Stimulation response performance prediction method of poly-eutectic fluorescent gel based on machine learning

Abstract

The invention relates to the technical field of fluorescent gel, and discloses a method for predicting stimulation response performance of poly-eutectic fluorescent gel based on machine learning, which comprises the following steps of: acquiring, constructing and preprocessing characteristic data such as chemical composition and microstructure of the poly-eutectic fluorescent gel; and performing physical, chemical and biological stimulation response performance prediction on the co-melting fluorescent gel according to the pre-processed co-melting fluorescent gel training data set. Compared with a traditional method, the method has the advantages of low cost, safe experiment and fast effect output, and has the advantages of high precision of quantitative prediction and high-dimensional nonlinear problem processing.

Description

Technical Field

[0001] This invention relates to the field of fluorescent gel technology, and in particular to a method for predicting the stimulus-response performance of polyeutectic fluorescent gels based on machine learning. Background Technology

[0002] Polyeutectic fluorescent gels are a composite concept, consisting of a polymerizable eutectic solvent and a fluorescent group. The polymerizable eutectic solvent is hailed as a special type of "novel ionic liquid," but it is inexpensive, has a simple preparation process, and better biocompatibility. Polyeutectic solvents typically consist of two or more solid or liquid compounds, such as hydrogen bond acceptors (e.g., choline chloride) and monomeric hydrogen bond donors (e.g., acrylic acid, acrylamide), which interact through hydrogen bonding to form a polymerizable low-melting-point mixture. A fluorescent group is then designed and introduced into the polymerizable eutectic solvent, followed by a chemical reaction (e.g., free radical polymerization or cross-linking) to form a three-dimensional network structure. This transforms the flowing low-melting-point liquid mixture into a solid or semi-solid polyeutectic fluorescent gel, combining the low toxicity and high biocompatibility of polyeutectic systems with the signal response characteristics of fluorescent materials.

[0003] The stimulus response of polyeutectic fluorescent gels is essentially based on the disruption / reconstruction of the gel network framework by external stimuli (such as hydrogen bonding, hydrophobic interactions, ion coordination, π-π stacking, etc.), thereby altering the microenvironment of the fluorophore and ultimately manifesting as tunable changes in fluorescence intensity, wavelength, and other signals. Currently, the types of responsive stimuli mainly revolve around the "structural sensitivity of the polyeutectic system" and the "environmental dependence of the fluorophore," and are mainly divided into three categories: physical stimuli (such as temperature, light, force, electricity, and humidity), chemical stimuli (such as redox stimuli, pH, salt or ion concentration, etc.), and biological stimuli (such as enzymes, antigens, and DNA, etc.).

[0004] Similar to traditional materials development, which relies on numerous repetitive "formulation-synthesis-testing" experiments, the formulations and synthesis conditions of polyeutectic fluorescent gels also number in the thousands. The experimental screening workload is extremely large, time-consuming, and resource-intensive. Furthermore, the relationship between the final performance of the material and its chemical composition and microstructure is highly complex and nonlinear. Human intuition struggles to grasp this, while machine learning excels at uncovering these hidden and complex patterns from massive amounts of data. Through predictive models, given a pre-set formulation and one of the three aforementioned stimulus conditions, the model can directly predict its performance, pre-determine the feasibility of the formulation, significantly reduce the consumption of expensive chemical reagents and experimental equipment time, avoid experimental pitfalls, and provide more accurate predictions. Therefore, a machine learning-based method for predicting the stimulus-response performance of polyeutectic fluorescent gels is proposed. Summary of the Invention

[0005] This invention overcomes the shortcomings of the prior art and provides a method for predicting the stimulus response performance of polyeutectic fluorescent gels based on machine learning.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention provides a method for predicting the stimulus-response performance of polyeutectic fluorescent gels based on machine learning, comprising the following steps: We collected characteristic data such as chemical composition and microstructure of polyeutectic fluorescent gels, and performed data preprocessing to construct a characteristic dataset of polyeutectic fluorescent gels. Data analysis was performed using the existing dataset of polyeutectic fluorescent gel properties to construct the microstructure features of the polyeutectic fluorescent gel. In the updated polyeutectic fluorescent gel property dataset, subset selection and data feature preprocessing were performed to obtain the polyeutectic fluorescent gel training dataset. By combining the training dataset of polyeutectic fluorescent gels, a machine learning model is constructed to simulate stimulation of polyeutectic fluorescent gels, and a Bayesian optimization model is used to fine-tune the hyperparameters of the model, outputting the stimulation response performance of polyeutectic fluorescent gels.

[0007] Furthermore, in a preferred embodiment of the present invention, the process of collecting characteristic data such as the chemical composition and microstructure of the polyeutectic fluorescent gel, and performing data preprocessing to construct a polyeutectic fluorescent gel characteristic dataset, specifically involves: Obtain the polyeutectic fluorescent gel for which the stimulus response performance needs to be predicted, and label it as the target polyeutectic fluorescent gel. At the same time, introduce the preparation formula of the target polyeutectic fluorescent gel during preparation and label it as the target preparation formula. The target eutectic fluorescent gel is composed of a polymerizable eutectic solvent and a combination of fluorophores. A terminal computing device is introduced, and the target preparation formula is analyzed within the terminal computing device. The formula analysis involves converting the SMILES strings of all hydrogen bond acceptors, hydrogen bond donors and fluorophores of the target polyeutectic fluorescent gel in the target preparation formula into 3D molecular descriptors, and batch calculating the feature information of different 3D molecular descriptors in the terminal computing device. Among them, 3D molecular descriptors include geometric descriptors, electronic descriptors, topological descriptors, and physicochemical descriptors; The interaction energies of the ternary complexes of all hydrogen bond acceptors, hydrogen bond donors and fluorophores in the target preparation formulation are calculated and calibrated as ternary complex interaction energies. The method for calculating the interaction energy of the ternary complex is to introduce the density functional tight-binding algorithm to quickly calculate the total energy and isolated energy of the ternary complex of hydrogen bond acceptor, hydrogen bond donor and fluorophore, and output the interaction energy of the ternary complex by the formula: total energy of ternary complex of hydrogen bond acceptor, hydrogen bond donor and fluorophore - total isolated energy of ternary complex. By combining the interaction energy of ternary complexes and the characteristic information of different 3D molecular descriptors, characteristic data of polyeutectic fluorescent gels are constructed. Multi-source data fusion is performed on the characteristic data of all polyeutectic fluorescent gels, and interpolation processing of the characteristic data of polyeutectic fluorescent gels is performed during the multi-source data fusion process to construct a characteristic dataset of polyeutectic fluorescent gels.

