A method for optimizing fluorescent probe concentration to improve signal-to-noise ratio

Abstract

This invention provides a method for optimizing the concentration of fluorescent probes to improve the signal-to-noise ratio, belonging to the field of fluorescence experimental technology. The invention involves pre-treating magnetic beads by incubating them with capture antibodies, then pre-mixing tracer antibodies with streptavidin-DNA conjugates to form immune complex magnetic beads. After magnetic separation and washing, these beads are added to a rolling circle amplification reaction solution containing fluorescently labeled probes and phi29 DNA polymerase, and amplified at 30°C in the dark. A comprehensive quality assessment value is then calculated using a signal-to-noise ratio adjustment function, and the amount of fluorescently labeled probes and phi29 DNA polymerase added in subsequent batches is adjusted accordingly. Simultaneously, based on Mie scattering theory, aggregate events are eliminated, and a graph-cut bead classification algorithm is used to output a corrected single-bead positive count. This solves the technical problem of low signal-to-noise ratio in digital detection of rolling circle amplification caused by a mismatch between fluorescent probe concentration and enzyme dosage.

Description

Technical Field

[0001] This invention belongs to the field of fluorescence experimental technology, and more specifically, relates to a method for optimizing the concentration of fluorescent probes to improve the signal-to-noise ratio. Background Technology

[0002] In the field of digital immunoassay, magnetic bead-based rolling circle amplification flow cytometry achieves ultra-high sensitivity quantification of low-abundance analytes by enriching fluorescence signals on individual magnetic beads. This process typically involves functional conjugation of magnetic beads with capture antibodies, formation of immune complexes, streptavidin-DNA conjugate-mediated circular DNA template anchoring, and PHI29 DNA polymerase-driven rolling circle amplification. Finally, the fluorescence signal of each magnetic bead is acquired by a flow cytometer and quantified digitally. However, current digital rolling circle amplification platforms rely on empirically determined amounts of fluorescently labeled probes and PHI29 DNA polymerase, lacking a dynamic optimization mechanism based on real-time signal quality feedback. This results in insufficient differentiation between background signals from non-specific hybridization of fluorescent probes and true positive signals under low-abundance analyte conditions. Furthermore, the combined interference from magnetic bead aggregation events and flow cytometry noise pulses misidentifying positive events compromises the signal-to-noise ratio of the final counting results. In other words, current technologies suffer from low signal-to-noise ratios in digital rolling circle amplification detection. Summary of the Invention

[0003] In view of this, the present invention provides a method for optimizing the concentration of fluorescent probes to improve the signal-to-noise ratio, which can solve the technical problem of low signal-to-noise ratio in the digital detection of rolling circle amplification in the prior art.

[0004] This invention is implemented as follows: This invention provides a method for optimizing the concentration of fluorescent probes to improve the signal-to-noise ratio, comprising the following steps: Incubating magnetic beads and capture antibodies in phosphate buffer, then magnetically separating and washing the magnetic beads with DNA buffer to remove free capture antibodies and non-specific adsorbates, completing the magnetic bead blocking pretreatment, and outputting the blocked pretreated magnetic beads; Premixing tracer antibodies with streptavidin-DNA conjugate, reacting the premix with the blocked pretreated magnetic beads incubated with the antigen, and outputting immune complex magnetic beads; After magnetically separating and washing the immune complex magnetic beads, adding them to rolling circle amplification reaction solution, wherein the amount of fluorescently labeled probe added to the rolling circle amplification reaction solution is... The amount of DNA polymerase added was set proportionally. Immediately after the reaction, DNA buffer was added to terminate the rolling circle amplification reaction, and fluorescently labeled magnetic beads were output. Flow cytometry data from the fluorescently labeled magnetic beads were input into a fluorescence scattering feature extraction and counting optimization model. This model fitted the fluorescence intensity distribution of each magnetic bead event using an expectation-maximization algorithm and automatically determined the mixed distribution components using the Bayesian information criterion, outputting the classification label and signal-to-noise ratio (SNR) estimate for each magnetic bead event. Based on the SNR estimate and the detection limit estimate, a comprehensive quality assessment value was calculated using a SNR adjustment function. When the comprehensive quality assessment value fell into different intervals, the amount of fluorescently labeled probe added in subsequent batches of experiments was adjusted accordingly. The amount of DNA polymerase added was gradient-regulated, and the adjusted amount of fluorescently labeled probe was output. DNA polymerase dosage; Based on Mie scattering theory, the ratio of forward scattering intensity to side scattering intensity in the flow cytometry data is calculated event-by-event. Mahalanobis distance is used to measure the deviation of each magnetic bead event from the theoretical prediction value for a single bead. Values ​​exceeding the Mahalanobis distance are... The magnetic bead events are marked as cluster events and removed from the count. The remaining magnetic bead events are then classified using a graph cut bead classification algorithm to output the corrected single bead positive count results.

[0005] The step of incubating the magnetic beads and the capture antibody in phosphate buffer specifically involves: the pH of the phosphate buffer being 7.0–8.0, and the DNA buffer containing 0.2% BSA, 10 mM EDTA, and 0.04% HCl. The magnetic separation and cleaning process is repeated 3 to 9 times.

[0006] Specifically, the step of premixing the tracer antibody with the streptavidin-DNA conjugate involves: conducting the premixing reaction at 37°C and 1000–1500 rpm for 20–60 min, with the tracer antibody concentration at 5–10 nM / ml and the streptavidin-DNA conjugate concentration at 10–20 nM / ml.

[0007] Specifically, the step of reacting the premixed solution with the blocking pretreated magnetic beads after antigen incubation involves reacting the premixed solution with the blocking pretreated magnetic beads at room temperature for 20–60 min, allowing the tracer antibody to form a ternary immune complex by recognizing the antigen epitope already bound to the capture antibody.

[0008] Specifically, the step of adding the rolling circle amplification reaction solution involves adding 0.5–1.0 μl of fluorescently labeled probe to every 650 μl of the reaction system. Add 5–10 μl of DNA polymerase and react at 30°C and 400–800 rpm in the dark for 20–60 min.

[0009] The preparation of the streptavidin-DNA conjugate specifically involves: replacing the streptavidin solution with phosphate buffer, and then mixing it with... (The sentence is incomplete and requires more context to translate accurately.) The ester was reacted at room temperature for 20–60 min, and free esters were removed by ultrafiltration through a 30 kD tube. The esterification and solution exchange yielded streptavidin-DBCO conjugates, which were then coupled with circular DNA templates overnight at 2–8°C.

[0010] The molar ratio of streptavidin-DBCO conjugate to circular DNA template was 1:1 to 5. The conjugate product was aliquoted and stored at -80℃ to obtain streptavidin-DNA conjugate.

[0011] The preparation of the circular DNA template is specifically as follows: the primer and the linear DNA template are mixed at a volume ratio of 1:1 to 2 and incubated at 95°C for 1 to 4 minutes. The mixture is then placed in hot water at 95°C and allowed to cool naturally to room temperature. A ligase solution is added and the mixture is allowed to stand at room temperature in the dark for 1 to 3 hours. Finally, the mixture is passed through a 5kD desalting column to change the medium and obtain the circular DNA template.

[0012] The signal-to-noise ratio adjustment function is defined as follows: Where SNR is the estimated signal-to-noise ratio for the current batch. The baseline signal-to-noise ratio is given, and the LOD is the estimated detection limit for the current batch. As the baseline detection limit, and The weighting coefficients are and satisfy the following conditions: Q is the dimensionless comprehensive quality assessment value.

[0013] Specifically, the gradient adjustment for the comprehensive quality assessment value falling into different ranges involves: when Q < 0.6, reducing the amount of fluorescently labeled probe added in subsequent batches by 20% and adjusting the number of magnetic separation and cleaning cycles to 8; when... At that time, The amount of DNA polymerase added was reduced by 10%; when When Q > 1.15, the amount of fluorescently labeled probe added is increased by 15%.

[0014] The underlying layer of the fluorescence scattering feature extraction and counting optimization model is a graph neural network module, which treats magnetic bead events within adjacent time windows during the flow cytometry detection process as graph nodes, and constructs a similarity graph using the Gaussian function of the fluorescence intensity difference between any two nodes as the edge weight.

[0015] In the graph neural network module, the initial feature vector of each node is composed of fluorescence intensity, forward scattering intensity, side scattering intensity and timestamp. The feature vectors of the neighboring nodes of each node are aggregated through a three-layer message passing mechanism. The aggregation method adopts weighted summation followed by linear transformation and ReLU activation function processing to output the graph-enhanced feature vector of each magnetic bead event.

[0016] The middle layer of the fluorescence scattering feature extraction and counting optimization model is a physical information neural network module. The input is a graph enhancement feature vector, the network structure is a 4-layer fully connected network, and the output is a physical constraint feature vector.

