Electrochemical microplate reader

Abstract

The invention provides an electrochemical microplate reader, which comprises: an acquisition module, which is used for acquiring an electric signal of a to-be-detected object; the interface module comprises an electrode interface unit and a communication interface unit, the electrode interface unit is used for forming a detection circuit of the acquisition module together with a detection electrode, and the communication interface unit is used for establishing control signal transmission and data transmission between the detection analysis module and the outside; the detection analysis module is used for performing quantitative analysis on the to-be-detected object according to the electric signal and optimizing a quantitative analysis result according to a machine learning algorithm; and the display module is used for displaying the result of the quantitative analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of detection instruments, and particularly relates to an electrochemical enzyme immunoassay analyzer. Background Art

[0002] In many fields such as life science research, clinical diagnosis, environmental monitoring, and food safety, the demand for accurate and rapid detection of biological samples or specific substances is increasing day by day. Traditional detection methods often have problems such as insufficient sensitivity, slow detection speed, and complex operation.

[0003] As an instrument that combines electrochemical detection technology and enzyme-linked immunosorbent assay technology, although electrochemical enzyme immunoassay analyzers have been applied to some extent, there are still many areas for improvement in current products on the market. For example, the detection sensitivity and accuracy of some electrochemical enzyme immunoassay analyzers are limited, making it difficult to detect target substances at extremely low concentrations, resulting in some trace analytes not being accurately identified; and the current electrochemical enzyme immunoassay analyzers are relatively large in size and not easy to carry. This causes excessive energy consumption for researchers and testers during use and affects work efficiency. Summary of the Invention

[0004] The purpose of this application is to provide an electrochemical enzyme immunoassay analyzer to solve the problems in related technologies, such as the low detection sensitivity and accuracy of existing electrochemical enzyme immunoassay analyzers and their inconvenience in carrying.

[0005] To achieve the above purpose, this application provides the following technical solutions:

[0006] The first aspect of the present invention discloses an electrochemical enzyme immunoassay analyzer, including:

[0007] A collection module for collecting the electrical signals of the analyte to be measured;

[0008] An interface module including an electrode interface unit and a communication interface unit. The electrode interface unit is used to form a detection circuit of the collection module with the detection electrode, and the communication interface unit is used to establish control signal transmission and data transmission between the detection and analysis module and the outside world;

[0009] A detection and analysis module for quantitatively analyzing the analyte to be detected according to the electrical signals and optimizing the results of the quantitative analysis according to machine learning algorithms;

[0010] A display module for presenting the results of the quantitative analysis.

[0011] In an optional embodiment, the quantitative analysis of the detection and analysis module includes the following steps:

[0012] Scanning the basic parameters of the standard sample through the collection module, where the basic parameters include the sample test current signal y, sample concentration x, test potential E, and test temperature T;

[0013] A multiple linear regression is used to fit the standard curve, and the standard curve is expressed as:

[0014] y = k 1 x + k 2 E + k 3 T + b

[0015] where k 1 , k 2 and k 3 represent the variation coefficients of the basic parameters, and b represents the basic current signal value when the sample concentration is 0;

[0016] The standard curves of different samples are uploaded to the standard database through the communication interface unit.

[0017] In an alternative embodiment, the machine learning algorithm built into the detection and analysis module includes the following steps:

[0018] Feature extraction and concentration prediction are performed on the real-time electrical signal through a lightweight neural network model;

[0019] Based on the reinforcement learning agent, the characteristic parameters of the concentration prediction are dynamically optimized to construct a candidate prediction model;

[0020] According to K-means clustering, personalized detection schemes are generated for different characteristic samples;

[0021] During the unit cycle, environmental interference factors and electrode aging factors in the detection are compensated through model fine-tuning.

[0022] In an alternative embodiment, the feature extraction and concentration prediction of the real-time electrical signal through the lightweight neural network model include:

[0023] The time-series signal of the detected object current and the electro-chemical reaction time is obtained through the acquisition module, and the time-series signal is preprocessed by noise reduction, feature extraction, and standardization;

[0024] A sample prediction model is constructed using a random forest classifier, the detected object is classified according to the preprocessed time-series signal characteristics and the sample curve, and a concentration prediction model is constructed through a lightweight neural network according to the classification result;

[0025] The real-time concentration of the time-series signal is estimated using the concentration prediction model The calculation formula for the concentration estimation is expressed as:

[0026]

[0027] where I(t) represents the time-series signal input to the model, and f CNN represents the prediction function for concentration estimation.