[0008] Furthermore, in a preferred embodiment of the present invention, the step of combining the constructed polyeutectic fluorescent gel characteristic dataset for data analysis and constructing the microstructural features of the polyeutectic fluorescent gel specifically involves: Within the polyeutectic fluorescent gel property dataset, based on the feature information of the 3D molecular descriptor, the molar ratios of hydrogen bond acceptors, hydrogen bond donors, and fluorophores are extracted. At the same time, based on the interaction energy of ternary complexes, the interaction energies of different binary complexes are calculated. Based on the molar ratio of hydrogen bond acceptor, hydrogen bond donor and fluorophore, the interaction weight of each binary complex is calculated, and the molar weighted global binding energy is calculated by combining the interaction energies of different binary complexes through a weighted calculation formula. The molar weighted global binding energy is used as a stability benchmark descriptor for the polyeutectic fluorescent gel. The analysis is performed to determine whether the molar weighted global binding energy is less than the standard value. If it is, the global strength of the hydrogen bond network of the polyeutectic fluorescent gel is deemed to be qualified. If not, the interaction energies of all binary complexes are aggregated to construct a list of binary complex interaction energies. The standard deviation and absolute mean of the list are calculated to generate the heterogeneity index of the hydrogen bond network of the polyeutectic fluorescent gel. The hydrogen bond network heterogeneity index of the polyeutectic fluorescent gel is a characteristic of the hydrogen bond network strength and a characteristic of the microstructure of the polyeutectic fluorescent gel. Gel network features were constructed for the polyeutectic fluorescent gel, and combined with hydrogen bond network strength features, all microstructural features of the polyeutectic fluorescent gel were generated.

[0009] Furthermore, in a preferred embodiment of the present invention, the construction of gel network features of the polyeutectic fluorescent gel, combined with hydrogen bond network strength features, to generate all the microstructural features of the polyeutectic fluorescent gel, specifically involves: Based on the binary complex interaction energy list, functional group type identification and functionality calculation of polyeutectic fluorescent gels were performed, and the probability of cross-linking between different functional groups was calculated. Among them, the cross-linking of functional groups involves calculating the probability of hydrogen bond formation between hydrogen bond acceptors and hydrogen bond donors, and analyzing the interaction energy of all hydrogen bonds in the polyeutectic fluorescent gel property dataset according to the binary complex interaction energy list, and screening hydrogen bonds with interaction energies lower than a predetermined value as target hydrogen bonds, while calculating the effective cross-linking density of the target hydrogen bonds. For all target hydrogen bonds, the total crosslinking density is calculated, where the total crosslinking density reflects the strength characteristics of the hydrogen bond network. The total crosslinking density of the polyeutectic fluorescent gel and the heterogeneity index of the hydrogen bond network of the polyeutectic fluorescent gel are combined to obtain all the microstructural features of the polyeutectic fluorescent gel, which are then used to update the polyeutectic fluorescent gel property dataset to obtain the updated polyeutectic fluorescent gel property dataset.

[0010] Furthermore, in a preferred embodiment of the present invention, the step of performing subset selection and data feature preprocessing in the updated polyeutectic fluorescent gel characteristic dataset to obtain the polyeutectic fluorescent gel training dataset specifically involves: For updating the polyeutectic fluorescent gel property dataset, it is divided into different data subsets based on different structural features within the dataset, and then further divided into training and test sets. A recursive feature elimination and cross-validation approach is introduced to preprocess the updated polyeutectic fluorescent gel property dataset. The data preprocessing method for updating the polyeutectic fluorescent gel property dataset is to recursively initialize the training set using a random forest approach and remove the feature with the lowest proportion among all features in the training set. Among them, the feature with the lowest proportion is that if the training subset is a subset corresponding to the microstructure feature, then the target hydrogen bond with the highest density in the effective cross-linking density of the target hydrogen bond is obtained, and the total cross-linking density is reconstructed until the effective cross-linking density of the target hydrogen bond is maintained within the predetermined threshold, then the recursive loop initialization is stopped. If the training subset consists of characteristic data of polyeutectic fluorescent gels, then a characteristic data proportion analysis is performed, and the characteristic data with the lowest proportion is removed until all characteristic data are maintained within the predetermined threshold, at which point the recursive loop initialization is stopped. Cross-validation tests were performed on all training subsets initialized by the recursive loop using the test set. All training subsets whose average scores from the cross-validation tests remained at the standard value were selected as output subsets and fused to obtain the polyeutectic fluorescent gel training dataset.

[0011] Furthermore, in a preferred embodiment of the present invention, the step of constructing a machine learning model based on the polyelastomer fluorescent gel training dataset to simulate stimulation by the polyelastomer fluorescent gel, and then combining this with a Bayesian optimization model to fine-tune the model hyperparameters and output the stimulation response performance of the polyelastomer fluorescent gel, specifically involves: A machine learning model capable of mathematically encoding stimulus-response mechanisms is introduced and labeled as the target machine learning model. A polyeutectic fluorescent gel training dataset is then imported into the target machine learning model. Mathematical encoding of three major categories of stimulus response mechanisms is performed within the target machine learning model. These three categories of stimulus response mechanisms include physical stimuli, chemical stimuli, and biological stimuli. Specifically, for physical stimuli, mathematical encoding of temperature, light, force, electricity, and humidity is performed within the target machine learning model; for chemical stimuli, mathematical encoding of pH, redox reactions, and ions is performed within the target machine learning model; and for biological stimuli, mathematical encoding of enzymes, antigens, and DNA is performed within the target machine learning model. Based on the mathematical encoding of three major types of stimulus response mechanisms, the target machine learning model is updated, and the updated machine learning model is output. The model is updated by introducing a gating network to train a classifier in the target machine learning model. The classifier divides the target machine learning model into three layers according to the stimulus response mechanism type encoded by mathematics, which respectively dominate the simulated physical, chemical and biological stimulus processing. By updating the machine learning model and combining it with a Bayesian optimization network, the training dataset of polyeutectic fluorescent gels is subjected to three major categories of stimulus response mechanisms to simulate stimulation, and the stimulus response performance of polyeutectic fluorescent gels is output.

