Gastric cancer cell drug sensitivity analysis system for tumor cleavage pathway modeling

Abstract

The present application relates to the technical field of health care informatics, and discloses a gastric cancer cell drug sensitivity analysis system for tumor shear channel modeling, comprising: a benchmark network construction module for constructing a benchmark signal channel topology graph; a fluid tensor mapping module for generating a sparse shear adjustment tensor based on fluid dynamics parameters; a dynamic topology evolution module for performing Hadamard product operation or superimposing a sparse latent crosstalk matrix according to the comparison result of shear stress scalar and skeleton collapse threshold, and generating a shear correction topology graph; and a drug efficacy propagation calculation module for calculating drug sensitivity scores, wherein the present application replaces multi-physical field coupling simulation with tensor mapping and matrix operation, introduces a latent crosstalk matrix and structure mutation decision logic, and defines structural switching rules between a conventional physiological state and an extreme pathological state at a data level, thereby solving the problem that a conventional topology structure cannot cover pathological signal crosstalk under extreme working conditions.

Description

Technical Field

[0001] This invention relates to a drug sensitivity analysis system for gastric cancer cells used for tumor cleavage pathway modeling, belonging to the field of healthcare informatics technology. Background Technology

[0002] Currently, in the field of precision medicine and drug screening for gastric cancer, mainstream data analysis relies on bioinformatics models based on static gene expression profiling. This involves acquiring high-throughput sequencing data from patient tumor samples, constructing intracellular signaling regulatory networks using standard protein-protein interaction databases, and then calculating the intervention effects of drug molecules on specific targets using network propagation algorithms or graph neural networks. When processing primary tumor samples in a relatively static microenvironment, this data-driven, high-throughput approach establishes a fundamental position in medical big data analysis. However, in clinical scenarios involving advanced gastric cancer with peritoneal metastasis or hematogenous dissemination, tumor cells are constantly exposed to ascites or blood circulation, generating fluid shear stress fields. Fluid shear force serves as a key physical environmental factor. The alteration of intracellular signaling pathway connectivity and activation thresholds through mechanotransduction mechanisms leads to prediction distortions in conventional prediction models based on static biochemical data under dynamic physical environments. Existing technologies introduce coupled simulation schemes of computational fluid dynamics and molecular dynamics to improve environmental fit. However, the enormous computational overhead required to solve complex fluid-structure interaction equations contradicts the need for rapid response in clinical diagnosis and treatment, making large-scale deployment in routine hospital information systems difficult. Existing homogenized network models do not establish a topological spatial attenuation index for the transmission of physical signals from the cell membrane to the cell nucleus, and lack a time dimension variable describing the adaptive passivation of biological systems under long-term mechanical stimulation, resulting in a lack of logical support for predicting drug resistance evolution.

[0003] Existing purely data-driven algorithms have logical blind spots when dealing with dynamic environments. For example, Chinese invention patent CN118280601B discloses a method and system for assessing the sensitivity of anticancer drugs based on semi-supervised learning. It uses a multi-task deep neural network framework and Monte Carlo algorithm to solve the problems of dimensionality reduction and interpretability of high-dimensional features. However, the modeling logic is limited to the static biochemical data space and ignores the hydrodynamic microenvironment in which tumor cells are located. Such models assume that the intracellular signal transduction pathways are constant and unchanging under any conditions. They cannot capture the structural reconstruction and adaptive drift of signal pathways triggered by mechanical transduction mechanisms in gastric cancer cells under extreme fluid shear conditions such as ascites flushing or hematogenous metastasis. The systematic neglect of physical field effects leads to fundamental deviations and distortions in predicting the evolution of drug resistance in complex fluid environments in late metastatic lesions.

[0004] Therefore, constructing an efficient computational model that integrates fluid physical characteristics and biological topology to achieve accurate analysis of drug sensitivity in dynamic fluid environments at the data level has become the technical problem to be solved by this invention. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A drug sensitivity analysis system for gastric cancer cells used for tumor cleavage pathway modeling, comprising: The baseline network construction module is used to acquire transcriptome sequencing data of gastric cancer cells to be analyzed, retrieve a pre-set protein interaction database, and construct a baseline signaling pathway topology consisting of a set of vertices representing protein nodes and a baseline adjacency matrix defining initial connectivity weights. The fluid tensor mapping module is used to receive fluid dynamic parameters characterizing the target physiological environment and calculate the shear stress scalar. Based on a preset mechanical sensitivity lookup table, it generates a sparse shear regulation tensor with the same dimension as the reference adjacency matrix. The dynamic topology evolution module contains a sparse latent crosstalk matrix that defines non-regular cross-path connections. It is used to execute the structural mutation decision logic: when the shear stress scalar is lower than the preset skeleton collapse threshold, only the Hadamard product operation of the sparse shear adjustment tensor and the reference adjacency matrix is ​​performed to generate a shear-corrected topology graph; when the shear stress scalar is not lower than the skeleton collapse threshold, a structural mutation command is triggered, and the non-zero elements of the sparse latent crosstalk matrix are superimposed on the matrix after the Hadamard product operation to establish cross-path connection edges that do not exist in the reference state in the shear-corrected topology graph. The drug efficacy propagation calculation module is used to map the molecular feature vector of the target drug to a shear-corrected topology graph, calculate the node perturbation propagation efficiency using a graph neural network algorithm, and output a drug sensitivity score.

[0006] Preferably, the superposition operation of the Hadamard product operation and the structural mutation instruction executed by the dynamic topology evolution module follows the following algebraic logic to achieve discrete jumps in the graph structure: ,in, For the reconstructed target matrix, Using the baseline adjacency matrix, To adjust the tensor for sparse shearing It represents the Hadamah accumulation. For indicator functions, For shear stress scalar, The threshold for skeletal collapse. It is a sparse latent crosstalk matrix.