[0028] In an optional embodiment, the characteristic parameters for dynamically optimizing the concentration prediction based on the reinforcement learning agent include:

[0029] Obtain the real-time potential E in the detection environment current , the real-time temperature T current , the real-time signal-to-noise ratio SNR, and the estimated concentration error Δx;

[0030] The action space of the reinforcement learning agent makes real-time adjustments to the potential feature and the temperature feature, where the adjustment value of the potential feature is ΔE = ±0.01V, and the adjustment value of the temperature feature is ΔT = ±1°C;

[0031] Define a reward function to evaluate the action effect of the reinforcement learning, and the reward function is expressed as:

[0032] R SNR = α·(SNR new - SNR old )

[0033] R error = β·(Δx new - Δx old )

[0034] R penalty = -γ·(SNR old - SNR new )

[0035] R total = R SNR + R error + R penalty

[0036] Wherein, R total represents the total reward function, R SNR represents the positive reward function for the increase in the signal-to-noise ratio, R error represents the positive reward function for the reduction in the concentration error, R penalty represents the negative reward function for the decrease in the signal-to-noise ratio or the increase in the error, SNR new represents the signal-to-noise ratio value after the real-time adjustment of the potential and temperature in the action space, SNR old represents the signal-to-noise ratio value before the real-time adjustment of the potential and temperature in the action space, Δx new represents the estimated concentration error after the adjustment of the potential and temperature, Δx old represents the estimated concentration error before the adjustment of the potential and temperature, and α, β, and γ represent reward coefficients;

[0037] Update the characteristic parameters of the concentration prediction model of the reinforcement learning agent by experience replay until the reinforcement learning agent converges, and output the environmental compensation coefficient of the candidate prediction model And the coupling function of potential E and temperature T.

[0038] In an alternative embodiment, the constructing of the candidate prediction model includes:

[0039] Determine the weight assignment of the concentration prediction model according to the classification result of the sample prediction model, and update the concentration prediction model to a candidate prediction model by using the environmental compensation coefficient and the coupling function. The concentration prediction calculation formula of the candidate prediction model is:

[0040]

[0041] Wherein, X represents the predicted concentration after calibration of the analyte detection, represents the input current signal value after calibration of the analyte detection, b represents the base current signal value when the sample concentration is 0, represents the environmental compensation coefficient, k i represents the change coefficient of the standard curve, N represents the number of candidate prediction models, w i represents the weight of the i-th candidate model, f(E i , T i ) represents the coupling function of potential and temperature.

[0042] In an alternative embodiment, the generating of the personalized detection scheme for different characteristic samples according to K-means clustering includes:

[0043] For each generated candidate prediction model, obtain the training data and historical usage records of the candidate prediction model, and load the pre-trained candidate prediction model for each sample of the input electrical signal;

[0044] Calculate the average value of the similarity between the required annotation of all detection data in the candidate prediction model and the sample detection requirements, and obtain the first similarity;

[0045] Calculate the average value of the similarity between the required annotation of all input detection data in the historical usage records of the candidate prediction model and the sample detection requirements, and obtain the second similarity;

[0046] Calculate the average value of the similarity between the input parameters of the standard curve in the historical usage records of the candidate prediction model and the input parameters of the current model, and obtain the third similarity;

[0047] Calculate the weighted sum average value of the first similarity, the second similarity and the third similarity to obtain the target similarity of different characteristic sample points. Use k-mean to perform unsupervised clustering on different samples of the historical usage records, and assign a unique personalized label to each cluster according to the clustering result and associate the historical detection data.

[0048] In an alternative embodiment, the clustering result dynamically adjusts the classification boundary according to newly input real-time sample electrical signals, continuously updates the clustering center of the personalized solution, verifies the clustering result through reinforcement learning, and if the detection error exceeds a set threshold, triggers retraining of the clustering model to generate a new clustering center.

[0049] In an alternative embodiment, during the unit period, environmental interference factors and electrode aging factors detected are compensated through model fine-tuning, including:

[0050] Model the environmental interference factors, obtain environmental interference parameters using multi-modal sensors, and quantify the impact of environmental interference factors on electrochemical signals through a multiple linear regression model, where the quantification of the impact is expressed as:

[0051]

[0052] where Δy represents the electrical signal deviation, T enu represents the environmental temperature, H represents the environmental humidity, N represents the noise level, represents the environmental interference coefficient;

[0053] For the electrode aging factors, define the aging index A according to the statistical law of electrode performance decay from historical data, and construct a battery aging prediction model using time series analysis and / or an LSTM network to generate the aging compensation coefficient λ(A).

