TMB developing solution formula optimization method and system, computer equipment and storage medium

Abstract

The invention discloses a TMB developing solution formula optimization method and system, computer equipment and a storage medium. The method comprises the steps that an experimental data set of formula samples is acquired, and an N-dimensional formula space is obtained; constructing a forward mapping relation from the N-dimensional formula space to the M-dimensional hidden space; constructing a Gaussian process regression agent model, and training the Gaussian process regression agent model based on the initial experimental data set; based on the trained Gaussian process regression agent model, calculating EI values of candidate points in the M-dimensional hidden space by using an expectation improvement point adding strategy to obtain an optimal hidden space point; and performing reverse mapping on the optimal hidden space point through a forward mapping relation of PCA to obtain a new N-dimensional suggestion formula. And performing experimental verification on the new N-dimensional suggestion formula, adding a verification result as a new experimental data point into the experimental data set, returning to the model training process for iterative optimization, and outputting the optimized suggestion formula.

Description

Technical Field

[0001] This invention relates to the field of analytical chemistry and artificial intelligence, and more specifically to a method, system, computer equipment, and storage medium for optimizing the formulation of TMB colorimetric solution. Background Technology

[0002] 3,3',5,5'-Tetramethylbenzidine (TMB), as a highly efficient and safe chromogenic substrate for horseradish peroxidase (HRP), has been widely used in many bioanalytical and clinical diagnostic fields such as enzyme-linked immunosorbent assay (ELISA), Western blotting, and immunohistochemistry (IHC). The performance of TMB chromogenic solutions, especially their detection sensitivity (usually expressed as absorbance OD value) and long-term storage stability, directly determines the reliability, accuracy, and detection limit of the entire analytical method. Therefore, developing TMB chromogenic solution formulations with higher sensitivity and better stability is of vital importance for improving the quality and market competitiveness of in vitro diagnostic (IVD) reagents.

[0003] Traditional TMB (Total Molecular Weight) solution formulation optimization heavily relies on the experience of experimental scientists and labor-intensive trial-and-error methods. A typical approach is the One-Variable-at-a-Time (OVAT) method, which involves fixing the concentrations of all other components and changing only the concentration of a single component to find its optimal value. While intuitive, this method has a fatal flaw: it completely ignores the complex synergistic or antagonistic effects between chemical components. For example, changes in the concentration of the stabilizer PVP may affect the optimal pH value; and the presence of the chelating agent EDTA may interact with metal ions in the buffer system. The OVAT method cannot capture these cross-effects, is prone to getting trapped in local optima, and struggles to discover truly high-performance formulations resulting from the synergistic effects of multiple components.

[0004] To overcome the limitations of OVAT, those skilled in the art have begun to employ more systematic Design of Experiments (DoE) methods, such as Response Surface Methodology (RSM) and the Taguchi method. These methods, by designing a relatively small, orthogonal experimental matrix, can study multivariate interactions to some extent. However, when the number of variable components in a formulation further increases, forming a high-dimensional (e.g., more than 8-10 variables) formulation space, these traditional DOE methods also require a large number of experimental samples. Therefore, to reduce the cost of experimental trial and error, some machine learning-assisted techniques have been introduced into fields such as analytical chemistry and biopharmaceutical development to achieve low-cost, rapid, and efficient surrogate-assisted experimental result prediction. Among these, Bayesian optimization (BO) is one of the representative global optimization methods capable of establishing surrogate models for expensive chemical experiments with small experimental sample sizes, demonstrating significant application potential.

[0005] The core essence of Bayesian optimization lies in its "surrogate model-driven intelligent exploration" paradigm. The surrogate model, as the core driving algorithm, directly determines the global optimization accuracy of Bayesian optimization. Gaussian Process Regression (GPR), with its unique theoretical advantages, is widely recognized as the most natural and powerful "golden partner" to the Bayesian optimization framework. This is mainly because the GPR model, like other models (such as neural networks or random forests), can not only fit the relationships between existing experimental data points but also provide the predicted mean for unknown points. μ It can also provide the "uncertainty" or confidence interval (i.e., prediction variance) of the predicted value for each prediction point. σ ²).

