Effectiveness analysis method for inhibiting exosomes of liver cancer cells and related equipment

Abstract

The embodiment of the invention provides an effectiveness analysis method for inhibiting exosomes of liver cancer cells and related equipment, and relates to the field of medicine and artificial intelligence. The method comprises the steps of obtaining a first inhibition strategy matched with a patient and strategy data of the first inhibition strategy from a pre-constructed inhibition strategy library based on individualized data of the patient; the individualized data comprises molecular data, clinical data and microenvironment data; inhibition strategies included in the inhibition strategy library are formed by adopting different parameters under a targeted removal strategy and / or a receptor end intervention strategy; performing feature extraction on the individualized data and the strategy data to obtain user features and strategy features; fusing the user feature and the strategy feature to obtain a fusion feature of the first inhibition strategy related to the biological state change of the patient; and determining a prediction index value for indicating the effectiveness of the first suppression strategy based on the fusion feature. The implementation of the invention can be adapted to individual conditions, and the effectiveness of the adopted suppression strategy can be accurately evaluated.

Description

Technical Field

[0001] The present disclosure relates to the fields of medicine and artificial intelligence technology, and in particular, to a method for analyzing the effectiveness of inhibiting exosomes of liver cancer cells and related equipment. Background Art

[0002] Considering that exosomes secreted by liver cancer cells can be transferred to recipient cells through specific miRNAs, inhibiting the expression of certain tumor suppressor genes in liver cancer tissues and thereby promoting the growth, migration, and invasion of liver cancer cells, it is therefore possible to effectively improve the cure rate of liver cancer by inhibiting the effectiveness of liver cancer cell exosomes. However, there are currently deficiencies in the separation and purification technology and targeted inhibition technology for exosome processing, which makes it impossible to stably improve the cure rate of liver cancer by controlling liver cancer cell exosomes. In addition, due to the large individual differences among liver cancer patients, the composition and function of their exosomes also vary. Even if the control method for liver cancer cell exosomes is effective in a group, it may not be applicable to every individual, resulting in low feasibility and poor effectiveness of the method. Summary of the Invention

[0003] The embodiments of the present disclosure provide a method for analyzing the effectiveness of inhibiting exosomes of liver cancer cells and related equipment, which are used to adapt to individual conditions and accurately evaluate the effectiveness of inhibition strategies for inhibiting exosomes of liver cancer cells.

[0004] According to one aspect of the embodiments of the present disclosure, a method for analyzing the effectiveness of inhibiting exosomes of liver cancer cells is provided, comprising: Based on the patient's personalized data, a first inhibition strategy and its strategy data matching the patient are obtained from a pre-built inhibition strategy library; the personalized data includes molecular data, clinical data, and microenvironmental data related to liver cancer cell exosomes; the inhibition strategy library includes multiple inhibition strategies, each of which is formed using different parameters under a targeted clearance strategy and / or a receptor-end intervention strategy; Performing feature extraction on the individualized data and the policy data respectively to obtain user features and policy features; fusing the user feature and the strategy feature to obtain a fusion feature related to the first inhibition strategy and the change in the patient's biological state; Based on the fusion signature, a predictive index value is determined to indicate the effectiveness of the first inhibition strategy in inhibiting exosomes from liver cancer cells.

[0005] In a feasible embodiment, the step of obtaining a first inhibition strategy and its strategy data that matches the patient from a pre-built inhibition strategy library based on the patient's individualized data includes: Based on the clinical data, excluding inhibition strategies that meet preset mutually exclusive conditions from the inhibition strategy library to obtain an initial inhibition strategy; performing similarity calculation on the user vector corresponding to the individualized data and the strategy vector corresponding to the initial suppression strategy, and determining the suppression strategy having a similarity greater than a preset threshold as the first suppression strategy; The suppression strategy library is constructed based on the acquired instance samples.

[0006] In a feasible embodiment, before extracting features from the individualized data and the policy data respectively, the method further includes: Preprocessing the individualized data and the strategic data according to different data types; Align the preprocessed individualized data with the preprocessed strategy data.

[0007] In a feasible embodiment, extracting features from the individualized data and the policy data to obtain user features and policy features includes: Performing feature extraction on the individualized data through the first network branch to obtain user features including multi-dimensional heterogeneous data; Through the second network branch, feature extraction is performed on the strategy feature to obtain the strategy feature that integrates parameter information and strategy information.

[0008] In a feasible embodiment, the extraction of user features includes: Performing feature extraction on the molecular data in spatial and temporal dimensions to obtain a first feature vector; Performing spatial mapping processing of discrete features on the clinical data to obtain a second feature vector; For the microenvironment data, graph convolution aggregation processing is performed to obtain the third eigenvector; The first feature vector, the second feature vector, and the third feature vector are fused to obtain user features.

[0009] In a feasible embodiment, the extraction of the policy features includes: Performing spatial attention processing on the strategy data to obtain a parameter feature vector; Performing semantic encoding on the policy data to obtain a semantic feature vector; The parameter feature vector and the semantic feature vector are fused to obtain a strategy feature.