[0012] Furthermore, in a preferred embodiment of the present invention, the step of updating the machine learning model, combining it with a Bayesian optimization network, simulating three major categories of stimulus response mechanisms on the polyeutectic fluorescent gel training dataset, and outputting the stimulus response performance of the polyeutectic fluorescent gel, specifically involves: In updating the machine learning model, three major types of stimulus response mechanisms were applied to the polyeutectic fluorescent gel training dataset, and the output data of the updated machine learning model were collected under different stimulus response mechanisms. The single application values ​​of the three major types of stimulus response mechanisms were different. The updated machine learning model outputs data including fluorescence intensity change values ​​and wavelength shift values. Within the updated machine learning model, a simulated stimulus report is generated, which describes the fluorescence intensity change values ​​and wavelength shift values ​​generated for different three major types of stimulus response mechanisms under different single-action values. A Bayesian optimization algorithm was introduced to perform forward and reverse tuning of the gating network in updating the machine learning model, resulting in a standard machine learning model. The simulated stimulus report generated by the standard machine learning model was then used as an evaluation report of the stimulus response performance of the polyeutectic fluorescent gel. Among them, positive tuning involves calling the corresponding response layer to simulate stimulation based on the type of stimulus response mechanism input; Reverse optimization involves pre-setting the target stimulus response performance of the polyeutectic fluorescent gel in the updated machine learning model, and combining it with a Bayesian optimization algorithm to screen single-action values ​​that meet the requirements in different stimulus response mechanisms.

[0013] A second aspect of this invention also provides a machine learning-based stimulus response performance prediction system for polyeutectic fluorescent gels. This system integrates a high-performance computing architecture and a data storage module, including a non-volatile memory consisting of an ECC-calibrated DDR4 RDIMM memory module and an NVMe solid-state storage array using 3D NAND flash memory, and a multi-core processor based on the Zen4 microarchitecture. The memory contains a stimulus response performance prediction method program with a prediction engine. When the program is executed in parallel via a superscalar pipeline execution unit within the processor, the following steps are achieved: We collected characteristic data such as chemical composition and microstructure of polyeutectic fluorescent gels, and performed data preprocessing to construct a characteristic dataset of polyeutectic fluorescent gels. Data analysis was performed using the existing dataset of polyeutectic fluorescent gel properties to construct the microstructure features of the polyeutectic fluorescent gel. In the updated polyeutectic fluorescent gel property dataset, subset selection and data feature preprocessing were performed to obtain the polyeutectic fluorescent gel training dataset. By combining the training dataset of polyeutectic fluorescent gels, a machine learning model is constructed to simulate stimulation of polyeutectic fluorescent gels, and a Bayesian optimization model is used to fine-tune the hyperparameters of the model, outputting the stimulation response performance of polyeutectic fluorescent gels.

[0014] This invention addresses the technical deficiencies in the prior art and offers the following advantages: It collects, constructs, and preprocesses characteristic data such as chemical composition and microstructure of polyeutectic fluorescent gels, and predicts the physical, chemical, and biological stimulus response performance of the polyeutectic fluorescent gels based on the preprocessed training dataset. Compared to traditional methods, this invention achieves advantages such as reduced cost, experimental safety, and faster results output. Furthermore, it offers advantages over traditional methods in terms of high-precision quantitative prediction and handling of high-dimensional nonlinear problems. Attached Figure Description

[0015] 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 some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0016] Figure 1 A flowchart is shown for a machine learning-based method for predicting the stimulus-response performance of polyeutectic fluorescent gels; Figure 2A flowchart illustrating the method for constructing microstructure features of polyeutectic fluorescent gels is shown. Figure 3 A program view of a machine learning-based system for predicting the stimulus-response performance of polyeutectic fluorescent gels is shown. Detailed Implementation

[0017] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0019] Figure 1 A flowchart illustrating a machine learning-based method for predicting the stimulus-response performance of polyeutectic fluorescent gels is shown, including the following steps: S102: Collect characteristic data such as chemical composition and microstructure of polyeutectic fluorescent gel, perform data preprocessing, and construct a characteristic dataset of polyeutectic fluorescent gel; S104: Combine the constructed dataset of polyeutectic fluorescent gel properties for data analysis, and construct the microstructure features of polyeutectic fluorescent gel; S106: Perform subset selection and data feature preprocessing on the updated polyeutectic fluorescent gel property dataset to obtain the polyeutectic fluorescent gel training dataset; S108: Using the training dataset of polyelastomer fluorescent gels, a machine learning model is constructed to simulate stimulation of polyelastomer fluorescent gels, and a Bayesian optimization model is used to fine-tune the hyperparameters of the model, outputting the stimulation response performance of polyelastomer fluorescent gels.