[0007] Preferably, the sparse shearing regulation tensor is a diagonally dominant matrix. The fluid tensor mapping module generates non-unit value regulation coefficients only for the diagonal elements corresponding to the nodes marked as mechanosensitive proteins in the reference signal pathway topology graph, while the remaining elements retain unit values, so as to maintain the topological stability of unrelated pathways in matrix operations.

[0008] Preferably, the fluid tensor mapping module further includes topology level correction logic, which is used to identify the root node as the mechanical receptor in the reference signal path topology graph, and to use a breadth-first search algorithm to traverse and calculate the minimum topology hop count of the remaining nodes relative to the root node; the logic generates a mask matrix containing the signal attenuation coefficient based on the minimum topology hop count, and uses the mask matrix to perform cascaded weight reduction processing on the sparse shearing adjustment tensor to characterize the signal attenuation characteristics along the topology depth at the data structure level.

[0009] Preferably, when generating the mask matrix, the topology level correction logic sets the corresponding signal attenuation coefficient to zero for nodes with a minimum topology hop count greater than a preset depth, thereby constructing a logical blocking boundary for fluid shear signals in the shear correction topology map to prevent deep noise interference.

[0010] Preferably, the fluid tensor mapping module is configured with fatigue accumulation calculation logic, which associates a non-zero element in the sparse shear regulation tensor with an initially zeroed fatigue state variable; in continuous time step calculation, if the fluid dynamic parameters are detected to remain within the effective stimulation range, a monotonically increasing operation is performed on the fatigue state variable, and the value of the sparse shear regulation tensor is dynamically reduced using an inverse inhibition function to simulate the signal path passivation effect under long-period action.

[0011] Preferably, the system further includes a pharmacological stiffness feedback module connected in series before the fluid tensor mapping module. This module is used to retrieve the pharmacological and mechanical property database based on the molecular identifier of the target drug and obtain the cell stiffness correction coefficient. This module uses the cell stiffness correction coefficient to perform scalar multiplication correction on the sparse shear regulation tensor in order to inversely compensate for the shear response sensitivity drift caused by the target drug changing the cell mechanical properties before generating the shear correction topology map.

[0012] Preferably, the baseline network construction module is configured with background stress fingerprint differential logic, which extracts the expression level of preset mechanical response markers in transcriptome sequencing data to calculate the intrinsic stress index and generate a background subtraction factor accordingly; when the dynamic topology evolution module performs Hadamard product operation, it uses the background subtraction factor to negatively gain correct the effect strength of the sparse shearing regulation tensor to eliminate the inherent mechanical stress background noise of the sample.

[0013] Preferably, the fluid dynamics parameters include pulsation frequency values. The fluid tensor mapping module retrieves a preset frequency response spectrum, determines the resonance gain factor of key nodes in the reference signal path topology diagram for the pulsation frequency values, and uses the resonance gain factor to perform multiplicative amplification on the corresponding elements in the sparse shear adjustment tensor to characterize the nonlinear activation of the signal path by fluid pulsation at a specific frequency.

[0014] Preferably, the system is deployed in an electronic computing device containing a processor and memory. Transcriptome sequencing data, baseline adjacency matrix, sparse shearing regulation tensor, and sparse latent crosstalk matrix are all stored in the memory in the form of floating-point arrays. The processor performs Hadamard product operations and matrix superposition operations to complete the numerical prediction of drug sensitivity of gastric cancer cells in a dynamic fluid environment.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In tumor shear pathway modeling, pre-defined mechanical transduction mapping rules are used to translate continuous fluid dynamics parameters into shear regulation matrices of the same dimension as the baseline signal pathway topology. The Hadamard product algorithm is used to perform weighted operations on the regulation matrix and the adjacency matrix. The complex physical process is solved by relying on computational fluid dynamics equations. At the data processing level, the dimensionality is reduced to a static algebraic operation process. In the pure digital space, the physical shear force is abstracted into a weight operator that changes the network connection strength. This allows the system to quickly complete the simulation of cell signal pathway reconstruction under specific fluid conditions in a standard computing environment without relying on physical microfluidic chips and high-performance simulation clusters. The mechanism enables the drug sensitivity analysis system to process biological network data containing a large number of nodes with extremely low time complexity, achieving high-throughput prediction under conventional medical information system hardware conditions.

[0016] 2. During the generation of the shear regulation matrix, a topology hierarchical correction logic based on a breadth-first search strategy is integrated. The system uses nodes marked as mechanosensitive receptors as the root node, traverses and calculates the shortest path hop count of the remaining nodes in the network relative to the root node, and generates an attenuation mask matrix containing signal transduction attenuation coefficients. Based on graph structure attribute data processing logic, a spatial depth index from cell membrane surface receptors to cell nucleus transcription factors is established at the logical level. The initial shear regulation tensor cascade is weighted and corrected using the attenuation mask matrix, enabling the computational model to distinguish the differences in the position of drug targets on the signal transduction path, automatically assigning lower shear response weights to deeper nodes, eliminating prediction bias caused by ignoring signal transduction distance, and improving the model's confidence in drug analysis involving nuclear mechanisms of action without introducing three-dimensional geometric modeling data.

[0017] 3. Fatigue accumulation calculation logic is configured in the shear stress vectorization process. The non-zero elements in the matrix are associated with the initial zero fatigue state variables. The system monitors fluid parameters in the time step calculation. When the stimulation threshold is exceeded, the state variables are triggered to increase monotonically. The shear adjustment matrix value is dynamically updated through the inverse inhibition function. Based on a simple counter and condition judgment algorithm, the time dimension memory function is embedded in the static graph data structure to simulate the mechanical adaptation phenomenon of cells under long-term high shear force due to receptor inactivation or endocytosis. The logic system predicts that drug sensitivity will drift nonlinearly with the extension of the treatment cycle, filling the logical gap in the long-term efficacy prediction based on static gene expression data model. Attached Figure Description

[0018] Figure 1 This is a system logic flowchart of the present invention that integrates fluid tensor mapping and dynamic topology evolution; Figure 2 This is a comparison chart of long-term drug sensitivity prediction errors under the spatiotemporal coupling correction logic of this invention; Figure 3 This is a system hardware deployment architecture diagram for the present invention that integrates microfluidics and high-performance computing clusters. Detailed Implementation

[0019] The present invention will be further described in detail below with reference to specific embodiments. The specific embodiments described in this section are only used to explain the present invention and are not limited to the scope of protection of the present invention.

