Machine learning driven collagen-based dna hydrogel optimization prediction method and apparatus

Abstract

This invention relates to the field of machine learning, providing a machine learning-driven method and apparatus for optimizing and predicting collagen-based DNA hydrogels. The method involves: collecting performance parameters of hydrogels at different collagen-to-DNA molar ratios; predicting and screening the optimal molar ratio range of collagen to DNA based on these performance parameters; controlling the crosslinking density of collagen and DNA by adjusting the concentration of the crosslinking agent and the crosslinking time according to the optimal molar ratio range; verifying and predicting the construction of a dual-crosslinked network structure with a specific pore size distribution using a synergistic approach of chemical and physical crosslinking; where chemical crosslinking is a covalent crosslinked network formed by the Schiff base reaction of collagen, and physical crosslinking is a self-assembled crosslinked network formed by complementary base pairing of DNA; and using the dual-crosslinked network structure to regulate the pore size distribution of the hydrogel, optimizing the hydrogel's colonization stability, glycyrrhizic acid sustained-release performance, and Wnt / β-catenin signaling pathway activation efficiency in the spinal cord injury microenvironment.

Description

Technical Field

[0001] This invention relates to the field of machine learning, and more particularly to a machine learning-driven method and apparatus for optimizing and predicting collagen-based DNA hydrogels. Background Technology

[0002] Spinal cord injury (SCI) is a severe central nervous system trauma that often leads to irreversible motor and sensory dysfunction. Its repair is limited by the weak regenerative capacity of the central nervous system and the intense inflammatory response, glial scarring, and high expression of inhibitory molecules (such as Nogo-A and CSPGs) in the injury microenvironment. Therefore, developing bioactive scaffold materials that can mimic the extracellular matrix (ECM), provide structural support, regulate key signaling pathways, and achieve sustained drug release has become a research hotspot.

[0003] Hydrogels are widely used in tissue engineering due to their high water content, good biocompatibility, and tunable mechanical properties. Collagen, as a major component of ECM, possesses excellent cell adhesion and biodegradability, making it an ideal substrate for neural repair scaffolds. Related technologies have attempted to load anti-inflammatory small molecules such as glycyrrhizic acid into collagen hydrogels to improve the microenvironment; however, traditional collagen hydrogels generally suffer from low mechanical strength, uncontrollable degradation, and drug burst release, making it difficult to stably colonize within the complex spinal cord microenvironment.

[0004] Currently, formulation optimization for such complex multi-component hydrogels still primarily relies on trial and error. This involves fixing some variables, repeatedly adjusting individual parameters, conducting numerous experiments, and then evaluating their mechanical properties, degradation behavior, drug release profiles, and biological effects through in vitro / in vivo testing. However, existing technologies suffer from the following drawbacks: firstly, long experimental cycles, high costs, and low throughput; secondly, difficulty in revealing nonlinear interactions between multiple parameters; and thirdly, monitoring key performance indicators relies on specially designed experimental measurements, resulting in high investment and low formulation validation efficiency. Especially in spinal cord injury repair scenarios, formulation materials must simultaneously meet multiple objectives such as mechanical compatibility, long-term residence, controlled drug release, and signal modulation. Traditional empirical development models are no longer sufficient to meet the actual R&D needs of material design, and cannot reduce experimental costs or improve material development efficiency through pre-experimental validation. Summary of the Invention

[0005] This invention provides a method and apparatus for optimizing collagen-based DNA hydrogels, aiming to solve the technical problem that related technologies mainly rely on empirical trial-and-error methods to optimize complex hydrogel formulations, which are difficult to meet the comprehensive performance requirements of materials for spinal cord injury repair.

[0006] In a first aspect, embodiments of the present invention provide a method for optimizing collagen-based DNA hydrogels, comprising: The performance parameters of hydrogels with different collagen to DNA molar ratios were collected. Based on the aforementioned performance parameters, the optimal molar ratio range of collagen to DNA was predicted and screened. Based on the optimal molar ratio range, the cross-linking density of collagen and DNA is controlled by adjusting the concentration of the cross-linking agent and the cross-linking time. A dual cross-linked network structure with a specific pore size distribution is verified and predicted by using a synergistic approach of chemical cross-linking and physical cross-linking. The chemical cross-linking is a covalent cross-linked network formed by collagen through Schiff base reaction, and the physical cross-linking is a self-assembled cross-linked network formed by DNA through complementary base pairing. The pore size distribution of the hydrogel was controlled by the aforementioned double cross-linked network structure to optimize the hydrogel's colonization stability, glycyrrhizic acid sustained-release performance, and Wnt / β-catenin signaling pathway activation efficiency in the spinal cord injury microenvironment.

[0007] In a second aspect, embodiments of the present invention provide a collagen-based DNA hydrogel optimization device, the device being used to implement a machine learning-driven collagen-based DNA hydrogel optimization prediction method as described in the first aspect or any embodiment of the present invention.

[0008] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device including a processor and a memory for storing a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the machine learning-driven collagen-based DNA hydrogel optimization and prediction method described in the first aspect or any embodiment of the present invention.

[0009] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer software program, which, when executed by a processor, implements the machine learning-driven collagen-based DNA hydrogel optimization and prediction method described in the first aspect or any embodiment of the present invention.

[0010] This invention provides a machine learning-driven method and apparatus for optimizing and predicting collagen-based DNA hydrogels. In this invention, performance parameters of the hydrogel are collected at different collagen-to-DNA molar ratios. Then, based on these performance parameters, the optimal molar ratio range of collagen to DNA is predicted and screened. Next, according to the optimal molar ratio range, the cross-linking density of collagen and DNA is controlled by adjusting the concentration of the cross-linking agent and the cross-linking time. A dual-crosslinked network structure with a specific pore size distribution is verified and predicted using a synergistic approach of chemical and physical cross-linking. The chemical cross-linking is a covalent cross-linked network formed by the Schiff base reaction of collagen, and the physical cross-linking is a self-assembled cross-linked network formed by complementary base pairing of DNA. Finally, the pore size distribution of the hydrogel is controlled using the dual-crosslinked network structure to optimize the hydrogel's colonization stability, glycyrrhizic acid sustained-release performance, and Wnt / β-catenin signaling pathway activation efficiency in the spinal cord injury microenvironment.

[0011] In this invention, by constructing a quantitative mapping relationship between the performance parameters of collagen and DNA molar ratio and a dynamic prediction model of crosslinking parameters and pore size distribution, the multidimensional performance of hydrogels in the spinal cord injury microenvironment can be accurately predicted. Based on this prediction result, the optimal material composition and structural parameters are determined, thereby improving the adaptability of hydrogel design to spinal cord repair needs and avoiding performance shortcomings. Furthermore, the crosslinking density and pore size distribution of the hydrogel are precisely controlled through a dual crosslinking network synergistic construction strategy, and the material structural parameters are corrected by a multidimensional performance verification and feedback optimization mechanism, ultimately forming a hydrogel with an ideal pore size distribution. This achieves precise adaptation and functional synergy of the hydrogel in the spinal cord injury microenvironment, improving the material's colonization stability, glycyrrhizic acid sustained-release accuracy, and Wnt / β-catenin signaling pathway activation efficiency. This helps to further optimize the performance of the hydrogel in complex spinal cord injury environments and reduces the experimental cost and time investment in material development. Attached Figure Description

[0012] Figure 1 A schematic flowchart of a collagen-based DNA hydrogel optimization method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating a scenario for optimizing a collagen-based DNA hydrogel according to an embodiment of the present invention. Detailed Implementation

[0013] This invention provides a method and apparatus for optimizing collagen-based DNA hydrogels. The method can be applied to terminal devices, such as mobile terminals, mobile phones, virtual reality devices, tablets, laptops, desktop computers, wearable devices, and other electronic devices. It can also be a server or server cluster connected to a cloud service device. This connection can be implemented through hardware circuitry or a communication module.

[0014] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Figure 1 This is a flowchart illustrating a machine learning-driven method for optimizing and predicting collagen-based DNA hydrogels, provided in an embodiment of the present invention. Please refer to... Figure 1 The method includes the following steps: Step S101: Collect the performance parameters of hydrogels with different molar ratios of collagen to DNA.

[0015] In this embodiment of the invention, the molar ratio of collagen to DNA is used as the main control variable to construct a set of collagen-based DNA hydrogel samples with continuously adjustable composition. Under uniform experimental conditions, their structural, mechanical, degradation, drug release, and signal regulation-related performance parameters are simultaneously collected, thereby forming a basic performance dataset that can be used for subsequent prediction and optimization. This step provides a data foundation for establishing a quantitative mapping relationship between material composition and multidimensional functions.