[0010] In a feasible embodiment, the fusing of the user feature and the strategy feature to obtain a fusion feature related to the first inhibition strategy and the change in the patient's biological state includes at least one of the following: Performing feature interaction based on the user features and the strategy features to obtain an interaction matrix; performing attention processing based on the interaction matrix to obtain a weight matrix; performing weighted fusion based on the weight matrix to obtain a fusion feature; Based on the graph corresponding to the user features, an interaction graph is constructed with the strategy features as node attributes of the graph; a graph convolution operation is performed on the interaction graph, and the updated interaction graph is aggregated to obtain a fusion feature.

[0011] In a feasible embodiment, determining a predictive index value indicating the effectiveness of the first inhibition strategy in inhibiting liver cancer cell exosomes based on the fusion signature includes: The fusion features are transformed through a fully connected layer; The output vector of the fully connected layer is processed through the activation function to obtain the predicted index value.

[0012] In a feasible embodiment, the predictive index value includes at least one of exosome clearance rate, receptor-end signaling pathway inhibition rate, or tumor growth inhibition rate; The method further comprises: Based on the predictive indicator value, a second suppression strategy matched to the patient is generated.

[0013] In a feasible embodiment, the method is performed by an artificial intelligence (AI) network, and updating of the AI ​​network includes: Acquiring a current first liver cancer state and a target inhibition strategy employed, wherein the target inhibition strategy is determined based on the second inhibition strategy; When the change in the patient's liver cancer status reaches a preset condition, obtaining a corresponding second liver cancer status; Based on the second liver cancer state, the first liver cancer state and the target suppression strategy, a reward value is determined, and network parameters of the AI ​​network are updated based on the reward value.

[0014] According to another aspect of an embodiment of the present disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method for analyzing the effectiveness of inhibiting exosomes of liver cancer cells as described in the above embodiment.

[0015] According to another aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for analyzing the effectiveness of inhibiting exosomes of liver cancer cells described in the above embodiment is implemented.

[0016] According to one aspect of an embodiment of the present disclosure, a computer program product is provided, including a computer program, which, when executed by a processor, implements the method for analyzing the effectiveness of inhibiting exosomes in liver cancer cells as described in the above embodiment.

[0017] The technical solutions provided by the embodiments of the present disclosure have the following beneficial effects: The present disclosure provides an effectiveness analysis method for inhibiting exosomes of liver cancer cells. Specifically, first, based on the patient's individualized data, a first inhibition strategy and its strategy data matching the patient can be obtained from a pre-built inhibition strategy library. The individualized data may include molecular data, clinical data, and microenvironmental data related to liver cancer cell exosomes. The inhibition strategy library may include several inhibition strategies, each of which is formed using different parameters under a targeted clearance strategy and / or a receptor-end intervention strategy. That is, the present disclosure can preliminarily screen out inhibition strategies that meet the individual needs of patients in the inhibition strategy library; then, feature extraction can be performed on the individualized data and strategy data respectively to obtain user features and strategy features, and the user features and strategy features can be fused to obtain fusion features related to the first inhibition strategy and the patient's biological state changes; finally, based on the fusion features, a predictive index value for indicating the effectiveness of the first inhibition strategy can be determined. The disclosed embodiments can, after screening for an initial inhibition strategy, predict the predictive index value of the inhibition strategy's effectiveness in inhibiting liver cancer cell exosomes through an artificial intelligence network. The predictive index value can directly and accurately reflect the effectiveness of the inhibition strategy for individual patients, thereby assisting medical staff to better adapt to individual patient conditions, determine more accurate treatment plans, and improve the effectiveness of controlling liver cancer cell exosomes. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following briefly introduces the drawings required for describing the embodiments of the present disclosure.

[0019] Figure 1 A schematic diagram of an analysis method for inhibiting the effectiveness of exosomes in liver cancer cells and related equipment provided in an embodiment of the present disclosure; Figure 2 A processing flow chart of an AI network provided in an embodiment of the present disclosure; Figure 3 A framework diagram of a first network branch provided in an embodiment of the present disclosure; Figure 4 A framework diagram of a second network branch provided in an embodiment of the present disclosure; Figure 5 A schematic structural diagram of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0020] The following describes embodiments of the present disclosure in conjunction with the accompanying drawings. It should be understood that the embodiments described below in conjunction with the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present disclosure and do not constitute a limitation on the technical solutions of the embodiments of the present disclosure.

[0021] Those skilled in the art will understand that, unless otherwise stated, the singular forms "a," "an," "said," and "the" used herein may also include the plural forms. It should be further understood that the terms "including" and "comprising" used in the embodiments of the present disclosure mean that the corresponding features can be implemented as the features, information, data, steps, operations, elements, and / or components presented, but do not exclude implementation as other features, information, data, steps, operations, elements, components, and / or combinations thereof supported by the present technical field. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, the element can be directly connected or coupled to the other element, or it can refer to the element and the other element establishing a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein indicates at least one of the items defined by the term, for example, "A and / or B" or "A, B" indicates implementation as "A," or implementation as "B," or implementation as "A and B."

