CN-122000090-A - Method for inverting drug diffusion coefficient by DLS time-varying autocorrelation

CN122000090ACN 122000090 ACN122000090 ACN 122000090ACN-122000090-A

Abstract

The application belongs to the technical field of biomedical engineering, and relates to a method for inverting a drug diffusion coefficient by DLS time-varying autocorrelation, which comprises the steps of constructing a time-varying autocorrelation function data set with enhanced structure through overlapping time window division, standard time-varying autocorrelation function calculation and correction factor fusion derived from structural sensitivity parameters, then constructing a dynamic parameterized physical model related to hydrogel structural parameters based on diffusion theory, enabling the diffusion coefficient to dynamically adjust along with structure evolution, generating a dimension-matched theoretical autocorrelation function sequence, combining regularization optimization and uncertainty estimation through a trained intelligent inversion model, accurately inverting an optimal time-varying diffusion coefficient, finally ensuring the accuracy of diffusion coefficient inversion and release efficiency estimation through theoretical and actual drug release curve comparison, dynamic error feedback adjustment and energy efficiency estimation, and providing accurate data support for hydrogel drug carrier formula optimization, release dynamics analysis and clinical drug delivery scheme design.

Inventors

JIANG JIANWEI
ZHANG HONGYAN
SHEN CHANGMING
LI JUNFEI
ZHENG XIAORONG
WANG CHUNLEI
WANG QIONG
HOU GUILAN
SHEN BIN

Assignees

浙江省肿瘤医院

Dates

Publication Date: 20260508
Application Date: 20260128

Claims (10)