[0020] This invention provides a drug sensitivity analysis system for gastric cancer cells used for tumor cleavage pathway modeling. Based on the algebraic mapping of fluid dynamics parameters and biological network topology, it quantifies the drug response characteristics of gastric cancer cells in a dynamic fluid environment at the data level. The system mainly includes a baseline network construction module, a fluid tensor mapping module, a dynamic topology evolution module, and a drug efficacy propagation calculation module. The baseline network construction module is used to establish a static topological foundation for the signaling pathway. This module acquires transcriptome sequencing data from the gastric cancer cell samples to be analyzed, parses it into gene expression vectors, and the processor retrieves a protein interaction database pre-stored in memory. It then maps the expression vectors to the connection relationships defined in the database, constructing a baseline signaling pathway topology graph containing a vertex set representing protein nodes and a baseline adjacency matrix defining initial connectivity weights. This module integrates background stress fingerprint differential logic to eliminate inherent mechanical stress noise in the samples. This logic retrieves the expression levels of preset mechanical response biomarkers in the transcriptome data, including CTGF, CYR61, and ANKRD1. The system uses a weighted average algorithm to calculate the dimensionless intrinsic stress index. When the inherent stress index When the value is higher than the preset baseline, for example, more than 3 times the baseline value, the system generates a background subtraction factor of 0.3. Background deduction factor Used to subsequently correct the shear strength and prevent signal baseline drift.

[0021] The fluid tensor mapping module is used to transform physical environmental parameters into mathematical regulation operators. This module receives fluid dynamic parameters characterizing the target physiological environment, including fluid velocity, pipe diameter, fluid viscosity, pulsation frequency, and pulsation index. The processor calculates the shear stress scalar under the current operating condition based on the Poiseuille flow formula. The system generates a sparse shearing adjustment tensor of the same dimension as the reference adjacency matrix based on a pre-set mechanical sensitivity lookup table. The sparse shearing modulating tensor As a diagonally dominant matrix, non-unit-valued adjustment coefficients are generated only at the diagonal positions corresponding to the mechanosensitive protein nodes marked in the baseline signal pathway topology graph, while the remaining elements remain at 1. During this process, the module executes frequency domain feature mapping logic, retrieves the preset mechanosensitive channel frequency response spectrum, and determines the resonance sensitivity of key nodes to the current input pulsation frequency value. These key nodes include Piezo1 and other ion channel proteins. If the input pulsation frequency value... If the system falls within the resonance interval of this node, it generates a resonance gain factor greater than 1. And adjust the tensor for sparse shearing The corresponding diagonal elements are multiplied and amplified, for example, when the pulsation frequency value... When the frequency is 1.2 Hz and is at the resonant center frequency, the corresponding adjustment coefficient is amplified by 1.5 times to characterize the nonlinear activation of the signal path by fluid pulsation at a specific frequency.

[0022] The fluid tensor mapping module also includes topology hierarchy correction logic to simulate spatial signal attenuation. This logic identifies nodes marked as mechanosensitive receptors such as integrins or ion channels in the reference signal path topology as level 0 root nodes, and uses a breadth-first search algorithm to traverse and calculate the minimum topology hop count of the remaining nodes relative to the nearest root node. The system is based on this minimum topology hop count. Generation and Sparse Shearing Adjustment Tensor A decay mask matrix of the same dimension, wherein the elements of the decay mask matrix follow the... The exponential decay law, in which As a preset signal attenuation constant, the system combines the attenuation mask matrix with the initially generated sparse shearing adjustment tensor. Perform the Hadamard product operation to cascade and reduce the weights of nuclear transcription factor nodes located deep in the topology, targeting the minimum topological hop count. For nodes exceeding a preset depth threshold, such as more than 6 hops, the system sets the corresponding signal attenuation coefficient to 0 and constructs a logical blocking boundary for the fluid shear signal at the data structure level. The fluid tensor mapping module is configured with fatigue accumulation calculation logic to characterize the adaptability in the time dimension. This logic is for the sparse shear adjustment tensor. Non-zero elements in the equation are associated with fatigue state variables initially set to 0. In continuous time-step calculations, if the input hydrodynamic parameters remain above the effective stimulus range, such as the shear stress scalar... If the value is consistently greater than 2 dyn / cm², then the fatigue state variable... The system executes a monotonically increasing behavior and utilizes an inverse inhibition function. Dynamically calculate the sensitivity retention ratio and update the sparse shearing modulating tensor The value of causes the signal path's response intensity to fluid shear to gradually decrease over simulation time.