[0017] The loss function of the physical information neural network module consists of two parts: data fitting loss and physical residual loss. The data fitting loss is the mean square error between the network output value and the training dataset label, and the physical residual loss is obtained by weighted summation of the residuals of the probe diffusion kinetic equation and the second-order binding rate equation.

[0018] The probe diffusion dynamics equation is established based on Fick's second law, the second-order binding rate equation is established based on the second-order binding dynamics between the probe and the repeating sequence, and the weight coefficients between the data fitting loss and the physical residual loss are automatically adjusted by the network during training through an adaptive weight mechanism.

[0019] The top layer of the fluorescence scattering feature extraction and counting optimization model is a multi-task learning network. The input is a physical constraint feature vector. After extracting common features through two shared fully connected layers, it branches into a signal-to-noise ratio prediction task head and a detection limit prediction task head, each consisting of two fully connected layers.

[0020] When the signal-to-noise ratio prediction task head and the detection limit prediction task head generate gradient conflicts during backpropagation, the gradient projection method is used to project the conflicting gradients to a direction that does not negatively interfere with the other task, and the Pareto front optimization strategy is used to find a non-dominated solution set between the signal-to-noise ratio estimate and the detection limit estimate.

[0021] The training dataset for the fluorescence scattering feature extraction and counting optimization model was established by performing a complete rolling circle amplification detection process on analyte samples with multiple concentration gradients, recording the fluorescence intensity, forward scattering intensity, side scattering intensity, and timestamp for each magnetic bead event, and using manually reviewed single-bead positive count results as labels.

[0022] The training dataset also records the corresponding amount of fluorescently labeled probes added. The parameters for DNA polymerase dosage, rolling circle amplification reaction time, and DNA buffer composition are as follows:

[0023] The training of the fluorescence scattering feature extraction and counting optimization model involves the following steps: First, the graph neural network module is pre-trained, with the accuracy of cluster event recognition as the loss function; then, the physical residual loss and data fitting loss of the physical information neural network module are weighted and summed for joint training; finally, the top layer of the multi-task learning network is jointly trained.

[0024] Among them, the joint training of the top layer of the multi-task learning network adopts the Pareto front optimization strategy and uses the gradient projection method to handle gradient conflicts between tasks until the validation set loss of both the signal-to-noise ratio prediction task head and the detection limit prediction task head converges.

[0025] Specifically, the graph cut bead classification algorithm involves: modeling each magnetic bead event as a graph node, constructing a similarity graph using the Gaussian function of the fluorescence intensity difference between any two nodes as the edge weight, transforming the magnetic bead positive / negative classification problem into a minimum cut problem, and using the Ford-Fulkerson algorithm to solve for the maximum flow to obtain the minimum cut.

[0026] The energy function of the graph cut bead classification algorithm consists of a data term and a smoothing term. The data term is the degree of agreement between the fluorescence intensity of each node and the prior positive and negative distributions, and the smoothing term is the consistency of classification of adjacent temporal magnetic bead events. The temporal local continuity constraint of the streaming detection data is used to suppress isolated noise points from being incorrectly classified as positive events.

[0027] Specifically, the Mie scattering theory refers to: a theoretical constant value used to derive the ratio of forward scattering intensity to lateral scattering intensity of monodisperse magnetic beads at a given particle size, and to establish theoretical isoscattering lines in a two-dimensional scatter plot of forward scattering intensity versus lateral scattering intensity.

[0028] Specifically, the Mahalanobis distance refers to the generalized distance by which each magnetic bead event deviates from the theoretical isoscattering line, considering the covariance structure of forward and side scattering intensities. The Mahalanobis distance threshold is set as follows: Events exceeding this threshold are marked as aggregate events and removed from the count.

[0029] Among them, the The ester is a heterobifunctional crosslinking agent containing a dibenzocyclooctylene group and a polyethylene glycol spacer arm. The NHS ester terminus reacts with the amino group of the protein, and the DBCO terminus undergoes copper-free click chemical coupling with azide-modified DNA.

[0030] This invention constructs a fluorescence scattering feature extraction and counting optimization model, combining a graph neural network module, a physical information neural network module, and a multi-task learning network in a three-layer architecture. It outputs a signal-to-noise ratio (SNR) estimate and a detection limit estimate for each batch of detection data. Then, a comprehensive quality assessment value is calculated using an SNR adjustment function. Based on the range of the comprehensive quality assessment value, the amount of fluorescently labeled probes and phi29 DNA polymerase added in subsequent batches is gradient-adjusted. Simultaneously, a Mie scattering theory-based agglomerate identification algorithm is used to eliminate agglomerate events, and a graph bead classification algorithm is used to suppress noise impulse misjudgments. Thus, background interference is suppressed and counting sources are purified at both the data acquisition and parameter decision levels, enabling the SNR assessment results to directly drive experimental parameters towards the optimal range. In summary, this invention solves the technical problem of low SNR in digital detection of rolling circle amplification mentioned in the background art. Attached Figure Description

[0031] Figure 1 This is a flowchart of the method of the present invention.

[0032] Figure 2 This is a two-dimensional scatter plot of forward scattering intensity and lateral scattering intensity based on Mie scattering theory, and an image showing the results of aggregate identification.

[0033] Figure 3 Comparison of single bead positive counts before and after correction for IL-6 samples at various concentration gradients.

[0034] Figure 4 The graph shows the fitting results of the multi-component mixed Gaussian distribution of fluorescence intensity.

[0035] Figure 5 This is a grouped histogram of the pulse height signal in Example 3.

[0036] Figure 6 This is a line graph showing the signal-to-noise ratio in Example 3.

[0037] Figure 7 This is a line graph of the median pulse height in Example 4.

[0038] Figure 8 This is a grouped histogram showing the proportion of pulses above the threshold in Example 4. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0040] like Figure 1 The diagram shown is a flowchart of a method for optimizing fluorescent probe concentration to improve the signal-to-noise ratio provided by the present invention. This method includes the following steps:

[0041] S01. After incubating the magnetic beads and capture antibodies in phosphate buffer at pH 70-8.0, the magnetic beads are magnetically separated and washed 3-9 times with DNA buffer containing 0.2% BSA, 10mM EDTA, and 0.04% NaN3 to remove free capture antibodies and non-specific adsorbents, thus completing the magnetic bead blocking pretreatment and outputting the blocked pretreated magnetic beads.

[0042] S02. The tracer antibody and streptavidin-DNA conjugate are premixed and reacted at 37°C and 1000-1500 rpm for 20-60 min. The concentration of the tracer antibody is 5-10 nM / ml and the concentration of the streptavidin-DNA conjugate is 10-20 nM / ml. The premixed solution is reacted with the blocking pretreated magnetic beads incubated with the antigen at room temperature for 20-60 min to output the immune complex magnetic beads.

[0043] S03. After the immune complex magnetic beads are magnetically separated and washed 3-9 times, they are added to the rolling circle amplification reaction solution. The amount of fluorescently labeled probe added to the rolling circle amplification reaction solution is set at a ratio of 0.5-1.0 μl per 650 μl of reaction system, and the amount of phi29 DNA polymerase added is set at a ratio of 5-10 μl per 650 μl of reaction system. After reacting at 30℃ and 400-800 rpm in the dark for 20-60 min, DNA buffer is immediately added to terminate the rolling circle amplification reaction, and the fluorescently labeled magnetic beads are output.

[0044] S04. Input the flow cytometry data of fluorescently labeled magnetic beads into the fluorescence scattering feature extraction and counting optimization model. The fluorescence scattering feature extraction and counting optimization model fits the fluorescence intensity distribution of each magnetic bead event using the expectation-maximization algorithm, and automatically determines the mixed distribution component number through the Bayesian information criterion, outputting the classification label and signal-to-noise ratio estimate of each magnetic bead event.

[0045] S05. Based on the signal-to-noise ratio estimate and detection limit estimate output from S04, calculate the comprehensive quality assessment value according to the signal-to-noise ratio adjustment function. When the comprehensive quality assessment value belongs to different intervals, perform gradient adjustment on the amount of fluorescently labeled probe and phi29 DNA polymerase added in subsequent batch experiments, and output the adjusted amount of fluorescently labeled probe and phi29 DNA polymerase added.

[0046] S06. Based on the Mie scattering theory, the ratio of forward scattering intensity to side scattering intensity in the flow cytometry detection data is calculated on an event-by-event basis. The Mahalanobis distance is used to measure the degree to which each magnetic bead event deviates from the theoretical prediction value of a single bead. Magnetic bead events with a Mahalanobis distance exceeding 3σ are marked as cluster events and removed from the count. For the remaining magnetic bead events, the graph cut bead classification algorithm is used to output the corrected single bead positive count result.