[0006] It is precisely this ability to simultaneously provide "predicted values" and "uncertainty" that makes the GPR model, combined with Expected Improvement (EI) methods, the most representative combination in Bayesian optimization. However, although the combination of GPR and BO is theoretically very mature, its direct application in real-world, high-dimensional chemical formulation optimization still faces two major technical challenges stemming from the inherent properties of the GPR model itself: First, there is model degradation under high-dimensional, sparse data. For a data-driven reagent optimization problem, establishing a surrogate model with the highest possible accuracy under the condition of the fewest possible experimental samples is of great engineering significance, and can greatly reduce R&D costs. However, when the dimensionality of the formulation design increases...D When the value is greater than 10, the "similarity" of expensive and scarce real experimental data samples decreases in the high-dimensional solution space, which leads to a decrease in the ability of the GPR surrogate model to describe the chemical space and an increase in prediction error.

[0007] Secondly, there is a contradiction between sequential decision-making and the need for parallel processing. The standard Bayesian optimization process is sequential, meaning that after the GPR model is updated, an optimal sample to be added is found through the EI function, an experiment is completed, and then the next iteration begins. This "recommending only one point at a time" approach results in other available experimental resources being idle and waiting during the sequential decision-making process, which greatly limits the efficiency of the entire optimization process.

[0008] Therefore, although GPR is an ideal surrogate model for achieving efficient Bayesian optimization, overcoming its vulnerability to high-dimensional, sparse data and extending it from sequential decision-making to efficient parallel decision-making has become a major challenge restricting the effectiveness of Bayesian optimization in the field of complex TMB colorimetric solution formulations. Summary of the Invention

[0009] In view of this, the present invention provides a method, system, computer equipment and storage medium for optimizing TMB colorimetric solution formulation, which can meet the needs of parallel experiments and enable the R&D team to obtain richer and more complementary information within a single experimental cycle, thereby significantly shortening the R&D cycle of the entire optimization project.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: A method for optimizing the formulation of a TMB colorimetric solution includes: Step 1: Obtain the experimental dataset of the formula samples. N Dimensional formula space; Step 2: Build from N Dimensional Formula Space M Forward mapping relationship in 3D implicit space; Step 3: Construct a Gaussian process regression surrogate model and train it based on the initial experimental dataset; Step 4: Based on the trained Gaussian process regression surrogate model, use the expected improvement point addition strategy in... M The EI value of candidate points is calculated in the latent space to obtain the optimal latent space point; Step 5: Reverse map the optimal latent space point using the forward mapping relationship of PCA to obtain a new... N Recommended formula; Step 6: For the new NThe recommended formula is experimentally verified, and the verification results are added to the experimental dataset as new experimental data points. The model is then returned to the training process for iterative optimization until the preset convergence condition is met, and the optimized recommended formula is output.

[0011] Preferably, the experimental dataset includes a training dataset of multiple initial formula samples and their corresponding performance metrics, each initial formula sample consisting of... N The concentration values ​​of each chemical component are defined, and the performance index is the maximum absorbance of the TMB colorimetric reaction at a specific wavelength, which is determined by... N The concentration values ​​of each chemical component constitute N Dimensional formula space.

[0012] Preferably, in step 2, principal component analysis is used to analyze the... N The N-dimensional formula space is processed to construct a positive mapping relationship from the N-dimensional formula space to the M-dimensional latent space, where M... <N。

[0013] Preferably, step 4 utilizes the desired improvement point addition strategy to calculate the EI value of candidate points and performs k-Means clustering on a preset number of candidate points with high EI values ​​to obtain the optimal latent space points.

[0014] Preferably, before obtaining the experimental dataset of the formulation samples in step 1, an experimental constraint definition step is also included, based on prior chemical knowledge or preliminary experimental observations, from... N Identify and immobilize undesirable reactions that lead to precipitation from a chemical composition to reduce the dimensionality of practical optimization. N .

[0015] Preferably, the N chemical components comprise at least eight selected from TMB, hydrogen peroxide, a pH adjuster, sodium acetate, polyvinylpyrrolidone, EDTA, dimethyl sulfoxide, trehalose, glycerol, citric acid, sodium thiosulfate, and 1,2,6-hexanetriol, wherein... N ≥10.

