CN-122024908-A - Quantitative test method for pharmacodynamic activity coefficient of pesticide auxiliary agent under multivariable coupling

CN122024908ACN 122024908 ACN122024908 ACN 122024908ACN-122024908-A

Abstract

The invention relates to the technical field of testing, and discloses a quantitative testing method for a drug effect active coefficient under the multi-variable coupling of pesticide auxiliaries. The method comprises the steps of defining 9 core variables of environment, application and organism 3 types, accurately regulating 8 variables through a multi-chamber high-throughput experiment system, generating an orthogonal experiment matrix through Latin hypercube sampling and automatically executing an experiment, constructing a gradient lifting decision tree model based on high-dimensional pharmacodynamic data, decoupling marginal contribution of an auxiliary agent through a Charpy value algorithm, and finally generating a multidimensional response curved surface of a pharmacodynamic activity coefficient through Gaussian process regression fitting. The invention realizes the accurate, dynamic and high reproducibility quantification of the synergistic effect of the auxiliary agent, and provides scientific basis for the optimization and accurate application of the pesticide formulation.

Inventors

ZHOU GUOWEI
ZHOU BIHONG
ZHANG KUN
ZHONG LI

Assignees

广州方中化工有限公司

Dates

Publication Date: 20260512
Application Date: 20260317

Claims (10)

