A method and system for predicting vomitoxin damage to sertoli cells

Abstract

The application provides a method and system for predicting the damage of testicular supporting cells by vomitoxin, and relates to the technical field of testicular supporting cell damage prediction, which comprises obtaining transcriptome sequencing, immunofluorescence quantification, electron microscope ultrastructure and flow detection data; based on the theory of persistent coherence, the topological characteristics of the gene expression space are extracted and encoded into persistent bar code representation; through a neural differential equation, the bar code is fused with protein and structure data for multi-modal fusion and dynamic evolution modeling, and topological kinetic characteristics representing barrier function are obtained; combined with flow data, a cell state phase transition model is constructed by applying a cellular automaton and percolation theory to simulate the critical damage threshold of the blood-testis barrier integrity; a potential well model is constructed based on the threshold and flow data; finally, the path integral and variation method are used to calculate the damage probability of cells under different toxin concentrations. The application realizes early warning and dynamic risk prediction of testicular supporting cell damage by vomitoxin.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of testicular Sertoli cell injury prediction, and in particular, relates to a method and system for predicting vomitoxin-induced testicular Sertoli cell injury. BACKGROUND

[0002] Vomitoxin is one of the most serious mycotoxins in the world, and its detection rate in feed and raw materials is high. It poses a serious threat to the reproductive performance of breeding livestock and poultry. Studies have shown that vomitoxin can penetrate the blood-testis barrier and directly act on Sertoli cells, causing damage to the barrier function by destroying the intercellular tight junction structure, and then causing sperm production disorders and male infertility. Currently, the evaluation of this toxic effect mainly relies on traditional in vitro cytotoxicity experiments and in vivo histopathological analysis. These methods have obvious limitations: on the one hand, they can only perform endpoint detection after exposure occurs, and cannot achieve early warning and dynamic risk prediction; on the other hand, traditional methods can only provide static and isolated toxic phenotype data, and cannot reveal the continuous dynamic evolution process from gene regulatory network disorder to cell population functional failure, and lack the ability to quantitatively and model predict the critical threshold of toxic injury, resulting in insufficient accuracy and timeliness of risk assessment.

[0003] Therefore, there is a need for a method and system for predicting vomitoxin-induced testicular Sertoli cell injury to solve the above technical problems. SUMMARY

[0004] The present application aims to provide a method and system for predicting vomitoxin-induced testicular Sertoli cell injury to improve the above problems. In order to achieve the above-mentioned purpose, the technical solution adopted by the present application is as follows:

[0005] In a first aspect, the present application provides a method for predicting vomitoxin-induced testicular Sertoli cell injury, comprising:

[0006] Obtaining the transcriptome sequencing data of testicular tissue of male animals under vomitoxin exposure conditions, the immunofluorescence quantitative data of Sertoli cell tight junction proteins, the electron microscope image data of the blood-testis barrier, and the flow cytometry detection data of Sertoli cell apoptosis and cell cycle under different toxin concentrations;

[0007] Performing feature extraction on the transcriptome sequencing data based on the persistent coherence theory, and encoding based on the extracted topological features to obtain a persistent barcode representation of the gene regulatory network;

[0008] Fusing the persistent barcode representation, the immunofluorescence quantitative data, and the electron microscope image data, and performing feature evolution modeling on the fused data by driving a pre-set neural differential equation to obtain a topological dynamics characterization of Sertoli cell barrier function;

[0009] Based on the topological dynamics characterization and the flow detection data, a cell state phase transition model is constructed through a cellular automaton and percolation theory, and cell state simulation is performed based on the cell state phase transition model to obtain a critical damage threshold of the blood-testis barrier integrity;

[0010] Based on the critical damage threshold and the flow detection data, a potential well model is constructed to obtain a potential well model of toxin dose-barrier damage;

[0011] Based on path integral and variational method, the potential well model is processed to generate damage probability values of testicular supporting cells under each toxin concentration.

[0012] In a second aspect, the present application also provides a prediction system for vomitoxin-induced damage to testicular supporting cells, comprising:

[0013] An acquisition unit is configured to acquire transcriptional sequencing data of testicular tissue of a male animal under vomitoxin exposure conditions, immunofluorescence quantification data of supporting cell tight junction proteins, electron microscope image data of the blood-testis barrier, and flow cytometry detection data of apoptosis and cell cycle of supporting cells under different toxin concentrations;

[0014] An extraction unit is configured to perform feature extraction on the transcriptional sequencing data based on the persistent homology theory, and encode the extracted topological features to obtain a persistent barcode representation of a gene regulatory network;

[0015] A modeling unit is configured to fuse and process the persistent barcode representation, the immunofluorescence quantification data, and the electron microscope image data, and perform feature evolution modeling on the fused data by driving a pre-set neural differential equation to obtain a topological dynamics characterization of the supporting cell barrier function;

[0016] A simulation unit is configured to construct a cell state phase transition model through a cellular automaton and percolation theory based on the topological dynamics characterization and the flow detection data, and perform cell state simulation based on the cell state phase transition model to obtain a critical damage threshold of the blood-testis barrier integrity;

[0017] A construction unit is configured to construct a potential well model based on the critical damage threshold and the flow detection data to obtain a potential well model of toxin dose-barrier damage;

[0018] A processing unit is configured to process the potential well model based on path integral and variational method to generate damage probability values of testicular supporting cells under each toxin concentration.

[0019] The present application has the following advantages:

[0020] The present application can simulate and predict the continuous dynamic process from gene expression disturbance to barrier function decline in the process of toxin exposure by integrating multi-dimensional data such as transcriptome, proteome and cell morphology, and constructing a dynamic evolution model by using neural differential equations. This can make risk warning when the damage has not fully appeared, and significantly improve the timeliness of evaluation.

[0021] The present application innovatively uses the theory of persistent homology to analyze transcriptome data and extracts the topological stability characteristics of the gene regulation network, thereby revealing the systematic disorder of the gene network structure under the action of toxins.

[0022] The present application calculates the probability of damage occurring by crossing the potential barrier of the system under different toxin concentrations by constructing a random dynamic potential well model and using path integral and variational method. The result is no longer "whether damaged", but "damaged with what probability". This probability form of prediction result can more scientifically and comprehensively reflect the individual differences and uncertainties of biological response, greatly improving the accuracy and reliability of risk assessment.