[0054] In an alternative embodiment, during the unit period, environmental interference factors and electrode aging factors detected are compensated through model fine-tuning, including:

[0055] Supplement the environmental quantification impact parameters and the aging compensation coefficient to the acquisition data input into the model during the unit period, and update the model weights through a lightweight neural network;

[0056] Embed environmental compensation and aging compensation in the concentration prediction model, and the compensation calculation formula is expressed as:

[0057] where Δy represents the compensation for the electrical signal deviation caused by environmental interference, and λ(A) represents the compensation coefficient for electrode aging factors;

[0058] Re-evaluate the compensation effect of the concentration prediction model using a reward function. If the estimated concentration error after compensation does not decrease, adjust the aging compensation coefficient;

[0059] If the electrode aging index A exceeds the threshold, enable automatic calibration to recalibrate the electrode performance curve and reset the aging prediction model.

[0060] Compared with the prior art, the present invention has the following advantages:

[0061] (1) At the hardware level, the electrochemical enzyme label analyzer in this application has a smaller overall volume through improved overall design, and the structure of the detection electrode is improved and optimized, so that while the electrochemical enzyme label analyzer is portable, it also maintains high detection sensitivity.

[0062] (2) At the software level, the electrochemical enzyme label analyzer in this application speeds up the processing efficiency of detection data through the built-in machine learning algorithm, and ensures detection sensitivity and accuracy at the software level through the algorithm, thereby reducing the overall volume of the electrochemical enzyme label analyzer. Brief Description of the Drawings

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0064] Figure 1 It is a schematic structural diagram of an electrochemical enzyme label analyzer according to an embodiment of this application;

[0065] Figure 2 It is a concentration broken line schematic diagram of an electrochemical enzyme label analyzer according to an embodiment of this application under different detection methods. Detailed Embodiments

[0066] In order to make the purpose, technical solutions and advantages of this application clearer, the following describes and explains this application in combination with the drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain this application and are not used to limit this application. Based on the embodiments provided by this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.

[0067] Obviously, the drawings in the following description are only some examples or embodiments of this application. For those of ordinary skill in the art, without creative efforts, this application can also be applied to other similar scenarios based on these drawings. In addition, it can also be understood that although the efforts made in this development process may be complex and lengthy, for those of ordinary skill in the art related to the content disclosed in this application, some designs, manufacturing or production changes based on the technical content disclosed in this application are only conventional technical means and should not be understood as the content disclosed in this application is insufficient.

[0068] References to "embodiments" in this application mean that specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of this application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive of other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application can be combined with other embodiments without conflict.

[0069] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the ordinary meaning understood by those of ordinary skill in the technical field to which this application pertains. The words "a", "an", "one", "the", and the like involved in this application do not indicate a limitation in quantity and can represent a singular or plural number. The terms "include", "comprise", "have", and any variations thereof involved in this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may further include unlisted steps or units, or may further include other steps or units inherent to these processes, methods, products, or devices. The terms "connected", "coupled", and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

[0070] The "plurality" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships can exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0071] The technical solutions of this application and how the technical solutions of this application solve the above technical problems will be described in detail below with specific embodiments. These several specific embodiments below can be combined with each other, and the same or similar concepts or processes will not be repeated in some embodiments. The embodiments of this application will be described below in conjunction with the accompanying drawings.

[0072] See Figure 1 , an embodiment of the present invention discloses an electrochemical enzyme label analyzer, including:

[0073] A collection module 101, configured to collect the electrical signal of the analyte;

[0074] The interface module 104 includes an electrode interface unit 1041 and a communication interface unit 1042. The electrode interface unit 1041 is used to form a detection circuit of the acquisition module 101 with the detection electrode 105, and the communication interface unit 1042 is used to establish control signal transmission and data transmission between the detection and analysis module 102 and the outside world;

[0075] The detection and analysis module 102 is used to perform quantitative analysis on the object to be detected according to the electrical signal and optimize the result of the quantitative analysis according to the machine learning algorithm;

[0076] The display module 103 is used to display the result of the quantitative analysis.

[0077] It should be noted that the electrochemical enzyme immunoassay analyzer is an instrument that combines electrochemical detection technology and the principle of enzyme-linked immunosorbent assay (ELISA), and is used to quickly and highly sensitively detect specific molecules (such as proteins, antibodies, antigens, enzymes, etc.) in biological samples. It quantitatively analyzes the concentration of the target substance through the change of electrochemical signals and is widely used in fields such as biomedical research, clinical diagnosis, and food safety detection. Usually, the object to be detected undergoes an enzymatic reaction with a detection marker and an electrochemical substrate. During the reaction, due to the gain and loss of electrons in the oxidation-reduction reaction, the current will change. The detection circuit can obtain the current change signal and output the content of the detection marker according to the current change signal. However, there are many experimental variables in the actual detection process, resulting in errors in the detection results. For example, the potential selection before detection, the reaction substrate selection, the detection scheme selection, the detection electrode selection, as well as the standard curve parameters, environmental interference (such as temperature, humidity, signal interference), electrode aging, and reaction rate during detection. In order to achieve a smaller volume, the electrochemical enzyme immunoassay analyzer needs to adjust the detection hardware and related analysis hardware. Therefore, how to achieve higher detection accuracy under the premise of reducing the hardware settings is the current challenge faced by the electrochemical enzyme immunoassay analyzer.