[0016] Preferably, a TMB colorimetric solution formulation optimization system includes: Data acquisition module: Used to acquire the experimental dataset of the formula samples, and obtain... N Dimensional formula space; Mapping module: used to build from N Dimensional Formula Space M Forward mapping relationship in 3D implicit space; Model training module: Used to build a Gaussian process regression surrogate model and train the Gaussian process regression surrogate model based on the initial experimental dataset; Optimal latent space computation module: Based on a trained Gaussian process regression surrogate model, using an expected improvement point addition strategy in...M The EI value of candidate points is calculated in the latent space to obtain the optimal latent space point; The reverse mapping module is used to reverse map the optimal latent space points using the forward mapping relationship of PCA, to obtain new... N Recommended formula; Optimization module: used for new... N The recommended formula is experimentally verified, and the verification results are added to the experimental dataset as new experimental data points. The model is then returned to the training process for iterative optimization until the preset convergence condition is met, and the optimized recommended formula is output.

[0017] Preferably, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the TMB colorimetric solution formulation optimization method.

[0018] Preferably, a computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the TMB colorimetric solution formulation optimization method.

[0019] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method, system, computer equipment, and storage medium for optimizing TMB colorimetric solution formulations, which has one or more of the following significant beneficial effects: 1. Effectively overcomes the "curse of dimensionality," significantly improving the performance and robustness of high-dimensional optimization. This invention innovatively introduces a "latent space exploration" strategy based on Principal Component Analysis (PCA). By intelligently mapping the original high-dimensional (e.g., 12 or more dimensions) chemical formulation space to a low-dimensional latent space that retains the main variation information, this invention enables the core GPR surrogate model to be trained and predicted in a lower-dimensional space with more "dense" data and a clearer structure. This fundamentally solves the problem in existing technologies where standard Bayesian optimization stagnates or fails when dealing with high-dimensional, sparse data due to GPR model overfitting, training failure, or ill-conditioned covariance matrix. Therefore, this invention can perform more stable, accurate, and reliable global optimization of high-dimensional complex formulations that are difficult to handle by traditional methods.

[0020] Achieving true parallel optimization and significantly shortening the R&D cycle, this invention proposes an efficient "batch addition strategy" based on k-Means clustering to resolve the contradiction between the "one point at a time" sequential decision-making model of standard Bayesian optimization and the requirements of modern parallel experiments. This strategy intelligently partitions all high-potential (high EI value) candidate points, generating a set of recommended formulations that combine "high potential" and "high diversity" in a single step. This allows R&D teams to fully utilize parallel experimental equipment such as 96-well plates and automated workstations, obtaining maximized and complementary experimental data within a single experimental cycle. Compared to the sequential optimization techniques of existing technologies, this invention upgrades "single-threaded" decision-making to "multi-threaded" exploration, greatly reducing resource idleness caused by algorithm waiting, thereby shortening the entire formulation optimization project's R&D cycle by several times. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0022] Figure 1 The TMB formulation optimization flowchart provided by this invention.

[0023] Figure 2 This is a visual diagram illustrating the medium-batch point addition strategy provided by the present invention.

[0024] Figure 3 The visualization of the prediction performance of the GPR model on a one-dimensional pH slice in a TMB formulation optimization example containing 12 variable components provided by the present invention.

[0025] Figure 4 The method flowchart provided by the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Example 1: like Figure 4 As shown in the figure, an embodiment of the present invention discloses a method for optimizing the formulation of TMB colorimetric solution, including: Step 1: Obtain the experimental dataset of the formula samples. N Dimensional formula space; Step 2: Constructing from N-dimensional recipe space to M Forward mapping relationship in 3D implicit space; Step 3: Construct a Gaussian process regression surrogate model and train it based on the initial experimental dataset; Step 4: Based on the trained Gaussian process regression surrogate model, use the expected improvement point addition strategy to calculate the EI value of the candidate points in the M-dimensional latent space to obtain the optimal latent space point; Step 5: Reverse map the optimal latent space point using the forward mapping relationship of PCA to obtain a new N-dimensional proposed formula; Step 6: For the new N The recommended formula is experimentally verified, and the verification results are added to the experimental dataset as new experimental data points. The model is then returned to the training process for iterative optimization until the preset convergence condition is met, and the optimized recommended formula is output.

[0028] Specifically, the experimental dataset includes a training dataset of multiple initial formula samples and their corresponding performance metrics. Each initial formula sample consists of... N The concentration values ​​of each chemical component are defined, and the performance index is the maximum absorbance of the TMB colorimetric reaction at a specific wavelength, which is composed of the concentration values ​​of N chemical components. N Dimensional formula space.