[0023] The system also includes a pharmacological stiffness feedback module, which is used to introduce the reverse regulation of cell mechanics by the drug. This module searches the pharmacological and mechanical property database based on the target drug molecule identifier to obtain the cell stiffness correction coefficient. For example, this coefficient is relevant for paclitaxel-based drugs that cause microtubule sclerosis. The value is 0.8; this coefficient is for cytochalasin drugs that cause cytoskeleton depolymerization. The value is set to 1.2, and the system utilizes this coefficient. Sparse shearing modulating tensor All elements undergo scalar multiplication correction to compensate for shear response sensitivity drift caused by changes in cytoskeleton stiffness due to drug alterations; the dynamic topology evolution module performs network reconstruction based on physical thresholds, and this module embeds a sparse latent crosstalk matrix. Defines non-regular cross-pathway connections that are activated only when the cytoskeleton undergoes severe deformation; the module executes structural mutation decision logic and compares the modified shear stress scalar. Compared with the preset skeleton collapse threshold ,when At that time, the system only performs sparse shearing adjustment tensor. Adjacency matrix with reference The Hadamard product operation, when For example, shear stress scalar When the density exceeds 15 dyn / cm², the system triggers a structural mutation instruction, following algebraic logic. Perform the operation, where For the reconstructed target matrix, It represents the Hadamah accumulation. As an indicator function, this operation transforms the sparse latent crosstalk matrix Non-zero elements are superimposed onto the weighted matrix to establish cross-pathway connections that do not exist in the baseline state in the shear correction topology graph; the drug efficacy propagation calculation module is used to output the final prediction results. This module maps the molecular feature vector of the target drug to the target matrix corresponding to the shear correction topology graph. The system utilizes a graph neural network algorithm to simulate the propagation of drug perturbations in the network, calculates the perturbation state changes of key effector protein nodes, and outputs a quantified drug sensitivity score, reflecting the predicted response of gastric cancer cells to the target drug under the combined effects of a specific hydrodynamic environment, a specific pulsation frequency, and specific pharmacokinetics. In the specific mapping process, this invention first extracts the molecular fingerprint features or SMILES sequence of the target drug, and identifies the directly acting protein node corresponding to the drug in the baseline signal pathway topology graph through a preset pharmacological target matching logic. Subsequently, the system retrieves the dissociation constant between the drug and the target protein from the pharmacokinetics database, and converts this biochemical affinity parameter into an initial perturbation intensity scalar for the corresponding node in the graph neural network input layer. If the drug has multiple targets, the feature vectors of the corresponding vertices in the topology graph are initialized according to the affinity weights of each target, thereby transforming the discrete molecular structure features into the initial value distribution of nodes on the shear-corrected topology graph, providing a physical benchmark for the perturbation propagation calculation of the graph neural network.

[0024] Example 1: In a clinical evaluation scenario for advanced gastric cancer with malignant ascites metastasis, the tumor cells in the patient's abdominal cavity are chronically exposed to a complex hydrodynamic environment caused by ascites flow and intestinal peristalsis. This manifests as fluid velocity fluctuations between 0.5 cm / s and 5.0 cm / s, accompanied by low-frequency mechanical pulsations with a frequency of approximately 0.2 Hz. The resulting shear stress scalar... The data exhibits non-constant dynamic changes, and the background contains a high baseline mechanical stress level caused by long-term stimulation by inflammatory factors. Under these conditions, conventional static culture or homogeneous fluid simulation models often fail to accurately reproduce the drug resistance drift of tumor cells due to long-term mechanical adaptation, and also struggle to capture abnormal crosstalk in signaling pathways induced by extreme shear forces. This leads to a serious disconnect between the predicted efficacy of cytoskeleton-regulating drugs such as paclitaxel and clinical reality. When the present invention's system processes transcriptome sequencing data in this scenario, the baseline network construction module identifies the abnormally high expression state of mechanical response markers such as CTGF and ANKRD1 in the sample through background stress fingerprint differential logic, and calculates the intrinsic stress index. The background subtraction factor is 0.75, generating a value of 0.4. The data is then transmitted to subsequent modules. The fluid tensor mapping module receives the measured fluid dynamics parameters from the field and calculates the real-time shear stress scalar. The pulsation frequency was 8 dyn / cm², and a pulsation frequency of 0.2 Hz was identified as falling precisely within the resonance-sensitive region of the Piezo1 channel protein, triggering the frequency domain feature mapping logic to generate a resonance gain factor of 1.8. The system generates a sparse shearing adjustment tensor based on the above parameters. Not only utilizing the resonance gain factor Nonlinear amplification is applied to the corresponding nodes of Piezo1, and a background subtraction factor is also used. The overall effect intensity is negatively corrected to remove inherent background noise from the sample and highlight the activation effect of fluid pulsation. Meanwhile, the pharmacological stiffness feedback module retrieves the cell stiffness correction coefficient based on the input paclitaxel drug label. The modulating tensor for sparse shearing is 0.85. Further scalar multiplication corrections are performed to compensate for the cytoskeleton sclerosis effect caused by the drug itself.

[0025] As the simulation timestepped to the 10th cycle, the fatigue accumulation calculation logic detected that the shear stress remained within the effective stimulation range, and the fatigue state variables of the relevant nodes... When the accumulation reaches a critical value, the inverse suppression function begins to take effect and suppresses the sparse shearing modulator tensor. The numerical values ​​simulate the adaptive desensitization of cells to continuous fluid stimulation. At a certain instant in the simulation, when the fluid velocity surges, leading to a shear stress scalar... Briefly exceeded the 15 dyn / cm² skeletal collapse threshold At that time, the dynamic topology evolution module triggers a structural mutation command, which changes the sparse latent crosstalk matrix. Non-zero elements are superimposed onto the baseline adjacency matrix, instantly reconnecting the abnormal connection between actin and the Bcl-2 anti-apoptotic pathway, which was in a disconnected state under the baseline condition. The pharmacodynamic propagation calculation module then dynamically reconstructs the shear correction topology based on this. The calculation showed that the drug sensitivity score of paclitaxel under this specific spatiotemporal condition was lower than the value predicted by the static model.