1. The method for quantitatively testing the drug effect active coefficient of the pesticide auxiliary agent under the multivariable coupling is characterized by comprising the following steps of: Defining and parameterizing a plurality of core variables affecting the efficacy of a pesticide; constructing a high-dimensional experimental data set; Constructing and training a gradient lifting decision tree prediction model based on the high-dimensional experimental data set, wherein the gradient lifting decision tree prediction model takes parameter values of the plurality of core variables as input, takes the scalar pesticide effect evaluation value as output, and generates a group of decision trees through sequential iterative training, so that the model can learn and characterize nonlinear mapping relations and high-order interaction effects between the plurality of core variables and the pesticide effect evaluation value; Applying the trained gradient lifting decision tree prediction model to each data point in the high-dimensional experimental data set, decomposing the predicted drug effect value of each data point by adopting an attribution analysis algorithm based on the additive interpretation of the summer, and calculating the marginal contribution value of the auxiliary agent mass concentration variable to the predicted drug effect value; And collecting marginal contribution values of the auxiliary agent mass concentration variables of all data points, combining the parameter values of the corresponding other core variables, and adopting a Gaussian process regression method to fit and generate a multidimensional response curved surface function, wherein the multidimensional response curved surface function is the pharmacodynamic activity coefficient of the auxiliary agent under the nine-degree variable coupling effect, and measuring the synergy contribution of the auxiliary agent as a continuous function of the states of the other variables.
2. The quantitative test method for the pharmacodynamic activity coefficient of the pesticide adjuvant under the multivariable coupling of claim 1, wherein the plurality of core variables comprise an environment parameter set, an application technical parameter set and a biological target parameter set, wherein the environment parameter set comprises environment temperature, environment relative humidity, illumination intensity and carbon dioxide concentration, the application technical parameter set comprises the median diameter of a droplet particle size spectrum, the deposition density of a unit area liquid medicine and the mass concentration of the adjuvant in the liquid medicine, and the biological target parameter set comprises the age of target pests and the surface wetting characteristics of target crop leaves.
3. The method for quantitatively testing the pharmacodynamic activity coefficient of a pesticide adjuvant under the multivariable coupling of claim 2, wherein constructing the high-dimensional experimental data set comprises the following steps: Constructing a multi-chamber high throughput experiment system comprising a plurality of independent environmental control units, wherein a sensor and an actuator are arranged in each independent environmental control unit; Generating a multidimensional orthogonal experimental matrix in a preset value range of the plurality of core variables by using a Latin hypercube sampling algorithm, wherein each row of the multidimensional orthogonal experimental matrix corresponds to a group of unique variable parameter combination set values; And controlling the multi-chamber high-throughput experiment system, automatically and sequentially executing all experiments according to the set values of the multi-dimensional orthogonal experiment matrix, in each experiment, acquiring the pharmacodynamic response image data of the target organism at fixed time through a high-resolution imaging module, and calculating by combining an image processing algorithm to obtain scalar pharmacodynamic evaluation values which are in one-to-one correspondence with the set values of each group of variable parameter combination, thereby forming a high-dimensional experiment data set.
4. The method for quantitative testing of the pharmacodynamic activity coefficient of a pesticide adjuvant under multivariable coupling according to claim 3, wherein the accurate, independent and dynamic regulation of the remaining variables except for the adjuvant mass concentration of the plurality of core variables specifically comprises: In each independent environment control unit, controlling the environment temperature within an error range of +/-0.1 ℃ of a preset value through a closed-loop temperature control system formed by a Peltier semiconductor refrigerating sheet array and a resistance wire heating array; The environment relative humidity is controlled within an error range of +/-1% of a preset value through a closed-loop humidity control system formed by an ultrasonic atomization humidifier and a molecular sieve drying dehumidifier; The illumination intensity and the spectral distribution are accurately regulated and controlled through a full-spectrum light-emitting diode area light source array and a pulse width modulation dimming controller; Precisely mixing high-purity carbon dioxide and air through a mass flow controller, and performing closed-loop feedback control by using a non-dispersive infrared sensor to stabilize the concentration of the carbon dioxide; The median diameter of the particle size spectrum of the fog drops is changed by adjusting the frequency of driving voltage through a micropore vibration atomizing nozzle driven by piezoelectric ceramics; the total volume of the sprayed liquid medicine is precisely controlled through a high-precision injection pump linked with the atomizing nozzle, and the deposition density of the liquid medicine in unit area is determined by combining the pre-calibrated spraying coverage area; Identifying and grading morphological characteristics of target pests through a biological microscope and image analysis software, and determining the age of the target pests; the surface wetting characteristics of the target crop leaves were determined by a contact angle meter and quantified as a static water contact angle value.
5. The method for quantitatively testing the pharmacodynamic activity coefficient of the pesticide adjuvant under the multivariable coupling of claim 4, wherein the calculating the target quantitative pharmacodynamic evaluation value specifically comprises: Collecting an initial state baseline image of a target organism before the application of the drug, and collecting a series of response images according to a preset time interval after the application of the drug; automatically identifying and counting the number of living target organisms and the number of dead target organisms in the image by adopting a semantic segmentation algorithm based on deep learning; calculating a drug effect evaluation value according to the formula: ; For the evaluation value of the efficacy of the drug, In order to kill the number of target organisms, Is the initial total target biomass.
6. The method for quantitative testing of pharmacodynamic activity coefficient under multivariate coupling of pesticidal adjuvant according to claim 5, wherein the training process of the gradient boosting decision tree prediction model comprises initializing a base learner only comprising constant values; In each iteration, calculating a negative gradient between the current model predicted value and the actual efficacy evaluation value, namely residual error; training a new weak learner, namely a depth-limiting decision tree, by taking the residual error as a target; adding the newly trained decision tree into the existing model at a preset learning rate, and updating the prediction capability of the whole model; Repeating the iterative process until the preset iterative times are reached or the performance of the model on the verification set is not improved, and finally forming the strong learner consisting of a plurality of decision trees.
7. The quantitative test method for the pharmacodynamic activity coefficient under the multivariable coupling of the pesticide auxiliary agent according to claim 6, wherein calculating the marginal contribution value by adopting an attribution analysis algorithm based on Xia Puli additive explanation specifically comprises: for any particular experimental data point in the dataset, consider all variable subsets that do not contain additive mass concentration variables; for each variable subset, calculating a predicted pesticide effect value of the model under the condition of the subset, and adding the auxiliary agent mass concentration variable on the basis of the subset; Carrying out weighted average on the difference value of the two predicted drug effect values, namely the predicted increment caused by the introduction of the auxiliary agent mass concentration variable; The weight is determined by the number of permutation combinations of the variable subset among all possible variable combinations; the final weighted average increment is the sharp value of the auxiliary agent mass concentration variable under the specific data point, namely the marginal contribution value.
8. The method for quantitatively testing the pharmacodynamic activity coefficient under the multivariable coupling of the pesticide auxiliary agent according to claim 7, wherein the multidimensional orthogonal experimental matrix generated by the Latin hypercube sampling algorithm ensures that exactly one sampling point exists in an equal probability interval of each variable, and sampling points among any 2 variables are uniformly distributed on two-dimensional projection.
9. The method for quantitative testing of the pharmacodynamic activity coefficient under the multivariable coupling of pesticide adjuvants according to claim 8, wherein the gaussian process regression method adopts a marton kernel function as a covariance function, and the optimal value of the length scale and the signal standard deviation super-parameter is determined by maximizing an edge likelihood function.
10. The quantitative test method for the pharmacodynamic activity coefficient of the pesticide adjuvant under the multivariable coupling of claim 9, wherein the multi-chamber high-throughput experiment system comprises 16 mutually isolated environment simulation bins, and the inner wall of each environment simulation bin adopts a polytetrafluoroethylene coating to prevent residual pollution of the pesticide liquid.