[0016] In this embodiment of the invention, multiple gradient ranges of the molar ratio of collagen to DNA are first established, such as 1:0, 1:0.05, 1:0.1, 1:0.2, 1:0.3, and 1:0.5. Hydrogel samples are prepared while maintaining consistent overall collagen concentration, solution pH, temperature, and gelation conditions. This method ensures that performance differences between different samples mainly stem from variations in the molar ratio of collagen to DNA.

[0017] In terms of pore size distribution parameter acquisition, the gelled hydrogel was fixed and freeze-dried to obtain cross-sectional images of its internal three-dimensional porous structure. By statistically analyzing the pore size, pore size range, and distribution uniformity, pore size distribution data reflecting the density of the hydrogel network were obtained. For example, the experiment showed that as the DNA molar ratio increased, the number of physical cross-linking points formed by complementary DNA base pairing increased, and the hydrogel pore size gradually changed from large and uneven to small and uniform, providing a basis for subsequent precise pore size control.

[0018] During the mechanical property acquisition process, small-amplitude, repeatable shear deformations were applied to the hydrogel samples, and the changes in their energy storage capacity during deformation were recorded to obtain storage modulus data. This parameter is used to characterize the elastic support capacity of the hydrogel and its degree of matching with the mechanical environment of spinal cord tissue. In practical examples, when the collagen to DNA molar ratio is in a moderate range, the hydrogel exhibits a stable storage modulus close to that of natural spinal cord tissue, which is beneficial to cell adhesion and axonal growth.

[0019] For degradation rate acquisition, hydrogels with different formulations were cultured in simulated body fluids or environments containing collagenase, and the changes in hydrogel mass or volume were recorded at multiple time points to construct degradation behavior curves over time. For example, samples with a high DNA molar ratio exhibited a significantly slower degradation process and longer structure retention time due to their denser double cross-linked network.

[0020] In the process of collecting data on the sustained-release performance of glycyrrhizic acid, glycyrrhizic acid was uniformly loaded into the network structure during the hydrogel preparation stage. The sample was then placed in the release medium, and the supernatant was periodically collected and its glycyrrhizic acid content was measured to obtain a curve showing the cumulative release over time. In practical applications, samples with a low DNA molar ratio exhibit the problem of excessively rapid initial release. Hydrogels with appropriate DNA content can significantly reduce burst release, achieving a smoother and more controllable release behavior.

[0021] In terms of collecting biological functional parameters, neural cells were seeded on or inside hydrogels with different molar ratios and cultured. The spatial distribution of β-catenin within the cells was detected, and the proportion entering the cell nucleus was statistically analyzed as a quantitative indicator of the activation level of the Wnt / β-catenin signaling pathway. For example, compared with pure collagen hydrogels, DNA-containing composite hydrogels can increase the nuclear translocation rate of β-catenin within a specific molar ratio range, suggesting its advantage in regulating neural repair-related signals.

[0022] Finally, the collected data on pore size distribution, storage modulus, degradation rate, glycyrrhizic acid sustained-release curve, and β-catenin nuclear translocation rate were uniformly sorted and screened to extract key performance parameters that contribute significantly to the overall performance of the material and have good stability. These parameters were then used for performance prediction and optimal range screening of the collagen to DNA molar ratio in subsequent steps.

[0023] In step S101, performance parameters of hydrogels at different collagen-to-DNA molar ratios are collected; key performance parameter data are extracted from these parameters. Optionally, the key performance parameter data includes pore size distribution, storage modulus, degradation rate, glycyrrhizic acid sustained-release curve, and β-catenin nuclear translocation rate.

[0024] For example, when performing step S101, a set of quantifiable and comparable specific performance data can be obtained around different collagen to DNA molar ratios for subsequent extraction and analysis of key performance parameters.

[0025] Regarding pore size distribution data, taking hydrogels with collagen:DNA molar ratios of 1:0, 1:0.1, 1:0.2, and 1:0.4 as examples, statistical results of pore size can be obtained by structural characterizing the cross-sections of the lyophilized samples. The average pore size of pure collagen hydrogels is typically concentrated in the range of 120–180 μm and has a relatively wide distribution. When the DNA molar ratio is 0.1, the average pore size decreases to approximately 80–120 μm, and the pore size distribution begins to converge. When the DNA molar ratio is further increased to 0.2–0.4, the average pore size can be stabilized in the range of 40–70 μm, and the pore size uniformity is significantly improved. This data directly reflects the role of DNA in regulating network density and pore structure after cross-linking.

[0026] Regarding storage modulus data, the elastic energy storage capacity can be obtained by recording the mechanical response of hydrogels with different formulations under small strain conditions. For example, the storage modulus of a collagen:DNA hydrogel with a ratio of 1:0 is approximately 80–120 Pa. When the DNA molar ratio is 0.1, the storage modulus increases to approximately 200–300 Pa; under a ratio of 1:0.2, it can be further increased to 400–600 Pa. However, when the DNA molar ratio is too high (e.g., 1:0.4), although the storage modulus can exceed 800 Pa, the material flexibility decreases. This data is used to evaluate the degree of matching between the hydrogel and the mechanical environment of spinal cord tissue.

[0027] Regarding degradation rate data, hydrogels with different molar ratios were placed in simulated body fluids or collagenase-containing systems, and the percentage change in mass over time was recorded. For example, pure collagen hydrogels experienced a mass loss of over 60% within 14 days. A sample with a collagen:DNA ratio of 1:0.1 experienced a mass loss of approximately 40% within 14 days; at a molar ratio of 1:0.2, the mass loss was controlled to around 25% within 14 days. Under a 1:0.4 ratio, over 60% of the initial mass was retained after 28 days. This data demonstrates the regulatory effect of the double cross-linked network on the long-term retention capacity of the material.

[0028] Regarding the glycyrrhizic acid sustained-release curve data, the cumulative release ratio at different time points can be obtained. For example, pure collagen hydrogel can release approximately 45% of glycyrrhizic acid within the first 24 hours, exhibiting a significant burst release; when the collagen:DNA ratio is 1:0.1, the release ratio decreases to approximately 30% after 24 hours; under the condition of 1:0.2, the release ratio can be further reduced to 18%–22% after 24 hours, and a near-linear cumulative release is observed over 14 days; however, when the DNA content is too high, the overall release rate slows down significantly, resulting in insufficient release in the later stages. This curve data is used to screen for the optimal release range that balances initial anti-inflammatory effects with long-term maintenance.

[0029] Regarding β-catenin nuclear translocation data, the degree of signaling pathway activation can be quantified by statistically analyzing the proportion of β-catenin entering the nucleus in nerve-related cells cultured in different hydrogels. For example, in pure collagen hydrogels, the β-catenin nuclear translocation rate is approximately 15%–20%; it increases to approximately 30% at a collagen:DNA ratio of 1:0.1; and reaches 45%–55% at a molar ratio of 1:0.2. While maintaining a high level at a 1:0.4 ratio, cell spreading and activity begin to decline. This data reflects the influence of material composition on the efficiency of Wnt / β-catenin signaling regulation.

[0030] By collecting and organizing the above-mentioned specific data, key performance parameters such as pore size range concentration, suitable energy storage modulus range, controllable degradation rate, stable drug release characteristics, and high β-catenin nuclear translocation rate can be extracted, providing reliable data support for the subsequent prediction and screening of the optimal molar ratio range of collagen and DNA.

[0031] Step S102: Predict and screen the optimal molar ratio range of collagen to DNA based on performance parameters.

[0032] As an optional embodiment, step S102, predicting and screening the optimal molar ratio range of collagen to DNA based on the performance parameters, includes: establishing a mapping relationship between different collagen and DNA molar ratios and the key performance parameters; calculating the predicted performance values ​​and confidence intervals of untested parameter combinations based on the mapping relationship; prioritizing parameter combinations whose predicted performance values ​​reach a set performance value and whose confidence intervals are greater than the set confidence interval width for simulation prediction; iteratively updating the simulation prediction results to the mapping relationship, and then jumping back to the step of calculating the predicted performance values ​​and confidence intervals of untested parameter combinations based on the mapping relationship, until the convergence condition is met and the iterative updating of the mapping relationship is stopped, and the optimal molar ratio range of collagen to DNA is screened from the mapping relationship.

[0033] In step S102, based on the collected collagen to DNA molar ratio-performance parameter data, a quantitative correlation model between material composition and multidimensional performance is constructed. By performing preliminary prediction and uncertainty assessment on the performance of untested formulations, the search space is gradually narrowed, thereby efficiently screening out the optimal molar ratio range that meets the needs of spinal cord injury repair, avoiding blind experimental verification of all parameter combinations.