[0078] Specifically, the acquisition module 101 includes, but is not limited to, sensors such as detection electrodes. Temperature sensors or humidity sensors can also be set in the electrochemical enzyme immunoassay analyzer. The sensors have a small impact on the volume of the electrochemical enzyme immunoassay analyzer, and the acquisition module is mainly used to convert the detection information into electrical signals for the detection and analysis model to process, so as to improve the accuracy of the detection results.

[0079] It can be understood that the detection and analysis module 102 constructs a communication connection with the cloud through the communication interface unit. The massive data in its detection process can be stored in the cloud, and the corresponding usage habits and personalized detection schemes of users can also be stored in the cloud for subsequent calls during detection. Therefore, the detection and analysis module 102 does not undertake the data storage function, and its data processing function can be regarded as the running CPU hardware, which will make the overall volume of the electrochemical enzyme immunoassay analyzer smaller.

[0080] In an optional embodiment, the quantitative analysis of the detection and analysis module includes the following steps:

[0081] By the acquisition module scanning the basic parameters of the standard sample, the basic parameters include the sample test current signal y, the sample concentration x, the test potential E, and the test temperature T;

[0082] Using multiple linear regression to fit the standard curve, the standard curve is expressed as:

[0083] y = k 1 x + k 2 E + k 3 T + b

[0084] Wherein, k 1 、k 2 and k 3 represent the change coefficients of the basic parameters, and b represents the basic current signal value when the sample concentration is 0;

[0085] Upload the standard curves of different samples to the standard database through the communication interface unit.

[0086] Specifically, the electrochemical microplate reader detects the electrochemical signals (such as current, potential or impedance changes) generated in the enzyme-catalyzed reaction through the electrode. These signals are proportional to the concentration of the target substance, so as to realize quantitative analysis. Therefore, it is necessary to pre-construct a standard curve. It can be understood that there are various influencing factors in actual detection, resulting in a deviation between the detected value and the actual value. The standard curve constructed conventionally cannot achieve ideal accuracy. Therefore, it can be saved as the original reference data or original experimental data for subsequent calls.

[0087] In an optional embodiment, the machine learning algorithm built in the detection and analysis module includes the following steps:

[0088] Extracting features and predicting the concentration of the real-time electrical signal through a lightweight neural network model;

[0089] Based on the reinforcement learning agent, dynamically optimizing the characteristic parameters of the concentration prediction to construct a candidate prediction model;

[0090] Generating personalized detection schemes for different characteristic samples according to K-means clustering;

[0091] Compensating for the environmental interference factors and electrode aging factors detected through model fine-tuning within a unit period.

[0092] Specifically, the detection and analysis module 102 can build in algorithms to process the detection data obtained by the acquisition module, synchronize with the cloud through the communication interface module, and can also call cloud computing resources to help the detection and analysis module execute the calculation tasks and decisions of the built-in algorithms.

[0093] In an optional embodiment, the feature extraction and concentration prediction of the real-time electrical signal by the lightweight neural network model include:

[0094] Obtain the time-series signal of the current of the analyte and the electrochemistry reaction time through the acquisition module, and preprocess the time-series signal by noise reduction, feature extraction, and standardization;

[0095] Use a random forest classifier to construct a sample prediction model, classify the analyte according to the characteristics of the preprocessed time-series signal and the sample curve, and construct a concentration prediction model through a lightweight neural network according to the classification result;

[0096] Use the concentration prediction model to perform real-time concentration estimation on the time-series signal The calculation formula of the concentration estimation is expressed as:

[0097]

[0098] where, I(t) represents the time-series signal input into the model, and f CNN represents the prediction function of the concentration estimation.

[0099] Specifically, since both the electrical signal and the reaction rate fed back by the acquisition module 101 are influencing variables for the detection result of the analyte concentration and are both time-series data, the lightweight neural network model can adopt the D1-CNN model, where wavelet transform is used for noise reduction processing of the original current and potential signals to retain the effective frequency band information, and the dimensional difference is eliminated after standardization processing by extracting key parameters.