[0023] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the present application as described in the written description and claims. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0025] Figure 1 The prediction method flow chart of the present application for the damage of testicular supporting cells by vomitoxin is shown in the figure;

[0026] Figure 2 The prediction system structure diagram of the present application for the damage of testicular supporting cells by vomitoxin is shown in the figure;

[0027] Figure 3 The KEGG enrichment analysis diagram of down-regulated differential expression genes after DON exposure in the prediction method of the present application for the damage of testicular supporting cells by vomitoxin is shown in the figure;

[0028] Figure 4 The transmission electron microscopy diagram of tight junction damage induced by DON in the prediction method of the present application for the damage of testicular supporting cells by vomitoxin is shown in the figure;

[0029] Figure 5 A scatter plot showing the effect of different concentrations of DON treatment on the apoptosis rate in the prediction method of DON-damaged testicular Sertoli cells described in the embodiments of the present application;

[0030] Figure 6 A column chart showing the effect of different concentrations of DON treatment on the apoptosis rate in the prediction method of DON-damaged testicular Sertoli cells described in the embodiments of the present application.

[0031] In the figure: 701, acquisition unit; 702, extraction unit; 703, modeling unit; 704, simulation unit; 705, construction unit; 706, processing unit. DETAILED DESCRIPTION

[0032] To make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art without creative labor based on the embodiments in the present application are within the scope of protection of the present application.

[0033] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0034] Embodiment 1

[0035] The present embodiment provides a prediction method for DON-damaged testicular Sertoli cells.

[0036] Referring to Figure 1 , the present method includes steps S1, S2, S3, S4, S5 and S6.

[0037] Step S1, acquiring the transcriptome sequencing data of testicular tissue of male animals under DON exposure conditions, the immunofluorescence quantitative data of Sertoli cell tight junction proteins, the electron microscope image data of the blood-testis barrier, and the apoptosis and cell cycle flow detection data of Sertoli cells under different toxin concentrations;

[0038] Understandably, the transcriptome sequencing data (RNA-seq) in this step provides a genome-wide expression profile of testicular tissue under DON exposure, enabling unbiased identification of key differentially expressed genes related to cell connectivity, inflammatory responses, and apoptosis pathways (such as...). Figure 3 As shown), where, Figure 3 The horizontal axis represents the pathway enrichment factor, i.e., the proportion of differentially expressed genes in a particular pathway out of the total number of genes in that pathway. The vertical axis represents the specific KEGG pathway name. This reveals the initial target and signaling pathway perturbation of toxin action at the molecular network level. Immunofluorescence quantitative data supporting tight junction proteins (such as ZO-1, CX43) directly correlate these molecular-level changes with protein localization and expression levels. High-resolution imaging and fluorescence intensity quantification precisely quantify the loss of key structural proteins in the blood-testis barrier. Electron microscopy images of the blood-testis barrier provide standard validation at the ultrastructural level, allowing direct observation of the breakage and loosening of tight junctions (TJs) (e.g., Figure 4 Morphological damage (as shown) including, among which, Figure 4 The scale bar for the first image (top left) is 5 µm, and for the second image (top right) it is 2 µm. Yellow arrows indicate intact tight junction desmosomes. The scale bar for the third image (bottom left) is 5 µm, and for the fourth image (bottom right) it is 2 µm. Red arrows indicate damaged or broken tight junction desmosomes. This provides structural confirmation and deeper supplementation to the immunofluorescence data. Flow cytometry data supporting apoptosis and cell cycle at different toxin concentrations quantitatively reveal the toxic effects of vomiting toxin (DON) at the cellular function level, such as increased apoptosis rate (e.g., ...). Figure 5 and Figure 6 (as shown) Figure 5 The colors represent the cell density at each location (combination of fluorescence signal intensities) (blue → green → yellow → red areas: cell number gradually increases). Figure 6 The different colored bars represent different concentrations of vomitoxin treatment. Gray indicates cells treated with 0 µM vomitoxin (control group), pink indicates cells treated with 0.4 µM vomitoxin, green indicates cells treated with 1.6 µM vomitoxin, and purple indicates cells treated with 6.4 µM vomitoxin (treatment group). Compared with the control group (0 µM), the apoptosis rate in the treatment group (6.4 µM) was significantly increased. These data directly reflect the survival status and proliferative capacity of the supporting cell population, serving as a crucial bridge connecting molecular damage and organ dysfunction (such as blood-testis barrier collapse).

[0039] Step S2, feature extraction is performed on the transcriptome sequencing data based on persistent homology theory, and the extracted topological features are encoded to obtain a persistent barcode representation of the gene regulatory network;

[0040] It can be understood that this step performs deep feature mining on high-throughput transcriptome data, and its advantage is that it can bypass the limitations of pre-set gene lists or pathways in traditional differential expression analysis, directly extract global and robust features from the geometry and topological structure of the data itself, and enable us to discover global network changes that may be ignored by traditional methods (for example, DON exposure may cause the topological connectivity of the gene module related to "cell adhesion" to become fragile, that is, its characteristic barcode is shortened), providing highly condensed and structure-rich feature inputs for subsequent fusion of other modal data and construction of prediction models. In this step, step S2 includes steps S21, S22, S23 and S24.

[0041] Step S21, the transcriptome sequencing data is subjected to gene expression relationship metric learning, wherein the difference and correlation between gene expression vectors in the sample space are calculated by Mahalanobis distance metric weighted by gene function annotation and known pathway prior knowledge, to obtain a high-dimensional distance matrix reflecting the functional correlation between genes;

[0042] It can be understood that this step first integrates prior knowledge from pre-set gene function annotations, gene ontology (GO) and KEGG pathways and other databases. Specifically, the system will construct a weighted gene function correlation network: if two genes are involved in the same biological process, located in the same organelle or belong to the same signal pathway, then the functional correlation weight between them is high. This weight matrix defined by prior knowledge is ingeniously embedded in the metric learning of Mahalanobis distance, used to constrain or guide the calculation of the covariance matrix. Its processing process can be understood as: the expression distance between gene pairs that are closely related in function is "contracted", while the expression distance between gene pairs that are not related in function is "expanded".

[0043] Wherein, first, the gene expression matrix is constructed, which has the dimension of m samples and n genes. Then, the gene function annotation information obtained from the authoritative database (such as GO, KEGG) is used to calculate a function correlation strength based on prior knowledge for each gene pair. Wherein, the strength is determined by calculating the Jaccard index, and then a weighted gene function correlation matrix is obtained. Then, the sample covariance matrix of the gene expression matrix is calculated, and the sample covariance matrix is modulated by using the weighted gene function correlation matrix, wherein the graph Laplace regularization processing is adopted to obtain the covariance estimation matrix, and then the Mahalanobis distance calculation is performed on the two vectors of the covariance estimation matrix to obtain the final high-dimensional distance matrix reflecting the functional correlation between genes, wherein the Mahalanobis distance calculation formula is as follows:

[0044] ;

[0045] Wherein, d cu is the Mahalanobis distance between genes, e c is the indicator vector of gene c, e u is the indicator vector of gene u, T is the transpose symbol, is the inverse matrix of the modulated covariance estimation matrix.