[0047] The preparation steps of the streptavidin-DNA conjugate are as follows: after changing the streptavidin solution to pH 7.0-8.0 phosphate buffer, mix it with DBCO at a molar ratio of 1:10-40. -NHS esters were reacted at room temperature for 20-60 min, and free DBCO- was removed by a 30kD ultrafiltration tube. -NHS ester was replaced with phosphate buffer containing 1 mM EDTA to obtain streptavidin-DBCO conjugate. The streptavidin-DBCO conjugate was then coupled with a circular DNA template overnight at 2-8°C. The molar ratio of the streptavidin-DBCO conjugate to the circular DNA template was 1:1-5. The conjugated product was aliquoted and stored at -80°C to obtain streptavidin-DNA conjugate.

[0048] The preparation steps of the circular DNA template are as follows: the primer and the linear DNA template are mixed at a volume ratio of 1:1-2 and incubated at 95℃ for 1-4 min. The mixture is then placed in hot water at 95℃ and allowed to cool naturally to room temperature. After cooling, ligase solution is added and the mixture is allowed to stand at room temperature in the dark for 1-3 h. The mixture is then desalted using a 5kD column and the solution is changed to phosphate buffer containing 1mM EDTA to obtain the circular DNA template.

[0049] The graph cut bead classification algorithm models each magnetic bead event as a graph node, constructs a similarity graph using the Gaussian function of the fluorescence intensity difference between any two nodes as the edge weight, and transforms the magnetic bead positive / negative classification problem into a minimum cut problem. The energy function consists of a data term and a smoothing term. The data term is the degree of agreement between the fluorescence intensity of each node and the prior positive / negative distribution, and the smoothing term is the consistency of classification of adjacent temporal magnetic bead events. The Ford-Fulkerson algorithm is used to solve the maximum flow to obtain the minimum cut, and the temporal local continuity constraint of the streaming detection data is used to suppress isolated noise points from being incorrectly classified as positive events.

[0050] The Mie scattering theory is used to derive the theoretical constant value of the ratio of forward scattering intensity to lateral scattering intensity of monodisperse magnetic beads under a given particle size. The theoretical isoscattering line is established in the two-dimensional scatter plot of forward scattering intensity-lateral scattering intensity. The Mahalanobis distance is used to measure the generalized distance of each magnetic bead event from the theoretical isoscattering line under the condition of considering the covariance structure of forward scattering intensity and lateral scattering intensity.

[0051] The specific structure of the fluorescence scattering feature extraction and counting optimization model is as follows: The bottom layer is a graph neural network module, which treats magnetic bead events within adjacent time windows during the flow cytometry detection process as graph nodes. A similarity graph is constructed using a Gaussian function of the fluorescence intensity difference between any two nodes as the edge weight. The initial feature vector of each node is composed of fluorescence intensity, forward scattering intensity, side scattering intensity, and timestamp. The feature vectors of the neighboring nodes of each node are aggregated through a 3-layer message passing mechanism. The aggregation method adopts weighted summation followed by linear transformation and ReLU activation function processing to output the graph-enhanced feature vector of each magnetic bead event, which is used to identify abnormally bright nodes caused by aggregates. The middle layer is a physical information neural network module, which takes the graph-enhanced feature vector as input. The network structure is a 4-layer fully connected network. The loss function consists of two parts: data fitting loss and physical residual loss. The data fitting loss is the mean square error between the network output value and the training dataset label. The physical residual loss is obtained by weighted summation of the residual of the probe diffusion kinetic equation and the residual of the second-order binding rate equation. The probe diffusion dynamics equation is established based on Fick's second law, and the second-order binding rate equation is established based on the second-order binding dynamics between the probe and the repeating sequence. The weight coefficients between the data fitting loss and the physical residual loss are automatically adjusted by the network during training through an adaptive weight mechanism, outputting a physical constraint feature vector. The top layer is a multi-task learning network, with the physical constraint feature vector as input. After extracting common features through two shared fully connected layers, it branches into a signal-to-noise ratio (SNR) prediction task head and a detection limit prediction task head. Each of the SNR prediction task head and the detection limit prediction task head consists of two fully connected layers, outputting SNR estimates and detection limit estimates, respectively. When gradient conflicts occur between the two task heads during backpropagation, a gradient projection method is used to project the conflicting gradients to a direction that does not negatively interfere with the other task. A Pareto front optimization strategy is used to find a non-dominated solution set between the SNR estimate and the detection limit estimate, providing two input parameters, the SNR estimate and the detection limit estimate, for the subsequent calculation of the comprehensive quality assessment value of the SNR adjustment function.

[0052] The steps for establishing the training dataset of the fluorescence scattering feature extraction and counting optimization model specifically include: performing a complete rolling circle amplification detection process on analyte samples with multiple concentration gradients, recording the fluorescence intensity, forward scattering intensity, side scattering intensity and timestamp of each magnetic bead event, and simultaneously recording the corresponding fluorescently labeled probe addition amount, phi29 DNA polymerase addition amount, rolling circle amplification reaction time and DNA buffer composition parameters, using manually reviewed single bead positive count results as labels to construct the training dataset;

[0053] The specific steps for training the fluorescence scattering feature extraction and counting optimization model include: first, pre-training the graph neural network module with the cluster event recognition accuracy as the loss function; then, jointly training the physical residual loss and data fitting loss of the physical information neural network module by weighted summation; finally, jointly training the top layer of the multi-task learning network, using the Pareto front optimization strategy, and handling gradient conflicts between tasks with the gradient projection method, until the validation set losses of the signal-to-noise ratio prediction task head and the detection limit prediction task head converge.

[0054] The technical effects of the fluorescence scattering feature extraction and counting optimization model are as follows: The graph neural network module models the temporal correlation between magnetic bead events through graph structure, distinguishing between real single bead signals and false positive signals of aggregates from the spatial topology level, thus making up for the deficiency of traditional thresholding methods in utilizing local contextual information; The physical information neural network module embeds the physical laws of the probe diffusion kinetics equation and the second-order binding rate equation into the model constraints, enabling the model to maintain physically consistent predictive ability in sparse experimental data regions, avoiding the overfitting problem of purely data-driven models; The multi-task learning network optimizes the simultaneous balance between the signal-to-noise ratio estimate and the detection limit estimate through Pareto front optimization, providing quantitative decision support for adjusting experimental parameters, and improving the overall accuracy and stability of low-abundance analyte detection;

[0055] The technical effects of the graph cut bead classification algorithm and the Mie scattering theory agglomeration identification algorithm are as follows: The graph cut bead classification algorithm, by introducing a smoothing constraint of temporally adjacent magnetic bead events, distinguishes isolated instrument noise pulses from real positive bead signals at the graph structure level, overcoming the neglect of local context information by the fixed threshold method, and effectively suppressing false positive counts under low abundance analyte conditions; The Mie scattering theory agglomeration identification algorithm, based on the principles of physical optics, utilizes the high sensitivity dependence of forward scattering intensity and side scattering intensity on magnetic bead size to identify and exclude agglomeration events from existing scattering channel data without adding additional experimental steps, fundamentally improving the purity and reliability of single bead digital counting;

[0056] The specific definition of the signal-to-noise ratio adjustment function is as follows:

[0057] ;

[0058] in, This is the estimated signal-to-noise ratio for the current batch. As the reference signal-to-noise ratio, This is the estimated detection limit for the current batch. As the baseline detection limit, and The weighting coefficients are and satisfy the following conditions: , This is a dimensionless comprehensive quality assessment value; when When this happens, the amount of fluorescently labeled probe added in subsequent batches will be reduced by 20% from the current value, and the number of magnetic separation and cleaning cycles will be adjusted to 8; when At the same time, while keeping the amount of fluorescently labeled probe added constant, the amount of phi29 DNA polymerase added in subsequent batches will be reduced by 10% from the current value; when At that time, the amount of fluorescently labeled probe and the amount of phi29 DNA polymerase added remained constant; when At that time, the amount of fluorescently labeled probe added in subsequent batches will be increased by 15% based on the current value;

[0059] The expectation-maximization algorithm is used to fit the multi-component mixed distribution of the fluorescence intensity of magnetic beads. The Bayesian information criterion automatically selects the optimal component by applying a component penalty term to the log-likelihood value of different component components, thereby avoiding the classification bias introduced by the double Gaussian hypothesis due to the overlap of the two peak distributions under low concentration analyte conditions.

[0060] Among them, the DBCO- -NHS ester is a heterobifunctional crosslinking agent containing a dibenzocyclooctylene group and a polyethylene glycol spacer arm. The NHS ester end reacts with the amino group of the protein, and the DBCO end undergoes copper-free click chemical coupling with azide-modified DNA.

[0061] The phi29 DNA polymerase has strand displacement activity and continuous synthesis ability. It continuously synthesizes on a circular DNA template to produce long single-stranded DNA multiply containing a large number of repetitive sequences. After the long single-stranded DNA multiply hybridizes with a fluorescently labeled probe, it enriches the fluorescent signal on a single magnetic bead.

[0062] The Pareto front optimization strategy refers to finding a non-dominated solution set between the two objectives of signal-to-noise ratio estimation and detection limit estimation, such that no single solution is superior to any solution in the non-dominated solution set for both objectives.