[0026] Example 2: To verify the predictive effectiveness and stability of the gastric cancer cell drug sensitivity analysis system proposed in this invention under real clinical samples and extreme physiological conditions, this experiment constructed a standardized multi-dimensional comparative verification platform. The basic data used in the experiment came from a clinical study involving 50 patients with advanced gastric cancer and peritoneal metastasis. All samples were collected using a standardized procedure and high-throughput transcriptome sequencing was performed. The core objective was to quantitatively evaluate the technical advantages of this invention in solving the challenges of dynamic fluid interference and long-term drug resistance prediction by comparing the differences between the system of this invention and existing mainstream technologies in predicting drug sensitivity, and by combining the experimental results of the in vitro microfluidic physical model. The experimental platform utilizes a high-performance computing cluster as the digital simulation environment, and is equipped with a microfluidic chip physical test bench capable of accurately simulating the peritoneal fluid dynamics environment. This microfluidic platform is equipped with a precision pressure pump and a pulsation generator, capable of reproducing the fluctuations in intraperitoneal fluid velocity within the range of 0.5 cm / s to 5.0 cm / s, as well as physiological pulsations at a frequency of 0.2 Hz. To ensure the experimental results closely reflect engineering realities, Gaussian white noise with a signal-to-noise ratio of 20 dB is introduced into the input fluid parameters to simulate complex biophysical disturbances within the body. Regarding the setting of key parameters in drug sensitivity analysis, this experiment follows a strict technical trade-off logic, particularly for shear stress scalars... The simulation range was set to 0 to 20 dyn / cm², covering all physiological conditions from static microenvironments to extreme high-velocity scouring, for the skeletal collapse threshold. The setting was not based on experience, but on the critical stress value of actin depolymerization in cell mechanics experiments, which was determined to be 15 dyn / cm² after statistical analysis.

[0027] The experimental design comprises four comparative schemes with clearly defined technical feature gradients: the first group is the static baseline group, employing a traditional G16H-type static gene regulation network model without introducing any fluid dynamics parameters, predicting solely based on initial gene expression levels; the second group is the linear correction group, introducing fluid shear stress parameters but only performing linear correction on signal pathway weights, lacking the dynamic topology evolution module and frequency domain feature mapping logic of this invention; the third group is the complete invention group, fully running the complete logic including fluid tensor mapping, dynamic topology evolution, frequency domain feature mapping, and fatigue accumulation calculation; the fourth group is the out-of-range verification group, incorporating fluid shear stress... The target concentration was set at 30 dyn / cm², exceeding the design range, to verify the model's response behavior under extreme boundary conditions. Paclitaxel was selected as the target drug, and the reciprocal of the half-maximal inhibitory concentration (IC50) was used as the drug sensitivity score. The predicted scores output by each model were compared with the measured apoptosis rate on the microfluidic physics platform to calculate the relative prediction error. The experiment showed that under low shear stress ( Within the range of dyn / cm², the predicted results of the static reference group and the linear correction group deviate little from the measured values, with the relative error controlled within 15%. As the shear stress... The increase, especially when Beyond 10 dyn / cm², the static baseline group, unable to perceive the remodeling effect of the physical environment on the signal pathway, experienced an exponential increase in prediction error, severely overestimating drug efficacy. The linear correction group, however, showed... It exhibits a certain correction effect within the range of 10 to 15 dyn / cm², but... After exceeding the skeleton collapse threshold of 15 dyn / cm², the measured error rapidly increased, which could not explain the sudden increase in drug resistance observed in the microfluidic experiment.

[0028] In stark contrast, the entire assembly of this invention is in When the density reaches 15 dyn / cm², the structural mutation logic is accurately triggered, and the prediction score shows a significant downward inflection point, which is highly consistent with the measured decreasing trend of cell apoptosis rate. The relative prediction error is consistently controlled within 8%. In the out-of-range validation group, the sensitivity score of the model output further decreases and tends to saturate, indicating that the system successfully simulates the survival limit state of cells under extreme stress and verifies the stability of the model in the nonlinear range. Table 1 details the comparison between the key prediction data and the measured true values ​​of each experimental group under different shear stress conditions.

[0029] Table 1: Comparison of Drug Sensitivity Prediction Errors of Different Models under Different Shear Stress Conditions

[0030] Data shows that in extreme fluid environments ( Traditional static models, lacking awareness of structural aberrations (dyn / cm²), suffer from prediction errors as high as 131.6%, rendering them completely ineffective. While linear correction models offer some improvement, they still exhibit a significant bias of 83.6% when dealing with nonlinear pathological responses. This invention addresses this issue by dynamically superimposing sparse latent crosstalk matrices. It successfully simulated the activation of abnormal anti-apoptotic pathways induced by high shear stress, keeping the prediction error at an extremely low level of 7.3%.

[0031] Example 3: This example combines Figures 1 to 3 A description of a drug sensitivity analysis system for gastric cancer cells used for tumor cleavage pathway modeling, such as... Figure 1As shown, the baseline network construction module receives transcriptome sequencing data to construct a baseline signaling pathway topology and eliminate background mechanical stress noise, outputting a baseline adjacency matrix. Next, the fluid tensor mapping module receives fluid dynamics parameters and generates a sparse shear regulation tensor containing topology hierarchy correction and fatigue accumulation logic. Simultaneously, the pharmacological stiffness feedback module searches a pharmacological and mechanical property database based on the target drug molecule characteristics to obtain cell stiffness correction coefficients, and uses these coefficients to perform reverse compensation shear sensitivity processing on the aforementioned sparse shear regulation tensor. The dynamic topology evolution module receives the baseline adjacency matrix and the corrected sparse shear regulation tensor, executes structural mutation decision logic, and superimposes a sparse latent crosstalk matrix when specific threshold conditions are met to generate a shear-corrected topology. Finally, the drug efficacy propagation calculation module receives this shear-corrected topology and maps the target drug molecule characteristics, using a graph neural network algorithm to calculate node perturbation propagation efficiency and outputting the final drug sensitivity score.