[0046] Through this processing, the final obtained high-dimensional distance matrix is no longer only a mathematical expression difference, but a biological distance matrix reflecting the functional correlation between genes. In this matrix, if two gene expression patterns are similar and biologically functionally related, the distance between them will be very small; on the contrary, even if the expression patterns are accidentally similar but functionally unrelated, the distance will be kept at a larger value. This effectively filters the noise in the high-dimensional data and highlights the biologically meaningful co-expression patterns.

[0047] Step S22, constructing the topological structure of the gene expression space based on the high-dimensional distance matrix, wherein each gene is taken as a vertex, and a filter parameter is set according to the distance matrix to construct a complex sequence evolving with the concentration of toxin, and the complex sequence is a sequence reflecting the birth and death process of gene co-expression communities under different concentrations of toxin;

[0048] It can be understood that this step first takes the high-dimensional distance matrix as input, and each element of the matrix represents the difference degree of a pair of gene expression patterns after biological function weighting. The system takes each gene as a vertex to construct an initial graph structure. Subsequently, a filter parameter is introduced, which is the maximum functional distance allowed between connected genes. The change process of the filter parameter value from small to large simulates the construction process of the gene co-expression network from fragmentation to integrity under gradually relaxed connection standards.

[0049] Then, for each specific toxin concentration (corresponding to a specific experimental group), the algorithm sets an initial filtration parameter and constructs a corresponding Rips complex. In the present invention, the rule of Rips complex is: when the distance between any two genes is less than or equal to the current filtration parameter, they form a simplex (e.g. two genes with distance ≤ filtration parameter, then connected as an edge, three genes with distance between any two ≤ filtration parameter, then form a triangular face). As the filtration parameter increases, more and more connections are established, and individual points and edges gradually form larger connected components, loops and cavities, thus forming a more and more complex complex structure.

[0050] Finally, a series of complexes constructed under different filtration parameters are arranged in order, that is, a complex sequence evolving with filtration parameters is obtained.

[0051] Step S23, extracting topological invariants based on the complex sequence, wherein the persistent homology values of homology groups of each dimension are calculated by persistent homology theory, the birth and death processes of topological features at different scales are tracked, and a set of topological features for describing the stability of the gene regulatory network is obtained, wherein the topological features are connected components, loops and cavities;

[0052] It can be understood that this step traverses each complex in the complex sequence (corresponding to a filtration parameter value). For each complex, this step calculates the homology groups of each dimension. Homology group is a core concept in algebraic topology, which is used to describe the "hole" structure of the complex in a specific dimension: among them, the generator of 0-dimensional homology group represents a connected component, and its number reflects the number of independent communities in the gene expression network; the generator of 1-dimensional homology group represents a loop, i.e. a closed ring formed by a series of edges connected end to end, which may correspond to a functional circuit or feedback regulation pathway; the generator of 2-dimensional homology group represents a cavity, i.e. a closed three-dimensional space surrounded by a face, which may symbolize a more complex gene regulatory module.

[0053] Among them, by tracking the "life cycle" of each topological feature: record its first appearance at which filtration parameter value (birth scale) and its "filling" or submersion due to the establishment of larger connections at which filtration parameter value (death scale). The persistent homology value of a feature is represented by this pair (birth scale, death scale). The length of its life cycle is the birth scale minus the death scale.

[0054] In the present application, a long-life 1D loop corresponds to a gene co-expression feedback loop that remains stable under a variety of connection conditions (such as a core regulatory circuit related to tight junction protein synthesis). Exposure to vomitoxin can cause the "death" scale of the loop to be advanced, i.e., its life cycle is significantly shortened, indicating that the stability of the functional loop is destroyed by the toxin and becomes fragile and easily broken.

[0055] The present application converts dynamic and difficult-to-directly-analyze complex sequences into a quantitative set of topological features containing birth, death, and life cycle information. This feature set is no longer the original gene expression amount, but a mathematical description of the high-order connection mode and structural stability of the gene network.

[0056] Step S24, topological feature coding based on the set of topological features, wherein by screening the persistent homology pairs with a life cycle length exceeding a preset threshold in the set of topological features, the persistent homology pairs are mapped to a point set on a two-dimensional plane to obtain a persistent barcode representation representing the stability of the gene regulatory network.

[0057] It can be understood that this step filters out topological features that exist for a short time and are likely to represent noise (such as features with a short life cycle) by setting a preset threshold (such as a life cycle length greater than the median of all feature life cycle lengths), and retains core topological features that are stable over a wide scale range.

[0058] Subsequently, each screened topological feature is coded as a point on a two-dimensional plane: the horizontal coordinate of the point is the birth scale, and the vertical coordinate of the point is the death scale. Therefore, the life cycle length of a feature is the vertical distance from the point to the diagonal. The point set formed by all these points is then stored on a two-dimensional plane by dimension grouping to obtain the final "persistent barcode" representation. In this representation, a point far from the diagonal represents a topological structure with a long life cycle and high stability, and a point close to the diagonal represents a short-lived and unstable structure.

[0059] Step S3, fusing the persistent barcode representation, immunofluorescence quantitative data, and electron microscope image data, and performing feature evolution modeling on the fused data by a preset neural differential equation drive to obtain a topological dynamics representation supporting cell barrier function;

[0060] Understandably, this step integrates and evolves three types of heterogeneous data—topological features characterizing macroscopic gene network stability (persistent barcodes), mesoscopic protein distribution (immunofluorescence), and microscopic ultrastructure (electron microscopy)—within a unified mathematical framework, thereby achieving a systematic quantitative modeling of the complex biological process supporting cell barrier function. In this step, step S3 includes steps S31, S32, S33, and S34.