[0034] Specifically, the experimental data obtained in step S101 are first organized and standardized. Using the molar ratio of collagen to DNA as the input variable, and key performance parameters such as pore size distribution characteristics, storage modulus, degradation rate, glycyrrhizic acid sustained-release characteristic parameters, and β-catenin nuclear translocation rate as output indicators, a mapping relationship is established between the two. This mapping relationship reflects the changing trends and interrelationships of various performance indicators under different molar ratios, such as the combined effect of increased DNA content on pore size reduction, modulus enhancement, and delayed drug release. Based on this, for molar ratio combinations that have not yet been validated in physical experiments (e.g., 1:0.15, 1:0.18, 1:0.22, etc.), the corresponding predicted performance values ​​are calculated based on the above mapping relationship. Simultaneously, the fluctuation range of each predicted result is given to characterize the uncertainty of the prediction. This fluctuation range is derived from the performance dispersion of existing samples within adjacent molar ratio intervals and reflects the reliability of the current prediction results. For example, for a molar ratio of 1:0.18, the average pore size can be predicted to be approximately 60–70 μm, and the energy storage modulus to be between 450 and 550 Pa, with a relatively clear upper and lower fluctuation range.

[0035] Subsequently, based on the comprehensive requirements of materials for spinal cord injury repair, multiple performance threshold ranges were pre-defined. These included pore size within a range conducive to cell migration, storage modulus close to that of spinal cord tissue, degradation cycle covering the critical repair window, no significant burst release of glycyrrhizic acid, and a β-catenin nuclear translocation rate higher than the baseline value. Molar ratio combinations that generally met the above-defined performance requirements and had a relatively clear predicted fluctuation range were prioritized as key candidates for the simulation prediction stage. During the simulation prediction process, multiple rounds of virtual evaluation were conducted on the selected candidate molar ratio combinations to simulate their performance changes under different crosslinking densities, culture times, and microenvironmental conditions, determining whether their overall performance was stable and whether there were any potential shortcomings. For example, if a certain molar ratio performed excellently under ideal conditions, but its drug release performance fluctuated significantly with slight changes in crosslinking density, its stability evaluation would be correspondingly lower.

[0036] Subsequently, the performance results obtained from the above simulation predictions are fed back into the original mapping relationship as new supplementary data to update the mapping relationship, making its description of the collagen to DNA molar ratio and performance more precise. After the update, the performance prediction and uncertainty assessment are performed again for the remaining untested molar ratio combinations, and candidate combinations that meet the conditions are screened again, forming an iterative cycle. When the variation of the prediction results gradually decreases after several consecutive updates, and the optimal performance range tends to stabilize within the adjacent molar ratio range, the convergence condition is determined, and the iterative update stops. Finally, a continuous collagen to DNA molar ratio range, such as 1:0.15 to 1:0.25, is screened from the stable mapping relationship. Hydrogels within this range can simultaneously meet the multiple requirements of spinal cord injury repair in terms of pore structure, mechanical properties, degradation behavior, glycyrrhizic acid sustained-release characteristics, and Wnt / β-catenin signal activation efficiency, thus serving as the output of the optimal collagen to DNA molar ratio range.

[0037] Further optionally, in the above embodiments, parameter combinations whose predicted performance values ​​reach the set performance values ​​and whose confidence intervals are greater than the set confidence interval width are preferentially selected for simulation prediction, including: establishing a material feature transformation matrix based on a pre-constructed basic material performance knowledge base; extracting and transferring structural performance relationship parameters related to pore size regulation in the basic material based on the material feature transformation matrix to obtain preliminary performance prediction values ​​of candidate collagen to DNA molar ratio and crosslinking parameter combinations under spinal cord injury microenvironment conditions, wherein the preliminary performance prediction values ​​include pore size distribution prediction values, degradation kinetics prediction values, and molecular diffusion coefficient prediction values; performing structural similarity weighting and uncertainty quantification correction on the preliminary performance prediction values ​​to generate a corrected performance score containing the predicted mean and standard deviation. The uncertainty quantification correction process is based on a dynamic weight allocation of similarity coefficients between the base material and the target material in terms of crosslinking chemical bond type, network topology, and intermolecular force characteristics. Multidimensional performance coupling analysis is performed based on the preliminary performance prediction values ​​to identify synergistic optimization regions for colonization stability, glycyrrhizic acid cumulative release rate, and β-catenin nuclear translocation efficiency. Based on the uncertainty assessment results of the synergistic optimization regions and corresponding parameter combinations, candidate parameter combinations located within the synergistic optimization regions and with uncertainties higher than a preset threshold are preferentially selected as simulation prediction results to be updated in the mapping relationship. High-uncertainty parameter combinations have higher priority than low-uncertainty but slightly better-performing parameter combinations to accelerate the optimization convergence process.

[0038] In this embodiment of the invention, a basic material performance knowledge base is first pre-constructed to provide transferable prior information for predicting the performance of collagen-DNA hydrogels. This knowledge base compiles experimental response data for hydrogels with various known cross-linking mechanisms under different component ratios and cross-linking conditions. This includes at least a dataset of hydrogels forming covalently cross-linked networks via Schiff base reactions, and a dataset of hydrogels forming self-assembled physically cross-linked networks via complementary nucleic acid base pairing. When constructing the material feature transformation matrix, the quantifiable physicochemical properties of the aforementioned basic materials are uniformly characterized and normalized, enabling them to be mapped to the feature space of collagen-DNA hydrogels. These physicochemical properties include, but are not limited to: the cross-linking free energy variation range, network porosity and pore size statistical characteristics, kinetic parameters of swelling rate over time, and the diffusion coefficient of small molecule drugs in the network.

[0039] For example, in the basic materials performance knowledge base, the pore size distribution of a certain type of Schiff base crosslinked hydrogel changes from a broadly dispersed distribution to a concentrated distribution when the crosslinking free energy decreases to a specific range; while in nucleic acid self-assembly crosslinking systems, when the base pairing density reaches a certain level, the network porosity and molecular diffusion coefficient exhibit a stable functional relationship. By uniformly encoding the above structure-performance relationships and mapping them to the adjustable molar ratio, crosslinking agent concentration, and crosslinking time dimensions in the collagen-DNA system, a material feature transformation matrix is ​​established to describe the transformation relationship between basic material features and target material features.

[0040] Optionally, the physicochemical properties parameters in the pre-constructed basic material performance knowledge base are mapped to the feature space of collagen and DNA hydrogels to establish a material feature transformation matrix. The physicochemical properties parameters include crosslinking free energy, network porosity, swelling kinetics parameters, and drug diffusion coefficient. The basic material performance knowledge base contains performance response data of hydrogels with various known crosslinking mechanisms at different component ratios. The known crosslinking mechanism hydrogels include at least independent datasets of Schiff base crosslinking mechanisms and base self-assembly crosslinking mechanisms.

[0041] It is worth noting that, in this embodiment of the invention, in order to realize the effective transfer and reuse of basic material experience to the collagen-DNA hydrogel system, the physicochemical property parameters in the pre-constructed basic material performance knowledge base are first mapped to the feature space of collagen and DNA hydrogel, thereby establishing a material feature transformation matrix, which is used to uniformly describe the intrinsic correlation between material structure and performance under different cross-linking mechanisms.

[0042] In practice, the basic materials performance knowledge base collects experimental data on various hydrogel systems under different component ratios and crosslinking conditions. These hydrogel systems include at least hydrogels that form covalently crosslinked networks via Schiff base reactions and hydrogels that form self-assembled crosslinked networks via complementary pairing of nucleic acid bases. For each of these different crosslinking mechanisms, quantifiable physicochemical property parameters are extracted, including the characteristics of the system's free energy change during crosslinking, the porosity distribution of the network after gelation, the swelling kinetics of the material in solution, and the diffusion capacity of small molecule drugs within the network. During parameter mapping, the physicochemical property parameters of different material systems are first standardized and dimensionally uniformly processed to ensure comparability at the same characteristic scale. For example, the crosslinking free energy changes measured under different crosslinking systems are converted into dimensionless indices reflecting network stability; network porosity and swelling ratio are converted into structural parameters characterizing spatial openness; and the drug diffusion coefficient is converted into a functional parameter reflecting network permeability. Subsequently, the standardized physicochemical properties were mapped to the compositional and structural characteristics of collagen-DNA hydrogels according to their effects on network structure formation and pore size regulation, establishing a correspondence between these parameters and adjustable parameters such as the molar ratio of collagen to DNA, crosslinking agent concentration, crosslinking time, and physical crosslinking density. This method allows for the construction of a material feature transformation matrix to describe the transformation relationships between the characteristics of the basic crosslinking system, collagen-DNA, and dual-crosslinking systems. For example, in the Schiff base crosslinking mechanism dataset, when the crosslinking free energy is within a specific range, the corresponding network porosity and swelling rate exhibit a stable correlation trend; in the base self-assembly crosslinking mechanism dataset, a clear response relationship exists between base pairing density and drug diffusion coefficient. By transferring these structure-performance relationships to the collagen-DNA hydrogel system using the material feature transformation matrix, it can be used to infer the pore structure stability, swelling behavior, and drug diffusion characteristics that a dual-crosslinked network may form under given molar ratios and crosslinking parameters.