[0022] The term "based on" used in various embodiments of the present disclosure can be interpreted as meaning that the premise, condition, or information on which the basis is based is not exclusive, but at least one or a portion of it. This means that there is at least one clear basis, and other possible bases are not excluded.

[0023] The following describes several exemplary embodiments to illustrate the technical solutions of the embodiments of the present disclosure and the technical effects produced by the technical solutions of the present disclosure. It should be noted that the following embodiments can refer to, draw on, or combine with each other, and the same terms, similar features, and similar implementation steps in different embodiments will not be repeated.

[0024] The following combination Figures 1 to 4 The method provided in the embodiment of the present disclosure is described in detail.

[0025] Specifically, if Figure 1 As shown, the method provided in the embodiment of the present disclosure includes S101 to S104: S101. Based on the patient's individualized data, obtain a first inhibition strategy and its strategy data that matches the patient from a pre-built inhibition strategy library; the individualized data includes molecular data, clinical data, and microenvironmental data related to liver cancer cell exosomes; the inhibition strategy library includes several inhibition strategies, each inhibition strategy is formed using different parameters under a targeted clearance strategy and / or a receptor-end intervention strategy.

[0026] Optionally, personalized data can include patient-specific data, such as molecular data related to liver cancer cell exosomes, clinical data, and microenvironmental data. Molecular data can include genomic data, such as liver cancer driver gene mutations and exosome-related gene expression; transcriptomic data, such as the specific expression profiles of exosome cargo (e.g., miRNA); and proteomic data, such as the quantitative levels of exosome surface markers and receptors on the surface of receptor cells. Clinical data can include tumor stage, vascular invasion, and liver function grade. Microenvironmental data can include the proportion of immune cell infiltration (e.g., through single-cell sequencing or multiplex immunofluorescence assays) and the degree of fibrosis.

[0027] Optionally, the inhibition strategy library is constructed based on acquired example samples. For example, treatment protocols for inhibiting exosomes in liver cancer cells from historical liver cancer treatment cases are obtained and processed to generate the inhibition strategy library. The inhibition strategies included in the inhibition strategy library can be examples or variations of examples.

[0028] Among them, the targeted clearance strategy aims to inhibit the biogenesis and secretion of exosomes, reducing the accumulation of exosomes in the tumor microenvironment by disrupting their synthesis, release, or clearance from the circulatory system. For example, genetic manipulation can be used, such as RNA interference, to disrupt key genes that regulate exosome biogenesis and secretion, inhibit the expression of ESCRT complex-related proteins or auxiliary proteins, and block the formation and release of exosomes. Drug inhibition can also be used, using small molecule drugs to inhibit key regulatory factors of exosome secretion. Physical clearance can also be used, such as through therapeutic plasma exchange or hemofiltration, to directly remove exosomes from the circulatory system.

[0029] Among them, receptor-intervention strategies aim to block the interaction between exosomes and target cells, inhibiting their pro-cancer effects by interfering with exosome uptake, signal transduction, or functional execution. For example, exosome uptake can be blocked by using competitive inhibitors or antibodies to prevent exosomes from binding to receptors on the surface of target cells. Alternatively, signal transduction can be interfered with by targeting signaling molecules (such as non-coding RNAs and proteins) carried by exosomes to block downstream signaling pathways. Alternatively, the immune microenvironment can be modulated, using exosomes as immunomodulators to activate anti-tumor immune responses.

[0030] Optionally, in the embodiments of the present disclosure, the targeted clearance strategy and the receptor-end intervention strategy are two different schemes, and different parameters are configured to adapt to different schemes to form several strategies. Exemplarily, for the targeted clearance strategy, in the drug combination, a first drug that inhibits exosome secretion and a second drug for targeted treatment of liver cancer are used in combination, and a low dose of the first drug + a low dose of the second drug, a medium dose of the first drug + a medium dose of the second drug, and the use of the first drug and the second drug alone are set, and the treatment time is set to different treatment time points of short-term (such as 24 hours) and long-term (such as 72 hours), which can adapt to derive a variety of inhibition strategies that belong to targeted clearance.

[0031] In the disclosed embodiment, the first inhibition strategy and its strategy data (such as dosage, medication time, drug concentration, etc.) that match the patient can be pre-determined from a pre-built inhibition strategy library. Independent of the experience configuration of professionals, the strategy that meets the individual situation of the user can be preliminarily screened intelligently and automatically.

[0032] S102: Extract features from the individualized data and the policy data to obtain user features and policy features.

[0033] Optionally, feature extraction can be viewed as a multimodal deep feature learning process. Considering the different sources and semantics of individualized data and policy data, independent network layers can be configured in the AI ​​network to perform feature extraction processing to avoid feature conflicts or information loss.