[0061] Step S31: Perform unified embedding of multimodal data based on the persistent barcode representation, wherein the lifecycle of feature points in the barcode is mapped to a high-dimensional vector and aligned in the latent space with the integrity index of the tight junction structure of the blood-testis barrier extracted from the immunofluorescence quantitative data and electron microscopy image data to obtain an initial joint feature vector;

[0062] Understandably, this step first maps the lifetime of each topological feature point of the persistent barcode to a set of basis functions (Gaussian kernels) using Betti curves, and then superimposes the contributions of all features to generate a smooth, fixed-dimensional function vector. Secondly, the immunofluorescence quantitative data (such as the fluorescence intensity of ZO-1 and CX43) are standardized to eliminate batch effects, followed by dimensionality reduction using principal component analysis (PCA) to extract the core feature vectors that best represent the expression level and distribution pattern of tight junction proteins. Finally, a pre-trained deep learning segmentation model (U-Net) is used to automatically identify and segment the connective structures of the blood-testis barrier. Subsequently, feature vectors such as continuous length, tortuosity, and the number and area of ​​broken gaps are extracted from the segmentation results. The deep learning segmentation model is learned from historical electron microscopy image data and connective structure data of the blood-testis barrier. This results in a semantically unified latent space with re-encoded and aligned fused representations.

[0063] Step S32: Based on the initial joint feature vector and the preset neural differential equation with toxin exposure time as the implicit time parameter, the dynamic interaction and evolution law among all topological features of the supporting cells under the action of toxin were simulated, and a dynamic model of the change of the barrier function of the supporting cells over time was obtained.

[0064] It is understood that this step uses the initial joint feature vector as the initial state of the neural differential equation. This vector contains the comprehensive state of the blood-testis barrier at the gene expression, protein distribution, and ultrastructure levels at the initial moment of toxin exposure (e.g., at the lowest concentration or zero time). The neural differential equation is as follows:

[0065] ;

[0066] in, f is the time derivative of the state vector, x θ f is a vector-valued function parameterized by a neural network, θ are the learnable parameters (weights and biases) of the neural network, Z(t) is the time-varying state vector, and t is time.

[0067] By integrating this neural differential equation using the fourth-order Runge-Kutta method, the continuous evolution trajectory of the state vector from the initial state to the target value of the toxin dose can be calculated from the initial state. This trajectory simulates the dynamic interaction process between the initial joint feature vectors under increasing toxin stress.

[0068] This step can query the features at any intermediate state by changing the input concentration parameter, thereby achieving fine characterization of the functional decline process and prediction of the intermediate state.

[0069] Step S33, the joint feature vector evolution of the kinetic model is solved, wherein the joint feature vector evolved over time is obtained by integrating the neural differential equation from the beginning of the toxin exposure time to the end of the toxin exposure time using the adjoint method, and the topological features of the joint feature vector are the result of the interaction of the topological features under toxin disturbance;

[0070] It can be understood that the trained kinetic model (differential equation) and the initial joint feature vector are input. The system state is integrated from the beginning of the toxin exposure time to the end of the toxin exposure time by the adjoint method. In the integration process, the adjoint state of the state variable is calculated in reverse and in parallel, thereby efficiently obtaining the state value of the system at each toxin exposure time. Further, the evolved joint feature vector is obtained, which contains the interaction results of all modal features of the system after the complete toxin exposure process, for example, gene network instability and protein degradation, and the dimension value representing tight junction integrity in the vector is significantly reduced.

[0071] It can be understood that this step solves the complete trajectory of the system state with respect to the toxin exposure concentration by the adjoint method, thereby obtaining the final state of the multi-modal features of the system at the end concentration, and obtaining the cumulative effect of the topological features of the joint feature vector under toxin disturbance.

[0072] Step S34, generating a topological dynamics representation according to the evolved joint feature vector, wherein the dimensions representing the function of the blood-testis barrier are decoded from the evolved joint feature vector, the dimensions representing the function of the blood-testis barrier include reflecting the expression dynamics of tight junction-related proteins, the topological stability of gene regulatory networks, and the physical structural integrity of intercellular connections, and obtaining the topological dynamics representation supporting the function of the Sertoli cell barrier.

[0073] It can be understood that the step is based on the evolution of the joint feature vector input into the preset decoder for processing. The decoder learns to give higher weights to specific dimensions in the joint vector that are strongly related to barrier function (topological stability dimension data of gene regulatory network, expression dynamics dimension of tight junction related proteins, and physical structure integrity dimension of intercellular connections), and suppresses noise dimensions unrelated to function, through supervised learning using samples labeled as "barrier intact" or "barrier severely damaged" in the training phase, wherein the decoder includes an input layer, a calculation layer, a semantic mapping layer, a normalization layer, and an output layer, wherein the semantic mapping layer is used to extract information related to the topological stability dimension data of the gene regulatory network, the expression dynamics dimension of the tight junction related proteins, and the physical structure integrity dimension of the intercellular connections from the samples. After training, only a small number of samples have non-zero weights, and the original feature dimensions corresponding to these non-zero weights are interpreted as the factors that contribute most to the specific dimensions, wherein the calculation layer is trained through L1 regularization, and the purpose is to let as many elements of the weight matrix as possible be 0.

[0074] Through the encoder, the step decodes a vector representing the dimensions of the blood-testis barrier function from the evolved joint feature vector, which includes data information of the topological stability dimension data of the gene regulatory network, the expression dynamics dimension of the tight junction related proteins, and the physical structure integrity dimension of the intercellular connections.

[0075] Step S4, based on the topological dynamics characterization and the flow detection data, a cell state phase transition model is constructed by a cellular automaton and percolation theory, and a cell state simulation is performed based on the cell state phase transition model to obtain a critical damage threshold of the blood-testis barrier integrity;

[0076] It can be understood that this step constructs a cross-scale model through a cellular automaton and percolation theory, thereby simulating in a computer the dynamic phase transition process of the supporting cell population from a healthy state to a functional collapse under the action of toxins, and accurately locating the critical damage threshold of the blood-testis barrier integrity. In this step, step S4 includes step S41, step S42, step S43, and step S44.

[0077] Step S41, mapping the topological dynamics characterization into a two-dimensional grid, wherein each cell represents a supporting cell, and the initial state is assigned by the flow detection data to obtain a spatial distribution model reflecting the initial state of the supporting cell population;

[0078] It can be understood that the present step firstly deconstructs the topological dynamics characterization. The "physical structural integrity of intercellular connection" dimension in the characterization is directly extracted and converted into an intercellular connection strength value between 0 and 1 through a Sigmoid function, which quantifies the ability of a specific cell to form and maintain functional connections with other cells in its local environment. At the same time, the other two dimensions of the characterization (expression dynamics of tight junction related proteins and physical structural integrity of intercellular connection) are combined into a cell function integrity index through a weighted formula, reflecting the intrinsic health degree of the cell under the condition of being separated from the population connection. The weighted formula is as follows:

[0079] ;

[0080] wherein, is the cell function integrity index, is the weight of the expression dynamics of tight junction related proteins, is the score of the expression dynamics of tight junction related proteins, is the weight of the topological stability of gene regulatory network, is the score of the topological stability of gene regulatory network.