[0063] The gradient projection method refers to projecting the conflicting gradients to a direction that does not produce negative interference when there is a conflict between the gradient directions of the signal-to-noise ratio prediction task head and the detection limit prediction task head during the training of a multi-task learning network, so as to ensure the synchronous optimization of the two tasks.

[0064] The specific implementation of step S01 is as follows: Magnetic beads and capture antibodies are incubated thoroughly in phosphate buffer (pH 7.0-8.0) to allow the capture antibodies to anchor onto the surface of the magnetic beads via electrostatic adsorption or covalent bonding, forming functionalized magnetic beads. After incubation, a solution containing 0.2% BSA, 10mM EDTA, and 0.04%... The magnetic beads were magnetically separated and washed 3-9 times with DNA buffer. Each wash involved using an external magnetic field to adsorb the beads onto the tube wall, discarding the supernatant, and resuspending them in fresh buffer. This process was repeated. BSA acts as a blocking agent, occupying non-specific binding sites on the surface of the magnetic beads to prevent non-specific adsorption of proteins or nucleic acids in subsequent steps. EDTA inhibits potential nuclease activity by chelating divalent metal ions in the solution, protecting the DNA template from degradation. As a preservative, it prevents microbial contamination. The purpose of 3-9 magnetic separation and washing cycles is to thoroughly remove free capture antibodies and non-specific adsorbates, ensuring the specificity of subsequent immune responses. After completing the blocking pretreatment, the blocking pretreated magnetic beads are output.

[0065] The specific implementation method of step S02 is as follows: the tracer antibody and streptavidin-DNA conjugate are premixed at 37°C and 1000-1500 rpm for 20-60 min at a ratio of 5-10 nM / ml for the tracer antibody and 10-20 nM / ml for the streptavidin-DNA conjugate. This allows for high-affinity non-covalent binding between the biotin on the tracer antibody and the streptavidin in the streptavidin-DNA conjugate. The molar ratio is set to 1:1-4 to ensure that each tracer antibody molecule has sufficient streptavidin-DNA conjugate for pairing, while avoiding competitive interference caused by excessive free streptavidin-DNA conjugate. After premixing, the premixed solution is added to the blocked pretreated magnetic bead system after antigen incubation. The reaction is carried out at room temperature for 20-60 minutes, allowing the tracer antibody to form a ternary immune complex by recognizing the antigen epitope already bound to the capture antibody. The antigen is simultaneously held by the capture antibody and the tracer antibody, thereby anchoring the circular DNA template carried by the streptavidin-DNA conjugate to the surface of the magnetic beads and outputting the immune complex magnetic beads.

[0066] The specific implementation of step S03 is as follows: After the immune complex magnetic beads are magnetically separated and washed 3-9 times, unbound tracer antibody-streptavidin-DNA conjugate complexes and other impurities are removed to reduce background fluorescence interference. Then, a rolling circle amplification reaction solution is prepared. In a 650 μl reaction system, the amount of fluorescently labeled probe added is 0.5-1.0 μl, the amount of phi29 DNA polymerase added is 5-10 μl, and the remaining components include... The reaction mixture consists of 10× reaction buffer and 10 mM dNTPs. The phi29 DNA polymerase uses a circular DNA template anchored on the surface of magnetic beads as a substrate. Leveraging its strand displacement activity and continuous synthesis capability, it extends along the circular template at 30°C, producing long single-stranded DNA polymers containing hundreds to thousands of repetitive sequences. These polymers serve as the hybridization targets for fluorescently labeled probes. The fluorescently labeled probes hybridize one-to-one with the repetitive sequences on the polymers, resulting in a high-density accumulation of fluorescent molecules on individual magnetic beads, amplifying the fluorescence signal. A shaking speed of 400-800 rpm and protection from light are used to maintain uniform suspension of the magnetic beads and prevent photobleaching of the fluorophores, respectively. The reaction is terminated immediately after 20-60 minutes by adding DNA buffer to prevent over-amplification and the resulting increase in non-specific signals, thus producing fluorescently labeled magnetic beads.

[0067] The specific implementation of step S04 is as follows: Fluorescently labeled magnetic beads are subjected to flow cytometry detection, and the fluorescence intensity, forward scattering intensity, side scattering intensity, and timestamp of each magnetic bead event are collected. The above multidimensional data are input into the fluorescence scattering feature extraction and counting optimization model. The bottom-level graph neural network module of this model uses magnetic bead events within adjacent time windows as graph nodes, constructs a similarity graph, and aggregates neighborhood features through a 3-layer message passing mechanism to output a graph-enhanced feature vector. The middle-level physical information neural network module uses the probe diffusion kinetics equation and the second-order binding rate equation described by Fick's second law as physical residual loss, and jointly constrains a 4-layer fully connected network with a data fitting loss in the form of mean square error, outputting a physical constraint feature vector. The top-level multi-task learning network branches into a signal-to-noise ratio prediction task head and a detection limit prediction task head after passing through 2 shared fully connected layers. Each task head consists of 2 fully connected layers and outputs the signal-to-noise ratio estimate and the detection limit estimate, respectively. Meanwhile, the expectation-maximization algorithm performs multi-component mixed Gaussian fitting on the fluorescence intensity distribution of each magnetic bead event, and the Bayesian information criterion automatically selects the optimal component number by superimposing a component number penalty term on the log-likelihood value, avoiding the classification bias introduced by the overlap of two peaks under low concentration analyte conditions, and outputs the classification label and signal-to-noise ratio estimate of each magnetic bead event.

[0068] The specific implementation of step S05 is as follows: Substitute the signal-to-noise ratio estimate (SNR) and the detection limit estimate (LOD) output in step S04 into the signal-to-noise ratio adjustment function. ,in and As the baseline value, and To meet The weighting coefficient, typically set as a reference value. , The dimensionless comprehensive quality assessment value Q was calculated. Based on the range of Q, corresponding adjustment strategies were implemented: when Q < 0.6, indicating excessively high background signal, the amount of fluorescently labeled probe added in subsequent batches was reduced by 20% from the current value, and the number of magnetic separation and cleaning cycles was increased from 6 to 8 to reduce residual free probe; when... When the enzyme dosage is too high, it indicates nonspecific amplification. The amount of phi29 DNA polymerase added should be reduced by 10% from the current value. When Q > 1.15, it indicates that the current parameter configuration is in the optimal range and should be kept unchanged; when Q > 1.15, it indicates that there is room for improvement in the amount of probe used. The amount of fluorescently labeled probe added should be increased by 15% based on the current value, and the adjusted amount of fluorescently labeled probe added and the amount of phi29 DNA polymerase added should be output for use in subsequent batch experiments.

[0069] The specific implementation of step S06 is as follows: Based on Mie scattering theory, a theoretical constant value for the ratio of forward scattering intensity to side scattering intensity is derived for monodisperse magnetic beads under a defined particle size condition. A theoretical isofragment is established in a two-dimensional scatter plot of forward scattering intensity versus side scattering intensity as the physical optical benchmark for a single bead event. For each magnetic bead event in the convective detection data, the ratio of forward scattering intensity to side scattering intensity is calculated individually. The Mahalanobis distance is used to measure the generalized distance of the event from the theoretical isofragment, considering the covariance structure of forward and side scattering intensities. The Mahalanobis distance threshold is set to 3σ. Magnetic bead events exceeding this threshold are marked as agglomeration events and removed from the count. The removal rate is typically between 3% and 8%. A graph-cut bead classification algorithm is used for the remaining magnetic bead events. Each magnetic bead event is modeled as a graph node. A similarity graph is constructed using the Gaussian function of the fluorescence intensity difference between any two nodes as the edge weight. The energy function consists of a data term (the degree of agreement between the fluorescence intensity of each node and the prior positive and negative distribution) and a smoothing term (the consistency of classification of adjacent temporal magnetic bead events). The Ford-Fulkerson algorithm is used to solve for the maximum flow to obtain the minimum cut. Temporal local continuity constraints are used to suppress isolated noise pulses from being misclassified as positive events. Finally, the corrected single bead positive count result is output.