[0034] S103: Fusing the user feature and the strategy feature to obtain a fusion feature related to the first inhibition strategy and the change in the patient's biological state.

[0035] Optionally, a fusion layer is configured in the AI ​​network, which can fuse user features (such as tumor molecular features) and strategy features (such as drug parameters) to simulate the dynamic process of interaction between intervention measures and individual biological states in biological systems and quantify the effectiveness of the first inhibition strategy for patients.

[0036] S104. Based on the fusion signature, determine a predictive index value for indicating the effectiveness of the first inhibition strategy in inhibiting exosomes of liver cancer cells.

[0037] Optionally, the first inhibition strategy may include multiple strategies, and the outputted prediction index value may be obtained separately for each inhibition strategy. Different prediction index values ​​may be outputted for different inhibition strategies. In one example, the prediction index value includes at least one of exosome clearance rate, receptor-side signaling pathway inhibition rate, or tumor growth inhibition rate.

[0038] In the embodiments of the present disclosure, the provided method can be implemented through an AI network, such as Figure 2 As shown, the AI ​​network can include an inhibition strategy screening layer (which obtains a first inhibition strategy and its strategy data based on an inhibition strategy library), a preprocessing layer, a feature extraction layer (including a first network branch and a second network branch in parallel), a fusion layer, and an output layer (also known as a prediction layer). The following embodiments will specifically describe the functions and operations of each network layer.

[0039] In a feasible embodiment, in S101, based on the individualized data of the patient, a first inhibition strategy and its strategy data matching the patient are obtained from a pre-built inhibition strategy library, including steps A1 to A2: Step A1: Based on the clinical data, exclude the inhibition strategies that meet the preset mutual exclusion conditions from the inhibition strategy library to obtain an initial inhibition strategy.

[0040] Among them, the predicted mutually exclusive condition can be a hard exclusion rule, such as excluding inapplicable strategies based on the patient's contraindications (such as liver and kidney dysfunction). For example, if the patient has an autoimmune disease, immune enhancement strategies are excluded.

[0041] Optionally, to adapt to individual differences, the predicted mutually exclusive conditions can be dynamically adjusted. Based on the patient's individual condition, the system can first screen out mutually exclusive conditions that match the individual patient, and then filter the inhibition strategies in the inhibition strategy library based on these mutually exclusive conditions. For example, if patient A's autoimmune disease and liver cancer share a common signaling pathway, exosome inhibitors targeting this pathway can be excluded to avoid double inhibition and severe toxicity.

[0042] Step A2: Calculate similarity between the user vector corresponding to the individualized data and the strategy vector corresponding to the initial suppression strategy, and determine the suppression strategy with a similarity greater than a preset threshold as the first suppression strategy.

[0043] Optionally, in the similarity calculation, the individualized data and strategy data can be first encoded into vectors, the matching degree between the patient and the suppression strategy can be calculated using cosine similarity or Euclidean distance, and the Top-N strategy can be selected as the first suppression strategy, where N is greater than or equal to 1.

[0044] Optionally, the similarity calculation described above can be assisted by machine learning, such as using historical case data to train a model, and then using the model to perform classification and ranking tasks to screen for an initial inhibitory strategy that is relatively compatible with the patient. For example, a classification model can predict strategy applicability (e.g., outputting a binary classification result of "applicable" or "not applicable"), and then a ranking model can be used to rank multiple strategies based on their expected efficacy.

[0045] Optionally, during the acquisition of the first inhibition strategy, dynamic adjustments and expert review can also be performed. For example, the weights of each inhibition strategy in the inhibition strategy library can be dynamically adjusted based on the latest research results (such as new drug approval status) and patient feedback (such as adverse reactions). Before the Top-N strategy is determined as the first inhibition strategy, it is output to the user end, and clinicians, pharmacists, etc. determine or adjust the corresponding strategy to make it clinically reasonable.

[0046] Exemplarily, the suppression strategy library may include the strategies shown in Table 1 below: Table 1

[0047] In one example, when building a library of inhibitory strategies for liver cancer exosomes, strategies can be categorized by mechanism of action, such as inhibition of exosome release, interference with exosome uptake, and exosome degradation. Strategies can also be categorized by drug, such as small molecule inhibitors, gene therapy, and combination strategies. Strategies can also be categorized by indication stage, such as focusing on preventing exosome-mediated metastasis for early-stage liver cancer or inhibiting exosome-driven immune escape or production for advanced liver cancer. Furthermore, metadata annotation can be performed to label each strategy, including target, practicality for liver cancer subtypes, level of clinical evidence (e.g., Phase X trial, case report, in vitro study), and adverse reactions.

[0048] In a feasible embodiment, considering that the acquired individualized data may contain problems such as noise, outliers, missing values, duplicate data, and the strategy data may contain contradictory data, large differences in parameter values, etc., in order to improve data quality, adapt to model requirements, optimize model performance, and ensure the accuracy and reliability of model output, the embodiment of the present disclosure also performs data preprocessing on the individualized data and strategy data.