[0081] Subsequently, the present step constructs a two-dimensional grid to simulate the actual monolayer spreading structure of supporting cells on the basement membrane. Each cell on the grid represents a supporting cell. The initial state of each cell is generated based on the probability model of flow detection data, obtaining a spatial distribution model (two-dimensional grid) reflecting the initial state of the supporting cell population. First, the apoptosis rate and cycle arrest proportion detected in the flow detection data are regarded as population statistical priors. For example, if the flow data shows that the initial apoptosis rate is 5%, about 5% of the cells in the grid are randomly initialized as "apoptosis" or "severe damage" state.

[0082] Step S42, defining the cell state evolution rule based on the spatial distribution model, wherein the neighbor coupling rule based on percolation theory is used to simulate the diffusion of apoptosis signal through cell connection in the population, obtaining the evolution function of dynamic interaction of supporting cell population under toxin stress;

[0083] It can be understood that the present step defines a state transition probability function, which determines the possibility of a cell changing state (such as from "healthy" to "stress" or "apoptosis") at the next moment. The state transition probability of the cell is the weighted sum of the initial apoptosis rate in the flow data and the neighbor pressure, and the weighted sum formula is as follows:

[0084] ;

[0085] wherein, is the state transition probability, σ is a logic function, and α is a preset self-vulnerability weight coefficient, is the self-vulnerability, which is a weighted average of the three dimensions of the topological stability dimension data of the gene regulatory network, the expression dynamics dimension of the tight junction related protein, and the physical structure integrity dimension of the intercellular connection, and β is a neighbor pressure weight coefficient, is a set of all neighbor cells of the current cell, is the cell connection strength, is a Kronecker delta function, which is 1 when the state of the neighbor cell j is “apoptosis”, and 0 otherwise, and γ is a bias term.

[0086] Step S43, based on the cellular automaton model, iteratively apply the evolution function to simulate the dynamic process of supporting the cell population from starting apoptosis to forming a connected apoptosis cluster larger than a preset area under low-dose to high-dose toxin exposure, and record the size of the largest connected apoptosis cluster in the population after each simulation, to obtain a data sequence reflecting the degree of collapse of the connectivity of the cell population under each toxin level;

[0087] It can be understood that this step starts the iterative simulation of the cellular automaton on a two-dimensional grid. For each set toxin level, the cellular automaton performs multiple independent simulations (to eliminate the influence of randomness). In each iteration of each simulation, the cellular automaton not only updates the cell state, but also records the grid area (i.e. the number of cells contained) occupied by the largest connected apoptosis cluster after each simulation reaches a stable state (or a preset number of iterations). When the size of the connected apoptosis cluster is very small, apoptosis is sporadic and isolated, and cannot affect the function of the population; when the size of the connected apoptosis cluster suddenly increases and crosses the entire system size, it indicates that the connectivity of the cell population has collapsed.

[0088] Finally, by traversing a series of toxin levels from low to high and simulating each level multiple times, a data sequence (size sequence of apoptosis clusters) reflecting the degree of collapse of the connectivity of the cell population under each toxin level is finally obtained.

[0089] Step S44, based on the data sequence, analysis is performed, wherein by identifying the mutation point of the data sequence, the dose corresponding to the mutation point is the critical point of the percolation phase transition, and the critical damage threshold of the integrity of the blood-testis barrier is obtained.

[0090] It can be understood that this step analyzes the data sequence generated by simulation, wherein the critical damage threshold is the mean value of all mutation points (points with a size change value of the apoptosis cluster greater than a preset threshold), so as to accurately identify the critical toxin dose at which the blood-testis barrier function systemically collapses. This step avoids the limitations of subjective pathological observation in traditional toxicology to determine the threshold, and provides a turning point obtained by data-driven calculation.

[0091] Step S5, constructing a potential well model based on the critical damage threshold and the flow detection data to obtain a toxin dose-barrier damage potential well model;

[0092] It can be understood that this step uses the critical damage threshold to fuse the population statistical noise information provided by the flow detection data, constructs a random dynamics potential well model, quantitatively describes the nonlinear relationship between the toxin dose and the blood-testis barrier function state, and reveals the inherent randomness. Further, the function state of the blood-testis barrier is abstracted as the position of a particle in an energy landscape. In the absence of toxin interference, the Sertoli cells are in a stable state (i.e., the bottom of the potential energy well) representing "barrier integrity". As the toxin dose increases, the energy landscape of the Sertoli cells deforms. In this step, step S5 includes step S51 and step S52.

[0093] Step S51, obtaining a correlation relationship coupling parameter set of the toxin dose and the barrier damage by taking the critical damage threshold as a bifurcation point and quantifying the barrier damage intensity by using the apoptosis rate and the cycle arrest proportion in the flow detection data;

[0094] It can be understood that this step directly takes the critical damage threshold as a control parameter in the potential function. A higher apoptosis rate indicates greater uncertainty in cell fate, and by calculating the variance of the apoptosis rate at different doses, the relationship between the apoptosis rate and the toxin dose can be obtained. An increase in the cycle arrest proportion indicates that the normal cycle process of the cells is disrupted, which is a direct manifestation of functional impairment. By curve fitting to establish a mapping function of the toxin concentration and the damage intensity, the critical damage threshold, the variance of the apoptosis rate at different doses, the mapping function of the toxin concentration and the damage intensity are coupled to obtain the parameter set.

[0095] Step S52, constructing a random dynamics potential well model according to the correlation relationship coupling parameter set, wherein a non-equilibrium state potential function is obtained by taking the toxin dose as a control parameter and the barrier damage as a state variable, to obtain a toxin dose-barrier damage potential well model.

[0096] The step is to input the coupling parameter set of the correlation relationship as the control parameter of the model; the functional state (from perfect to complete damage) of the blood-testis barrier is defined as the state variable of the system, and then a non-equilibrium potential function is obtained and used as a potential well model of toxin dose-barrier damage, which accurately reproduces and predicts the sharp change in system behavior near the critical dose, providing a kinetic explanation for the suddenness and irreversibility of toxic effects. The non-equilibrium potential function is as follows:

[0097] ;

[0098] wherein, is the non-equilibrium potential function, x is the state variable of the blood-testis barrier, C is the toxin dose, C* is the critical damage threshold, and γ(C) is the potential well shape parameter. The value of this parameter is calibrated by flow cytometry data.