[0020] Furthermore, in a preferred embodiment of the present invention, the process of collecting characteristic data such as the chemical composition and microstructure of the polyeutectic fluorescent gel, and performing data preprocessing to construct a polyeutectic fluorescent gel characteristic dataset, specifically involves: Obtain the polyeutectic fluorescent gel for which the stimulus response performance needs to be predicted, and label it as the target polyeutectic fluorescent gel. At the same time, introduce the preparation formula of the target polyeutectic fluorescent gel during preparation and label it as the target preparation formula. The target eutectic fluorescent gel is composed of a polymerizable eutectic solvent and a combination of fluorophores. A terminal computing device is introduced, and the target preparation formula is analyzed within the terminal computing device. The formula analysis involves converting the SMILES strings of all hydrogen bond acceptors, hydrogen bond donors and fluorophores of the target polyeutectic fluorescent gel in the target preparation formula into 3D molecular descriptors, and batch calculating the feature information of different 3D molecular descriptors in the terminal computing device. Among them, 3D molecular descriptors include geometric descriptors, electronic descriptors, topological descriptors, and physicochemical descriptors; The interaction energies of the ternary complexes of all hydrogen bond acceptors, hydrogen bond donors and fluorophores in the target preparation formulation are calculated and calibrated as ternary complex interaction energies. The method for calculating the interaction energy of the ternary complex is to introduce the density functional tight-binding algorithm to quickly calculate the total energy and isolated energy of the ternary complex of hydrogen bond acceptor, hydrogen bond donor and fluorophore, and output the interaction energy of the ternary complex by the formula: total energy of ternary complex of hydrogen bond acceptor, hydrogen bond donor and fluorophore - total isolated energy of ternary complex. By combining the interaction energy of ternary complexes and the characteristic information of different 3D molecular descriptors, characteristic data of polyeutectic fluorescent gels are constructed. Multi-source data fusion is performed on the characteristic data of all polyeutectic fluorescent gels, and interpolation processing of the characteristic data of polyeutectic fluorescent gels is performed during the multi-source data fusion process to construct a characteristic dataset of polyeutectic fluorescent gels.

[0021] It should be noted that the target preparation formulation reflects the process and materials used in the fabrication of the polyeutectic fluorescent gel. Analysis of the formulation reveals that it contains all the hydrogen bond acceptors, hydrogen bond donors, and fluorophores of the target polyeutectic fluorescent gel. Specifically, hydrogen bond acceptors include, but are not limited to, the chemical information of choline chloride and betaine; hydrogen bond donors include, but are not limited to, the chemical information of urea, glycerol, and citric acid. The fluorophore is an organic fluorescent dye. The SMILES strings of the hydrogen bond acceptors, hydrogen bond donors, and fluorophores are converted into 3D molecular descriptors. The purpose is to generate microscopic-level chemophysical descriptors for the formulation, which are crucial features for performance prediction. 3D molecular descriptors include geometric descriptors, electronic descriptors, topological descriptors, and physicochemical descriptors. Geometric descriptors include molecular weight and van der Waals volume; electronic descriptors represent photoelectric properties; topological descriptors represent connectivity indices and the degree of molecular branching; and physicochemical descriptors represent polar surface area, etc. Different 3D molecular descriptors describe different characteristics. Simultaneously, it is necessary to rapidly estimate the hydrogen bond interaction energy, that is, the interaction energy of the complex of the three chemical substances. The larger the value, the more stable the hydrogen bond network is, which is likely to be a key microscopic factor affecting gel stability and fluorescence responsiveness.

[0022] Furthermore, in a preferred embodiment of the present invention, the step of performing subset selection and data feature preprocessing in the updated polyeutectic fluorescent gel characteristic dataset to obtain the polyeutectic fluorescent gel training dataset specifically involves: For updating the polyeutectic fluorescent gel property dataset, it is divided into different data subsets based on different structural features within the dataset, and then further divided into training and test sets. A recursive feature elimination and cross-validation approach is introduced to preprocess the updated polyeutectic fluorescent gel property dataset. The data preprocessing method for updating the polyeutectic fluorescent gel property dataset is to recursively initialize the training set using a random forest approach and remove the feature with the lowest proportion among all features in the training set. Among them, the feature with the lowest proportion is that if the training subset is a subset corresponding to the microstructure feature, then the target hydrogen bond with the highest density in the effective cross-linking density of the target hydrogen bond is obtained, and the total cross-linking density is reconstructed until the effective cross-linking density of the target hydrogen bond is maintained within the predetermined threshold, then the recursive loop initialization is stopped. If the training subset consists of characteristic data of polyeutectic fluorescent gels, then a characteristic data proportion analysis is performed, and the characteristic data with the lowest proportion is removed until all characteristic data are maintained within the predetermined threshold, at which point the recursive loop initialization is stopped. Cross-validation tests were performed on all training subsets initialized by the recursive loop using the test set. All training subsets whose average scores from the cross-validation tests remained at the standard value were selected as output subsets and fused to obtain the polyeutectic fluorescent gel training dataset.

[0023] It should be noted that the updated polyeutectic fluorescent gel property dataset contains microstructural features and other characteristics of the updated polyeutectic fluorescent gel. These need to be combined for analysis to select the most relevant and effective subset from the numerous features to avoid the curse of dimensionality and overfitting. First, subset selection is performed. Based on different structural features within the dataset, it is divided into different data subsets, and then into training and test sets. The test set is completely isolated throughout the training and tuning process and is used only for the final evaluation of the model's generalization ability. Recursive feature elimination and cross-validation is a powerful and stable feature selection method. It finds the optimal feature combination by recursively removing the least important features while using cross-validation to evaluate the model performance of different feature subsets. In other words, it finds the most suitable data subset. First, the training set is recursively initialized using a random forest, eliminating features with the lowest proportion among all features in the training set to prevent overfitting. Different recursive initialization methods are used for different subsets. After each feature elimination step, cross-validation is performed on the test set to obtain the feature subset with the highest average score in the cross-validation. This subset contains the features with the strongest predictive power and the lowest redundancy, and is used for model training.