[0043] The material characteristic transformation matrix established in the above manner provides reliable prior constraints and quantitative references for the structural prediction, pore size distribution control and functional performance optimization of untested crosslinking parameter combinations in subsequent steps, which helps to reduce experimental trial and error costs and improve the overall efficiency of double crosslinked hydrogel design.

[0044] After obtaining the material feature transformation matrix, this embodiment of the invention extracts and transfers structural performance parameters highly correlated with pore size regulation from the base material based on this matrix, forming an initial parameter combination prediction framework for collagen-DNA hydrogels. This framework uses the molar ratio of collagen to DNA, cross-linking agent concentration, and cross-linking time as input variables, and pore size distribution, degradation kinetics, and molecular diffusion capacity as output variables. In specific implementation, for candidate collagen-to-DNA molar ratios and cross-linking parameter combinations, the material feature transformation matrix is ​​first used to map their structural responses under the microenvironmental conditions of spinal cord injury, obtaining preliminary performance prediction values. Among these, the predicted pore size distribution value reflects the spatial structural stability of the network under physiological conditions. The predicted degradation kinetics value describes the mass or volume decay trend of the material over time in inflammatory and enzymatic environments. The predicted molecular diffusion coefficient value characterizes the migration ability of small molecules such as glycyrrhizic acid within the hydrogel. For example, for collagen, a combination of parameters such as a DNA molar ratio of 1:0.18 and a moderate concentration of cross-linking agent can predict that it will form a network structure with pore sizes concentrated in the 50-70 μm range in the microenvironment of spinal cord injury, with a degradation rate covering the critical repair window of 2-4 weeks, while the molecular diffusion capacity is within a reasonable range that can both avoid burst release and ensure continuous release.

[0045] After obtaining the preliminary performance predictions, this embodiment of the invention further introduces a structural similarity weighting and uncertainty quantification correction step to improve the reliability of the prediction results. Specifically, for candidate parameter combinations, their crosslinking chemical bond types, network topology characteristics, and intermolecular force characteristics are compared one by one with various hydrogels in the basic material performance knowledge base, and their structural similarity coefficients are calculated. Based on this, dynamic weighting is applied to prediction information from different sources according to the structural similarity coefficients. When a candidate combination is highly similar to the basic material in terms of crosslinking mechanism and network structure, its prediction result weight is increased. When the similarity is low, the fluctuation range of the prediction result is correspondingly expanded. In this way, the uncertainty of the preliminary performance predictions is quantified and corrected, generating a corrected performance distribution that simultaneously includes the prediction mean and standard deviation.

[0046] For example, for a candidate combination that simultaneously contains Schiff base covalent crosslinking and DNA self-assembly physical crosslinking features, if it has moderate similarity to two independent datasets in the basic materials, its pore size prediction mean may remain stable while the standard deviation is relatively large, reflecting that this parameter combination still has high exploratory value under the current data conditions.

[0047] After obtaining the corrected performance distribution, this embodiment of the invention performs multidimensional performance coupling analysis on pore structure, degradation behavior, and molecular diffusion ability, and further correlates biological effect indicators such as colonization stability, glycyrrhizic acid cumulative release rate, and β-catenin nuclear translocation efficiency to identify parameter regions that synergistically enhance overall performance. Specifically, by analyzing the matching relationship between pore size distribution and degradation rate, structural regions conducive to long-term material residence are screened. Combining molecular diffusion ability and release kinetics, the cumulative release characteristics of glycyrrhizic acid within the critical repair time window are determined. Simultaneously, the above-mentioned structural and drug release characteristics are correlated with the predicted trend of β-catenin nuclear translocation efficiency to identify synergistically optimized regions that can simultaneously promote structural stability and signaling pathway activation. For example, the analysis results show that within the range of collagen:DNA molar ratio of 1:0.15 to 1:0.25, the material can achieve slow drug release and significantly improve β-catenin nuclear translocation efficiency while maintaining suitable pore size and degradation rate; this range is identified as a potential synergistically optimized region.

[0048] After determining the collaborative optimization region, this embodiment of the invention does not select parameters solely based on the absolute superiority or inferiority of their predictive performance, but rather prioritizes them by combining uncertainty assessment results. Specifically, within the collaborative optimization region, candidate parameter combinations with predictive uncertainty higher than a preset threshold are preferentially selected and updated into the mapping relationship as simulated prediction results. Further optionally, the collaborative optimization region is defined as a subset of the parameter space that simultaneously meets clinical requirement thresholds in at least two performance dimensions and is not lower than the baseline level in the third performance dimension. This strategy allows parameter combinations with high uncertainty but greater potential for improvement to receive higher priority when performance is similar, thereby accelerating the convergence of the overall optimization process. For example, if a certain molar ratio combination is slightly inferior to the current optimal point in predictive performance, but its standard deviation is large, indicating that performance improvement may be achieved through further structural adjustments, then this combination will be prioritized for the next round of updates.

[0049] The above methods gradually guide the composition and structural parameters of collagen DNA hydrogels toward the optimal range of high performance and low trial-and-error cost, thereby achieving efficient prediction and optimization of complex spinal cord injury repair materials.

[0050] Further optionally, in the above embodiments, based on the material feature transformation matrix, extracting and transferring structural performance parameters related to pore size regulation in the base material to obtain preliminary performance prediction values ​​of the candidate collagen to DNA molar ratio and crosslinking parameter combination under spinal cord injury microenvironment conditions includes: extracting and transferring structural performance parameters related to pore size regulation in the base material based on the material feature transformation matrix to form an initial parameter combination prediction model; using historical experimental data of collagen and DNA hydrogels, locally calibrating the parameters related to the spinal cord microenvironment response in the initial parameter combination prediction model, while retaining the global parameters related to the basic crosslinking principle unchanged; and calculating the preliminary performance prediction values ​​of the candidate collagen to DNA molar ratio and crosslinking parameter combination under spinal cord injury microenvironment conditions based on the initial parameter combination prediction model.

[0051] In this embodiment of the invention, to pre-evaluate the structural and functional performance of candidate collagen-DNA molar ratios and crosslinking parameter combinations in the spinal cord injury microenvironment before experiments, a parameter combination prediction model is constructed based on an established material feature transformation matrix. This model extracts and transfers structure-performance relationship parameters highly correlated with pore size regulation from the basic materials, and obtains preliminary performance prediction values. Specifically, firstly, when constructing the initial parameter combination prediction model, the material feature transformation matrix is ​​used as the core constraint to extract structure-performance relationships directly related to pore size formation mechanisms from the basic material performance knowledge base. These relationships mainly include the correlation between crosslinking free energy changes and network density, the correspondence between network porosity and fiber connection point density, and the influence paths of swelling kinetics parameters and molecular diffusion capacity on pore size stability. These relationships are not directly replicated from the basic materials, but rather transferred to the collagen-DNA dual crosslinking system in the form of structural response trends and relative change directions, forming an initial prediction framework that can describe the changes in crosslinking conditions and pore size structure response.

[0052] In terms of model structure, the initial parameter combination prediction model adopts a hierarchical design. The first layer uses the molar ratio of collagen to DNA, the concentration of cross-linking agent, and the cross-linking time as input variables to characterize the composition and cross-linking strength of the dual cross-linking system. The second layer is the structural response layer, used to characterize key intermediate features in the network formation process under the synergistic effect of chemical and physical cross-linking, such as effective cross-linking density, fiber aggregation degree, and variable pore space ratio. The third layer is the performance output layer, used to provide preliminary performance prediction results such as pore size distribution range, pore size concentration, degradation trend, and molecular diffusion ability. The layers are connected through the mapping relationship defined in the material feature transformation matrix, thereby ensuring that the prediction results conform to the basic physical and chemical laws of cross-linking.

[0053] Subsequently, using historical experimental data from collagen and DNA hydrogels, local parameters related to the microenvironmental response to spinal cord injury in the initial parameter combination prediction model were calibrated. Specifically, parameters related to degradation rate, swelling behavior, and pore stability in the model were locally modified to address specific conditions of spinal cord injury, such as the presence of inflammatory factors, enhanced enzyme activity, and changes in the body fluid environment. However, fundamental parameters describing the formation of Schiff base covalent crosslinks and the complementary pairing mechanism of DNA bases remained unchanged to ensure that the model consistently adheres to the basic crosslinking principles of material formation. This approach allows the prediction model to both inherit fundamental material knowledge and possess adaptability to target application scenarios.

[0054] After completing local calibration, based on the initial parameter combination prediction model, the molar ratio of candidate collagen to DNA and the cross-linking parameter combination are calculated to obtain preliminary performance predictions under the microenvironment conditions of spinal cord injury. For example, for a parameter combination with a collagen:DNA molar ratio of 1:0.2, a moderate cross-linking agent concentration, and a cross-linking time of 40 minutes, the model can predict that the pore size of the formed double cross-linked network is mainly concentrated in the 100–300 nm range, with high uniformity of pore size distribution, while maintaining a stable network structure and moderate molecular diffusion ability in an inflammatory environment.