[0099] Step S6, processing the potential well model based on path integration and variational method to generate the damage probability value of testicular supporting cells under each toxin concentration.

[0100] It can be understood that for any given toxin concentration, a theoretical damage probability value can be calculated. The traditional toxicology "yes or no" binary judgment mode is changed, and quantitative risk assessment is realized. In this step, step S6 includes step S61 and step S62.

[0101] Step S61, weighting and summing all possible paths connecting the healthy potential well and the damage potential well in the potential well model by the path integration method, and the weight is the Onsager-Machlup action of each path, to obtain the probability amplitude of the transition from the healthy state to the damage state under each toxin concentration;

[0102] It can be understood that this step regards the system evolution described by the potential well model as a random process. There are infinite paths connecting the healthy state to the damage state, and each path contributes to the final transition probability. The Onsager-Machlup action is defined as the weight factor of each path. It is as follows:

[0103] ;

[0104] wherein, is the Onsager-Machlup action, t a is the initial time, t b is the end time, D is the diffusion coefficient, is the first order derivative of the state variable with respect to time, is the potential function gradient, V(x) is the potential function, x is the state variable of the blood-testis barrier, t represents time, and d is the differential symbol.

[0105] Then, the path integral calculates the probability amplitude of the transition by weighted summation of all possible paths, which is used to quantify the probability of the cell state transition under the action of toxins, where the weighting formula is as follows:

[0106] ;

[0107] where, represents the probability amplitude of the system being in state x a (healthy state) at time t a and being found in state x b (damaged state) at time t b , is the path integral, which represents the summation of all possible paths connecting the initial state and the final state, x(t) represents the function of the state variable changing with time, x is the state variable of the blood-testis barrier, and t represents time, represents the Boltzmann weight of each path.

[0108] Step S62, the action of Onsager-Machlup is determined by applying the variational method to find the extreme path, the quasi-potential function value of the extreme path is calculated to determine its transition probability, and the most possible path and its probability of the blood-testis barrier function collapse under each concentration of toxins are obtained.

[0109] It can be understood that this step finds the most possible (probability contribution is the largest) evolution path from all possible paths by the variational method, and quantitatively calculates the probability of the system state transition based on the path, so as to realize the accurate and interpretable probability prediction of the blood-testis barrier collapse risk, wherein the most possible (probability contribution is the largest) evolution path is found from all possible paths, and the corresponding Euler-Lagrange equation is solved, and the formula is as follows:

[0110] ;

[0111] wherein L is the Lagrangian, x is the state variable of the blood-testis barrier, which continuously changes from "healthy" to "damaged", is the first-order derivative of the state variable with respect to time, t represents time, represents the partial derivative of the Lagrangian with respect to the state variable, represents the total derivative of the generalized momentum pair, represents the partial derivative of the Lagrangian with respect to the state change velocity.

[0112] Subsequently, the rate constant of the path from the healthy state to the damaged state is calculated by Kramers formula, which is used as the most possible path and its probability value of the blood-testis barrier function collapse under each concentration of toxins.

[0113] Embodiment 2

[0114] As Figure 2 shown, the embodiment provides a system for predicting the damage of testicular supporting cells by vomitoxin, see Figure 2 The system comprises

[0115] An acquisition unit 701 is configured to acquire, for a male animal, transcriptome sequencing data of testicular tissue under vomitoxin exposure conditions, immunofluorescence quantitative data of supporting cell tight junction proteins, electron microscope image data of the blood-testis barrier, and flow cytometry detection data of apoptosis and cell cycle of supporting cells under different toxin concentrations;

[0116] An extraction unit 702 is configured to perform feature extraction on the transcriptome sequencing data based on the persistence homology theory, and encode based on the extracted topological features to obtain a persistence barcode representation of a gene regulatory network;

[0117] A modeling unit 703 is configured to fuse the persistence barcode representation, the immunofluorescence quantitative data, and the electron microscope image data, and perform feature evolution modeling on the fused data by driving a preset neural differential equation to obtain a topological dynamics characterization of the supporting cell barrier function;

[0118] A simulation unit 704 is configured to construct a cell state phase transition model based on the topological dynamics characterization and the flow cytometry detection data by using a cellular automaton and a percolation theory, and perform cell state simulation based on the cell state phase transition model to obtain a critical damage threshold of the blood-testis barrier integrity;

[0119] A construction unit 705 is configured to construct a potential well model based on the critical damage threshold and the flow cytometry detection data to obtain a potential well model of toxin dose-barrier damage;

[0120] A processing unit 706 is configured to process the potential well model based on path integral and variational method to generate damage probability values of testicular supporting cells under each toxin concentration.

[0121] It should be noted that, as for the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment relating to the method, and will not be described in detail here.

[0122] The above only describes preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0123] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of predicting the damage of vomitoxin to Sertoli cells, characterized by, The application relates to a method for predicting the critical damage threshold of blood-testis barrier (BTB) under different toxin concentrations, comprising the following steps: acquiring the transcriptome sequencing data of testis tissue of male animals under the condition of vomitoxin exposure, the immunofluorescence quantitative data of tight junction proteins of Sertoli cells, the electron microscope image data of the blood-testis barrier, and the flow cytometry detection data of apoptosis and cell cycle of Sertoli cells under different toxin concentrations; feature extraction is performed on the transcriptome sequencing data based on the persistent homology theory, and the extracted topological features are coded to obtain the persistent barcode representation of the gene regulatory network; the persistent barcode representation, the immunofluorescence quantitative data and the electron microscope image data are fused, and feature evolution modeling is performed on the fused data by driving a preset neural differential equation to obtain the topological dynamics characterization of the barrier function of Sertoli cells; based on the topological dynamics characterization and the flow cytometry detection data, a cell state phase transition model is constructed by using a cellular automaton and a percolation theory, and cell state simulation is performed based on the cell state phase transition model to obtain the critical damage threshold of the blood-testis barrier integrity; based on the critical damage threshold and the flow cytometry detection data, a potential well model is constructed to obtain the potential well model of toxin dose-barrier damage; the potential well model is processed based on path integral and variational method to generate the damage probability value of testis Sertoli cells under each toxin concentration.