[0100] Furthermore, the random forest classifier and the concentration prediction model are sample prediction models trained according to the standard curve and experimental data. For different analytes, the concentration prediction model can be selected for preliminary classification. Compared with the detection value, the predicted value output is closer to the actual value. It can be understood that the predicted value output by the neural network model still does not reach the ideal accuracy. Therefore, it is necessary to further improve the accuracy, sensitivity, and real-time efficiency.

[0101] In an optional embodiment, the dynamic optimization of the feature parameters for concentration prediction based on the reinforcement learning agent includes:

[0102] Obtain the real-time potential E current 、real-time temperature T current 、real-time signal-to-noise ratio SNR, and estimated concentration error Δx;

[0103] The action space of the reinforcement learning agent adjusts the potential feature and the temperature feature in real time, where the adjustment value of the potential feature is ΔE = ±0.01 V, and the adjustment value of the temperature feature is ΔT = ±1 °C;

[0104] Define a reward function to evaluate the action effect of the reinforcement learning. The reward function is expressed as:

[0105] R SNR = α·(SNR new - SNR old )

[0106] R error = β·(Δx new - Δx old )

[0107] R penalty = -γ·(SNR old - SNR new )

[0108] R total = R SNR + R error + R penalty

[0109] Among them, R total represents the total reward function, R SNR represents the positive reward function for the increase in signal-to-noise ratio, R error represents the positive reward function for the reduction in concentration error, R penalty represents the negative reward function for the decrease in signal-to-noise ratio or the increase in error, SNR new represents the signal-to-noise ratio value after the action space adjusts the potential and temperature in real time, SNR old represents the signal-to-noise ratio value before the action space adjusts the potential and temperature in real time, Δx new represents the estimated concentration error after adjusting the potential and temperature, Δx old represents the estimated concentration error before adjusting the potential and temperature, and α, β, and γ represent reward coefficients;

[0110] Update the feature parameters of the concentration prediction model of the reinforcement learning agent by experience replay until the reinforcement learning agent converges, and output the environmental compensation coefficient of the candidate prediction model

[0111] Specifically, obtain the feature parameters under the concentration prediction in real time and incorporate them into the influencing factors that affect the concentration prediction result. Optimize the parameter selection of the influencing factors through the reward function until the predicted value gets closer and closer to the true value. Among them, the reinforcement learning optimizes the data generalization ability of the prediction model, and the lightweight 1D-CNN and RL agent reduce the detection response time and reduce the error rate to less than 2%.

[0112] In an optional embodiment, constructing the candidate prediction model includes:

[0113] Determine the weight allocation of the concentration prediction model according to the classification result of the sample prediction model, and update the concentration prediction model to a candidate prediction model by using the environmental compensation coefficient and the coupling function. The concentration prediction calculation formula of the candidate prediction model is:

[0114]

[0115] Wherein, X represents the predicted concentration after calibration of the analyte detection, represents the input current signal value after calibration of the analyte detection, b represents the base current signal value when the sample concentration is 0, represents the environmental compensation coefficient, k i represents the change coefficient of the standard curve, N represents the number of candidate prediction models, w i represents the weight of the i-th candidate model, f(E i ,T i ) represents the coupling function of potential and temperature.

[0116] It can be understood that the updated candidate prediction model can improve the robustness of the model after iteration through the concentration prediction model, and can output more accurate concentration values in a detection environment with multiple influencing factors. For example, there may be multiple candidate prediction models in the detection of highly active enzymes, and each candidate prediction model is correspondingly set with a weight to aggregate the final prediction result to achieve the accurate purpose. At the same time, personalized detection scheme combinations for the analyte to be detected can also be generated according to different candidate prediction models, improving the generalization detection ability and response sensitivity of the electrochemical enzyme labeler to samples with different characteristics.

[0117] In an optional embodiment, generating a personalized detection scheme for different characteristic samples according to K-means clustering includes:

[0118] For each generated candidate prediction model, obtain the training data and historical usage records of the candidate prediction model, and load the pre-trained candidate prediction model for each sample of the input electrical signal;

[0119] Calculate the average value of the similarity between the required annotation of all detection data in the candidate prediction model and the sample detection requirement, and obtain the first similarity;

[0120] Calculate the average value of the similarity between the required annotation of all input detection data in the historical usage record of the candidate prediction model and the sample detection requirement, and obtain the second similarity;

[0121] Calculate the average similarity between the input parameters of the standard curve in the historical usage record of the candidate prediction model and the input parameters of the current model to obtain the third similarity;

[0122] Calculate the weighted average of the first similarity, the second similarity, and the third similarity to obtain the target similarity of different characteristic sample points. Use k-mean to perform unsupervised clustering on different samples in the historical usage record, and assign a unique personalized label to each cluster according to the clustering result and associate it with the historical detection data.