[0055] The initial parameter combination prediction model constructed and applied in the above manner enables the prediction of the structure and performance of candidate formulations before actual experiments, providing a reliable basis for subsequent uncertainty assessment, selection of experimental verification points and further optimization of crosslinking parameters, and effectively reducing the trial and error cost in the material development process.

[0056] Step S103: Based on the optimal molar ratio range, the cross-linking density of collagen and DNA is controlled by adjusting the concentration of the cross-linking agent and the cross-linking time. The construction of a double cross-linked network structure with a specific pore size distribution is verified and predicted by using a synergistic approach of chemical cross-linking and physical cross-linking.

[0057] In this embodiment of the invention, the chemical cross-linking refers to the covalent cross-linking network formed by collagen through a Schiff base reaction. Chemical cross-linking refers to the reaction between the amino groups on the collagen molecular chain and the aldehyde groups in the cross-linking agent molecule to form stable covalent bonds. This process, through the formation of Schiff base structures, creates irreversible covalent cross-linking points between collagen molecules, thereby constructing a continuous and stable three-dimensional network framework. This network is mainly responsible for providing the overall mechanical support and long-term structural stability of the hydrogel. For example, under conditions of high cross-linking agent concentration and long reaction time, dense covalent connections are formed between collagen fibers, increasing the overall rigidity of the network, and the pore structure tends to be stable but the pore size decreases.

[0058] The physical cross-linking refers to the self-assembled cross-linked network formed by DNA through complementary base pairing. Physical cross-linking means that DNA molecules spontaneously form reversible cross-linking nodes through complementary base pairing. This cross-linking method does not rely on chemical bond formation but rather on hydrogen bonds and intermolecular interactions to achieve network connection, exhibiting dynamic reversibility. The physical cross-linked network is embedded within the covalent network of collagen, primarily used for fine-tuning pore size and enhancing the adaptive capacity of the network structure. For example, when the DNA molar ratio is within a suitable range, the cross-linking points formed by DNA self-assembly can effectively limit pore size expansion, while undergoing local dissociation or recombination when the microenvironment changes, thereby endowing the hydrogel with a certain degree of dynamic regulation.

[0059] In actual construction, collagen Schiff base cross-links to form the main network, and DNA base pairing forms the auxiliary network. The two work together to enable the hydrogel to have both structural stability and adjustable micropore size.

[0060] As an optional embodiment, in step S103, a historical database combining crosslinking parameters and structural characteristics is constructed. This historical database records the correlation data between different crosslinking agent concentrations, crosslinking times, and the resulting network structural characteristics. These network structural characteristics include fiber diameter distribution, network connection point density, pore size distribution, and swelling ratio. Based on this historical database, a crosslinking kinetic mapping function is established to calculate predicted structural characteristics and prediction confidence intervals for untested crosslinking parameter combinations. According to the predicted structural characteristics and prediction confidence intervals, an optimal sequence of crosslinking parameters is generated, and parameter combinations are selected as experimental verification points. The actual network structural characteristic data obtained from the experimental verification points are fed back to the crosslinking kinetic mapping function, and the selection of crosslinking parameters is iteratively optimized to obtain the optimal crosslinking parameter combination. Using the optimal crosslinking parameter combination, the crosslinking agent concentration and crosslinking time are adjusted to control the crosslinking density between collagen and DNA. A dual-crosslinked network structure with a specific pore size distribution range of 50-500 nm is constructed using a synergistic approach of chemical and physical crosslinking.

[0061] Specifically, in step S103, a historical database combining crosslinking parameters and network structure characteristics is first constructed to support the prediction and optimization of crosslinking parameters. In practice, within a defined range of collagen to DNA molar ratio, the concentration of the crosslinking agent and the crosslinking time are varied to create multiple sets of experimental samples. For each set of crosslinking parameter combinations, a double-crosslinked hydrogel is prepared, and its network structure characteristics are characterized. These structural characteristics include collagen fiber diameter distribution, network connection point density, pore size distribution range, and swelling ratio variation. For example, in samples with lower crosslinking agent concentration and shorter crosslinking time, larger fiber diameters, sparser connection points, wider pore size distribution, and higher swelling ratios are observed. Conversely, in samples with increased crosslinking agent concentration and longer crosslinking time, fiber diameters tend to be more uniform, connection point density increases significantly, pore size distribution concentrates towards the nanoscale, and the swelling ratio decreases. This data is systematically organized and stored, forming a one-to-one correspondence between crosslinking parameters and structural characteristics.

[0062] After obtaining the historical database, this embodiment of the invention predicts the structure of untested crosslinking parameter combinations based on their internal correlations. Specifically, for any unverified combination of crosslinking agent concentration and crosslinking time, the possible fiber structure, pore size distribution, and swelling behavior are inferred based on its position in the historical parameter space, and a confidence interval for the prediction results is given to reflect the degree of uncertainty in the prediction. For example, for a parameter combination with a crosslinking agent concentration at a moderate level and a crosslinking time between tested time points, its pore size can be predicted to be mainly distributed in the 100–300 nm range. However, since there is relatively little experimental data near this combination, its prediction confidence interval is relatively wide, indicating that this parameter combination has value for further experimental exploration.

[0063] Optionally, after obtaining the predicted structural values ​​and prediction confidence intervals, a preferred sequence of crosslinking parameters is generated based on multi-objective criteria. These multi-objective criteria comprehensively evaluate the following factors: the degree of matching between the predicted pore size distribution and the target pore size range (50–500 nm), the width of the prediction confidence interval, and the need for parameter space exploration. The need for parameter space exploration is inversely proportional to the data density of that region in the historical database; that is, the sparser the historical data, the higher the exploration priority for that region. In practice, crosslinking parameter combinations with predicted pore sizes near the center of the target interval (e.g., 150–300 nm) and prediction confidence interval widths greater than a preset threshold are preferentially selected as experimental verification points for the current optimization iteration. For example, if a parameter combination predicts a pore size of approximately 220 nm, and there are few historical data samples in that region, then this parameter combination will be given priority for the experimental verification stage.

[0064] After selecting experimental verification points, double-crosslinked hydrogels with corresponding parameter combinations were prepared, and their actual network structure characteristics data, including actual pore size distribution, fiber diameter, and swelling ratio, were collected. Subsequently, this actual data was fed back into the mapping relationship between crosslinking parameters and structural characteristics to update the predicted relationship. After the update, the predicted structural values ​​and prediction confidence intervals of the remaining untested parameter combinations were re-evaluated, and a new optimal crosslinking parameter sequence was generated again. This process was repeated iteratively. When the average width of the prediction confidence interval is lower than the preset convergence threshold in three consecutive optimization iterations, and the pore size prediction results tend to stabilize in the local parameter space, the crosslinking parameter optimization process is considered to have converged. Finally, from the converged optimal crosslinking parameter sequence, the combination of crosslinking agent concentration and crosslinking time with the pore size distribution closest to the target center value was selected as the optimal crosslinking parameter combination.

[0065] Using the optimal crosslinking parameter combination obtained from the above screening, a double-crosslinked hydrogel was constructed under a determined collagen to DNA molar ratio. By precisely controlling the concentration of the crosslinking agent and the crosslinking time, the density of covalent crosslinking of collagen Schiff bases was adjusted. Simultaneously, a reversible physical crosslinking network was formed using complementary DNA base pairing, allowing the two types of crosslinks to be spatially synergistically distributed. The resulting hydrogel network had pore sizes primarily concentrated in the 50–500 nm range, exhibiting a uniform and stable structure. This not only meets the requirements for long-term material colonization and osmotic exchange in the spinal cord injury microenvironment but also provides an ideal structural basis for subsequent glycyrrhizic acid sustained release and Wnt / β-catenin signal modulation.

[0066] Further optionally, in the above embodiments, generating a preferred sequence of crosslinking parameters based on the predicted structural characteristics and the predicted confidence interval, and selecting parameter combinations as experimental verification points, includes: defining a multi-objective optimization criterion based on the predicted structural characteristics and the predicted confidence interval, wherein the multi-objective optimization criterion comprehensively evaluates the target conformity of pore size distribution, the predicted confidence interval width, and the parameter space exploration requirement, wherein the parameter space exploration requirement is inversely proportional to the density of historical data in local regions; generating a preferred sequence of crosslinking parameters based on the multi-objective optimization criterion, and preferentially selecting crosslinking parameter combinations whose predicted values ​​are close to the target center within the target pore size range of 50-500 nm and whose predicted confidence interval width is greater than a preset threshold, as experimental verification points for the current optimization iteration.