[0024] Furthermore, in a preferred embodiment of the present invention, the step of constructing a machine learning model based on the polyelastomer fluorescent gel training dataset to simulate stimulation by the polyelastomer fluorescent gel, and then combining this with a Bayesian optimization model to fine-tune the model hyperparameters and output the stimulation response performance of the polyelastomer fluorescent gel, specifically involves: A machine learning model capable of mathematically encoding stimulus-response mechanisms is introduced and labeled as the target machine learning model. A polyeutectic fluorescent gel training dataset is then imported into the target machine learning model. Mathematical encoding of three major categories of stimulus response mechanisms is performed within the target machine learning model. These three categories of stimulus response mechanisms include physical stimuli, chemical stimuli, and biological stimuli. Specifically, for physical stimuli, mathematical encoding of temperature, light, force, electricity, and humidity is performed within the target machine learning model; for chemical stimuli, mathematical encoding of pH, redox reactions, and ions is performed within the target machine learning model; and for biological stimuli, mathematical encoding of enzymes, antigens, and DNA is performed within the target machine learning model. Based on the mathematical encoding of three major types of stimulus response mechanisms, the target machine learning model is updated, and the updated machine learning model is output. The model is updated by introducing a gating network to train a classifier in the target machine learning model. The classifier divides the target machine learning model into three layers according to the stimulus response mechanism type encoded by mathematics, which respectively dominate the simulated physical, chemical and biological stimulus processing. By updating the machine learning model and combining it with a Bayesian optimization network, the training dataset of polyeutectic fluorescent gels is subjected to three major categories of stimulus response mechanisms to simulate stimulation, and the stimulus response performance of polyeutectic fluorescent gels is output.

[0025] It should be noted that the stimulus-response performance prediction of polyeutectic fluorescent gels is performed by simulating stimuli, since a training dataset for polyeutectic fluorescent gels already exists. The stimulus response of polyeutectic fluorescent gels is essentially based on external stimuli disrupting / reconstructing the gel network framework, hydrogen bonds, hydrophobic interactions, ion coordination, and π-π stacking, thereby altering the microenvironment of the fluorophore and ultimately manifesting as tunable changes in fluorescence intensity, wavelength, and other signals. Currently, the types of responsive stimuli mainly revolve around the "structural sensitivity of the polyeutectic system" and the "environmental dependence of the fluorophore," thus being mainly divided into three categories: physical, chemical, and biological stimuli. Mathematical encoding quantifies the physicochemical effects of these three categories of stimuli into numerical features understandable to machine learning models.

[0026] For physical stimuli, such as aggregation-induced emission (AIE) type polyeutectic fluorescent gels, hydrogen bonds break upon heating, causing the gel network to disintegrate and the AIE fluorophores to change from a "confined aggregated state" to a "freely dispersed state," resulting in fluorescence quenching. Conversely, upon cooling, hydrogen bonds recombine, the fluorophores re-aggregate, and the fluorescence is restored. For example, polyeutectic fluorescent gels containing photosensitive groups or photosensitive fluorophores such as coumarin, spiropyran, and azobenzene can exhibit changes in their fluorescence response due to photoinduced electron transfer, isomerization, and photocrosslinking reactions.

[0027] In response to chemical stimuli, polyeutectic fluorescent gels containing quantum dots, rhodamine, and pH-sensitive groups such as amino, carboxyl, and pyridyl groups exhibit phenomena such as gel network contraction and protonation of fluorophores under acidic conditions, thus promoting fluorescence activation; while under alkaline conditions, gel swelling and deprotonation lead to redshift in fluorescence. Another example is that anions in the polyeutectic gel system (such as chloride and carboxyl ions) can react with Na+. + Ca 2+ Fe 3+ Isocational coordination, or its internal hydrogen bond donors such as -OH and -NH, and F - CN - SO4 2- When anions combine, they disrupt the hydrogen bond network or form complexes, causing changes in the energy level of the fluorophore, which in turn quenches or enhances fluorescence.

[0028] In response to biological stimuli, such as eutectic gels containing ester bonds, peptide bonds, etc., which are specifically recognized by enzymes, the gel network can be broken by enzymatic hydrolysis, allowing fluorophores to be released and thus enhancing fluorescence.

[0029] Gated networks are used to train classifiers, which encode and assign input stimulus types to corresponding layers for predicting stimulus response performance.

[0030] Furthermore, in a preferred embodiment of the present invention, the step of updating the machine learning model, combining it with a Bayesian optimization network, simulating three major categories of stimulus response mechanisms on the polyeutectic fluorescent gel training dataset, and outputting the stimulus response performance of the polyeutectic fluorescent gel, specifically involves: In updating the machine learning model, three major types of stimulus response mechanisms were applied to the polyeutectic fluorescent gel training dataset, and the output data of the updated machine learning model were collected under different stimulus response mechanisms. The single application values ​​of the three major types of stimulus response mechanisms were different. The updated machine learning model outputs data including fluorescence intensity change values ​​and wavelength shift values. Within the updated machine learning model, a simulated stimulus report is generated, which describes the fluorescence intensity change values ​​and wavelength shift values ​​generated for different three major types of stimulus response mechanisms under different single-action values. A Bayesian optimization algorithm was introduced to perform forward and reverse tuning of the gating network in updating the machine learning model, resulting in a standard machine learning model. The simulated stimulus report generated by the standard machine learning model was then used as an evaluation report of the stimulus response performance of the polyeutectic fluorescent gel. Among them, positive tuning involves calling the corresponding response layer to simulate stimulation based on the type of stimulus response mechanism input; Reverse optimization involves pre-setting the target stimulus response performance of the polyeutectic fluorescent gel in the updated machine learning model, and combining it with a Bayesian optimization algorithm to screen single-action values ​​that meet the requirements in different stimulus response mechanisms.

[0031] It should be noted that by applying the three major stimulus-response mechanisms separately to the training dataset of the polyeutectic fluorescent gel for simulated stimulation, generating fluorescence intensity changes and wavelength shifts, and then combining the values ​​of the three major stimulus-response mechanisms at different single-action times, a simulated stimulus report can be generated. For example, a choline chloride (ChCl)-acrylic acid-tetraphenylethylene sulfonate sodium derivative. The tetraphenylethylene sulfonate sodium derivative is a fluorophore; its fluorescence is weak in the dispersed state, but significantly enhanced in a hydrophobic environment or in an aggregated state. Its sulfonate group increases water solubility and has ionic interactions with ChCl, making it extremely sensitive to changes in environmental polarity, and thus it can be used as a fluorophore for aggregation-induced emission. The stimulation condition is: under stimulation with pH decreasing from 7 to 4, the predicted fluorescence enhancement is 180% ± 15%. The purpose of introducing a Bayesian optimization algorithm is to optimize and accurately simulate stimulation, improve the accuracy of the report, and output the stimulus-response performance of the polyeutectic fluorescent gel.