[0123] Specifically, in order to improve the detection accuracy and consistency of different samples, use the target similarity as the distance between each class to cluster the candidate model and the target detection samples, and group samples with similar electrochemical response patterns into the same category. The optimization of the clustering parameters can use the elbow method or the silhouette coefficient to determine the optimal number of clusters to ensure the rationality of the category division. Then, complete the category distinction according to the target similarity, and perform label management on each category (such as high-activity enzyme category and low-concentration substrate category). The similarity of each label can be directly associated with the detection historical data stored in the cloud through the communication interface unit, which is convenient for data calling and accuracy verification. At the same time, the personalized detection scheme can automatically adjust the detection parameters according to the clustering result. For example, for the high-activity enzyme category, the detection time can be shortened, and the potential scanning range can be reduced to save energy consumption. For the low-concentration substrate category, the sampling frequency can be increased, and the reaction time can be extended to enhance signal stability. In addition, a pre-trained lightweight neural network sub-model can be loaded for each category of samples to optimize the weight distribution of concentration prediction, so as to atomize the detection parameters of each sample and improve the detection accuracy under the personalized scheme to adapt to the complex scenarios of multiple samples.

[0124] In an alternative embodiment, the classification boundary of the clustering result is dynamically adjusted according to the newly input sample electrical signal in real time, the clustering center of the personalized scheme is continuously updated, and the clustering result is verified by reinforcement learning. If the detection error exceeds the set threshold, the retraining of the clustering model is triggered to generate a new clustering center.

[0125] In an alternative embodiment, during the unit cycle, compensation for environmental interference factors and electrode aging factors is performed through model fine-tuning, including:

[0126] Model the environmental interference factors, use a multi-modal sensor to obtain environmental interference parameters, and quantify the impact of environmental interference factors on electrochemical signals through a multiple linear regression model. The quantification of the impact is expressed as:

[0127]

[0128] where, Δy represents the electrical signal deviation, T enuT represents the ambient temperature, H represents the ambient humidity, and N represents the noise level. represents the environmental interference coefficient;

[0129] For the electrode aging factor, define the aging index A according to the statistical law of electrode performance decay based on historical data, and use time series analysis and / or LSTM network to construct a battery aging prediction model to generate the aging compensation coefficient λ(A).

[0130] In an optional embodiment, compensating for the detected environmental interference factors and electrode aging factors through model fine-tuning within the unit period includes:

[0131] Supplement the environmental quantization impact parameters and aging compensation coefficients to the acquisition data input into the model within the unit period, and update the model weights through a lightweight neural network;

[0132] Embed environmental compensation and aging compensation in the concentration prediction model, and the compensation calculation formula is expressed as:

[0133] where Δy represents the compensation for the electrical signal deviation of environmental interference, and λ(A) represents the compensation coefficient for electrode aging factors;

[0134] Re-evaluate the compensation effect of the concentration prediction model using a reward function. If the estimated concentration error after compensation does not decrease, then adjust the aging compensation coefficient;

[0135] If the electrode aging index A exceeds the threshold, then enable automatic calibration to recalibrate the electrode performance curve and reset the aging prediction model.

[0136] Specifically, within the unit period, update the model weights through incremental learning to compensate for environmental interference and electrode aging. The environment and aging compensation model can significantly improve the long-term stability of the electrochemical enzyme immunoassay analyzer. It can be understood that corresponding improvements have been made to the modification of the detection electrode in the prior art. This application mainly focuses on the detection error problem caused by environmental interference and electrode aging in the general electrode scenario. If the detection electrode is activated and modified (such as nanomaterials, standard solutions, and polypeptides, etc.), then the corresponding aging index and aging prediction model can be selected according to the electrode improvement points, and the detection compensation for environmental factors and aging factors can also be achieved.

[0137] In an optional embodiment, the structure of the electrochemical enzyme immunoassay analyzer includes, but is not limited to, columnar, conical, cuboid, and cube-shaped structures, etc. Among them, the length of the electrochemical enzyme immunoassay analyzer is preferably between 0 - 220 mm, the width is preferably between 0 - 140 mm, and the corresponding height is preferably between 0 - 50 mm. It can be understood that the specific size parameters can be adjusted according to the size of the built-in module, but the volume will be greatly reduced compared to the existing enzyme immunoassay analyzer, which is convenient for users to carry and use.