[0067] In the above embodiments, a preferred sequence of crosslinking parameters is generated based on the predicted values ​​of structural characteristics and the corresponding predicted confidence intervals, using a hierarchical evaluation and ranking approach. Specifically, for each untested combination of crosslinking parameters, the degree of matching between its predicted pore size distribution and the target pore size range of 50–500 nm is calculated. By comparing the deviation between the predicted pore size center value and the target range center value, the target pore size distribution conformity is quantified. Subsequently, the width of the predicted confidence interval corresponding to the parameter combination is simultaneously evaluated, and parameter combinations with wider confidence intervals are marked as candidate points with higher uncertainty but potential information gain. Based on this, an evaluation of parameter space exploration requirements is further introduced, i.e., the density of validated data in the neighborhood of the parameter combination in the historical database is statistically analyzed. When historical data is sparse in a certain region, the exploration priority of parameter combinations in that region is increased. By comprehensively considering the above three evaluation results, all candidate crosslinking parameter combinations are ranked to form a preferred sequence of crosslinking parameters. For example, in a parameter combination with a moderate crosslinking agent concentration and a crosslinking time between adjacent tested time points, the predicted pore size is approximately 240 nm, located near the center of the target pore size range. Furthermore, this combination exhibits a relatively wide prediction confidence range and has few historical experimental points in its parameter region. Therefore, this parameter combination ranks highly in the preferred sequence and is preferentially selected as the experimental verification point for the current optimization iteration. This approach ensures that the pore size target is gradually approached while achieving efficient exploration of the parameter space.

[0068] Further optionally, in the above embodiments, the actual network structure characteristic data obtained from the experimental verification points is fed back to update the crosslinking kinetic mapping function, and the selection of crosslinking parameters is iteratively optimized to obtain the optimal crosslinking parameter combination. This includes: collecting the corresponding actual network structure characteristics through the experimental verification points of the current optimization iteration; feeding back the obtained actual network structure characteristic data to update the crosslinking kinetic mapping function, and recalculating the prediction confidence interval of the remaining untested crosslinking parameter combinations; based on the recalculated prediction confidence interval, returning to the step of generating a preferred crosslinking parameter sequence according to the predicted structural characteristic value and the prediction confidence interval, and selecting parameter combinations as experimental verification points; iteratively optimizing the preferred crosslinking parameter sequence until the average width of the prediction confidence interval obtained in three consecutive iterations is less than a preset convergence threshold, at which point the crosslinking parameter optimization process is determined to have converged; and selecting the crosslinking parameter combination whose pore size distribution is closest to the target center value from the converged preferred crosslinking parameter sequence as the optimal crosslinking parameter combination.

[0069] In the above embodiments, during the current optimization iteration, collagen-DNA double-crosslinked hydrogels are first prepared for the selected experimental verification points according to the corresponding crosslinking agent concentration and crosslinking time. Actual network structure characteristic data, including pore size distribution, fiber diameter distribution, and network connection point density, are collected using electron microscopy, swelling experiments, and other methods. Subsequently, the measured network structure characteristic data are fed back into the crosslinking kinetic mapping function as new samples to correct the correlation between different crosslinking parameters and structural characteristics. After updating, structural characteristics are re-predicted for untested crosslinking parameter combinations, and their prediction confidence intervals are calculated simultaneously to ensure the mapping relationship reflects the latest experimental information. Based on this, a new optimal sequence of crosslinking parameters is generated according to the updated predicted structural characteristic values ​​and prediction confidence intervals, and new parameter combinations are selected as the next round of experimental verification points. This process is repeated cyclically, continuously reducing the prediction uncertainty of untested parameter combinations in each iteration. When the average width of the prediction confidence intervals corresponding to three consecutive iterations is less than a preset convergence threshold, the crosslinking parameter optimization process is considered to have reached convergence. Finally, from the converged optimal sequence of crosslinking parameters, the deviations between the predicted pore size distribution and the target pore size center value corresponding to each parameter combination are compared. The crosslinking agent concentration and crosslinking time combination with the pore size distribution closest to the target center is selected as the optimal crosslinking parameter combination. For example, if after multiple iterations it is found that the pore size distribution corresponding to a certain crosslinking agent concentration and crosslinking time combination is stably concentrated at approximately 250 nm, and the prediction confidence interval has significantly converged, then this parameter combination is selected as the optimal solution for constructing the double crosslinked network structure.

[0070] For example, the mapping function update formula in the above embodiments can be expressed as the following formula: .in, Indicates the first The crosslinking kinetics mapping function established during the next iteration is used to crosslinking parameters... Predicting the network structure characteristics of hydrogels under specific conditions. This function comprehensively reflects the historical patterns of the formation process of double-crosslinked networks influenced by crosslinking agent concentration, crosslinking time, and the molar ratio of collagen to DNA. It is a predictive model built based on existing experimental data. The vector of crosslinking parameters to be evaluated includes the crosslinking agent concentration, crosslinking time, and the molar ratio of collagen to DNA, used to characterize the preparation conditions of the double crosslinked hydrogel. This indicates the result obtained after feedback correction following the introduction of new experimental verification data. The iterative mapping function improves the prediction accuracy of local parameter regions while maintaining the original prediction continuity, and is used for subsequent prediction and optimization of the structural characteristics of crosslinking parameter combinations. This represents the combination of crosslinking parameters selected and actually used in the preparation and testing during the current iteration. This parameter point corresponds to the actual prepared hydrogel sample and serves as the reference center for mapping function correction, used to correct the model's prediction bias in this region.

[0071] This is an adjustment coefficient for correcting the magnitude of the mapping function, used to control the intensity of the influence of experimental feedback on the overall mapping function. When the value is small, the mapping function updates gradually, which helps maintain model stability; when... When the value is large, the mapping function becomes more sensitive to experimental data, which can accelerate local convergence. In implementation, It can be set according to the experimental noise level and the density of the historical database.

[0072] This is used to limit the range of corrections made by the experimental feedback to the mapping function, so that the corrections are mainly concentrated in the region that is similar to the parameters at the experimental verification points. Used to control the rate at which experimental feedback decays as the distance to the parameter increases. When the value is large, the correction effect is more concentrated in a small range of parameters near the experimental verification point; When the value is small, experimental feedback has an impact on a larger range of parameter spaces. This parameter ensures that the mapping function update conforms to the local continuity characteristics of crosslinking kinetics, avoiding non-physical over-corrections to parameter regions far from experimental conditions.

[0073] This deviation vector represents the discrepancy between the experimentally measured structural properties and the model's predictions, and is the core driving force for the mapping function's correction. The introduction of this deviation vector enables the mapping function to gradually reduce prediction errors, achieving an adaptive approximation of the relationship between crosslinking parameters and structural properties. Indicates the experimental verification point Under the given conditions, the actual measured characteristics vector of the hydrogel network structure are obtained through experimental means, including real physical quantities such as fiber diameter distribution, network connection point density, pore size distribution, and swelling ratio. This indicates that before the introduction of experimental feedback, the first... Sub-iteration mapping function For experimental verification points The predicted structural characteristic vector is used to compare with the actual experimental results in order to calculate the prediction error.

[0074] Step S104: The pore size distribution of the hydrogel is controlled by the double cross-linked network structure to optimize the colonization stability, glycyrrhizic acid sustained-release performance and Wnt / β-catenin signaling pathway activation efficiency of the hydrogel in the spinal cord injury microenvironment.

[0075] Step S104 utilizes the collagen-DNA dual crosslinked network structure constructed above to finely regulate the pore size distribution within the hydrogel, ensuring it simultaneously meets the comprehensive requirements of structural stability, sustained drug release, and biological signal regulation within the spinal cord injury microenvironment. Through the synergistic effect of covalent crosslinking and self-assembled crosslinking networks, a hierarchical and tunable micro / nanoporous structure is formed, thereby achieving synergistic optimization of multiple biological processes. Specifically, the chemical crosslinking network in the dual crosslinking network provides a long-term stable three-dimensional scaffold structure for the hydrogel, enabling it to resist local mechanical disturbances and fluid erosion after implantation at the spinal cord injury site, maintaining its overall shape and preventing collapse. The physical crosslinking network forms a reversible and dynamic connection structure within the hydrogel, allowing for a degree of flexible adjustment in the pore size distribution. When changes occur in the external microenvironment (such as enzyme degradation, pH, or ionic strength changes), some physical crosslinks can rearrange, thereby buffering structural stress and improving colonization stability. At the pore size distribution level, by controlling the pore size primarily within the 50–500 nm range, the hydrogel allows for the efficient diffusion of nutrients, oxygen, and metabolites while limiting the excessive invasion of inflammatory cells such as macrophages, thereby improving the long-term stability of implanted materials in the spinal cord injury microenvironment. Simultaneously, this pore size structure provides suitable diffusion channels for glycyrrhizic acid molecules, enabling their slow release within the hydrogel network through a combination of restricted diffusion and network dissociation, achieving a sustained and stable release effect and avoiding burst release phenomena. Furthermore, the controlled pore size distribution and stable three-dimensional microstructure provide a favorable physical environment for nerve cell adhesion, growth, and signal transduction. During sustained release, glycyrrhizic acid acts on the local cellular microenvironment, synergistically promoting the activation of the intracellular Wnt / β-catenin signaling pathway with the mechanical support and spatial constraints provided by the hydrogel. In this way, the double-crosslinked network structure not only optimizes the structural properties of the hydrogel at the physical level but also enhances its comprehensive effect on the regulation of cell behavior and signaling pathways at the functional level, thereby improving the overall therapeutic potential of the hydrogel in spinal cord injury repair applications.