2. The method of claim 1, wherein the testicular support cells are damaged by a toxin. Based on the transcriptome sequencing data, multi-scale topological features of gene expression space are extracted by a topological data analysis method of persistent homology, comprising the following steps: the transcriptome sequencing data is subjected to metric learning of gene expression relationship, wherein the difference and correlation between gene expression vectors in the sample space are calculated by using Mahalanobis distance measurement weighted by gene function annotation and known pathway prior knowledge to obtain a high-dimensional distance matrix reflecting the functional correlation between genes; topological structure of the gene expression space is constructed based on the high-dimensional distance matrix, wherein each gene is taken as a vertex, and a complex sequence evolving with the toxin concentration is constructed by setting a filtering parameter according to the distance matrix, and the complex sequence is a sequence reflecting the birth and death process of gene co-expression communities under different toxin concentrations; topological invariants are extracted based on the complex sequence, wherein the persistent homology values of homology groups in different dimensions are calculated by using the persistent homology theory to track the birth and death process of topological features in different scales, and a topological feature set used for describing the stability of the gene regulatory network is obtained, wherein the topological features are connected branches, ring structures and cavities; topological features are coded based on the topological feature set, wherein the persistent homology pairs with a life cycle length exceeding a preset threshold are screened out from the topological feature set, and are mapped to a point set on a two-dimensional plane to obtain a persistent barcode representation representing the stability of the gene regulatory network.

3. The method of claim 1, wherein the testicular support cells are damaged by vomitoxin. the persistent barcode representation, the immunofluorescence quantitative data and the electron microscope image data are fused, and feature evolution modeling is performed by driving a preset neural differential equation, comprising the following steps: unified embedding of the multi-modal data based on the persistence barcode representation, wherein, by mapping the life cycle of feature points in the barcode into a high-dimensional vector, and aligning with the integrity indicators of the tight junction structure of the blood-testis barrier extracted from the immunofluorescence quantitative data and the electron microscope image data in the latent space, an initial joint feature vector is obtained; based on the initial joint feature vector and a preset neural differential equation with toxin exposure time as an implicit time parameter, the dynamic interaction and evolution law among all topological features of the supporting cells under the action of the toxin is simulated, and a kinetic model of the barrier function of the supporting cells changing with time is obtained; evolution of the joint feature vector is solved by the kinetic model, wherein, by using the adjoint method to integrate the neural differential equation from the beginning of the toxin exposure time to the end of the toxin exposure time, the joint feature vector evolving with the time parameter is obtained, and the topological features of the joint feature vector interact with each other as a result of the disturbance of the toxin; topological kinetic representation is generated according to the evolved joint feature vector, wherein, the dimensions representing the function of the blood-testis barrier are decoded from the evolved joint feature vector, the dimensions representing the function of the blood-testis barrier include the expression dynamics of the tight junction related proteins, the topological stability of the gene regulation network and the physical structure integrity of the intercellular connection, and the topological kinetic representation of the barrier function of the supporting cells is obtained.

4. The method of claim 1, wherein the testicular support cells are damaged by vomitoxin. , based on the topological kinetic representation and the flow detection data, a cell state phase transition model is constructed by using a cellular automaton and a percolation theory, and cell state simulation is performed based on the cell state phase transition model, and a critical damage threshold of the blood-testis barrier integrity is obtained, including: the topological kinetic representation is mapped to a two-dimensional grid, wherein each cell represents a supporting cell, and the initial state is assigned by the flow detection data, and a spatial distribution model reflecting the initial state of the supporting cell population is obtained; based on the spatial distribution model, the definition of the evolution rule of the cell state is performed, wherein, based on the neighbor coupling rule of the percolation theory, the diffusion of the apoptosis signal through the cell connection in the population is simulated, and an evolution function of the dynamic interaction of the supporting cell population under the stress of the toxin is obtained; based on the cellular automaton model, the evolution function is iteratively applied to simulate the dynamic process of the supporting cell population from the beginning of apoptosis to the formation of a connected apoptosis cluster larger than a preset area under the exposure of the toxin from a low dose to a high dose, and the size of the largest connected apoptosis cluster in the population after each simulation is recorded, and a data sequence reflecting the collapse degree of the connectivity of the cell population under each toxin level is obtained; based on the data sequence, analysis is performed, wherein, by identifying the mutation point of the data sequence, the dose corresponding to the mutation point is the critical point of the percolation phase transition, and the critical damage threshold of the blood-testis barrier integrity is obtained.

5. The method of claim 1, wherein the testicular support cells are damaged by vomitoxin. , based on the critical damage threshold and the flow detection data, a potential well model is constructed, and a potential well model of the toxin dose-barrier damage is obtained, including: The critical damage threshold is taken as a bifurcation point, and the apoptosis rate and cycle arrest proportion in the flow detection data are used to quantify the barrier damage intensity, to obtain a correlation coupling parameter set of the toxin dose and the barrier damage; According to the correlation coupling parameter set, a stochastic dynamics potential well model is constructed, wherein, by taking the toxin dose as a control parameter and the barrier damage as a state variable, a non-equilibrium state potential function is obtained, and a potential well model of the toxin dose-barrier damage is obtained.

6. A system for predicting emetic toxin damage to Sertoli cells, characterized by, It comprises: An acquisition unit is configured to acquire transcriptional sequencing data of testicular tissue of male animals under the condition of vomitoxin exposure, immunofluorescence quantitative data of tight junction proteins of Sertoli cells, electron microscope image data of the blood-testis barrier, and flow cytometry detection data of apoptosis and cell cycle of Sertoli cells under different toxin concentrations; An extraction unit is configured to perform feature extraction on the transcriptional sequencing data based on the persistent homology theory, and encode the extracted topological features to obtain a persistent barcode representation of the gene regulatory network; A modeling unit is configured to fuse the persistent barcode representation, the immunofluorescence quantitative data, and the electron microscope image data, and perform feature evolution modeling on the fused data by driving a pre-set neural differential equation to obtain a topological dynamics characterization of the barrier function of Sertoli cells; An analog unit is configured to construct a cell state phase transition model based on the topological dynamics characterization and the flow cytometry detection data by using a cellular automaton and a percolation theory, and perform cell state simulation based on the cell state phase transition model to obtain a critical damage threshold of the blood-testis barrier integrity; A construction unit is configured to construct a potential well model based on the critical damage threshold and the flow cytometry detection data to obtain a potential well model of the toxin dose-barrier damage; A processing unit is configured to process the potential well model based on path integration and variational method to generate damage probability values of testicular Sertoli cells under each toxin concentration.