[0029] Specifically, in step 2, principal component analysis is used to analyze the... N The dimensional recipe space is processed to construct a space from the above. N Dimensional Formula Space M The positive mapping relation of the latent space, where M < N .

[0030] Specifically, step 4 uses the expected improvement point addition strategy to calculate the EI value of the candidate points, and performs k-Means clustering on a preset number of candidate points with high EI values ​​to obtain the optimal latent space points.

[0031] Specifically, before obtaining the experimental dataset of the formulation samples in step 1, there is also an experimental constraint definition step, which, based on prior chemical knowledge or preliminary experimental observations, defines the experimental constraints. N Identify and fix components that cause adverse reactions leading to precipitation from a chemical composition to reduce the dimension N of the actual optimization.

[0032] Specifically, it is characterized by, NEach chemical component includes at least eight of the following: TMB, hydrogen peroxide, pH adjuster, sodium acetate, polyvinylpyrrolidone, EDTA, dimethyl sulfoxide, trehalose, glycerol, citric acid, sodium thiosulfate, and 1,2,6-hexanetriol, wherein N ≥ 10.

[0033] In a specific embodiment of the present invention, the data preparation step includes: obtaining multiple initial formulation samples and their corresponding performance indicators (such as...). OD max The training dataset; Dimensionality reduction mapping step: Principal component analysis (PCA) is used to process the N-dimensional formulation space and construct a dimension reduction mapping. M dimension( M < N The positive mapping relationship of the latent space; (1) Where X represents the original N-dimensional formula data, and Z represents... M Latent space coordinates of dimension, μ X for X The mean of W is the N×M projection matrix obtained by PCA. Proxy model construction steps: In M In the latent space, based on the training dataset D train Train a Gaussian process regression (GPR) surrogate model; (2) in, D train For the training dataset, μ ( z ( ) represents the predicted mean. σ 2 ( z ) represents the prediction variance.

[0034] Batch point finding steps: In M In the latent space, based on the GPR model, the EI values ​​of a large number of candidate points are calculated using the expected improvement (EI) point addition strategy, and the candidate points with high EI values ​​are clustered by k-Means to obtain a set of optimal latent space points with diversity. (3) in, y ( z ) =( μ ( z )- y best ) / σ (z),y best Φ and are the best known performance metrics. These are the cumulative distribution function and probability density function of the standard normal distribution, respectively.

[0035] Inverse mapping step: The set of optimal latent space points is inversely mapped using PCA to obtain a new set of N-dimensional suggested recipes; (4) Among them, Z new For a newly found optimal set of points in the latent space, W’ Projection matrix W The transpose of .

[0036] Iterative optimization steps: Add the experimental results of the new formula to the training dataset, return to step b) and iterate until the convergence condition is met.

[0037] Specifically, a TMB colorimetric solution formulation optimization system includes: Data acquisition module: used to acquire experimental datasets of formulation samples, resulting in an N-dimensional formulation space; Mapping module: used to build from N Dimensional Formula Space M Forward mapping relationship in 3D implicit space; Model training module: Used to build a Gaussian process regression surrogate model and train the Gaussian process regression surrogate model based on the initial experimental dataset; Optimal latent space computation module: Based on a trained Gaussian process regression surrogate model, using an expected improvement point addition strategy in... M The EI value of candidate points is calculated in the latent space to obtain the optimal latent space point; Inverse mapping module: used to reverse map the optimal latent space point through the forward mapping relationship of PCA to obtain a new N-dimensional proposed formula; The optimization module is used to experimentally verify the new N-dimensional proposed formula, and add the verification results as new experimental data points to the experimental dataset. It returns to the model training process for iterative optimization until the preset convergence condition is met, and outputs the optimized proposed formula.

[0038] Specifically, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the TMB colorimetric solution formulation optimization method.

[0039] Specifically, a computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the TMB colorimetric solution formulation optimization method.

[0040] In summary, this embodiment effectively overcomes the difficulty of training surrogate models under high-dimensional sparse data by optimizing in a low-dimensional latent space, thereby improving the model's prediction accuracy and optimization stability. At the same time, the batch addition strategy meets the needs of parallel experiments in modern chemical research and development, significantly shortens the research and development cycle, and can efficiently and robustly find the global optimal solution for high-dimensional complex TMB colorimetric solution formulations.