[0076] As an optional embodiment, in step S104, based on the synergistic relationship between the chemical cross-linking network and the physical cross-linking network in the dual cross-linked network structure, a pore size function mapping relationship model is established. This model represents the quantitative correlation between the 50-500 nm pore size distribution and colonization stability, glycyrrhizic acid sustained-release performance, and Wnt / β-catenin signaling pathway activation efficiency. A combination of regulatory parameters is determined according to the pore size function mapping relationship model, including gradient swelling treatment time, ionic strength, and temperature parameters. The relative density of the collagen network and DNA network in the dual cross-linked network is adjusted using the regulatory parameter combination to form a hydrogel with a target pore size distribution, and the pore size distribution change is monitored in real time. The comprehensive performance of the hydrogel in a simulated spinal cord injury microenvironment is evaluated based on the real-time monitoring data. When the comprehensive performance does not reach a preset threshold, the pore size function mapping relationship model is updated based on performance feedback data, and a new combination of regulatory parameters is generated. When the comprehensive performance reaches the preset threshold, the optimal pore size distribution structure is fixed through a freeze-drying rehydration process to obtain a hydrogel with a 50-500 nm pore size distribution.

[0077] In the above embodiments, the collagen-DNA double cross-linked network structure constructed above is used to finely control the pore size distribution inside the hydrogel, so that it can simultaneously meet the comprehensive requirements of structural stability, sustained drug release, and biological signal regulation in the microenvironment of spinal cord injury. Through the synergistic effect of covalent cross-linked networks and self-assembled cross-linked networks, a hierarchical and tunable micro-nanoporous structure is formed, thereby achieving synergistic optimization of multiple biological processes.

[0078] Specifically, the chemical cross-linking network in the dual cross-linking network provides a long-term stable three-dimensional scaffold structure for the hydrogel, enabling it to resist local mechanical disturbances and fluid erosion after implantation at the spinal cord injury site, maintaining its overall shape and preventing collapse. The physical cross-linking network forms a reversible, dynamic connection structure within the hydrogel, allowing for flexible adjustment of the pore size distribution. When the external microenvironment changes (such as enzyme degradation, pH, or ionic strength alterations), some physical cross-links can rearrange, thereby buffering structural stress and improving implantation stability. At the pore size distribution level, by controlling the pore size primarily within the 50–500 nm range, the hydrogel allows for the effective diffusion of nutrients, oxygen, and metabolites while limiting the excessive invasion of inflammatory cells such as macrophages, thus improving the long-term stability of the implanted material in the spinal cord injury microenvironment. Simultaneously, this pore size structure provides suitable diffusion channels for glycyrrhizic acid molecules, allowing for slow release within the hydrogel network through a combination of restricted diffusion and network dissociation, achieving a continuous and stable sustained-release effect and avoiding sudden release phenomena within a short period. Furthermore, the controlled pore size distribution and stable three-dimensional microstructure provide a favorable physical environment for nerve cell adhesion, growth, and signal transduction. During sustained release, glycyrrhizic acid acts on the local cellular microenvironment, synergistically promoting the activation of the intracellular Wnt / β-catenin signaling pathway with the mechanical support and spatial constraints provided by the hydrogel. In this way, the double-crosslinked network structure not only optimizes the structural properties of the hydrogel at the physical level but also enhances its comprehensive effect on the regulation of cell behavior and signaling pathways at the functional level, thereby improving the overall therapeutic potential of the hydrogel in spinal cord injury repair applications.

[0079] In this invention, by constructing a quantitative mapping relationship between the performance parameters of collagen and DNA molar ratio and a dynamic prediction model of crosslinking parameters and pore size distribution, the multidimensional performance of hydrogels in the spinal cord injury microenvironment can be accurately predicted. Based on this prediction result, the optimal material composition and structural parameters are determined, thereby improving the adaptability of hydrogel design to spinal cord repair needs and avoiding performance shortcomings. Furthermore, the crosslinking density and pore size distribution of the hydrogel are precisely controlled through a dual crosslinking network synergistic construction strategy, and the material structural parameters are corrected by a multidimensional performance verification and feedback optimization mechanism, ultimately forming a hydrogel with an ideal pore size distribution. This achieves precise adaptation and functional synergy of the hydrogel in the spinal cord injury microenvironment, improving the material's colonization stability, glycyrrhizic acid sustained-release accuracy, and Wnt / β-catenin signaling pathway activation efficiency. This helps to further optimize the performance of the hydrogel in complex spinal cord injury environments and reduces the experimental cost and time investment in material development.

[0080] This invention provides a machine learning-driven collagen-based DNA hydrogel optimization and prediction device, with reference to... Figure 2As shown, the device includes: a data acquisition module for acquiring performance parameters of the hydrogel at different collagen-to-DNA molar ratios; a screening module for predicting and screening the optimal collagen-to-DNA molar ratio range based on the performance parameters; a prediction module for controlling the crosslinking density of collagen and DNA by adjusting the crosslinking agent concentration and crosslinking time according to the optimal molar ratio range, and verifying and predicting the construction of a dual-crosslinked network structure with a specific pore size distribution using a synergistic approach of chemical and physical crosslinking, wherein the chemical crosslinking is a covalent crosslinked network formed by the Schiff base reaction of collagen, and the physical crosslinking is a self-assembled crosslinked network formed by complementary base pairing of DNA; and an optimization module for regulating the pore size distribution of the hydrogel using the dual-crosslinked network structure to optimize the colonization stability, glycyrrhizic acid sustained-release performance, and Wnt / β-catenin signaling pathway activation efficiency of the hydrogel in the spinal cord injury microenvironment. In some embodiments, the collagen-based DNA hydrogel optimization device can be applied to terminal devices. It should be noted that, for the sake of convenience and brevity, the specific working process of the collagen-based DNA hydrogel optimization device described above can be referred to the corresponding process in the aforementioned embodiment of the machine learning-driven collagen-based DNA hydrogel optimization prediction method, and will not be repeated here.

[0081] This invention provides a terminal device. The terminal device 300 includes a processor 301 and a memory 302, connected via a bus 303, such as an I2C bus. Specifically, the processor 301 provides computing and control capabilities to support the operation of the entire terminal device. The processor 301 can be a central processing unit, or it can be other general-purpose processors, digital signal processors, application-specific integrated circuits, field-programmable gate arrays (FPGAs), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Those skilled in the art will understand that the structures shown in the above embodiments are merely block diagrams of some structures related to the embodiments of this invention and do not constitute a limitation on the terminal device to which the embodiments of this invention are applied. A specific server may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. The processor is used to run a computer program stored in the memory and, when executing the computer program, implements any of the collagen-based DNA hydrogel optimization methods provided in this invention. It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the terminal device described above can be referred to the aforementioned collagen-based DNA hydrogel optimization method embodiment, and will not be repeated here.

[0082] This invention also provides a storage medium for computer-readable storage, wherein the storage medium stores one or more programs that can be executed by one or more processors to implement the steps of any of the collagen-based DNA hydrogel optimization methods provided in the specification of this invention.

Claims

1. A machine learning-driven method for optimizing and predicting collagen-based DNA hydrogels, characterized in that, The method includes: The performance parameters of hydrogels with different collagen to DNA molar ratios were collected. Based on the aforementioned performance parameters, the optimal molar ratio range of collagen to DNA was predicted and screened. Based on the optimal molar ratio range, the cross-linking density of collagen and DNA is controlled by adjusting the concentration of the cross-linking agent and the cross-linking time. A dual cross-linked network structure with a specific pore size distribution is verified and predicted by using a synergistic approach of chemical cross-linking and physical cross-linking. The chemical cross-linking is a covalent cross-linked network formed by collagen through Schiff base reaction, and the physical cross-linking is a self-assembled cross-linked network formed by DNA through complementary base pairing. The pore size distribution of the hydrogel was controlled by the aforementioned double cross-linked network structure to optimize the hydrogel's colonization stability, glycyrrhizic acid sustained-release performance, and Wnt / β-catenin signaling pathway activation efficiency in the spinal cord injury microenvironment.