[0138] As Figure 2 shown, to verify the detection effect of the electrochemical enzyme-labeling instrument according to the embodiment of the present invention, the electrochemical enzyme-labeling instrument according to the embodiment of the present invention is used to detect CEA (carcinoembryonic antigen) with a reaction system of TMB (tetramethylbenzidine) + H 2 O 2 , and the sample concentration is 8 ng / mL. After the detection, the actual measurement values of the detection electrodes in the electrochemical enzyme-labeling instrument, the estimated values of the concentration prediction model, and the predicted values of the candidate prediction models are experimented with 15 sets of control group data. According to the control experiment data within the unit reaction time, the precision and accuracy of the detection electrode and the electrochemical enzyme-labeling instrument optimized by the machine learning algorithm for detecting the analyte are compared, as shown in Table (1) below:

[0139] Table (1) - Precision and Accuracy Table of Experimental Control Groups

[0140]

[0141]

[0142] As can be seen from Table (1) and Figure 2 , the quantitative analysis results obtained by the electrochemical enzyme-labeling instrument through the detection and analysis module are more accurate than the detection results obtained by the existing detection device through the detection electrode, and it has a smaller volume compared with the existing detection device, and its comprehensive performance is better than that of the existing detection device.

[0143] The second aspect of the present invention discloses a computer-readable storage medium, which stores computer-executable instructions for causing a computer to execute the built-in algorithm of the electrochemical enzyme-labeling instrument according to any one of the first aspect of the present invention.

[0144] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above embodiments of each electrochemical enzyme-labeling instrument. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0145] Alternatively, if the above modules of the present invention are implemented in the form of software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solutions of the embodiments of the present invention, in essence, or the parts that contribute to the related technologies can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a terminal, or a network device, etc.) to execute all or part of the methods of the various embodiments of the present invention. The foregoing storage medium includes: various media such as removable storage devices, RAM, ROM, magnetic disks, or optical discs that can store program codes.

[0146] The foregoing is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An electrochemical microplate reader, characterized in that: include: An acquisition module, used for acquiring electrical signals of the object to be tested; An interface module, comprising an electrode interface unit and a communication interface unit, wherein the electrode interface unit is used to form a detection circuit of the acquisition module with the detection electrode, and the communication interface unit is used to establish control signal transmission and data transmission between the detection and analysis module and the outside world; A detection and analysis module, used to perform quantitative analysis on the object to be detected according to the electrical signal, and optimize the result of the quantitative analysis according to a machine learning algorithm; The display module is used to display the results of the quantitative analysis.

2. The electrochemical microplate reader according to claim 1, characterized in that: The quantitative analysis of the detection and analysis module includes the following steps: Scanning the basic parameters of the standard sample through the acquisition module, the basic parameters including the sample test current signal y, the sample concentration x, the test potential E, and the test temperature T; The standard curve was fitted using multiple linear regression, and the standard curve was expressed as: y=k1x+k2E+k3T+b Among them, k1, k2 and k3 represent the coefficients of variation of the basic parameters, and b represents the basic current signal value when the sample concentration is 0; The standard curves for different samples are uploaded to the standard database through the communication interface unit.

3. The electrochemical microplate reader according to claim 1, characterized in that: The machine learning algorithm built into the detection and analysis module includes the following steps: Feature extraction and concentration prediction of real-time electrical signals through lightweight neural network models; Dynamically optimize the characteristic parameters of concentration prediction based on reinforcement learning agents and build candidate prediction models; Generate personalized detection solutions for samples with different characteristics based on K-means clustering; The environmental interference factors and electrode aging factors detected are compensated through model fine-tuning within a unit cycle.

4. The electrochemical microplate reader according to claim 3, characterized in that: The method of extracting features and predicting concentration of real-time electrical signals by using a lightweight neural network model includes: Acquire the time series signal of the detection object current and electrochemical reaction time through the acquisition module, and pre-process the time series signal by using noise reduction, feature extraction and standardization; A random forest classifier is used to construct a sample prediction model, the detected object is classified according to the preprocessed time series signal characteristics and sample curves, and a concentration prediction model is constructed through a lightweight neural network according to the classification results; Using the concentration prediction model to estimate the concentration of the time series signal in real time The calculation formula for the concentration estimation is expressed as: Where I(t) represents the time series signal of the input model, f CNN Represents the prediction function for concentration estimation.