[0049] Optionally, before extracting features from the individualized data and the policy data respectively, the process further includes S100 to S200: S100 , pre-processing the individualized data and the policy data according to different data types.

[0050] S200: Align the pre-processed individualized data and the pre-processed strategy data.

[0051] Optionally, preprocessing of individualized data may include missing value processing, feature normalization, classification feature encoding, high-dimensional data dimensionality reduction, etc.

[0052] For example, for individualized data processing, missing values ​​in clinical data (e.g., liver function markers) can be filled using multiple imputation or the mean of similar patients. For molecular data (e.g., gene expression), missing values ​​can be filled using the mean of neighboring samples using KNN. For feature normalization, continuous features (e.g., exosome concentration, tumor volume, etc.) can be scaled to the [0, 1] range. For categorical feature encoding, discrete features (e.g., TNM stage) can be processed using one-hot encoding or target encoding. For high-dimensional data dimensionality reduction, PCA can be used for genomic / transcriptomic data (e.g., miRNA expression profiles).

[0053] Optionally, preprocessing of strategy data may include parameter discretization and mechanism labeling. For example, continuous strategy data (e.g., drug concentration) can be divided into intervals (e.g., 0-10 μM, 10-50 μM) and converted into categorical features to reduce overfitting. Each strategy can be labeled with its mechanism of action (e.g., "receptor blocking") to facilitate model learning of the relationship between mechanism and effect.

[0054] Optionally, after preprocessing the individualized data and the strategy data, a multimodal alignment operation can be performed. For example, the temporal and spatial resolutions of the strategy data and the individualized data can be unified (e.g., mapping in vitro experimental data to patient tumor microenvironment simulation parameters).

[0055] In a feasible embodiment, in S102, feature extraction is performed on the individualized data and the policy data respectively to obtain user features and policy features, including steps B1 to B2: Step B1: extracting features from the individualized data through the first network branch to obtain user features including multi-dimensional heterogeneous data.

[0056] Step B2: extract the policy features through the second network branch to obtain policy features that integrate parameter information and policy information.

[0057] Optional, such as Figure 2 As shown in the figure, the AI ​​network is equipped with a first network branch for processing individualized data and a second network branch for processing policy data. The network structure shows that the two branches are parallel. The first network branch is designed to integrate multi-dimensional heterogeneous data, extract complementary features of the patient's pathophysiological state from the three dimensions of molecular data, clinical data, and microenvironmental data, and construct a comprehensive representation. The second network branch is designed to perform multimodal policy semantic understanding, extract the properties of the inhibition strategy from the two levels of parameters and mechanisms, and realize the computable representation of the inhibition strategy.

[0058] Optional, such as Figure 3As shown, in the first network branch, there are at least three sub-networks for processing different types of data respectively. Specifically, the extraction of user features in step B1 includes steps B11 to B14: Step B11: Perform feature extraction on the molecular data in the spatial and temporal dimensions to obtain a first feature vector.

[0059] For example, in the molecular data subnetwork, a network structure combining one-dimensional convolution (e.g., kernel size = 3, number of channels = 64), maximum pooling, and LSTM can be used to capture both spatially localized patterns and temporal dynamics of gene expression. The convolutional layer captures spatially localized patterns of gene co-expression, the maximum pooling layer reduces feature dimensionality, and the LSTM layer constructs the temporal dynamics of gene expression, ultimately outputting a molecular dynamic feature vector that reflects tumor biological behavior.

[0060] Step B12: Perform spatial mapping processing of discrete features on the clinical data to obtain a second feature vector.

[0061] For example, in the clinical data subnetwork, a network structure with fully connected layers and activation functions (such as ReLU) can be used to learn the combined features of discrete features such as staging and grade. The fully connected layer (FC) can map discrete features to a multidimensional continuous space, and then introduce nonlinearity through ReLU activation to output a clinical risk feature vector to quantify patient-related performance.

[0062] Step B13: Perform graph convolution aggregation processing on the microenvironment data to obtain a third eigenvector.

[0063] For example, in the microenvironment data subnetwork, a graph convolutional network can be used, such as constructing a graph of immune cell-exosome interactions to extract topological features of intercellular communication in the microenvironment. The input of the graph convolutional network can be a graph of immune cell-exosome interactions (e.g., with cell / exosome types as nodes and edge weights as colocalization probabilities or signaling pathway strengths). A two-layer graph convolution aggregates node neighborhood information to capture the topological features of intercellular communication. The features of all nodes in the graph are averaged or weighted summed to generate global microenvironment features. The output is a microenvironment interaction feature vector that reflects the state of the tumor immune microenvironment.

[0064] Step B14: Fusing the first feature vector, the second feature vector, and the third feature vector to obtain user features.

[0065] Optionally, the molecular dynamic feature vector, clinical risk feature vector, and microenvironment interaction feature vector can be concatenated or weighted summed to form a comprehensive feature representation of the patient, resulting in a user feature. Feature fusion can integrate static pathological data with dynamic biological processes, improving feature discriminability and thus the accuracy of prediction results.