[0032] like Figure 2 As shown, the horizontal axis represents simulation time in hours, covering a time span of 0 to 48 hours, and the vertical axis represents the prediction relative error in percentage. The graph contains four trend lines. The error of the fully transparent reference group continuously increases over time and remains above 100%, eventually exceeding 200%, indicating the failure of the uncorrected model. The spatial correction group has a low error in the initial stage, but the error increases to 90% over time, indicating a lack of time-adaptive simulation. The time correction group has a high error in the initial stage, gradually decreasing and stabilizing, indicating a lack of spatial depth simulation. The spatiotemporal coupling correction group, which is the complete solution of this invention, maintains a low error curve throughout the entire 0 to 48 hour time span. Figure 3 As shown, the left side depicts the clinical physics laboratory environment, equipped with a high-throughput sequencer for generating transcriptome data and a microfluidic physics testing platform for generating fluid dynamics parameters. The sequencing data and fluid parameters are uploaded via a network. The middle section shows the core high-performance computing cluster server, configured with floating-point array storage and parallel computing. The underlying hardware resources include multi-core processors and high-speed memory. The server deploys a baseline network construction component for background stress fingerprint differentiation, a fluid tensor mapping component for generating sparse shear regulation tensors, a pharmacological stiffness feedback component for calculating cell stiffness correction coefficients, a dynamic topology evolution component for structural mutation judgment and matrix reconstruction, and a pharmacodynamic propagation calculation component for graph neural network sensitivity scoring. The right side shows the data storage center, which interacts with the server for data retrieval and access. It internally stores a protein interaction database with predefined topological connections and a pharmacological and mechanical property database containing drug stiffness and molecular characteristics. The server's calculation results are ultimately sent to the medical workstation at the bottom.

[0033] Example 4: In a supplementary experiment to verify the effectiveness of the topology hierarchy correction logic and fatigue accumulation calculation logic of the present invention in long-term, deep signal transduction scenarios, a spatiotemporal analysis simulation platform focused on simulating the dynamic response of nuclear transcription factors was constructed. The core objective of the experiment was to quantitatively evaluate the improvement in the model's accuracy in predicting nuclear targets such as transcription factors, DNA repair enzymes, and drug efficacy after introducing a signal attenuation mechanism based on topology depth and a fatigue adaptation mechanism based on time dimension. The basic data and fluid dynamics parameters used in the experiment were consistent with those in Example 2, but additionally, fluorescently labeled measured data for the nuclear translocation rate of specific nuclear transcription factors such as YAP / TAZ were introduced as verification values. To specifically verify the independent and synergistic effects of the two core logic modules, four comparative schemes with clear functional gradients were designed: the first group was the full permeation baseline group, in which the topology hierarchy correction logic and fatigue accumulation calculation logic were removed, assuming that the fluid shear signal could be transmitted to the nuclear cavity without attenuation and that the cell did not produce fatigue adaptation to continuous stimulation; the second group was the spatial correction group, in which a minimum topology hop count was introduced. The attenuation mask matrix performs cascaded weight reduction on deep nodes, but fatigue accumulation logic is not enabled; the third group is a time-only correction group, which introduces fatigue state variables. The first group has a dynamic suppression mechanism, but the topology level correction logic is not enabled; the fourth group is the spatiotemporal coupling correction group, which is the complete solution of this invention, and simultaneously enables spatial decay and time fatigue mechanisms.

[0034] The experiment selected platinum-based chemotherapy drugs, such as oxaliplatin, that act on nuclear DNA as the target drug. The focus was on examining the model's predictive ability for drug-induced apoptosis rates under continuous fluid shearing for up to 48 hours. The experiment showed that in the initial simulation phase (0-12 hours), both the full permeation baseline group and the time-corrected group had higher predictions than the measured values. This was because the model ignored spatial attenuation during signal transmission from the cell membrane to the nucleus, incorrectly equating the strong shearing stimulation of membrane surface receptors directly to the activation intensity of nuclear transcription factors, leading to an overestimation of the activation level of nuclear drug resistance pathways. In contrast, the spatial correction group and the spatiotemporal coupling correction group, by introducing an attenuation constant, demonstrated superior performance. (Set to 0.3), reasonable signal weighting is applied to nuclear nodes with a topological depth greater than 4 hops, significantly reducing the deviation between the initial prediction results and the measured values; the attenuation constant α=0.3 is calculated based on the signal conduction dissipation rate in the protein-protein interaction network (PPN). Based on the known statistical laws of signal cascade amplification and negative feedback regulation in this invention, the effective perturbation intensity of the physical stress signal to the downstream target point decreases by an average of about 25%-30% for each topological hop. This invention determines 0.3 as the empirical constant that best matches the characteristics of the real signal attenuation gradient by performing linear regression fitting on the conduction link from integrin to nuclear YAP protein in multiple gastric cancer cell lines; as the simulation time is extended to 24 hours and 48 hours, the prediction error of only the spatial correction group begins to gradually increase, manifested as the predicted drug sensitivity being lower than the measured value, i.e., overestimating drug resistance. This is because this model cannot simulate the mechanical adaptation of cells under long-term stimulation and maintain a constant shear response weight. The spatiotemporal coupling correction group plays a decisive role at this time. Its fatigue accumulation calculation logic monitors the continuous effective stimulation and drives the fatigue state, variable Monotonically increasing, the shear regulation tensor value is dynamically reduced through the inverse inhibition function. This mechanism successfully simulates the gradual desensitization of the mechanosensitive pathway in the cell nucleus, causing the predicted drug sensitivity to rebound in the later stage. It is highly consistent with the measured cell apoptosis rate curve. Table 2 lists in detail the relative error comparison of the predicted drug sensitivity of each experimental group for the intranuclear target under different simulation durations.

[0035] Table 2: Comparison of Prediction Error of Intranuclear Target Drug Sensitivity over Time under Different Correction Logics

[0036] In long-term drug action scenarios involving intranuclear targets, single-dimensional corrections cannot meet the requirements for high-precision prediction. The full-transmission reference group, due to the complete lack of spatiotemporal constraints, consistently exhibits extremely high prediction errors (>150%). The spatial correction group, while performing well initially, fails over long periods due to neglecting fatigue effects (48-hour error +88.7%). The temporal correction group, although capable of simulating later adaptation, still suffers from significant overall deviations due to initial spatial signal overload. Only the spatiotemporal coupling correction group, by simultaneously introducing spatial hierarchical attenuation and temporal fatigue accumulation, achieves physical constraints on the signal transmission path, stabilizing the prediction error throughout the entire cycle to within 10%.