2. The machine learning-driven collagen-based DNA hydrogel optimization and prediction method according to claim 1, characterized in that, The performance parameters of the hydrogels collected at different collagen to DNA molar ratios include: The performance parameters of hydrogels with different collagen to DNA molar ratios were collected. Key performance parameter data are extracted from the performance parameters, including pore size distribution, storage modulus, degradation rate, glycyrrhizic acid sustained-release curve, and β-catenin nucleotranslocation rate.

3. The machine learning-driven collagen-based DNA hydrogel optimization and prediction method according to claim 1, characterized in that, The process of predicting and screening the optimal molar ratio range of collagen to DNA based on the performance parameters includes: Establish a mapping relationship between different collagen and DNA molar ratios and the aforementioned key performance parameters; Based on the mapping relationship, the predicted performance value and confidence interval of the untested parameter combination are calculated; Prioritize parameter combinations that predict performance values ​​that reach the set performance values ​​and whose confidence intervals are greater than the set confidence interval width for simulation prediction; The simulation prediction results are iteratively updated to the mapping relationship. The process then jumps back to the step of calculating the predicted performance value and confidence interval of the untested parameter combination based on the mapping relationship until the convergence condition is met. The iterative update of the mapping relationship is then stopped, and the optimal molar ratio range of collagen to DNA is selected from the mapping relationship.

4. The machine learning-driven collagen-based DNA hydrogel optimization and prediction method according to claim 3, characterized in that, The step of prioritizing the selection of parameter combinations that yield predicted performance values ​​that reach a set performance value and whose confidence intervals are greater than the set confidence interval width for simulation prediction includes: A material characteristic transformation matrix is ​​established based on a pre-constructed basic material property knowledge base; Based on the material feature transformation matrix, structural performance parameters related to pore size regulation in the basic material are extracted and transferred to obtain preliminary performance predictions of candidate collagen to DNA molar ratio and crosslinking parameter combinations under spinal cord injury microenvironment conditions. The preliminary performance predictions include pore size distribution predictions, degradation kinetics predictions, and molecular diffusion coefficient predictions. The preliminary performance prediction values ​​are corrected by structural similarity weighting and uncertainty quantification to generate a corrected performance distribution containing the prediction mean and standard deviation; Based on the preliminary performance predictions, a multidimensional performance coupling analysis was performed to identify the synergistic optimization region of colonization stability, glycyrrhizic acid cumulative release rate, and β-catenin nuclear translocation efficiency. Based on the uncertainty assessment results of the collaborative optimization region and the corresponding parameter combination, candidate parameter combinations that are located within the collaborative optimization region and have an uncertainty higher than a preset threshold are preferentially selected as simulation prediction results to be updated in the mapping relationship. Among them, parameter combinations with high uncertainty have higher priority than parameter combinations with low uncertainty but slightly better performance, so as to accelerate the optimization convergence process.

5. The machine learning-driven collagen-based DNA hydrogel optimization and prediction method according to claim 4, characterized in that, Based on the material feature transformation matrix, the structural performance parameters related to pore size regulation in the basic material are extracted and transferred to obtain preliminary performance predictions of candidate collagen to DNA molar ratios and crosslinking parameter combinations under spinal cord injury microenvironment conditions, including: Based on the material feature transformation matrix, structural performance parameters related to pore size control in the basic material are extracted and transferred to form an initial parameter combination prediction model. Using historical experimental data of collagen and DNA hydrogels, the parameters related to the spinal cord microenvironment response in the initial parameter combination prediction model are locally calibrated, while the global parameters related to the basic cross-linking principle are kept unchanged. Based on the initial parameter combination prediction model, preliminary performance prediction values ​​of candidate collagen to DNA molar ratio and crosslinking parameter combination under spinal cord injury microenvironment conditions are calculated.

6. The machine learning-driven collagen-based DNA hydrogel optimization and prediction method according to claim 1, characterized in that, The process involves controlling the crosslinking density of collagen and DNA by adjusting the concentration of the crosslinking agent and the crosslinking time according to the optimal molar ratio range, and verifying and predicting the construction of a double-crosslinked network structure with a specific pore size distribution using a synergistic approach of chemical and physical crosslinking. A historical database combining crosslinking parameters and structural characteristics is constructed. The historical database records the correlation data between different crosslinking agent concentrations, crosslinking times and the characteristics of the formed network structure. The network structure characteristics include fiber diameter distribution, network connection point density, pore size distribution and swelling ratio. Based on a historical database of crosslinking parameters and structural properties, a crosslinking kinetic mapping function is established to calculate predicted values ​​and prediction confidence intervals for untested combinations of crosslinking parameters. Based on the predicted structural characteristics and the prediction confidence interval, a preferred sequence of crosslinking parameters is generated, and parameter combinations are selected as experimental verification points. The actual network structure characteristic data obtained from the experimental verification points are fed back to update the crosslinking kinetic mapping function, and the selection of crosslinking parameters is iteratively optimized to obtain the optimal combination of crosslinking parameters. By using the optimal combination of crosslinking parameters, adjusting the concentration of crosslinking agent and the crosslinking time to control the crosslinking density of collagen and DNA, and constructing a double crosslinked network structure with a specific pore size distribution range of 50-500 nm using a synergistic approach of chemical crosslinking and physical crosslinking, a double crosslinked network structure was obtained.

7. The machine learning-driven collagen-based DNA hydrogel optimization and prediction method according to claim 6, characterized in that, The step of generating a preferred sequence of crosslinking parameters based on the predicted structural characteristics and the predicted confidence interval, and selecting parameter combinations as experimental verification points, includes: Based on the predicted values ​​of the structural characteristics and the prediction confidence interval, a multi-objective optimization criterion is defined. The multi-objective optimization criterion comprehensively evaluates the target conformity of aperture distribution, the prediction confidence width, and the parameter space exploration requirement. The parameter space exploration requirement is inversely proportional to the density of historical data in local areas. Based on the multi-objective optimization criteria, a preferred sequence of crosslinking parameters is generated. The crosslinking parameter combinations with predicted values ​​close to the target center within the target pore size range of 50-500 nm and a prediction confidence interval width greater than a preset threshold are preferentially selected as the experimental verification points for the current optimization iteration.

8. The machine learning-driven collagen-based DNA hydrogel optimization and prediction method according to claim 7, characterized in that, The step of updating the crosslinking kinetic mapping function with the actual network structure characteristic data obtained from the experimental verification points, and iteratively optimizing the selection of crosslinking parameters to obtain the optimal combination of crosslinking parameters, includes: Through the experimental verification points of the current optimization iteration, the corresponding actual network structure characteristics are collected; The obtained actual network structure characteristic data are fed back to update the crosslinking kinetic mapping function, and the prediction confidence interval of the remaining untested crosslinking parameter combinations is recalculated; Based on the recalculated prediction confidence interval, the process returns to the step of generating a preferred sequence of crosslinking parameters based on the predicted values ​​and prediction confidence intervals of the structural characteristics, and selecting parameter combinations as experimental verification points. The preferred sequence of crosslinking parameters is iteratively optimized until the average width of the prediction confidence interval obtained from three consecutive iterations is less than a preset convergence threshold, at which point the crosslinking parameter optimization process is determined to have converged. The optimal crosslinking parameter combination is selected from the converged crosslinking parameter optimization sequence, with the pore size distribution closest to the target center value.

9. The machine learning-driven collagen-based DNA hydrogel optimization and prediction method according to claim 1, characterized in that, The method of controlling the pore size distribution of the hydrogel using the aforementioned double cross-linked network structure includes: Based on the synergistic relationship between the chemical cross-linking network and the physical cross-linking network in the aforementioned dual cross-linking network structure, a pore size function mapping relationship model is established. This model is used to represent the quantitative correlation between the 50-500 nm pore size distribution and colonization stability, glycyrrhizic acid sustained-release performance, and Wnt / β-catenin signaling pathway activation efficiency. The combination of control parameters is determined based on the pore size-functional mapping model. The combination of control parameters includes gradient swelling treatment time, ionic strength and temperature parameters. The relative densities of the collagen network and DNA network in the double crosslinked network are adjusted using the aforementioned combination of control parameters to form a hydrogel with a target pore size distribution, and the changes in pore size distribution are monitored in real time. The comprehensive performance of hydrogels in a simulated spinal cord injury microenvironment was evaluated based on real-time monitoring data. When the overall performance does not reach the preset threshold, the aperture function mapping relationship model is updated based on the performance feedback data, and the combination of control parameters is regenerated. When the overall performance reaches the preset threshold, the optimal pore size distribution structure is fixed by freeze-drying and rehydration process to obtain a hydrogel with a pore size distribution of 50-500nm.

10. A machine learning-driven collagen-based DNA hydrogel optimization prediction device, characterized in that, The device is used to implement the machine learning-driven collagen-based DNA hydrogel optimization prediction method as described in claims 1 to 9.

Images