[0041] Example 2: like Figure 1 As shown, in a specific embodiment of the present invention, the following steps are included: Step 100: Data preparation.

[0042] Collect or design an initial experimental dataset. For example, design 18 initial TMB formulation samples using Latin hypercube sampling (LHS) or based on chemical experience. Each sample is a 12-dimensional vector. Experiment with these 18 formulations, measuring their properties under specific conditions. OD max The values ​​form an 18×12 input matrix. X orig and an 18×1 output vector Y OD .

[0043] Step 200: Dimensionality reduction mapping.

[0044] For an 18×12 input matrix X orig Standardization process is required to obtain X scaled Then, principal component analysis (PCA) was used. This method involves... X scaled The covariance matrix is ​​decomposed into eigenvalues ​​to obtain the projection matrix. coeff The coordinates of the sample in the latent space. X latent Obtained through the following transformation: (5) in, coeff M Projection matrix coeff The largest corresponding to the middle M The first eigenvalues M Column. In this embodiment, a column is set. M =5.

[0045] Step 300: Proxy model construction.

[0046] In the 5-dimensional latent space, using X latent and standardization YOD_scaled To train a GPR surrogate model, the model employs an anisotropic squared exponential kernel function and optimizes its hyperparameters (such as length scale and signal variance) by maximizing the marginal likelihood function.

[0047] Step 400: Batch point search.

[0048] Within the boundary of the 5-dimensional latent space, a large number (e.g., 50,000) of candidate points are generated through Latin hypercube sampling.

[0049] Using the trained GPR model, predict the mean and variance of the OD for these 50,000 candidate points.

[0050] Based on the prediction results and the currently known optimal OD value f best Calculate the expected improvement (EI) value for each candidate point. The EI value is calculated using the following method: (6) in, y =( μ ( z cand )- y best _s caled ) / σ ( z cand ).

[0051] Note that all calculations here are performed on a standardized scale.

[0052] The 1000 candidate points with the highest EI values ​​were selected, and these points constituted the "high-potential area".

[0053] Run the k-Means clustering algorithm on these 1000 points (in 5-dimensional latent space), setting the number of clusters to 10. The algorithm will output the centroids of the 10 clusters; these centroids are our proposed set of optimal exploration points with diversity. This process can be found in [link to documentation]. Figure 2 A visual representation.

[0054] Step 500: Reverse mapping. The 10 5D latent space center points obtained in step 400 are... C latent By using the inverse transformation of PCA, the chemical formulas are mapped back to the 12-dimensional real chemical formula space, resulting in 10 new suggested formulas. C real The inverse transformation process consists of two steps: first, coordinate inversion, and then inverse normalization.

[0055] (7) (8) in, coeff M ' is the transpose of the 5D projection matrix. σ X and μ X The standard deviation and mean of the original 12-dimensional data are calculated in step 200.

[0056] Step 600: Iterative optimization. Perform parallel experiments on these 10 new formulations to obtain their true OD (Original Demand). max Add these 10 new {recipe-OD value} data points to the initial dataset to form a larger, more informative new dataset. Return to step 200 and begin the next round of dimensionality reduction, model training, and point addition recommendation with the new dataset. Repeat this process until the found recipes no longer offer significant performance improvements, or the preset experimental budget is reached.

[0057] Furthermore, Figure 3 This diagram illustrates the predictions of the GPR model along the most important latent space dimensions in a given iteration. The figure clearly shows the model's prediction mean, 95% confidence interval, and how the EI addition function guides the next direction of exploration.

[0058] Furthermore, this embodiment utilizes PCA dimensionality reduction to transform the original high-dimensional optimization problem into a low-dimensional latent space, greatly mitigating the negative impact of high-dimensional data similarity on GPR model training. This allows the surrogate model to learn the performance characteristics of the chemical space more robustly and accurately under sparse data. Compared with existing technologies, this avoids optimization stagnation or getting trapped in local optima due to model failure.