[0070] The key technical ideas and their effects of this invention are analyzed as follows. The first key technical idea is a closed-loop parameter optimization mechanism based on the signal-to-noise ratio (SNR) adjustment function. Traditional methods rely on offline experience to set the dosage of fluorescently labeled probes and enzymes, failing to perceive the actual distribution of background and real signals in the current batch of experiments. This invention integrates the SNR estimate and the detection limit estimate using a weighted normalization method to form a comprehensive quality assessment value, providing a quantitative basis for both the direction and magnitude of parameter adjustment. Reducing the dosage of fluorescently labeled probes directly reduces free fluorescence generated by non-specific hybridization, while reducing the dosage of enzymes inhibits non-specific amplification, forming a convergent closed-loop control logic. The second key technical idea is SNR estimation under the constraint of a physical information neural network. Purely data-driven models are prone to overfitting in sparse sample regions, leading to distorted SNR estimates and subsequent incorrect parameter adjustment decisions. This invention embeds the probe diffusion kinetics equation and the second-order binding rate equation described by Fick's second law as physical residual losses into the neural network training, ensuring that the model's predictions under low-concentration analyte conditions still conform to the physical laws of diffusion and binding, guaranteeing the reliability of the comprehensive quality assessment value. The third key technical approach is to combine Mie scattering theory with graph cut algorithms to purify counting sources. Aggregation events and noise impulses are the two main sources of interference leading to false positives in counting. The former manifests as scattering characteristics deviating from the single-bead theoretical prediction, while the latter manifests as isolated high-fluorescence events in time sequence. The two are fundamentally different in their physical mechanisms. This invention identifies and eliminates these two types of interference sources from both the physical optics level (Mahathano distance screening) and the time sequence graph structure level (graph cut smoothing constraint), resulting in higher purity of samples entering the signal-to-noise ratio (SNR) estimation and more accurate input to the SNR adjustment function. The three technical approaches work synergistically, ensuring that the decision input for parameter adjustment, the adjustment logic, and the counting purification mutually support each other, jointly improving the SNR to a stable and reliable level.

[0071] It should be noted that this invention also solves the following technical problem: In existing technologies, magnetic bead agglomeration events and instrument noise pulses can mix into the flow cytometry counting results during the rolling ring amplification digital detection process. Traditional fixed threshold methods cannot effectively distinguish between true positive signals from single beads, false positive signals from agglomerations, and false positive signals from noise pulses, resulting in high and unstable single bead positive counts. This invention establishes theoretical isoscattering lines based on the principles of physical optics using Mie scattering theory. It identifies and eliminates agglomeration events in the scattering feature space using Mahalanobis distance, and then introduces a graph cut bead classification algorithm to introduce temporal smoothing constraints to suppress misjudgments of isolated noise pulses. This purifies the counting sources from both scattering features and temporal structure dimensions, solving the technical problem of inaccurate single bead positive counts caused by the mixing of agglomerations and noise pulses. In existing technologies, the assumption of a double Gaussian distribution of fluorescence intensity leads to classification bias due to overlapping positive and negative peaks under low analyte concentration conditions, making the magnetic bead positive / negative determination results highly sensitive to the classification threshold. This invention uses the expectation-maximization algorithm to fit a multi-component mixed Gaussian distribution and automatically determines the optimal number of components using the Bayesian information criterion. It abandons the fixed bimodal assumption, which enables accurate differentiation of components even in sparse analyte samples with severe overlap of positive and negative peaks. This solves the technical problem of magnetic bead classification deviation caused by missetting the number of components in the fluorescence intensity distribution under low concentration conditions.

[0072] Specifically, the principle of this invention is as follows: The fundamental reason why this invention can solve the above-mentioned technical problems is that the signal-to-noise ratio (SNR) adjustment function combines the SNR estimate and the detection limit estimate in a weighted normalization manner into a comprehensive quality assessment value. This assessment value directly reflects the overall level of fluorescence signal quality under the current experimental parameter configuration, rather than relying on a single indicator in isolation. When the comprehensive quality assessment value is low, the background fluorescence caused by non-specific hybridization is reduced by decreasing the amount of fluorescently labeled probe, while the number of magnetic separation and washing cycles is increased to further remove free probes. When the comprehensive quality assessment value is high, the probe amount is adjusted upwards to fully utilize the amplification capacity, forming a closed-loop feedback control logic. In the fluorescence scattering feature extraction and counting optimization model, the physical information neural network module uses the probe diffusion kinetics equation and the second-order binding rate equation established by Fick's second law as physical constraints to ensure that the prediction in the sparse experimental data region still conforms to physical laws, avoiding the overestimation or underestimation of the SNR due to insufficient samples in the pure data-driven model. This ensures that the comprehensive quality assessment value has a reliable decision basis, and overall ensures that the adjustment direction and amplitude of the fluorescent probe concentration and enzyme dosage conform to physicochemical laws.

[0073] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0074] The specific implementation of step S01 is as follows: The magnetic beads and the capture antibody are fully incubated in phosphate buffer at pH 7.0-8.0, allowing the capture antibody to bind to the surface of the magnetic beads via covalent or affinity methods. Then, a solution containing 0.2%... 10 0.04% The magnetic beads are subjected to 3-9 magnetic separation and washing cycles with DNA buffer. An external magnetic field is applied to the magnetic beads to settle during each washing cycle. The supernatant is discarded and the beads are resuspended. This cycle is repeated to fully remove free capture antibodies and non-specific adsorbents, thus completing the magnetic bead blocking pretreatment. The blocked pretreatment magnetic beads are then output.

[0075] The specific implementation method of step S02 is as follows: the tracer antibody and streptavidin-DNA conjugate are mixed at a volume ratio of 1000-1500 at 37°C. Premixed reaction under the conditions of 20-60 The concentration of the tracer antibody is 5-10. The concentration of streptavidin-DNA conjugate is 10-20. The molar ratio is 1:1-4, allowing biotin on the tracer antibody to fully bind with streptavidin, forming a tracer antibody-streptavidin-DNA premixed complex. The premixed solution is then reacted with the pretreated blocking magnetic beads incubated with the antigen at room temperature for 20-60 minutes. It forms immune complexes through immune sandwich recognition and outputs immune complex magnetic beads.

[0076] The specific implementation of step S03 is as follows: After the immune complex magnetic beads are magnetically separated and washed 3-9 times, they are added to the rolling circle amplification reaction solution, and the amount of fluorescently labeled probe added is calculated at 650 mg / L. The reaction system has a capacity of 0.5-1.0. The proportion setting, DNA polymerase addition amount: per 650 mg / L Reaction system 5-10 The ratio is set at 30℃ and 400-800. The reaction takes 20-60 seconds under light-protected conditions. , DNA polymerase uses a circular DNA template as a substrate for continuous strand displacement synthesis, producing long single-stranded DNA multiply containing a large number of repetitive sequences. After the fluorescently labeled probe hybridizes with the repetitive sequences, it enriches the fluorescent signal on a single magnetic bead. After the reaction is completed, DNA buffer is added immediately to terminate the reaction, and the fluorescently labeled magnetic beads are output.

[0077] The specific implementation of step S04 is as follows: The flow cytometry detection data of the fluorescently labeled magnetic beads are input into the fluorescence scattering feature extraction and counting optimization model. For the first... Each magnetic bead event has its initial feature vector. Based on fluorescence intensity Forward scattering intensity Lateral scattering intensity and timestamp The assembly process is described in the following formula:

[0078] ;

[0079] In the formula, For the first The fluorescence intensity of each magnetic bead event, in arbitrary fluorescence units, is expressed as follows: express For the first Forward scattering intensity of a single magnetic bead event, in units of For the first The lateral scattering intensity of a single magnetic bead event, in units of For the first The timestamp of each magnetic bead event, in units of After sequential processing by the graph neural network module, the physical information neural network module, and the multi-task learning network, the fluorescence intensity distribution of each magnetic bead event is fitted with a multi-component mixed distribution using the expectation-maximization algorithm. The probability density function formula for the mixed distribution is as follows:

[0080] ;

[0081] In the formula, In the parameter set Under the conditions Fluorescence intensity of individual magnetic beads The probability density, with dimensions of The number of components in the mixed distribution is automatically determined by the Bayesian information criterion. For the first The mixing weights of the components satisfy Dimensionless For the first The mean of the Gaussian components, in units of For the first The variance of each Gaussian component, in units of For the set of all parameters to be estimated The mean ,variance The normal distribution density function, with dimensions of Its formula is expressed as follows:

[0082] ;

[0083] The formula for scoring the Bayesian information criterion is expressed as follows:

[0084] ;

[0085] In the formula, The number of groups is Bayesian information criterion scoring, dimensionless The number of groups is The maximum likelihood value after the expected value maximization algorithm converges is dimensionless, and its log-likelihood formula is as follows:

[0086] ;

[0087] The number of groups is The number of free parameters in the time model, Dimensionless The total number of magnetic bead events participating in the fitting is dimensionless. A selection is made that... smallest As the optimal grouping score, the mixed distribution grouping score is automatically determined, avoiding the classification bias introduced by the double Gaussian assumption due to the overlap of the two peaks under low analyte concentration conditions. Finally, the classification label and signal-to-noise ratio of each magnetic bead event are output. Estimated value and detection limit Estimated value.