[0032] Figure 2 A flowchart illustrating a method for constructing microstructural features of polyeutectic fluorescent gels is shown, including the following steps: S202: Combine the constructed dataset of polyeutectic fluorescent gel properties for data analysis to construct the microstructure features of polyeutectic fluorescent gel; S204: Construct gel network features for polyeutectic fluorescent gels and combine them with hydrogen bond network strength features to generate all microstructural features of polyeutectic fluorescent gels.

[0033] Furthermore, in a preferred embodiment of the present invention, the step of combining the constructed polyeutectic fluorescent gel characteristic dataset for data analysis and constructing the microstructural features of the polyeutectic fluorescent gel specifically involves: Within the polyeutectic fluorescent gel property dataset, based on the feature information of the 3D molecular descriptor, the molar ratios of hydrogen bond acceptors, hydrogen bond donors, and fluorophores are extracted. At the same time, based on the interaction energy of ternary complexes, the interaction energies of different binary complexes are calculated. Based on the molar ratio of hydrogen bond acceptor, hydrogen bond donor and fluorophore, the interaction weight of each binary complex is calculated, and the molar weighted global binding energy is calculated by combining the interaction energies of different binary complexes through a weighted calculation formula. The molar weighted global binding energy is used as a stability benchmark descriptor for the polyeutectic fluorescent gel. The analysis is performed to determine whether the molar weighted global binding energy is less than the standard value. If it is, the global strength of the hydrogen bond network of the polyeutectic fluorescent gel is deemed to be qualified. If not, the interaction energies of all binary complexes are aggregated to construct a list of binary complex interaction energies. The standard deviation and absolute mean of the list are calculated to generate the heterogeneity index of the hydrogen bond network of the polyeutectic fluorescent gel. The hydrogen bond network heterogeneity index of the polyeutectic fluorescent gel is a characteristic of the hydrogen bond network strength and a characteristic of the microstructure of the polyeutectic fluorescent gel. Gel network features were constructed for the polyeutectic fluorescent gel, and combined with hydrogen bond network strength features, all microstructural features of the polyeutectic fluorescent gel were generated.

[0034] It's important to note that the prediction of stimulus-response performance is achieved by analyzing the microstructure of polyeutectic fluorescent gels. The microstructural features elevate the interactions between multiple molecules to the "network" level, quantifying the hydrogen-bonding environment of the entire gel matrix. First, considering the molar proportions of the components in the gel, a weighted average interaction energy is calculated. Since ternary interaction energies already exist, binary interaction energies need to be analyzed to better represent the overall environment. The hydrogen-bonding network heterogeneity index of the polyeutectic fluorescent gel describes the heterogeneity of the hydrogen-bonding network. A homogeneous network may lead to a more consistent response; while a highly heterogeneous network may have weaker, more sensitive responses. Therefore, analyzing the hydrogen-bonding network heterogeneity index of the polyeutectic fluorescent gel generates all the microstructural features of the polyeutectic fluorescent gel. Simultaneously, it quantifies the ratio of the dispersion between different interaction energies to the average binding strength. A high heterogeneity index indicates the simultaneous presence of very strong and very weak interactions in the system, which may predict that under external stimuli, the network will preferentially break and recombine from the weaker bonds. The purpose of the hydrogen-bonding network heterogeneity index is to find this point.

[0035] Furthermore, in a preferred embodiment of the present invention, the construction of gel network features of the polyeutectic fluorescent gel, combined with hydrogen bond network strength features, to generate all the microstructural features of the polyeutectic fluorescent gel, specifically involves: Based on the binary complex interaction energy list, functional group type identification and functionality calculation of polyeutectic fluorescent gels were performed, and the probability of cross-linking between different functional groups was calculated. Among them, the cross-linking of functional groups involves calculating the probability of hydrogen bond formation between hydrogen bond acceptors and hydrogen bond donors, and analyzing the interaction energy of all hydrogen bonds in the polyeutectic fluorescent gel property dataset according to the binary complex interaction energy list, and screening hydrogen bonds with interaction energies lower than a predetermined value as target hydrogen bonds, while calculating the effective cross-linking density of the target hydrogen bonds. For all target hydrogen bonds, the total crosslinking density is calculated, where the total crosslinking density reflects the strength characteristics of the hydrogen bond network. The total crosslinking density of the polyeutectic fluorescent gel and the heterogeneity index of the hydrogen bond network of the polyeutectic fluorescent gel are combined to obtain all the microstructural features of the polyeutectic fluorescent gel, which are then used to update the polyeutectic fluorescent gel property dataset to obtain the updated polyeutectic fluorescent gel property dataset.

[0036] It is important to note that after outputting the heterogeneity index of the hydrogen bond network, it is also necessary to transform the chemical composition into features describing the physical structure of the three-dimensional gel network, particularly the crosslinking density, to acquire the strength characteristics of the hydrogen bond network. First, the functional groups in each component molecule that can participate in the formation of the coach network are identified and quantified, namely hydrogen bond acceptor groups and hydrogen bond donor groups, and the functionality, i.e., the number of molecules in the groups, is calculated. The probability of crosslinking between different functional groups is calculated to determine the probability of hydrogen bond formation between hydrogen bond acceptors and hydrogen bond donors. Different hydrogen bond acceptors and donors have pairing probabilities, and the formation of hydrogen bonds after pairing also has a probability. The lower the interaction energy of the hydrogen bonds, the higher the pairing probability. Finally, by combining all target hydrogen bonds, the total crosslinking density is calculated as a microstructural feature, resulting in an updated dataset of polyeutectic fluorescent gel properties.