5. The electrochemical microplate reader according to claim 4, characterized in that: The characteristic parameters of the concentration prediction dynamically optimized based on the reinforcement learning agent include: Get the real-time potential E under the detection environment current , Real-time temperature T current , real-time signal-to-noise ratio SNR and estimated concentration error Δx; The action space of the reinforcement learning agent adjusts the potential feature and the temperature feature in real time, wherein the adjustment value of the potential feature is ΔE=±0.01V, and the adjustment value of the temperature feature is ΔT=±1°C; A reward function is defined to evaluate the action effect of the reinforcement learning, and the reward function is expressed as: R SNR =α·(SNR new -SNR old ) R error =β·(Δx new -Dx old ) R penalty =-γ·(SNR old -SNR new ) R total =R SNR +R error +R penalty Among them, R total represents the total reward function, R SNR Represents a positive reward function for improved signal-to-noise ratio, R error represents a positive reward function for decreasing concentration error, R penalty Negative reward function for decreased signal-to-noise ratio or increased error, SNR new It represents the signal-to-noise ratio value after the action space adjusts the potential and temperature in real time, SNR old represents the signal-to-noise ratio value before the action space adjusts the potential and temperature in real time, Δx new represents the estimated concentration error after adjusting for potential and temperature, Δx old represents the estimated concentration error before adjusting the potential and temperature, α, β, and γ represent the reward coefficients; Use experience replay to update the concentration prediction model characteristic parameters of the reinforcement learning agent until the reinforcement learning agent converges and outputs the environmental compensation coefficient of the candidate prediction model And the coupling function of potential E and temperature T.

6. The electrochemical microplate reader according to claim 5, characterized in that: The step of constructing a candidate prediction model comprises: The weight distribution of the concentration prediction model is determined according to the classification result of the sample prediction model, and the concentration prediction model is updated to a candidate prediction model using the environmental compensation coefficient and the coupling function. The concentration prediction calculation formula of the candidate prediction model is: Where X represents the predicted concentration of the analyte after calibration. represents the input current signal value after the object to be tested is calibrated, b represents the basic current signal value when the sample concentration is 0, represents the environmental compensation coefficient, k i represents the coefficient of variation of the standard curve, N represents the number of candidate prediction models, and w i represents the weight of the i-th candidate model, f(E i ,T i ) represents the coupling function between potential and temperature.

7. The electrochemical microplate reader according to claim 6, characterized in that: The method of generating a personalized detection scheme for samples with different characteristics based on K-means clustering includes: For each generated candidate prediction model, the training data and historical usage records of the candidate prediction model are obtained, and the pre-trained candidate prediction model is loaded for each sample of the input electrical signal; Calculate the average of the similarities between the requirement annotations of all the test data in the candidate prediction model and the sample test requirements to obtain a first similarity; Calculate the average of the similarities between the requirement annotations of all input test data and the sample test requirements in the historical use record of the candidate prediction model to obtain a second similarity; Calculating an average value of similarities between input parameters of the standard curve in the historical usage record of the candidate prediction model and input parameters of the current model to obtain a third similarity; Calculate the weighted average of the first similarity, the second similarity and the third similarity to obtain the target similarity of different characteristic sample points, use k-mean to perform unsupervised clustering on different samples of the historical usage records, assign a unique personalized label to each cluster according to the clustering results and associate it with historical detection data.

8. The electrochemical microplate reader according to claim 7, characterized in that: The clustering result dynamically adjusts the classification boundary according to the new sample electrical signal input in real time, continuously updates the clustering center of the personalized solution, and verifies the clustering result through reinforcement learning. If the detection error exceeds the set threshold, the retraining of the clustering model is triggered to generate a new clustering center.

9. The electrochemical microplate reader according to claim 7, characterized in that: The environmental interference factors and electrode aging factors detected are compensated by fine-tuning the model within the unit period, including: The environmental interference factors are modeled, and the environmental interference parameters are obtained using multimodal sensors. The impact of environmental interference factors on electrochemical signals is quantified using a multivariate linear regression model, where the quantified impact is expressed as: Where Δy represents the electrical signal deviation, T enu represents the ambient temperature, H represents the ambient humidity, N represents the noise level, and represents the environmental interference coefficient; The electrode aging factor is defined according to the electrode performance attenuation law based on historical data statistics. The battery aging prediction model is constructed using time series analysis and / or LSTM network to generate the aging compensation coefficient λ(A).

10. The electrochemical microplate reader according to claim 9, characterized in that: The environmental interference factors and electrode aging factors detected are compensated by fine-tuning the model within the unit period, including: The collected data of the unit cycle input model is supplemented with environmental quantitative impact parameters and aging compensation coefficients, and the model weights are updated through a lightweight neural network; Environmental compensation and aging compensation are embedded in the concentration prediction model, and the compensation calculation formula is expressed as: Among them, Δy represents the electrical signal deviation compensation of environmental interference, and λ(A) represents the compensation coefficient of electrode aging factor; The compensation effect of the concentration prediction model is re-evaluated using a reward function, and if the estimated concentration error after compensation is not reduced, the aging compensation coefficient is adjusted; If the electrode aging index A exceeds the threshold, automatic calibration is enabled to recalibrate the electrode performance curve and reset the aging prediction model.

Images