[0037] Example 5: To eliminate the uncertainty of the data source in the construction process of the mechanical sensitivity lookup table and the frequency response spectrum of the mechanical sensitivity channel, this example implements a standardized offline parameter calibration and meta-knowledge base filling procedure, defining the shear stress scalar The orthogonal calibration space formed by the target signal path nodes, where The value range covers 0 to 30 dyn / cm², with a step size set at 0.5 dyn / cm². For each discrete working condition within this space, a constant flow field stimulation is applied to the standardized gastric cancer cell line using a high-throughput microfluidic chip. The transcription fold change values ​​of key nodes such as integrin, YAP / TAZ, and PI3K are measured using real-time quantitative PCR technology. These change values ​​are normalized and converted into dimensionless response coefficients, which are then filled into the corresponding storage units of the lookup table. For the frequency domain feature mapping logic, a programmable pulsatile pump is used to generate a swept-frequency flow field from 0.1 Hz to 2.0 Hz. Combined with an intracellular calcium ion fluorescent probe and a high-speed confocal microscope, the channel opening probability is recorded in real time and mapped to a resonance sensitivity coefficient. The measured data shows that the Piezo1 channel has a resonance peak at 1.2 Hz, and its corresponding sensitivity coefficient is set to a peak value of 1.8. This physical experimental data constitutes the metadata basis for the system's frequency domain analysis.

[0038] To ensure consistent output across different sequencing platforms or heterogeneous computing environments, the system is configured with a pre-deployment calibration procedure. Before formally running clinical samples, the system loads a benchmark dataset containing known drug sensitivity phenotypes for trial operation. If the deviation between the system's predicted score and the standard value exceeds a preset tolerance... If this happens, the parameter fine-tuning algorithm is triggered, and the global gain coefficient of the shear adjustment tensor is iteratively corrected using the gradient descent method until the prediction error converges to an acceptable range.

[0039] Example 6: To ensure the stable operation of the fluid tensor mapping module and the dynamic topology evolution module in practical engineering applications, and to eliminate potential computational biases introduced by sample differences or environmental changes, this example establishes a standardized offline parameter calibration and online operation calibration engineering procedure as a necessary prerequisite for system deployment. By constructing a controlled physical-digital mapping benchmark, it provides key algorithm modules with clearly traceable initialization parameters. In the offline parameter calibration stage, a parameter extraction platform based on a standardized microfluidic chip is constructed. A step-like shear stress field is generated using a programmable fluid pump, and the shear stress scalar... The calibration range was set to 0 to 30 dyn / cm², with a step size refined to 0.5 dyn / cm². For each shear stress step, the deformation degree of the actin backbone and the nucleocytoplasmic distribution ratio of key signaling proteins such as YAP in standardized cell lines such as AGS were monitored simultaneously using real-time fluorescence imaging technology. The system collected the above physical response data and used the least squares method to fit and generate a baseline curve for the mechanical sensitivity lookup table.

[0040] In addition to addressing environmental adaptability issues during online operation, this procedure integrates an adaptive baseline calibration logic. Before each clinical sample analysis is initiated, the system runs a set of pre-set calibration quality control sequences, including standard simulated signals under known fluid parameters. The system calculates the output response of this sequence and compares it with a pre-set standard response curve to calculate the baseline drift. If the drift exceeds... Within the allowable tolerance, the system automatically triggers a gain compensation algorithm, which performs reverse compensation by adjusting the global scaling factor of the shear adjustment tensor until the system response returns to the standard range. This closed-loop calibration mechanism effectively isolates the interference caused by floating-point errors in computing hardware or batch effects in sequencing data, ensuring the consistency and reproducibility of drug sensitivity analysis results in different spatiotemporal dimensions.

[0041] Example 7: This example constructs a standard topology discovery process for the sparse latent crosstalk matrix, establishes the physical authenticity of cross-pathway connections under extreme fluid conditions, and establishes a paired sample library including a static control group and a high-shear experimental group. The high-shear experimental group was cultured for 24 hours in a microfluidic environment maintaining a constant shear stress of 30 dyn / cm² to induce a steady-state pathological response. The whole transcriptome of the two groups of samples was sequenced, and the weighted co-expression coefficients between gene pairs were calculated. Connections with a correlation coefficient lower than 0.2 in the static group and a correlation coefficient higher than 0.85 in the high-shear group were retained as candidate differential topologies. The Bayesian network inference algorithm was used to eliminate indirect regulatory redundancy, and the remaining direct action paths were mapped to non-zero elements of the sparse latent crosstalk matrix to ensure that each connection in the matrix corresponds to a high-stress induced real signal bypass rather than random noise. The system is configured with dynamic confidence verification logic for the sparse latent crosstalk matrix. In the clinical application stage, the drug efficacy propagation calculation module monitors drug sensitivity in real time. If the root mean square error of the residuals between the predicted value of the sensitivity score and the patient follow-up data exceeds 15% for three consecutive treatment cycles, the system automatically triggers a gradient descent-based weight correction instruction to fine-tune the numerical strength of the non-zero elements of the sparse latent crosstalk matrix while maintaining the topological structure. In specific implementation, the non-zero elements of this sparse latent crosstalk matrix represent abnormal connections activated under extreme mechanical stimulation. For example, in gastric cancer cells, this matrix defines the non-regular direct enhancement pathway from integrin to the anti-apoptotic protein Bcl-2, as well as the mechanical coupling edge between actin-binding protein and the YAP / TAZ nuclear translocation pathway. The initial weights of these non-zero elements are quantified based on the difference in co-expression coefficients of the aforementioned high-shear experimental groups, and their distribution positions in the matrix correspond to potential mechanosensitive sites in the protein interaction database, thereby ensuring that the reconstructed target matrix can accurately reflect the drug resistance signaling loop under high flow rate conditions.