[0088] The specific implementation of step S05 is as follows: based on the output of S04 and Calculate the overall quality assessment value according to the signal-to-noise ratio adjustment function. The formula is expressed as follows:

[0089] ;

[0090] In the formula, Dimensionless comprehensive quality assessment value This is the estimated signal-to-noise ratio for the current batch, dimensionless. The baseline signal-to-noise ratio is dimensionless and is the average of historical batches. This is an estimated detection limit for the current batch, with units consistent with the analyte concentration units. As the baseline detection limit, the unit is... Same, take the average of historical batches. and All ratios are of the same dimension, therefore the result is dimensionless. and Let be the weighting coefficient, satisfying Dimensionless, empirical values ​​are usually taken as , .when When this happens, the amount of fluorescently labeled probe added in subsequent batches will be reduced by 20% from the current value, and the number of magnetic separation and cleaning cycles will be adjusted to 8; when At the same time, keep the amount of fluorescently labeled probe added constant, and then add subsequent batches. The amount of DNA polymerase added is reduced by 10% from the current value; when At the same time, maintain the amount of fluorescently labeled probe added and The amount of DNA polymerase added remains constant; when At that time, the amount of fluorescently labeled probe added in subsequent batches will be increased by 15% based on the current value, and the adjusted amount of fluorescently labeled probe added will be output. DNA polymerase addition amount.

[0091] The specific implementation of step S06 is as follows: Based on the Mie scattering theory, for each magnetic bead event in the convection detection data... and The ratio is calculated event-by-event. Under the condition of monodisperse magnetic beads with a defined particle size, the theoretical ratio is... Derived from Mie scattering theory, the theoretical isofragments in a two-dimensional space of forward scattering intensity and side scattering intensity are quantified by Mahalanobis distance. The generalized distance by which an event of a magnetic bead deviates from the theoretically equal scattered radiation can be expressed by the following formula:

[0092] ;

[0093] In the formula, For the first The squared Mahalanobis distance of each magnetic bead event, dimensionless. For the first The scattering feature vector of a magnetic bead event, The unit is The scattering eigenvectors are predicted by the single-bead theory derived from Mie scattering theory. The unit is ,in This represents the average forward scattering intensity of a single bead. The values ​​represent the average side scattering intensity of a single bead, calculated using Mie scattering theory under known bead size, refractive index, and laser wavelength. This is the covariance matrix of forward scattering intensity and side scattering intensity, in units of 1. Estimated from single-bead reference sample experimental data, middle Dimensions are , Dimensions are The overall result is dimensionless. The specific form is as follows:

[0094] ;

[0095] In the formula, The variance of the forward scattering intensity is expressed in units of . The variance of the lateral scattering intensity is expressed in units of . The covariance of forward scattering intensity and side scattering intensity is expressed in units of . All three were estimated from reference experimental data. At that time, among them The quantile of the chi-square distribution with 2 degrees of freedom at a confidence level of 0.9973 is approximately 11.83, corresponding to 3. The confidence interval is used to mark the magnetic bead event as a clustering event and remove it from the count. For the remaining magnetic bead events, a graph-cut bead classification algorithm is used for positive and negative classification. Each magnetic bead event is modeled as a graph node, where any two nodes... and Between the boundary weights Defined by a Gaussian function of the fluorescence intensity difference, the formula is as follows:

[0096] ;

[0097] In the formula, For nodes With nodes The boundary weight between them is dimensionless. For the first Fluorescence intensity of each magnetic bead event, in units of This is the Gaussian kernel width parameter, in units of... Empirical values ​​are usually taken as the standard deviation of fluorescence intensity across all magnetic bead events. and All dimensions are The exponential term is dimensionless. The energy function consists of the data terms. With smoothing terms The composition and total energy formula are expressed as follows:

[0098] ;

[0099] ;

[0100] ;

[0101] In the formula, Total energy, dimensionless For data items, dimensionless The term is a smooth term and is dimensionless. This is the weighting coefficient for the smoothing term; it is dimensionless, and its empirical value is usually taken as 0.1 to 1.0. For nodes The classification labels are dimensionless, with 1 indicating positive and 0 indicating negative. For nodes fluorescence intensity The dimensionless negative log-likelihood with respect to the prior positive distribution For nodes fluorescence intensity The negative log-likelihood of the prior negative distribution is dimensionless; their formulas are as follows:

[0102] ;

[0103] ;

[0104] In the formula, for The probability density function under a prior positive distribution, with dimensions of . for The probability density function under a prior negative distribution, with dimensions of . Both were obtained by fitting the fluorescence intensity distribution of positive and negative magnetic bead events from historical experimental data that were manually reviewed, and were modeled as follows: and ,in This represents the mean of the prior positive distribution, in units of... The variance of the prior positive distribution is given in units of 1. This is the mean of the prior negative distribution, in units of... The variance of the prior negative distribution is expressed in units of 1. All four were determined by fitting data from a reference experimental batch. Because and All dimensions are After taking its natural logarithm and For logarithmic values, and All are dimensionless log-likelihood values For the set of temporally adjacent node pairs For the characteristic function, when The value is 1 if the condition is met, and 0 otherwise; it is dimensionless. The Ford-Fulkersen algorithm is used to solve for the maximum flow to obtain the minimum cut, and the corrected single-bead positive count result is output.

[0105] In the preparation of streptavidin-DNA conjugates, after replacing the streptavidin solution with phosphate buffer at pH 7.0-8.0, it is mixed with... (The sentence is incomplete and requires more context to translate accurately). - - The ester reacts at room temperature for 20-60 seconds. , The ester terminus undergoes an amidation reaction with the amino group of the protein, after 30 minutes. Ultrafiltration tubes remove free radicals - - Ester and replace the solution to contain 1 Phosphate buffer was used to obtain streptavidin- The conjugate, then streptavidin- The conjugate is coupled to the circular DNA template at a molar ratio of 1:1-5 overnight at 2-8°C. The streptavidin-DNA conjugate was obtained by copper-free click chemical coupling with azide-modified DNA. The coupling product was aliquoted and stored at -80°C. For the preparation of the circular DNA template, primers and linear DNA template were mixed at a volume ratio of 1:1-2 and incubated at 95°C for 1-4 days. Place in 95℃ hot water and allow to cool naturally to room temperature. After cooling, add ligase solution and let stand at room temperature in the dark for 1-3 minutes. After 5 Desalting column solution replacement to 1 phosphate buffer was used to obtain a circular DNA template.

[0106] To better understand and implement this invention, the following is a specific application scenario of the invention, Example 2: To verify the effect of the invention, technicians set up a test environment, using interleukin-6 (IL-6) as the analyte, and processed a batch of test samples in a single-molecule digital detection scenario using the complete process of the invention, recording the key parameters and output results of each step.

[0107] Technicians first completed the magnetic bead sealing pretreatment according to step S01. (The text abruptly ends here.) Magnetic beads modified with anti-IL-6 capture antibody were incubated with the capture antibody in phosphate buffer (pH 7.0-8.0) for a period of time, followed by incubation with a solution containing 0.2% BSA, 10 mM EDTA, and 0.04%... The DNA buffer was used for 3-9 magnetic separation and washing cycles. BSA effectively occupied the non-specific binding sites on the surface of the magnetic beads, and EDTA chelated divalent metal ions to prevent nuclease activity. After washing, the amount of residual free antibody on the surface of the magnetic beads was greatly reduced, and the preparation of the blocking pretreated magnetic beads was completed.

[0108] Preparation of streptavidin-DNA conjugate: Take 1 mg of streptavidin, change the medium to pH 7.0-8.0 phosphate buffer, and then mix with... (The sentence is incomplete and requires more context to translate accurately.) The ester reacts at room temperature for 20-60 minutes. The NHS ester terminus of the ester is covalently bound to the amino group of streptavidin. Free esters are removed by centrifugation 2-5 times (6000-11000 rpm, 4-8 min / cycle) using a 30 kD ultrafiltration tube. Esterification was performed by replacing the buffer with phosphate buffer containing 1 mM EDTA. The concentration of the streptavidin-DBCO conjugate was determined using Nanodrop protein A280 mode, and the measured value was divided by 3.4 to obtain the actual molar concentration. For the preparation of circular DNA templates, 8 μl of primer (Marker1, 11932.89 g / mol) and 8-16 μl of linear DNA template (Marker2, 24372.82 g / mol) were mixed at a volume ratio of 1:1-2. The mixture was incubated at 95°C for 1-4 min, then allowed to cool naturally to room temperature (approximately 11-3 h) in 95°C hot water. Ligase solution (TaKaRa DNA Ligation Kit Ver.2.1) was added, and the mixture was incubated at room temperature in the dark for 1-4 h. The solution was then replaced with phosphate buffer containing 1 mM EDTA using a 5 kD desalting column. The concentration was determined using Nanodrop single-stranded DNA mode, and the measured value was typically around 365 ng / μl. The streptavidin-DBCO conjugate was mixed with a circular DNA template at a molar ratio of 1:1-5 and coupled overnight at 2-8°C. The product was aliquoted and stored at -80°C to obtain the streptavidin-DNA conjugate.