[0059] Furthermore, this invention innovatively introduces the k-Means clustering algorithm into the point addition stage of Bayesian optimization, forming an efficient batch point addition strategy. This strategy clusters all high-potential (high EI value) candidate points and selects the center point of each cluster as the recommended sample, ensuring that the recommended batch of experimental points is not only highly potential but also spatially separated, exhibiting high diversity. Compared to the sequential "one point at a time" mode in existing technologies, this invention meets the needs of parallel experiments, enabling the R&D team to obtain richer and more complementary information within a single experimental cycle, thereby significantly shortening the overall R&D cycle of the optimization project.

[0060] In summary, this invention provides a complete, efficient, and robust systematic solution to the problem of optimizing complex high-dimensional chemical formulations by combining "latent space exploration" with "batch point addition strategy".

[0061] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0062] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing the formulation of a TMB colorimetric solution, characterized in that, include: Step 1: Obtain the experimental dataset of the formula samples. N Dimensional formula space; Step 2: Constructing from N-dimensional recipe space to M Forward mapping relationship in 3D implicit space; Step 3: Construct a Gaussian process regression surrogate model and train it based on the initial experimental dataset; Step 4: Based on the trained Gaussian process regression surrogate model, use the expected improvement point addition strategy in... M The EI value of candidate points is calculated in the latent space to obtain the optimal latent space point; Step 5: Reverse map the optimal latent space point using the forward mapping relationship of PCA to obtain a new... N Recommended formula; Step 6: For the new N The recommended formula is experimentally verified, and the verification results are added to the experimental dataset as new experimental data points. The model is then returned to the training process for iterative optimization until the preset convergence condition is met, and the optimized recommended formula is output.

2. The method for optimizing the formulation of TMB colorimetric solution according to claim 1, characterized in that, The experimental dataset includes a training dataset of multiple initial formula samples and their corresponding performance metrics. Each initial formula sample consists of... N The concentration values ​​of each chemical component are defined, and the performance index is the maximum absorbance of the TMB colorimetric reaction at a specific wavelength. The N-dimensional formulation space is composed of the concentration values ​​of N chemical components.

3. The method for optimizing the formulation of a TMB colorimetric solution according to claim 2, characterized in that, In step 2, principal component analysis is used to process the N-dimensional formulation space to construct a model from the N-dimensional formulation space. N Dimensional Formula Space M The positive mapping relation of the 3D implicit space, where M < N .

4. The method for optimizing the formulation of a TMB colorimetric solution according to claim 2, characterized in that, Step 4 utilizes the expected improvement point addition strategy to calculate the EI value of candidate points, and performs k-Means clustering on a preset number of candidate points with high EI values ​​to obtain the optimal latent space points.

5. The method for optimizing the formulation of a TMB colorimetric solution according to claim 1, characterized in that, Before obtaining the experimental dataset of the formulation samples in step 1, there is also an experimental constraint definition step, which, based on prior chemical knowledge or preliminary experimental observations, determines the constraints from... N Identify and immobilize undesirable reactions that lead to precipitation from a chemical composition to reduce the dimensionality of practical optimization. N .

6. A method for optimizing the formulation of a TMB colorimetric solution according to any one of claims 1-5, characterized in that, N The chemical components include at least eight of the following: TMB, hydrogen peroxide, pH adjuster, sodium acetate, polyvinylpyrrolidone, EDTA, dimethyl sulfoxide, trehalose, glycerol, citric acid, sodium thiosulfate, and 1,2,6-hexanetriol. N ≥10.

7. A TMB colorimetric solution formulation optimization system, characterized in that, include: Data acquisition module: Used to acquire the experimental dataset of the formula samples, and obtain... N Dimensional formula space; Mapping module: used to build from N Dimensional Formula Space M Forward mapping relationship in 3D implicit space; Model training module: Used to build a Gaussian process regression surrogate model and train the Gaussian process regression surrogate model based on the initial experimental dataset; Optimal latent space computation module: Based on a trained Gaussian process regression surrogate model, using an expected improvement point addition strategy in... M The EI value of candidate points is calculated in the latent space to obtain the optimal latent space point; The reverse mapping module is used to reverse map the optimal latent space points using the forward mapping relationship of PCA, to obtain new... N Recommended formula; Optimization module: used for new... N The recommended formula is experimentally verified, and the verification results are added to the experimental dataset as new experimental data points. The model is then returned to the training process for iterative optimization until the preset convergence condition is met, and the optimized recommended formula is output.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images