[0066] Optional, such as Figure 4 As shown, the second network branch includes at least a parameter subnetwork and a strategy subnetwork. Specifically, the extraction of strategy features in step B2 includes steps B21 to B23: Step B21: perform spatial attention processing on the strategy data to obtain a parameter feature vector.

[0067] Optionally, in the parameter sub-network, a network structure with a fully connected layer, an activation function (such as ReLU), and an attention mechanism can be used to improve attention to key parameters.

[0068] Optionally, the input of the parameter sub-network can be continuous parameters in the inhibition strategy, such as numerical variables such as drug concentration, dosage, and dosing frequency. The parameter sub-network can map the parameters to a multidimensional space, introduce nonlinearity through ReLU function activation, and calculate the weights of each parameter (such as generating normalized weights through Softmax), focusing on parameters that have a greater impact on the effect, and outputting a weighted parameter feature vector.

[0069] Step B22: Perform semantic encoding on the policy data to obtain a semantic feature vector.

[0070] Optionally, in the policy subnetwork, a network structure of text embedding, average pooling, and fully connected layers can be used to describe the policy mechanism text and convert natural language into a computable vector.

[0071] Optionally, the input of the policy sub-network can be the suppression policy description text. The policy sub-network can encode the text into a multi-dimensional word vector sequence, then take the average of the word vector sequence to generate a sentence-level representation, and then compress the multi-dimensional vector through a fully connected layer to reduce the computational complexity, and finally output the mechanism semantic feature vector to quantify the policy's action path and biological goals.

[0072] Step B23: Fuse the parameter feature vector and the semantic feature vector to obtain a strategy feature.

[0073] Optionally, the parameter feature vector and the semantic feature vector can be concatenated to form a comprehensive feature representation of the strategy, which can combine quantitative parameters with qualitative mechanism information to achieve accurate expression of the strategy.

[0074] In a feasible embodiment, the fusion layer provided by the AI ​​network can fuse user features and policy features to generate fusion features related to changes in biological state. The fusion process can eliminate the modal differences between user features and policy features, capture the dynamic relationship between patient status and policy effects, and ensure that the fusion features reflect the actual biological state change process.

[0075] Optional S103 includes fusing the user feature and the strategy feature to obtain a fusion feature related to the first inhibition strategy and the change in the patient's biological state, including at least one of the following steps C1 and C2: Step C1: performing feature interaction based on the user features and the strategy features to obtain an interaction matrix; performing attention processing based on the interaction matrix to obtain a weight matrix; and performing weighted fusion based on the weight matrix to obtain a fusion feature.

[0076] Alternatively, an interaction matrix (also known as a similarity matrix) can be generated by first performing a dot product of user features and policy features. This interaction matrix can then be processed using a Softmax function to generate a weight matrix representing the contribution of each policy dimension to the patient features. This weighted summation can then be performed to calculate the policy-aware patient features, yielding the final fused features. The processing of the disclosed embodiments can effectively identify the impact of key policy parameters or mechanisms on patient-specific biomarkers.

[0077] Step C2: Based on the graph corresponding to the user features, an interaction graph is constructed with the strategy features as node attributes of the graph; a graph convolution operation is performed on the interaction graph, and the updated interaction graph is aggregated to obtain a fusion feature.

[0078] Optionally, when the patient's microenvironment is represented by a graph structure, the policy features can be injected into the patient's microenvironment graph as node attributes, and then processed using the GCN propagation strategy. The updated node features are averaged or attention pooled to generate fused features. For example, the processing of the embodiment of the present disclosure can simulate how the policy changes the intercellular communication pattern.

[0079] In the disclosed embodiment, the fusion layer can generate a feature representation that is both patient-specific, strategically interventional, and biologically plausible, providing support for dynamic decision-making in medical treatment.

[0080] In a feasible embodiment, determining a predictive index value indicating the effectiveness of the first inhibition strategy in inhibiting liver cancer cell exosomes based on the fusion feature in S104 includes steps D1 to D2: Step D1: The fusion features are transformed through a fully connected layer.

[0081] Step D2: Process the output vector of the fully connected layer through the activation function to obtain the prediction index value.

[0082] Optionally, a fully connected layer and an activation function (such as Sigmoid) can be used in the output layer of the AI ​​network. For example, when outputting the inhibition rate prediction result, the output continuous value ranges from [0, 1] and can be mapped to a percentage, such as 0.75 being converted to a 75% inhibition rate expression.

[0083] Optionally, in obtaining the predicted index value, the fused feature vector can be input into a fully connected layer, and the fused features can be mapped to a new feature space through a linear combination of the weight matrix and the bias vector as the output of the fully connected layer. The output vector of the fully connected layer is then processed through an activation function, such as a Sigmoid function or a Softmax function. For example, if the predicted index value is a continuous value, the Sigmoid function can be used to map the output to between 0 and 1. If the predicted index value is multiple categories (such as indicating the effectiveness of different levels or parameters), the Softmax function can be used to convert the output into a probability distribution.