[0109] In step S02, the tracer antibody and streptavidin-DNA conjugate are prepared in Buffer 3 (after half dilution with Buffer 1) at concentrations of 5-10 nM / ml and 10-20 nM / ml, respectively. The mixture is premixed and reacted in a metal bath at 37°C and 1000-15000 rpm for 20-60 min to allow the biotin on the tracer antibody to bind with the streptavidin-DNA conjugate with high affinity. 50-200 μl of the premix is ​​added to blocking pretreated magnetic beads incubated with IL-6 antigen, and the mixture is reacted at room temperature for 20-60 min to form a quaternary immune complex of capture antibody-IL-6-tracer antibody-streptavidin-DNA conjugate. The immune complex magnetic beads are then discharged.

[0110] In step S03, after the immune complex magnetic beads are magnetically separated and washed 3-9 times, 650 μl of rolling circle amplification reaction solution is added. The formulation includes... The following ingredients were added: 65 μl of 10× reaction buffer, 2.03 μl of 10 mM dNTP, 0.5-1.0 μl of fluorescently labeled probe (Marker3, 6017.98 g / mol), and 5-10 μl of phi29 DNA polymerase (Beyotime, 10 U / μl). The reaction was carried out at 30℃ and 400-800 rpm in a metal bath in the dark for 20-60 min. Phy29 DNA polymerase continuously synthesized long single-stranded DNA polymers along the circular DNA template anchored on the surface of the magnetic beads by means of strand displacement activity. The fluorescently labeled probe hybridized with the repetitive sequences on the polymers, and the fluorescent signal was densely enriched on a single magnetic bead. After strictly controlling the reaction for 20-60 min, 300 μl of DNA buffer was added to terminate the reaction, and the fluorescently labeled magnetic beads were output.

[0111] In step S04, fluorescently labeled magnetic beads are fed into a flow cytometer to collect fluorescence intensity (FL channel), forward scattering intensity (FSC), side scattering intensity (SSC), and timestamp for each magnetic bead event. The raw data is then input into the fluorescence scattering feature extraction and counting optimization model. The bottom-level graph neural network module constructs a similarity graph using magnetic bead events within adjacent time windows as nodes and aggregates neighborhood features. The middle-level physical information neural network module is jointly trained using Fick's second law probe diffusion kinetics equation and the second-order binding rate equation as physical constraints. The top-level multi-task learning network outputs signal-to-noise ratio estimates and detection limit estimates. Simultaneously, the expectation-maximization algorithm performs multi-component Gaussian mixture fitting on the fluorescence intensity distribution, and the Bayesian information criterion automatically selects the optimal component score. The signal-to-noise ratio estimates and detection limit estimates for this batch of detections are shown in Table 1.

[0112] Table 1. Calculation Results of Key Parameters and Signal-to-Noise Ratio Adjustment Function for This Batch of Tests

[0113]

[0114] As shown in Table 1, the overall quality assessment value Q = 0.77, which belongs to... If the current concentration of phi29 DNA polymerase is too high, leading to non-specific amplification, then adjust the concentration according to step S05: keep the amount of fluorescently labeled probe added constant at 0.65 μl, and reduce the amount of phi29 DNA polymerase added by 10% from 6.5 μl to 5.85 μl for subsequent batches.

[0115] In step S06, based on Mie scattering theory, isoscattering lines of monodisperse magnetic beads are established in the two-dimensional scatter plot of FSC-SSC, such as... Figure 2 As shown, the Mahalanobis distance is calculated for each magnetic bead event. Events with a Mahalanobis distance exceeding 3σ are classified as clusters. In this batch, cluster events, accounting for 4.7% of the total number of events, are removed. The remaining magnetic bead events are then processed by the graph cut bead classification algorithm. The Ford-Fulkerson maximum flow algorithm is used to solve for the minimum cut, and the corrected single bead positive count result is output. Figure 3 The figure shows a comparison between the single bead positive count and the theoretical value for samples at each concentration gradient.

[0116] The detection results of IL-6 samples with different concentration gradients are shown in Table 2.

[0117] Table 2. Results of positive counts of single beads in samples with different concentration gradients of IL-6.

[0118]

[0119] As shown in Table 2, after the removal of aggregates and correction by the graph-cut bead classification algorithm, the deviation between the measured number of positive beads and the theoretical value for each concentration gradient is significantly narrowed, indicating that the counting purification process of the present invention can effectively suppress false positives under different concentration conditions.

[0120] The results of multi-component Gaussian fitting of fluorescence intensity distribution are as follows: Figure 4 As shown, the expectation-maximization algorithm automatically identifies three distribution components. The Bayesian information criterion determines that the three-component model is superior to the two-component model. The classification boundary between the negative peak and the positive peak is accurately located, avoiding the classification bias caused by the overlap of the two peaks under low concentration conditions.

[0121] The following provides specific examples of experiments conducted in two different directions.

[0122] Example 3: Validation of IL-6 detection performance under different antibody / DNA ratios. The relevant figures for this example are as follows. Figure 5 and Figure 6 As shown.

[0123] To further optimize the ratio of antibody to DNA conjugate in the detection system, technicians used IL-6 as the analyte and examined the detection performance under three conditions: an antibody / DNA molar ratio of 1 / 2, 1 / 3, and 1 / 5, in a single-molecule digital detection scenario.

[0124] The experimental procedure was performed according to the aforementioned standard procedure, with the magnetic bead blocking pretreatment, immune complex formation, and rolling circle amplification steps remaining unchanged, and the RCA reaction time fixed at 30 min. Each group was set up with parallel replicates, and pulse height signals and the percentage of pulse height signals exceeding the specified threshold were collected in channel 3 for samples at three concentration gradients: 0 pg / ml, 1 pg / ml, and 10 pg / ml.

[0125] In the post-processing stage of streaming data, isoscattering lines based on Mie scattering theory are established in the FSC-SSC two-dimensional scatter plot using monodisperse magnetic beads. Mahalanobis distances are calculated for each magnetic bead event, and events exceeding 3σ are classified as aggregates and removed. The remaining events are then processed using a graph-cut bead classification algorithm with temporal smoothing constraints to suppress misclassification of isolated noise impulses, resulting in a corrected single-bead positive count. Fluorescence intensity distribution is fitted using an expectation-maximization algorithm with a multi-component Gaussian mixture, and the Bayesian information criterion automatically selects the optimal component number, avoiding classification bias caused by overlapping positive and negative peaks under low-concentration conditions under the fixed bimodal hypothesis.

[0126] Table 3. Pulse height signal and signal-to-noise ratio of IL-6 channel 3 under different antibody / DNA ratios.

[0127]

[0128] Table 4. Percentage of pulse height signals exceeding the specified threshold under different IL-6 antibody / DNA ratios (%)

[0129]

[0130] After aggregate removal and correction using the graph-cut bead classification algorithm, the positive rate of the blank group was consistently controlled at around 3% under all ratio conditions, effectively eliminating the interference of aggregates and noise pulses on the counting results. The results showed that as the anti / DNA ratio increased from 1 / 2 to 1 / 5, the signal-to-noise ratio at a concentration of 10 pg / ml generally increased, with Rep5 reaching a signal-to-noise ratio of 10.75 at the 1 / 5 ratio, achieving a positive rate of 97.8%. Simultaneously, the expectation-maximization algorithm combined with the Bayesian information criterion accurately identified fluorescence intensity distribution components at a low concentration of 1 pg / ml, ensuring classification accuracy even in cases of overlapping positive and negative peaks. These results indicate that appropriately increasing the DNA coupling ratio helps enhance the enrichment efficiency of single magnetic bead fluorescence signals and improves the detection sensitivity of low-concentration samples.

[0131] Example 4: Detection performance verification of tau217 protein under different RCA durations and washing cycles. The relevant figures for this example are as follows. Figure 7 and Figure 8 As shown.

[0132] To investigate the effects of rolling circle amplification duration and magnetic separation washing cycles on the detection performance of tau217 protein, technicians used tau217 as the analyte and set up a two-factor combination of RCA duration (30 min, 1 h) and washing cycles (3 times, 6 times) to evaluate the signal-to-noise ratio and positive percentage under each condition.

[0133] The experimental procedure was performed according to the aforementioned standard procedure, with the immune complex formation steps remaining unchanged. Only the RCA reaction time and number of washes were adjusted in step S03. The sample concentration gradients for each group were set to 0, 1 pg / ml, 10 pg / ml, and 100 pg / ml. The median pulse height signal and the percentage of pulse height signals above the specified threshold were collected in channel 3.

[0134] In the post-processing stage of streaming data, the theoretical isofragmentation based on Mie scattering theory combined with Mahalanobis distance (threshold 3σ) is used to identify and remove aggregate events. Then, the temporal smoothing constraint of the graph-cut bead classification algorithm is used to correct for misclassification of isolated noise impulses. This process purifies the counting sources from both scattering characteristics and temporal structure dimensions, ensuring the accuracy and stability of single-bead positive counts under various parameter combinations. For fluorescence intensity distribution, the expectation-maximization algorithm is used to fit a multi-component Gaussian mixture model, and the Bayesian information criterion automatically determines the optimal component number, effectively addressing the classification bias caused by overlapping positive and negative peaks in low-concentration samples.

