data: the density peak of the "<4 hours" group appears around 22%, showing a unimodal symmetrical distribution; while the "≥4 hours" group forms a main peak at 22%, and a secondary density peak appears around 35%, reflecting that after the storage time is extended, a new distribution pattern of platelet activation state appears.

[0108] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for detecting platelet-rich plasma based on biosensors, characterized in that, The method includes: Prepare a biosensing interface that can recognize platelet surface specific antigens; A platelet-rich plasma sample to be tested is obtained, and the platelet-rich plasma sample to be tested is applied to the biosensor interface to induce a specific binding reaction between platelets and the biosensor interface. Real-time monitoring of the physicochemical signal changes at the biosensing interface during the specific binding reaction process to obtain raw sensing response data; The original sensor response data is preprocessed to eliminate baseline drift and environmental noise interference, resulting in a purified sensor response curve. Characteristic response parameters are extracted from the purified sensor response curve, including reaction binding rate, equilibrium response signal value, and dissociation rate constant. The extraction of characteristic response parameters from the purified sensor response curve includes: Identify the starting point of the binding phase, the ending point of the binding phase, the starting point of the dissociation phase, and the signal equilibrium plateau region on the purified sensor response curve; Within the binding phase range, the reaction binding rate is calculated by curve fitting; In the signal balance plateau region, the average response signal value is calculated as the balance response signal value; Within the dissociation phase interval, the dissociation rate constant is calculated through curve fitting; The calculation of the reaction binding rate by curve fitting includes: Identify data points in the binding phase interval of the purified sensor response curve, the binding phase interval being the stage from the start of the reaction to the signal reaching the plateau period; A first-order combined dynamic model is selected as the fitting function, which describes the mathematical relationship between the response signal and time. The fitting function is fitted to the data points in the binding phase interval using the nonlinear least squares method, and the fitting parameters are optimized. The binding rate constant is extracted from the optimized fitting parameters, and the binding rate constant is the reaction binding rate; Based on a pre-established quantitative relationship model between characteristic response parameters and platelet concentration and platelet activation state, the extracted characteristic response parameters are analyzed to calculate the platelet concentration and platelet activation index of the platelet-rich plasma sample to be tested.

2. The method for detecting platelet-rich plasma based on biosensors as described in claim 1, characterized in that, The preparation of the biosensing interface with recognition function for platelet surface specific antigens includes: A substrate sensor chip is provided, and the surface of the substrate sensor chip is cleaned and activated. Recognition molecules that can specifically bind to platelet surface target antigens are immobilized on the surface of the activated substrate sensor chip to form a recognition molecule layer. The surface areas of the substrate sensor chip not covered by the recognition molecules are sealed with a sealing agent to reduce non-specific adsorption. The biosensing interface is obtained by stabilizing the substrate sensor chip on which the recognition molecule is fixed and after sealing treatment.

3. The method for detecting platelet-rich plasma based on biosensors as described in claim 2, characterized in that, The step of immobilizing recognition molecules capable of specifically binding to platelet surface target antigens onto the activated substrate sensor chip surface includes: Select specific antibodies or aptamers targeting platelet membrane glycoproteins as the recognition molecules; The specific functional groups of the recognition molecule are covalently coupled to the active groups on the surface of the activated substrate sensor chip using a chemical crosslinking agent. The conditions for controlling the coupling reaction, including the pH of the reaction solution, ionic strength, reaction temperature, and reaction time, are used to optimize the fixation density and orientation of the recognition molecules on the surface of the substrate sensor chip.

4. The method for detecting platelet-rich plasma based on biosensors as described in claim 3, characterized in that, The real-time monitoring of changes in the physicochemical signals of the biosensing interface during the specific binding reaction includes: The biosensing interface is signal acquired using one of the following techniques: surface plasmon resonance, quartz crystal microbalance, or electrochemical impedance spectroscopy. Throughout the entire process of the specific binding reaction, signal values ​​reflecting changes in interface quality, optical properties, or electrical properties are continuously recorded at set time intervals. The signal values ​​constitute the original sensing response data that varies over time.

5. The method for detecting platelet-rich plasma based on biosensors as described in claim 4, characterized in that, Preprocessing the raw sensor response data includes: The original sensor response data is smoothed using a digital filtering algorithm to suppress high-frequency random noise; A baseline correction algorithm is used to identify and subtract background signals caused by instrument drift or nonspecific binding. The data, after smoothing and baseline correction, is normalized to a uniform dimensional range to obtain the purified sensor response curve.

6. The method for detecting platelet-rich plasma based on biosensors as described in claim 5, characterized in that, The process of establishing the pre-established quantitative relationship model between the characteristic response parameters and platelet concentration and platelet activation state includes: Prepare a series of standard platelet-rich plasma samples with known platelet concentrations and known activation states; For each of the standard platelet-rich plasma samples, the specific binding reaction, real-time monitoring, data preprocessing, and feature response parameter extraction steps are performed to obtain the feature response parameter dataset corresponding to each standard sample. By applying multiple regression analysis or machine learning algorithms, a mathematical mapping relationship is established between the feature response parameter dataset and the corresponding known platelet concentration values ​​and known platelet activation state indicators; The accuracy and robustness of the mathematical mapping relationship are verified and optimized to form the quantitative relationship model.

7. The method for detecting platelet-rich plasma based on biosensors as described in claim 6, characterized in that, The application of multiple regression analysis or machine learning algorithms establishes a mathematical mapping relationship between the feature response parameter dataset and the corresponding known platelet concentration values ​​and known platelet activation state indicators, including: The feature response parameter dataset is used as the input variable, and the known platelet concentration value and the known platelet activation state index are used as the output variables. Use one of the following algorithms: partial least squares regression, support vector regression, or artificial neural network to train and learn the input and output variables. Determine the model coefficients or network weights that connect the input and output variables to construct the mathematical mapping relationship.

8. The method for detecting platelet-rich plasma based on biosensors as described in claim 7, characterized in that, The extracted characteristic response parameters are analyzed based on a pre-established quantitative relationship model between characteristic response parameters and platelet concentration and platelet activation state, including: The extracted feature response parameters are input into the quantitative relationship model; Using the algorithm built into the quantitative relationship model, preliminary platelet concentration estimates and platelet activation state estimates are calculated based on the input feature response parameters. Based on the collection information or preprocessing conditions of the platelet-rich plasma sample to be tested, the preliminary platelet concentration estimate and platelet activation state estimate are corrected. Output the corrected platelet concentration and platelet activation index of the platelet-rich plasma sample to be tested.

9. The method for detecting platelet-rich plasma based on biosensors as described in claim 8, characterized in that, After calculating the platelet concentration and platelet activation index of the platelet-rich plasma sample to be tested, the method further includes: The platelet concentration value, the platelet activation index, and related characteristic response parameters are stored in a database. The quantitative relationship model is periodically updated and retrained using newly measured standard sample data to maintain the model's predictive accuracy.

Images