[0084] Optionally, the disclosed embodiment further provides a visualization mechanism for generating a patient-strategy interaction heat map to illustrate the matching status of the first inhibition strategy with the patient.

[0085] In a prediction example, the predictions for patient A are as follows: Exosome data: low expression of miR-21 (FPKM=30), wild type of RAB27A.

[0086] Receptor cell data: M1 macrophages account for 60% (high anti-tumor activity).

[0087] AI network prediction results: GW4869 inhibition rate = 45%, because the basal level of exosome release is low and the anti-tumor ability of recipient cells is strong.

[0088] In a feasible embodiment, the method provided by the embodiment of the present disclosure further includes S105: generating a second inhibition strategy matching the patient based on the prediction index value.

[0089] In one example, after obtaining the prediction index values ​​corresponding to each first suppression strategy, the weight of each prediction index can be determined according to the type of each first suppression strategy, and then the various parameters in the first suppression strategy can be adjusted according to the weight value to obtain the second suppression strategy.

[0090] For example, when the first inhibition strategy is obtained by setting corresponding parameters under the targeted clearance strategy, based on prior knowledge, the weight of the tumor growth inhibition rate can be set to 0.6, the weight of the exosome clearance rate can be set to 0.3, and the weight of the receptor-side signaling pathway inhibition rate can be set to 0.1. This is only an example and can be adjusted according to actual conditions. The present disclosure is not limited to this.

[0091] In another example, the prediction indicator can be used as the optimization target, and the parameter settings that achieve the optimal level of each indicator can be found by searching in the strategy space with the constructed target optimization model (which can use genetic algorithm, particle swarm optimization algorithm, etc.).

[0092] In another example, based on the prediction results, a strategy that meets clinical needs and has better indicator performance can be selected as the second inhibition strategy.

[0093] In a feasible embodiment, the method provided by the embodiment of the present disclosure is executed through an artificial intelligence (AI) network. Considering the problem of data scarcity, a generative adversarial network (GAN) can be used to generate patient data, or model iteration can be performed through transfer learning to achieve end-to-end prediction from patient data to strategy effectiveness, providing accurate and dynamic decision support for liver cancer exosome inhibition.

[0094] Optionally, the update of the AI ​​network includes S001 to S003: S001. Acquire a current first liver cancer state and a adopted target suppression strategy, where the target suppression strategy is determined based on the second suppression strategy.

[0095] S002. When the change in the patient's liver cancer status reaches a preset condition, obtain a corresponding second liver cancer status.

[0096] S003. Determine a reward value based on the second liver cancer state, the first liver cancer state, and the target suppression strategy, and update network parameters of the AI ​​network based on the reward value.

[0097] Optionally, the target inhibition strategy can be an inhibition strategy formulated by medical staff based on the actual situation of the patient with reference to the second inhibition strategy, and can be used as a part of the patient's liver cancer treatment.

[0098] Optionally, liver cancer conditions vary widely and may present differently in different individuals. Therefore, preset conditions can be dynamically set based on the patient's actual condition. In one example, preset conditions can be set based on pathological characteristics, such as tumor size and number, tumor differentiation, or vascular invasion; or based on stage, such as the transition from mid-stage to late-stage; or based on the patient's overall condition.

[0099] Optionally, AI networks can be updated using transfer learning, which extracts common knowledge (such as features, parameters, model structure, etc.) from the source task and migrates it to the target task to reduce the demand for data and computing resources.

[0100] In one example, the reward value may be a weighted combination of multiple sub-reward items, each of which corresponds to a medical or model optimization objective.

[0101] For example, a sub-reward value 1 related to the treatment effect is set to quantify the degree of improvement of the liver cancer status by the inhibition strategy; a sub-reward value 2 related to the punishment is also set to avoid serious side effects (such as liver toxicity, etc.) caused by the inhibition strategy; and a sub-reward value 3 related to the rationality of the state transition can also be set to ensure that the next state predicted by the AI ​​complies with medical laws (such as avoiding unreasonable rapid shrinkage or growth of tumors).

[0102] Among them, sub-reward value 1 includes the following two items: Tumor volume change: If the tumor volume V2 of the second liver cancer state is smaller than that of the first state V1, the reward value R11 = α × (V1 - V2) / V1 (α is the weight coefficient).

[0103] Decreased tumor markers: If the levels of liver cancer-related markers (such as AFP) decrease, the reward value R12 = β × (AFP1-AFP2) / AFP1 (β is the weight coefficient).

[0104] Among them, the sub-reward value 2 can include dividing the side effects into levels 1-5 according to the common adverse reaction event evaluation criteria, and assigning a penalty R21=- × Level ( is the weight coefficient); if the side effects lead to treatment interruption, an additional penalty R22=-1 can be assigned.