[0135] Table 5. Median pulse height and signal-to-noise ratio of channel 3 under different RCA durations and washing cycles for tau217.

[0136]

[0137] Table 6. Percentage of pulse height signals exceeding the specified threshold under different RCA durations and washing cycles for tau217 (%)

[0138]

[0139] The results showed that when the RCA duration was extended to 1 hour and the number of washes was only 3, the median signal in the blank group reached 12139, and the background noise increased significantly. Even after aggregate removal and time-series smoothing correction, the signal-to-noise ratio of low-concentration samples was still below 1. This indicates that under insufficient washing conditions, the residue of non-specific amplification products was the main factor limiting detection performance, rather than the limitation of the counting and purification process. Under the condition of 1 hour RCA combined with 6 washes, the positive rate of 100 pg / ml samples reached 65.6%, with a P / N value of 5.032, and the detection performance was significantly improved. This shows that sufficient washing can effectively suppress background interference and release the signal gain brought by extending the RCA duration. In summary, 30 minutes RCA combined with 6 washes achieves a better balance between signal-to-noise ratio stability and background control, and is the recommended parameter combination for subsequent batches. If higher sensitivity is required, 1 hour RCA combined with 6 washes can be used, but attention should be paid to the linkage adjustment of the corresponding background noise level and counting and purification parameters.

[0140] Compared to traditional methods, this invention brings the following technological advancements: Traditional methods rely on fixed thresholds and offline empirical parameters, failing to perceive the actual state of non-specific amplification and probe residue in each batch of experiments. This invention embeds the diffusion dynamics constraints of a physical information neural network into the signal-to-noise ratio estimation, ensuring that the evaluation results remain physically consistent even in sparse sample regions, providing a reliable quantitative basis for parameter adjustment. The graph-cut bead classification algorithm introduces temporal smoothing constraints, distinguishing isolated noise pulses from true positive signals at the graph structure level, overcoming the neglect of local contextual information by fixed threshold methods. Mie scattering theory identifies aggregation events based on physical optics principles, eliminating the need for additional experimental steps and purifying the counting source from existing scattering channel data, fundamentally improving the purity and reliability of single-bead digital counting. It should be noted that the user data involved in the embodiments of this application has been authorized, acquired, processed, and transmitted in accordance with legal and regulatory requirements.

[0141] It should be noted that the variables involved in this invention are explained in detail in Tables 7 and 8.

[0142] Table 7. Variable Explanation Table (Part 1)

[0143]

[0144] Table 8. Variable Explanation Table (Part Two)

[0145]

[0146] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing fluorescent probe concentration to improve signal-to-noise ratio, characterized in that, Includes the following steps: After incubating the magnetic beads with the capture antibody in phosphate buffer, the magnetic beads were magnetically separated and washed with DNA buffer to remove free capture antibody and non-specific adsorbates, completing the magnetic bead blocking pretreatment. The pretreated magnetic beads were then output. The tracer antibody was premixed with streptavidin-DNA conjugate, and the premix was reacted with the pretreated magnetic beads incubated with the antigen to output the immune complex magnetic beads. After magnetic separation and washing, the immune complex magnetic beads were added to rolling circle amplification reaction solution. The amount of fluorescently labeled probe added to the rolling circle amplification reaction solution was... The amount of DNA polymerase added was set proportionally. Immediately after the reaction, DNA buffer was added to terminate the rolling circle amplification reaction, and fluorescently labeled magnetic beads were output. Flow cytometry data from the fluorescently labeled magnetic beads were input into a fluorescence scattering feature extraction and counting optimization model. This model fitted the fluorescence intensity distribution of each magnetic bead event using an expectation-maximization algorithm and automatically determined the mixed distribution components using the Bayesian information criterion, outputting the classification label and signal-to-noise ratio (SNR) estimate for each magnetic bead event. Based on the SNR estimate and the detection limit estimate, a comprehensive quality assessment value was calculated using a SNR adjustment function. When the comprehensive quality assessment value fell into different intervals, the amount of fluorescently labeled probe added in subsequent batches of experiments was adjusted accordingly. The amount of DNA polymerase added was gradient-regulated, and the adjusted amount of fluorescently labeled probe was output. DNA polymerase dosage; Based on Mie scattering theory, the ratio of forward scattering intensity to side scattering intensity in the flow cytometry data is calculated event-by-event. Mahalanobis distance is used to measure the deviation of each magnetic bead event from the theoretical prediction value for a single bead. Values ​​exceeding the Mahalanobis distance are... The magnetic bead events are marked as cluster events and removed from the count. The remaining magnetic bead events are then classified using a graph cut bead classification algorithm to output the corrected single bead positive count results.

2. The method for optimizing fluorescent probe concentration to improve signal-to-noise ratio according to claim 1, characterized in that, The step of incubating the magnetic beads and the capture antibody in phosphate buffer specifically involves: the pH of the phosphate buffer being 7.0–8.0, and the DNA buffer containing 0.2% BSA, 10 mM EDTA, and 0.04% HCl. The magnetic separation and cleaning process is repeated 3 to 9 times.

3. The method for optimizing fluorescent probe concentration to improve signal-to-noise ratio according to claim 2, characterized in that, The step of premixing the tracer antibody with the streptavidin-DNA conjugate specifically involves: the premixing reaction being carried out at 37°C and 1000–1500 rpm for 20–60 min, with the tracer antibody concentration being 5–10 nM / ml and the streptavidin-DNA conjugate concentration being 10–20 nM / ml.

4. The method for optimizing fluorescent probe concentration to improve signal-to-noise ratio according to claim 3, characterized in that, The step of reacting the premixed solution with the blocking pretreated magnetic beads after antigen incubation specifically involves reacting the premixed solution with the blocking pretreated magnetic beads at room temperature for 20–60 min, so that the tracer antibody can form a ternary immune complex by recognizing the antigen epitope that has been bound to the capture antibody.

5. The method for optimizing fluorescent probe concentration to improve signal-to-noise ratio according to claim 4, characterized in that, The step of adding the rolling circle amplification reaction solution specifically involves adding 0.5–1.0 μl of fluorescently labeled probe to every 650 μl of the reaction system. Add 5–10 μl of DNA polymerase and react at 30°C and 400–800 rpm in the dark for 20–60 min.

6. The method for optimizing fluorescent probe concentration to improve signal-to-noise ratio according to claim 5, characterized in that, The preparation of the streptavidin-DNA conjugate specifically involves: replacing the streptavidin solution with phosphate buffer, and then mixing it with... (The sentence is incomplete and requires more context to translate accurately). The ester was reacted at room temperature for 20–60 min, and free esters were removed by ultrafiltration through a 30 kD tube. The esterification and solution exchange yielded streptavidin-DBCO conjugates, which were then coupled with circular DNA templates overnight at 2–8°C.

7. The method for optimizing fluorescent probe concentration to improve signal-to-noise ratio according to claim 6, characterized in that, The molar ratio of streptavidin-DBCO conjugate to circular DNA template was 1:1 to 5. After aliquoting, the conjugate product was stored at -80°C to obtain streptavidin-DNA conjugate.

8. The method for optimizing fluorescent probe concentration to improve signal-to-noise ratio according to claim 7, characterized in that, The preparation of the circular DNA template is as follows: the primer and the linear DNA template are mixed at a volume ratio of 1:1 to 2 and incubated at 95°C for 1 to 4 minutes. The mixture is then placed in hot water at 95°C and allowed to cool naturally to room temperature. A ligase solution is added and the mixture is allowed to stand at room temperature in the dark for 1 to 3 hours. Finally, the mixture is passed through a 5kD desalting column to change the medium and obtain the circular DNA template.

9. The method for optimizing fluorescent probe concentration to improve signal-to-noise ratio according to claim 8, characterized in that, The signal-to-noise ratio adjustment function is defined as follows: ,in This is the estimated signal-to-noise ratio for the current batch. As the reference signal-to-noise ratio, This is the estimated detection limit for the current batch. As the benchmark detection limit, and The weighting coefficients are and satisfy the following conditions: , This is a dimensionless comprehensive quality assessment value.

10. The method for optimizing fluorescent probe concentration to improve signal-to-noise ratio according to claim 9, characterized in that, The gradient adjustment for the comprehensive quality assessment value falling into different ranges is as follows: when Q < 0.6, the amount of fluorescently labeled probe added in subsequent batches is reduced by 20%, and the number of magnetic separation and cleaning cycles is adjusted to 8; when... At that time, The amount of DNA polymerase added was reduced by 10%; when When Q > 1.15, the amount of fluorescently labeled probe added is increased by 15%.

Images