[0042] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A drug sensitivity analysis system for gastric cancer cells used for tumor cleavage pathway modeling, characterized in that, include: The baseline network construction module is used to acquire transcriptome sequencing data of gastric cancer cells to be analyzed, retrieve a pre-set protein interaction database, and construct a baseline signaling pathway topology consisting of a set of vertices representing protein nodes and a baseline adjacency matrix defining initial connectivity weights. The fluid tensor mapping module is used to receive fluid dynamic parameters characterizing the target physiological environment and calculate the shear stress scalar. Based on a preset mechanical sensitivity lookup table, it generates a sparse shear regulation tensor with the same dimension as the reference adjacency matrix. The dynamic topology evolution module has an embedded sparse latent crosstalk matrix that defines non-regular connections across pathways. It is used to execute the structural mutation decision logic: when the shear stress scalar is lower than the preset skeleton collapse threshold, only the Hadamard product operation of the sparse shear adjustment tensor and the reference adjacency matrix is ​​performed to generate a shear correction topology graph. When the shear stress scalar is not lower than the skeleton collapse threshold, a structural mutation command is triggered, and the non-zero elements of the sparse latent crosstalk matrix are superimposed on the matrix after the Hadamard product operation, and cross-path connection edges that do not exist in the baseline state are established in the shear correction topology graph. The drug efficacy propagation calculation module is used to map the molecular feature vector of the target drug to a shear-corrected topology graph, calculate the node perturbation propagation efficiency using a graph neural network algorithm, and output a drug sensitivity score.

2. The gastric cancer cell drug sensitivity analysis system for tumor cleavage pathway modeling according to claim 1, characterized in that, The superposition of the Hadamard product operation and the structural mutation instruction executed by the dynamic topology evolution module follows the following algebraic logic to achieve discrete jumps in the graph structure: ,in, For the reconstructed target matrix, Using the baseline adjacency matrix, To adjust the tensor for sparse shearing It represents the Hadamah accumulation. For indicator functions, For shear stress scalar, The threshold for skeletal collapse. It is a sparse latent crosstalk matrix.

3. The gastric cancer cell drug sensitivity analysis system for tumor cleavage pathway modeling according to claim 1, characterized in that, The sparse shearing regulation tensor is a diagonally dominant matrix. The fluid tensor mapping module generates non-unit regulation coefficients only for the diagonal elements corresponding to the nodes marked as mechanosensitive proteins in the baseline signal pathway topology graph, while the remaining elements retain unit values ​​to maintain the topological stability of unrelated pathways in matrix operations.

4. The gastric cancer cell drug sensitivity analysis system for tumor cleavage pathway modeling according to claim 1, characterized in that, The fluid tensor mapping module also includes topology hierarchy correction logic, which is used to identify the root node as the mechanical receptor in the reference signal path topology graph and use a breadth-first search algorithm to traverse and calculate the minimum topology hops of the remaining nodes relative to the root node. The topology-level correction logic generates a mask matrix containing the signal attenuation coefficient based on the minimum topology hop count, and uses the mask matrix to perform cascaded weight reduction processing on the sparse shear adjustment tensor to characterize the signal attenuation characteristics along the topology depth at the data structure level.

5. The gastric cancer cell drug sensitivity analysis system for tumor cleavage pathway modeling according to claim 4, characterized in that, When generating the mask matrix, the topology-level correction logic sets the corresponding signal attenuation coefficient to zero for nodes with a minimum topology hop count greater than a preset depth. This constructs a logical blocking boundary for fluid shear signals in the shear correction topology map, preventing deep noise interference.

6. The gastric cancer cell drug sensitivity analysis system for tumor cleavage pathway modeling according to claim 1, characterized in that, The fluid tensor mapping module is configured with fatigue accumulation calculation logic, which associates a fatigue state variable that is initially set to zero with a non-zero element in the sparse shear adjustment tensor. In continuous time-step calculations, if the hydrodynamic parameters are detected to remain within the effective stimulation range, a monotonically increasing operation is performed on the fatigue state variable, and the value of the sparse shear adjustment tensor is dynamically reduced using the inverse inhibition function to simulate the signal pathway passivation effect under long-period action.

7. The gastric cancer cell drug sensitivity analysis system for tumor cleavage pathway modeling according to claim 1, characterized in that, The system also includes a pharmacological stiffness feedback module connected in series before the fluid tensor mapping module. This pharmacological stiffness feedback module is used to search the pharmacological and mechanical property database based on the molecular identifier of the target drug to obtain the cell stiffness correction coefficient. This module uses a cell stiffness correction coefficient to perform scalar multiplication correction on the sparse shear regulation tensor, in order to inversely compensate for the shear response sensitivity drift caused by the target drug altering the cell's mechanical properties before generating the shear correction topology.

8. The gastric cancer cell drug sensitivity analysis system for tumor cleavage pathway modeling according to claim 1, characterized in that, The baseline network construction module is equipped with background stress fingerprint differential logic, which extracts the expression level of preset mechanical response markers in transcriptome sequencing data to calculate the intrinsic stress index and generate a background subtraction factor accordingly. When performing the Hadamard product operation, the dynamic topology evolution module uses the background subtraction factor to negatively correct the effect strength of the sparse shearing regulation tensor.

9. The gastric cancer cell drug sensitivity analysis system for tumor cleavage pathway modeling according to claim 1, characterized in that, The fluid dynamics parameters include pulsation frequency values. The fluid tensor mapping module retrieves the preset frequency response spectrum, determines the resonance gain factor of key nodes in the reference signal path topology diagram for the pulsation frequency values, and uses the resonance gain factor to perform multiplicative amplification on the corresponding elements in the sparse shearing adjustment tensor to characterize the nonlinear activation of the signal path by fluid pulsation at a specific frequency.

10. A drug sensitivity analysis system for gastric cancer cells for tumor cleavage pathway modeling according to claim 1, characterized in that, The system is deployed in an electronic computing device that includes a processor and memory. Transcriptome sequencing data, baseline adjacency matrix, sparse splicing regulation tensor and sparse latent crosstalk matrix are all stored in memory in the form of floating-point arrays. The processor performs Hadamard product operations and matrix superposition operations to complete numerical prediction of drug sensitivity of gastric cancer cells in a dynamic fluid environment.

Images