Description

Quantitative test method for pharmacodynamic activity coefficient of pesticide auxiliary agent under multivariable coupling Technical Field The invention belongs to the technical field of testing, and particularly relates to a quantitative testing method for a drug effect active coefficient under the multi-variable coupling of pesticide auxiliaries. Background The pesticide auxiliary agent is used as a key component for improving the pesticide effect and the physicochemical property of the pesticide liquid, and is widely applied to modern agriculture plant protection systems. The core functions of the active ingredient are to amplify the bioavailability of the active ingredient by regulating the surface tension, enhancing the wetting spreadability, promoting the osmotic absorption and other mechanisms. However, the actual synergistic effect of the adjuvant does not exist in isolation, but rather is highly dependent on the combined action of the application environment and the operating parameters. The current pesticide effect evaluation system is mostly based on single or few variable control experiments, and is difficult to truly reflect the dynamic response characteristics under the complex field conditions. Nine-degree variables including temperature, humidity, illumination intensity, wind speed, pH value of liquid medicine, spraying pressure, concentration of auxiliary agent, leaf surface characteristics of crops and active ingredient types together form a core parameter set influencing the pharmacodynamic activity coefficient. These variables do not act independently during actual administration, but rather exhibit a strong nonlinear coupling relationship. For example, the volatilization rate and the leave-on time of the auxiliary agent are mutually restricted in a high-temperature and high-humidity environment, and the spray pressure can change the particle size distribution of liquid drops, so that the deposition uniformity of the auxiliary agent on different leaf structures is affected. This multidimensional interaction effect makes the traditional static calibration method face fundamental limitations in quantifying the contribution of the auxiliaries. In the prior art, the efficiency of the auxiliary agent is usually evaluated by adopting a fixed variable method or an orthogonal test design, and the essence of the auxiliary agent still belongs to a static decoupling idea. The method has certain applicability when the coupling strength between the variables is low, but under a high dynamic coupling scene, the interaction interference between the variables cannot be effectively separated, so that the synergistic contribution of the auxiliary agent is systematically overestimated or underestimated. Especially under the conditions of concentrated variable coupling frequency bands and severe interaction effect, the measurement error is obviously amplified, and misjudgment on the effectiveness of the auxiliary agent is extremely easy to be caused, so that formula optimization and field application decision are misled. Therefore, there is a need for a quantitative test method for the pharmacodynamic activity coefficient that can identify variable coupling strength and implement selective dynamic compensation to achieve accurate attribution and scientific evaluation of the efficacy of the adjuvant. Disclosure of Invention In order to solve the problems, the invention provides a quantitative test method for the drug efficacy activity coefficient under the multi-variable coupling of pesticide auxiliary agents, which comprises the following steps: Defining and parameterizing a plurality of core variables affecting the efficacy of a pesticide; constructing a high-dimensional experimental data set; Constructing and training a gradient lifting decision tree prediction model based on the high-dimensional experimental data set, wherein the gradient lifting decision tree prediction model takes parameter values of the plurality of core variables as input, takes the scalar pesticide effect evaluation value as output, and generates a group of decision trees through sequential iterative training, so that the model can learn and characterize nonlinear mapping relations and high-order interaction effects between the plurality of core variables and the pesticide effect evaluation value; Applying the trained gradient lifting decision tree prediction model to each data point in the high-dimensional experimental data set, decomposing the predicted drug effect value of each data point by adopting an attribution analysis algorithm based on the additive interpretation of the summer, and calculating the marginal contribution value of the auxiliary agent mass concentration variable to the predicted drug effect value; And collecting marginal contribution values of the auxiliary agent mass concentration variables of all data points, combining the parameter values of the corresponding other core variables, and adopting a Gaussian p