[0037] like Figure 3 As shown, the second aspect of the present invention also provides a stimulus response performance prediction system for polyeutectic fluorescent gels based on machine learning. The stimulus response performance prediction system integrates a high-performance computing architecture and a data storage module, including a non-volatile memory consisting of a DDR4 RDIMM memory module with ECC verification and an NVMe solid-state storage array using 3D NAND flash memory, and a multi-core processor based on the Zen4 microarchitecture. The memory contains a stimulus response performance prediction method program with a prediction engine. When the program is executed in parallel through a superscalar pipeline execution unit within the processor, the following steps are implemented: We collected characteristic data such as chemical composition and microstructure of polyeutectic fluorescent gels, and performed data preprocessing to construct a characteristic dataset of polyeutectic fluorescent gels. Data analysis was performed using the existing dataset of polyeutectic fluorescent gel properties to construct the microstructure features of the polyeutectic fluorescent gel. In the updated polyeutectic fluorescent gel property dataset, subset selection and data feature preprocessing were performed to obtain the polyeutectic fluorescent gel training dataset. By combining the training dataset of polyeutectic fluorescent gels, a machine learning model is constructed to simulate stimulation of polyeutectic fluorescent gels, and a Bayesian optimization model is used to fine-tune the hyperparameters of the model, outputting the stimulation response performance of polyeutectic fluorescent gels.

[0038] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting the stimuli-responsive performance of a machine learning-based poly-eutectic fluorescent gel, characterized by, The method comprises the following steps: Collecting the chemical composition and microstructure of the poly-eutectic fluorescent gel, and performing data preprocessing to construct a poly-eutectic fluorescent gel characteristic data set; Combining the constructed poly-eutectic fluorescent gel characteristic data set to perform data analysis and construct the microstructure characteristics of the poly-eutectic fluorescent gel; In the update of the poly-eutectic fluorescent gel characteristic data set, the subset is screened and the data characteristics are preprocessed to obtain a poly-eutectic fluorescent gel training data set; Combining the poly-eutectic fluorescent gel training data set, a machine learning model is constructed to simulate the stimulation of the poly-eutectic fluorescent gel, and a Bayesian optimization model is used to optimize the model hyperparameters, and the stimulation response performance of the poly-eutectic fluorescent gel is output.

2. The method for predicting the stimuli-responsive performance of a machine learning-based deep eutectic fluorescent gel according to claim 1, characterized by, The method comprises the following steps: Obtain the poly-eutectic fluorescent gel whose stimulation response performance needs to be predicted, and label it as the target poly-eutectic fluorescent gel. At the same time, introduce the preparation formula of the target poly-eutectic fluorescent gel during preparation, and label it as the target preparation formula. The target poly-eutectic fluorescent gel is composed of polymerizable low-eutectic solvent and fluorescent group. Introduce a terminal computing device, and perform formula analysis on the target preparation formula in the terminal computing device. The formula analysis is to convert the SMILES strings of all hydrogen bond acceptors, hydrogen bond donors and fluorescent groups in the target preparation formula into 3D molecular descriptors, and calculate the feature information of different 3D molecular descriptors in the terminal computing device. The 3D molecular descriptors include geometric descriptors, electronic descriptors, topological descriptors and physical and chemical descriptors. Calculate the interaction energy of the ternary complex of all hydrogen bond acceptors, hydrogen bond donors and fluorescent groups in the target preparation formula, and label it as the ternary complex interaction energy. The method for calculating the ternary complex interaction energy is to introduce the density functional tight binding algorithm to quickly calculate the total energy and isolated energy of the ternary complex of hydrogen bond acceptors, hydrogen bond donors and fluorescent groups, and output the ternary complex interaction energy through the formula total energy of the ternary complex of hydrogen bond acceptors, hydrogen bond donors and fluorescent groups-isolated energy. Combine the ternary complex interaction energy and the feature information of different 3D molecular descriptors to construct the characteristic data of the poly-eutectic fluorescent gel, perform multi-source data fusion on the characteristic data of all poly-eutectic fluorescent gels, and perform interpolation processing on the characteristic data of the poly-eutectic fluorescent gel during the multi-source data fusion process to construct the poly-eutectic fluorescent gel characteristic data set.

3. The method for predicting the stimuli-responsive performance of a machine learning-based deep eutectic fluorescent gel according to claim 1, characterized by, The method comprises the following steps: In the poly-eutectic fluorescent gel characteristic data set, according to the feature information of the 3D molecular descriptor, the molar ratio of the hydrogen bond acceptor, the hydrogen bond donor and the fluorescent group is extracted, and according to the ternary complex interaction energy, the interaction energy of different binary complexes is calculated. According to the molar ratio of hydrogen bond acceptor, hydrogen bond donor and fluorophore, the interaction weight of each binary complex is calculated, and the molar weighted global binding energy is calculated by the weighted calculation formula combined with the interaction energy of different binary complexes; Wherein, the molar weighted global binding energy is used as the stability benchmark descriptor of the poly-eutectic fluorescent gel, and whether the molar weighted global binding energy is less than the standard value is analyzed, if yes, it is judged that the hydrogen bond network global strength of the poly-eutectic fluorescent gel is qualified, if not, the interaction energy of all binary complexes is collected to construct the binary complex interaction energy list, and the standard deviation and absolute value mean of the list are calculated, thereby generating the hydrogen bond network heterogeneity index of the poly-eutectic fluorescent gel; Wherein, the hydrogen bond network heterogeneity index of the poly-eutectic fluorescent gel is the hydrogen bond network strength characteristic, which is the microstructure characteristic of the poly-eutectic fluorescent gel; The gel network characteristics of the poly-eutectic fluorescent gel are constructed, and combined with the hydrogen bond network strength characteristic, all the microstructure characteristics of the poly-eutectic fluorescent gel are generated.

4. The method for predicting the stimuli-responsive performance of a machine learning-based deep eutectic fluorescent gel according to claim 3, characterized by, The gel network characteristics of the poly-eutectic fluorescent gel are constructed, and combined with the hydrogen bond network strength characteristic, all the microstructure characteristics of the poly-eutectic fluorescent gel are generated, specifically: Based on the binary complex interaction energy list, the functional group type recognition and the functionality calculation of different functional groups are carried out, and the probability of crosslinking between different functional groups is calculated; Wherein, the functional group crosslinking is to calculate the probability of hydrogen bond formation between hydrogen bond acceptor and hydrogen bond donor, and according to the binary complex interaction energy list in the poly-eutectic fluorescent gel characteristic data set, the interaction energy of all hydrogen bonds is analyzed, and the hydrogen bonds with interaction energy lower than the predetermined value are selected as the target hydrogen bonds, and the effective crosslinking density of the target hydrogen bonds is calculated; For all target hydrogen bonds, the total crosslinking density is calculated, wherein the total crosslinking density reflects the hydrogen bond network strength characteristic, and the total crosslinking density of the poly-eutectic fluorescent gel is combined with the hydrogen bond network heterogeneity index of the poly-eutectic fluorescent gel to obtain all the microstructure characteristics of the poly-eutectic fluorescent gel, which is used for data updating of the poly-eutectic fluorescent gel characteristic data set to obtain the updated poly-eutectic fluorescent gel characteristic data set.