1. A method for inverting a drug diffusion coefficient by time-varying autocorrelation of DLS, comprising the steps of: step 1, synchronously acquiring multi-mode data in the hydrogel drug release process, denoising and normalizing the acquired data, and then performing multi-mode data fusion and feature extraction to obtain a fusion time sequence data set; Dividing the fusion time sequence data set into overlapping time windows, calculating a standard time-varying autocorrelation function in each window, deriving a structural correction factor based on structural sensitivity parameters in the multi-mode data, and obtaining a time-varying autocorrelation function data set with enhanced structure through fusion operation; step 3, constructing a dynamic parameterized physical model based on a diffusion theory, introducing a time-varying diffusion coefficient and establishing association with hydrogel structural parameters to enable the diffusion coefficient to dynamically change along with the structural evolution, and generating a theoretical autocorrelation function sequence matched with the structural enhancement time-varying autocorrelation function data set in a dimension way by combining a parameter evolution rule; Step 4, after an intelligent inversion model is built and trained, a theoretical autocorrelation function sequence is used as core reference data of inversion, a time-varying autocorrelation function data set with reinforced structure is input to obtain an initial time-varying diffusion coefficient sequence, adjustment is carried out through a regularization optimization method, an optimal time-varying diffusion coefficient is obtained in stages, and uncertainty estimation is completed; And 5, calculating a theoretical drug release curve based on the optimal time-varying diffusion coefficient, comparing the theoretical drug release curve with an actual drug release curve to obtain a dynamic error, correcting related parameters or acquisition strategies through a feedback adjustment mechanism until the error reaches the standard, and finally evaluating the drug release efficiency and outputting an evaluation report.
2. The method of claim 1, wherein the multi-modal data includes dynamic light scattering data, environmental parameters including pH and temperature, and hydrogel elastic modulus obtained by low frequency raman spectroscopy or micromechanics sensors, the hydrogel elastic modulus being a structure sensitive parameter for quantifying hydrogel degradation, phase change, or crosslink density change.
3. The method of inverting drug diffusion coefficients with DLS time-varying autocorrelation of claim 1 wherein said multi-modal data fusion and feature extraction comprises the steps of: The normalized multi-mode data matrix is used as input, the centering processing is firstly carried out, the covariance matrix of the centering data is calculated, after the covariance matrix is subjected to eigenvalue decomposition, a plurality of front main components are screened according to the eigenvalue accumulation contribution rate, a composite eigenvector is generated based on the screened main components, the composite eigenvector is used as a fusion time sequence data set, the differentiation of diffusion and structural effect is realized, and meanwhile, the data dimension reduction processing is completed.
4. The method of inverting a drug diffusion coefficient with DLS time-varying autocorrelation as set forth in claim 1, wherein said calculating a standard time-varying autocorrelation function within each window includes the steps of: The overlapping rate of adjacent windows is determined by a comparison test based on the time window length, the data acquisition frequency and the signal time sequence continuity requirement, so that the standard time-varying autocorrelation functions generated by the adjacent windows are connected in the time dimension; In each time window, firstly, calculating a cross-correlation matrix fusing all the characteristics in the time sequence data set, analyzing the association degree with the drug diffusion behavior, selecting a dynamic light scattering intensity component as a core calculation basis, obtaining a standardized time-varying autocorrelation function based on the dynamic light scattering intensity component, and representing the basic statistical characteristics of the drug molecular diffusion behavior.
5. The method for time-varying autocorrelation inversion of a drug diffusion coefficient of a DLS according to claim 1, wherein the structural correction factor is derived based on the elastic modulus of the hydrogel or/and the pH value, and the structural correction factor coefficient is correspondingly set when the structural correction factor coefficient is pushed, and the structural correction factor coefficient is calibrated by combining with the characteristics of the hydrogel system through a control experiment; When the elastic modulus is adopted for deduction, the structural correction factor adopts an exponential form, and the structural correction factor coefficient is used for adapting the nonlinear influence of the elastic modulus change on the diffusion behavior by taking the initial elastic modulus as a reference; When the pH value is adopted for deduction, the structural correction factor adopts a linear form, and the structural correction factor coefficient is used for adapting to the linear influence of pH fluctuation on the diffusion environment by taking the initial pH value as a reference; And multiplying the standard time-varying autocorrelation function by a structural correction factor through a multiplication model, adjusting the signal attenuation or attenuation rate, compensating the signal interference caused by the evolution of the hydrogel structure, and obtaining the time-varying autocorrelation function data set with the enhanced structure.
6. The method of inverting a drug diffusion coefficient with time-varying autocorrelation of DLS of claim 1 wherein the generation of the sequence of theoretical autocorrelation functions comprises the steps of: The dynamic parameterized physical model takes a generalized Fick diffusion law as a framework, a time-varying diffusion coefficient is set as a time integral function of a structural sensitive parameter, an association equation is constructed through an initial diffusion coefficient and a degradation rate constant, and the association equation is dynamically adjusted along with the accumulated evolution of the structural sensitive parameter; the evolution of the structural sensitivity parameter is described by a differential equation containing a structural degradation rate coefficient and an ion concentration environmental degradation factor, so that linkage with environmental conditions is realized; Substituting the dynamic diffusion coefficient into a light scattering theory autocorrelation function model, and calculating a time dimension attenuation rate by combining a scattering vector parameter to generate a theory autocorrelation function sequence matched with the structure enhanced time-varying autocorrelation function data set in delay time and time dimension.
7. The method of inverting a drug diffusion coefficient with time-varying autocorrelation of DLS of claim 1 wherein said intelligent inversion model employs a lightweight recurrent neural network, wherein training of the lightweight recurrent neural network comprises the steps of: Based on a dynamic parameterized physical model, constructing a plurality of groups of sample pairs of different structural evolution scenes by adjusting initial diffusion coefficients and degradation rate constants, adding Gaussian white noise related to measured data noise level into the samples for data enhancement, dividing a training set, a verification set and a test set according to preset proportions, training the lightweight cyclic neural network by taking mean square error as a loss function, and training termination conditions are that the loss of the verification set does not drop in a continuous preset iteration period or reaches the maximum iteration times; taking an initial time-varying diffusion coefficient sequence output by the lightweight recurrent neural network as a starting point, and adopting a self-adaptive regularization least square method to optimize and construct a total objective function; The total objective function is weighted sum of fitting error term, time smooth constraint term and structure consistency constraint term, the sum of weight coefficients corresponding to the terms is 1, wherein the weight of the time smooth constraint term and the structure consistency constraint term is increased in the early swelling stage, the weight of the fitting error term is increased in the late degradation stage, and based on the weight, the total objective function is optimized to the state that the change amount of the total objective function is lower than a threshold value through staged iteration.
8. The method of inverting a drug diffusion coefficient with time varying autocorrelation of DLS of claim 1 wherein said uncertainty evaluation comprises the steps of: Carrying out statistical analysis on the residual sequence of the fitting error item in the optimization process, and modeling the distribution characteristic of the residual sequence by adopting a normal distribution hypothesis; Calculating a residual standard deviation of a residual sequence, and calculating a confidence interval of the optimal time-varying diffusion coefficient under a preset confidence level based on the standard deviation combined with normal distribution characteristics, wherein the confidence interval is determined by combining the optimal time-varying diffusion coefficient with the product of the residual standard deviation and normal distribution quantiles of the corresponding confidence level; and evaluating the reliability of the inversion result through the statistical characteristics of the residual sequence, wherein the closer the residual sequence is to normal distribution and the smaller the residual standard deviation is, the higher the reliability of the inversion result is.
9. The method of inverting a drug diffusion coefficient by time-varying autocorrelation of DLS of claim 1 wherein said feedback adjustment mechanism is as follows: the dynamic error is calculated first before the feedback adjustment mechanism is executed, wherein the specific steps of calculating the dynamic error are as follows: The method comprises the steps of adopting root mean square error of a sliding window as an evaluation index, wherein the length of the window is odd, the window in an early swelling stage is smaller than the window in a late degradation stage, defining a window with a corresponding length by taking a corresponding time point as a center for each time point, calculating the difference value between the theoretical drug release amount and the actual drug release amount in the window, respectively squaring all the difference values, summing the difference values, taking an average value, squaring again to obtain the root mean square error of the sliding window at the corresponding time point, setting the size of an error threshold according to the preset proportion of the total initial drug load, wherein the theoretical drug release amount is obtained based on a theoretical drug release curve, and the actual drug release amount is obtained through an actual drug release curve; After the dynamic miscalculation is performed, a feedback adjustment mechanism is performed, comprising the steps of: When the root mean square error of the sliding window exceeds an error threshold, if the error is caused by the association deviation of the structural evolution and the diffusion coefficient, the degradation rate constant in the dynamic parameterized physical model is adjusted, and the influence weight of the structural parameter on the diffusion coefficient is corrected; If the error occurs in the early swelling stage, the sampling interval is reduced, and the sampling frequency is increased; And (3) when the parameters are adjusted, controlling the parameter adjustment amplitude of the dynamic parameterized physical model in a confidence interval of uncertainty estimation, synchronously adjusting all mode data sampling intervals, repeating the steps (2) to (4) after the adjustment is finished, regenerating the optimal time-varying diffusion coefficient, and calculating the dynamic error until the error is continuously lower than a preset error threshold value.
10. The method for inverting the drug diffusion coefficient by time-varying autocorrelation of DLS according to claim 9, wherein the theoretical drug release curve is obtained by deriving based on generalized Fick diffusion law and combining the optimal time-varying diffusion coefficient, the initial drug load total amount and the characteristic size of hydrogel, and the actual drug release curve is formed by plotting the accumulated release amount obtained by quantifying the evolution degree of a structure through elastic modulus and combining dynamic light scattering intensity data conversion.