7. The system for predicting emetic toxin-damaged Sertoli cells according to claim 6, wherein The extraction unit comprises: A first extraction subunit is configured to perform metric learning of gene expression relationships on the transcriptional sequencing data, wherein the difference and correlation between gene expression vectors in a sample space are calculated by using Mahalanobis distance measurement weighted by gene function annotation and known pathway prior knowledge to obtain a high-dimensional distance matrix reflecting the functional correlation between genes; A second extraction subunit is configured to construct a topological structure of a gene expression space based on the high-dimensional distance matrix, wherein each gene is taken as a vertex, and a complex sequence evolving with toxin concentration is constructed by setting a filtering parameter according to the distance matrix, and the complex sequence is a sequence reflecting the birth and death process of gene co-expression communities under different toxin concentrations; A third extraction subunit is configured to extract topological invariants based on the complex sequence, wherein the persistent homology values of each dimension homology group are calculated by using the persistent homology theory to track the birth and death process of topological features at different scales, and a topological feature set for describing the stability of the gene regulatory network is obtained, wherein the topological features are connected branches, ring structures, and cavities. A fourth extraction subunit is configured to perform topology feature coding based on the set of topology features, wherein a persistent coherent pair with a life cycle length exceeding a preset threshold is filtered out from the set of topology features, mapped as a point set on a two-dimensional plane, and a persistence barcode representing the stability of the gene regulatory network is obtained.

8. The system for predicting emetic toxin-damaged Sertoli cells according to claim 6, wherein The modeling unit comprises: A first modeling subunit is configured to perform unified embedding of the multi-modal data based on the persistence barcode, wherein the life cycle of a feature point in the barcode is mapped as a high-dimensional vector, and the integrity of the tight junction structure of the blood-testis barrier extracted from the immunofluorescence quantitative data and the electron microscope image data is aligned in a hidden space to obtain an initial joint feature vector; A second modeling subunit is configured to simulate the dynamic interaction and evolution rule between all topology features of the supporting cells under the action of the toxin based on the initial joint feature vector and a preset neural differential equation with the toxin exposure time as an implicit time parameter, and obtain a dynamic model reflecting the change of the barrier function of the supporting cells over time; A third modeling subunit is configured to perform evolution solving of the joint feature vector based on the dynamic model, wherein the neural differential equation is integrated from the start of the toxin exposure time to the end of the toxin exposure time by using the adjoint method to obtain the joint feature vector evolved over time, and the topology features of the joint feature vector interact with each other as a result of the disturbance of the toxin; A fourth modeling subunit is configured to generate a topology dynamics representation according to the evolved joint feature vector, wherein a dimension representing the function of the blood-testis barrier is decoded from the evolved joint feature vector, the dimension representing the function of the blood-testis barrier includes the expression dynamics of the tight junction related protein, the topology stability of the gene regulatory network, and the physical structure integrity of the intercellular connection, and a topology dynamics representation of the barrier function of the supporting cells is obtained.

9. The system for predicting emetic toxin-damaged Sertoli cells according to claim 6, wherein The modeling unit comprises: A first modeling subunit is configured to perform unified embedding of the multi-modal data based on the persistence barcode, wherein the life cycle of a feature point in the barcode is mapped as a high-dimensional vector, and the integrity of the tight junction structure of the blood-testis barrier extracted from the immunofluorescence quantitative data and the electron microscope image data is aligned in a hidden space to obtain an initial joint feature vector; A second modeling subunit is configured to simulate the dynamic interaction and evolution rule between all topology features of the supporting cells under the action of the toxin based on the initial joint feature vector and a preset neural differential equation with the toxin exposure time as an implicit time parameter, and obtain a dynamic model reflecting the change of the barrier function of the supporting cells over time; A third modeling subunit is configured to perform evolution solving of the joint feature vector based on the dynamic model, wherein the neural differential equation is integrated from the start of the toxin exposure time to the end of the toxin exposure time by using the adjoint method to obtain the joint feature vector evolved over time, and the topology features of the joint feature vector interact with each other as a result of the disturbance of the toxin; A fourth modeling subunit is configured to generate a topology dynamics representation according to the evolved joint feature vector, wherein a dimension representing the function of the blood-testis barrier is decoded from the evolved joint feature vector, the dimension representing the function of the blood-testis barrier includes the expression dynamics of the tight junction related protein, the topology stability of the gene regulatory network, and the physical structure integrity of the intercellular connection, and a topology dynamics representation of the barrier function of the supporting cells is obtained. The modeling unit comprises: A first modeling subunit is configured to perform unified embedding of the multi-modal data based on the persistence barcode, wherein the life cycle of a feature point in the barcode is mapped as a high-dimensional vector, and the integrity of the tight junction structure of the blood-testis barrier extracted from the immunofluorescence quantitative data and the electron microscope image data is aligned in a hidden space to obtain an initial joint feature vector; A second modeling subunit is configured to simulate the dynamic interaction and evolution rule between all topology features of the supporting cells under the action of the toxin based on the initial joint feature vector and a preset neural differential equation with the toxin exposure time as an implicit time parameter, and obtain a dynamic model reflecting the change of the barrier function of the supporting cells over time; A third modeling subunit is configured to perform evolution solving of the joint feature vector based on the dynamic model, wherein the neural differential equation is integrated from the start of the toxin exposure time to the end of the toxin exposure time by using the adjoint method to obtain the joint feature vector evolved over time, and the topology features of the joint feature vector interact with each other as a result of the disturbance of the toxin; A fourth modeling subunit is configured to generate a topology dynamics representation according to the evolved joint feature vector, wherein a dimension representing the function of the blood-testis barrier is decoded from the evolved joint feature vector, the dimension representing the function of the blood-testis barrier includes the expression dynamics of the tight junction related protein, the topology stability of the gene regulatory network, and the physical structure integrity of the intercellular connection, and a topology dynamics representation of the barrier function of the supporting cells is obtained.

10. The system for predicting emetic toxin-damaged Sertoli cells according to claim 6, wherein The construction unit comprises: A first construction sub-unit, configured to obtain a coupling parameter set of a correlation between the toxin dose and the barrier damage by taking the critical damage threshold as a bifurcation point and quantifying the barrier damage intensity by using the apoptosis rate and the cycle arrest proportion in the flow detection data; A second construction sub-unit, configured to construct a random kinetic potential well model according to the coupling parameter set of the correlation, wherein the potential well model of the toxin dose and the barrier damage is obtained by using the toxin dose as a control parameter and the barrier damage as a state variable of a non-equilibrium state potential function.

Images