5. The method for predicting the stimuli-responsive performance of a machine learning-based deep eutectic fluorescent gel according to claim 1, characterized by, The subset screening and data feature preprocessing are carried out in the updated poly-eutectic fluorescent gel characteristic data set to obtain the poly-eutectic fluorescent gel training data set, specifically: For the updated poly-eutectic fluorescent gel characteristic data set, different data subsets are divided according to different structure characteristics in the data set, and different data subsets are divided into training set and test set; Recursive feature elimination and cross-validation method is introduced for data preprocessing of the updated poly-eutectic fluorescent gel characteristic data set; Wherein, the data preprocessing method of the updated poly-eutectic fluorescent gel characteristic data set is to remove the feature with the lowest proportion in all features of the training set by recursively initializing the training set through random forest method; The feature with the lowest proportion is that if the training subset is a subset corresponding to the microstructure feature, the target hydrogen bond with the maximum effective cross-linking density in the target hydrogen bond is obtained, and the total cross-linking density is reconstructed until the effective cross-linking density of the target hydrogen bond is maintained within a predetermined threshold, and the recursive loop initialization is stopped; If the training subset is a characteristic data of the poly-eutectic fluorescent gel, the proportion analysis of the characteristic data is performed, and the characteristic data with the lowest proportion is removed until all characteristic data is maintained within a predetermined threshold, and the recursive loop initialization is stopped; All training subsets initialized by the recursive loop are cross-validated by the test set, and all training subsets with an average score maintained at a standard value output by the cross-validation test are selected as output subsets and are fused to obtain a poly-eutectic fluorescent gel training dataset.

6. The method for predicting the stimuli-responsive performance of a machine learning-based deep eutectic fluorescent gel according to claim 1, characterized by, The poly-eutectic fluorescent gel training dataset is combined to construct a machine learning model for simulating the stimulation of the poly-eutectic fluorescent gel, and a Bayesian optimization model is combined for model hyperparameter tuning to output the stimulation response performance of the poly-eutectic fluorescent gel, specifically: A machine learning model capable of mathematical coding of the stimulation response mechanism is introduced and calibrated as a target machine learning model, and the poly-eutectic fluorescent gel training dataset is imported into the target machine learning model; Mathematical coding of three major types of stimulation response mechanisms is performed in the target machine learning model, including physical stimulation, chemical stimulation, and biological stimulation; Among them, for physical stimulation, mathematical coding of temperature, light, force, electricity, and humidity is performed in the target machine learning model; for chemical stimulation, mathematical coding of pH, oxidation-reduction, and ions is performed in the target machine learning model; for biological stimulation, mathematical coding of enzymes, antigens, and DNA is performed in the target machine learning model; Based on the mathematical coding of the three major types of stimulation response mechanisms, the target machine learning model is updated to output an updated machine learning model; The model update is performed by introducing a gating network to train a classifier in the target machine learning model, which divides the target machine learning model into three layers according to the type of mathematical coding of the stimulation response mechanism, and respectively dominates the simulation of physical, chemical, and biological stimulation; Through the updated machine learning model, the poly-eutectic fluorescent gel training dataset is simulated for the three major types of stimulation response mechanisms by combining the Bayesian optimization network, and the stimulation response performance of the poly-eutectic fluorescent gel is output.

7. The method for predicting the stimuli-responsive performance of a machine learning-based deep eutectic fluorescent gel according to claim 6, characterized by, Through the updated machine learning model, the poly-eutectic fluorescent gel training dataset is simulated for the three major types of stimulation response mechanisms by combining the Bayesian optimization network, and the stimulation response performance of the poly-eutectic fluorescent gel is output. In the updated machine learning model, the three major types of stimulation response mechanisms are respectively applied to the poly-eutectic fluorescent gel training dataset, and the output data of the updated machine learning model under different stimulation response mechanisms is respectively collected, and the number of single actions of different three major types of stimulation response mechanisms is different. The output data of the updated machine learning model includes the fluorescence intensity change value and the wavelength shift value, and a simulation stimulation report is generated in the updated machine learning model; wherein the simulation stimulation report describes the corresponding generated fluorescence intensity change value and wavelength shift value under different single action values of different three major types of stimulation response mechanisms; The Bayesian optimization algorithm is introduced to forward and backward optimize the gating network in the updated machine learning model to obtain a standard machine learning model, and the simulation stimulation report generated by the standard machine learning model is used as a stimulation response performance evaluation report of the eutectic fluorescent gel; The forward optimization is to simulate the stimulation according to the input stimulation response mechanism type and call the corresponding reaction layer; The backward optimization is to preset the target stimulation response performance of the eutectic fluorescent gel in the updated machine learning model, and combine the Bayesian optimization algorithm to screen the single action value that meets the requirements in different stimulation response mechanisms.

8. A system for predicting the stimuli-responsive performance of a machine learning-based poly-eutectic fluorescent gel, characterized by, The stimulation response performance prediction system integrates a high-performance computing architecture and a data storage module, including a non-volatile memory composed of an ECC-verified DDR4 RDIMM memory module and a 3D NAND flash-based NVMe solid-state storage array, and a multi-core processor based on Zen4 microarchitecture; the memory is solidly disposed with a stimulation response performance prediction method program with a prediction engine, when the program is executed in parallel by the superscalar pipeline execution unit in the processor, the stimulation response performance prediction steps of any one of claims 1-7 are realized.

Images