Description

Method for inverting drug diffusion coefficient by DLS time-varying autocorrelation Technical Field The application belongs to the technical field of biomedical engineering, and particularly relates to a method for inverting a drug diffusion coefficient by DLS time-varying autocorrelation. Background The hydrogel is used as a high polymer material with a three-dimensional reticular structure, has excellent biocompatibility, hydrophilicity, structural adjustability and drug loading capacity, becomes one of core carriers of a controllable drug release system, and has wide acceptance in the industry in the application fields of biological medicines such as cancer treatment, chronic disease management and the like by regulating and controlling the drug release rate and the release period, thereby realizing targeted enrichment of drugs at focus positions, reducing toxic and side effects of systemic drug administration, remarkably improving treatment effect and tolerance of patients. The dynamic evolution of the hydrogel is mainly reflected in degradation reaction (such as high molecular chain fracture caused by enzymolysis and hydrolysis), phase change behavior (such as sol-gel transition of temperature sensitive hydrogel) and associated density change (such as formation and fracture of cross-linking bonds), the structural evolution can directly change the porosity, pore size and network compactness of a hydrogel matrix, further the migration channel of drug molecules is obviously influenced, finally the diffusion coefficient is caused to present a complex nonlinear change rule, the nonlinear characteristic is particularly prominent in the long-term drug release process, and great challenges are brought to the accurate characterization of the diffusion coefficient. At present, the method for obtaining the diffusion coefficient of the medicine in the hydrogel is mainly divided into two types of traditional experimental measurement methods and model inversion methods, wherein the traditional experimental measurement methods are mainly based on the Fick diffusion law, and the diffusion coefficient is calculated by monitoring the accumulated release amount of the medicine, but the methods generally assume that the diffusion coefficient is a constant value, neglect the dynamic structure evolution of the hydrogel caused by swelling, degradation, crosslinking degree change and the like in the medicine release process, so that the time-varying characteristic of the diffusion coefficient can not be captured, only the average diffusion coefficient can be obtained, and the deviation exists between the average diffusion coefficient and the dynamic rule of the actual medicine release process. The model inversion method makes up the limitations of the traditional method to a certain extent by constructing the association of a physical model and actual monitoring data and back-pushing a diffusion coefficient, however, the traditional inversion technology still has a plurality of technical bottlenecks that firstly, a data processing level is provided, most methods only depend on single-mode monitoring data, complementary information of multi-mode data is not fully fused, a targeted structure sensitivity correction mechanism is lacked, feature extraction precision is insufficient easily caused by data noise and structure interference, and reliable inversion reference data is difficult to form, secondly, the physical model construction level is provided, the traditional model cannot always establish dynamic association of the diffusion coefficient and hydrogel structural parameters, and the regulation effect of structural evolution on the diffusion coefficient cannot be accurately reflected, so that the adaptability of a theoretical model and an actual release process is poor. Disclosure of Invention The invention provides a DLS time-varying autocorrelation inversion method for a drug diffusion coefficient, which aims to solve the technical problem that the existing method cannot accurately acquire the time-varying diffusion coefficient of a drug in the hydrogel release process, and further cannot reliably evaluate the drug release efficiency. The invention provides a DLS time-varying autocorrelation inversion method for drug diffusion coefficients, which comprises the following steps: step 1, synchronously acquiring multi-mode data in the hydrogel drug release process, denoising and normalizing the acquired data, and then performing multi-mode data fusion and feature extraction to obtain a fusion time sequence data set; Dividing the fusion time sequence data set into overlapping time windows, calculating a standard time-varying autocorrelation function in each window, deriving a structural correction factor based on structural sensitivity parameters in the multi-mode data, and obtaining a time-varying autocorrelation function data set with enhanced structure through fusion operation; step 3, constructing a dynamic parameterized