[0105] Regarding sub-reward value 3, it should be noted that the AI ​​network can predict a third liver cancer state under the first liver cancer state and the target suppression strategy. In sub-reward value 3, the third liver cancer state predicted by the AI ​​network (as predicted) can be compared with the second liver cancer state (as used in the example). If the deviation is less than a threshold value θ, the reward value R31 = λ × (1-difference / θ) (λ is the severity coefficient). If the state transition violates known biological laws (such as the tumor spontaneously disappearing without treatment), the reward value R32 is set to -1.

[0106] The weighted sum of the above sub-reward values ​​can be used to obtain the total reward value, which can be used to update the network parameters of the AI ​​network based on the reward value. ,λ) can be adjusted according to the priority of clinical experience.

[0107] In the disclosed embodiment, considering that the patient's liver cancer treatment is long-term, in order to make the output of the AI ​​network more in line with the patient's individual situation, the AI ​​network can be updated during the patient's treatment process to improve the effectiveness of controlling the exosomes of liver cancer cells and provide medical staff with more valuable auxiliary information.

[0108] In an embodiment of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored on the memory, wherein the processor executes the computer program to implement the steps of the method provided in any optional embodiment of the present disclosure. Compared with the prior art, the present disclosure provides a method for analyzing the effectiveness of inhibiting exosomes of liver cancer cells. Specifically, based on the patient's individualized data, a first inhibition strategy and its strategy data matching the patient can be obtained from a pre-built inhibition strategy library. The individualized data can include molecular data, clinical data, and microenvironmental data related to liver cancer cell exosomes. The inhibition strategy library can include multiple inhibition strategies, each of which is formed using different parameters under a targeted clearance strategy and / or a receptor-end intervention strategy. In other words, the present disclosure can preliminarily screen out an inhibition strategy that meets the individual needs of the patient from the inhibition strategy library; then, feature extraction can be performed on the individualized data and strategy data respectively to obtain user features and strategy features, and the user features and strategy features can be fused to obtain a fusion feature related to the first inhibition strategy and the patient's biological state change; finally, based on the fusion feature, a predictive index value for indicating the effectiveness of the first inhibition strategy can be determined. The disclosed embodiments can, after screening for an initial inhibition strategy, predict the predictive index value of the inhibition strategy's effectiveness in inhibiting liver cancer cell exosomes through an artificial intelligence network. The predictive index value can directly and accurately reflect the effectiveness of the inhibition strategy for individual patients, thereby assisting medical staff to better adapt to individual patient conditions, determine more accurate treatment plans, and improve the effectiveness of controlling liver cancer cell exosomes.

[0109] In an alternative embodiment, an electronic device is provided, such as Figure 5 As shown, Figure 5 The electronic device 4000 shown includes: a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which may be used for data exchange between the electronic device and other electronic devices, such as data transmission and / or data reception. It should be noted that in actual applications, the number of transceivers 4004 is not limited to one, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present disclosure.

[0110] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the present disclosure. Processor 4001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, or a combination of a DSP and a microprocessor.

[0111] Bus 4002 may include a path for transmitting information between the above components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. Bus 4002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0112] The memory 4003 may be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium that can be used to carry or store computer programs and can be read by a computer, without limitation herein.

[0113] The memory 4003 is used to store the computer program for executing the embodiments of the present disclosure, and the execution is controlled by the processor 4001. The processor 4001 is used to execute the computer program stored in the memory 4003 to implement the steps shown in the above method embodiments.

[0114] Among them, electronic equipment includes but is not limited to: terminal equipment and servers.

[0115] An embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.

[0116] The embodiments of the present disclosure further provide a computer program product, including a computer program, which can implement the steps and corresponding contents of the aforementioned method embodiments when executed by a processor.

[0117] It should be understood that, although the flowcharts of the embodiments of the present disclosure indicate the various operation steps by arrows, the order of implementation of these steps is not limited to the order indicated by the arrows. Unless otherwise clearly stated herein, in some implementation scenarios of the embodiments of the present disclosure, the implementation steps in each flowchart can be performed in other orders as required. In addition, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage in these sub-steps or stages can also be executed at different times. In scenarios where the execution times are different, the order of execution of these sub-steps or stages can be flexibly configured as required, and the embodiments of the present disclosure do not limit this.

[0118] The above description is only an optional implementation method for some implementation scenarios of the present disclosure. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the solution of the present disclosure, other similar implementation methods based on the technical ideas of the present disclosure also fall within the protection scope of the embodiments of the present disclosure.

Claims

1. A method for analyzing the effectiveness of inhibiting exosomes in liver cancer cells, characterized in that: include: Based on the patient's personalized data, a first inhibition strategy and its strategy data matching the patient are obtained from a pre-built inhibition strategy library; the personalized data includes molecular data, clinical data, and microenvironmental data related to liver cancer cell exosomes; the inhibition strategy library includes multiple inhibition strategies, each of which is formed using different parameters under a targeted clearance strategy and / or a receptor-end intervention strategy; Performing feature extraction on the individualized data and the policy data respectively to obtain user features and policy features; fusing the user feature and the strategy feature to obtain a fusion feature related to the first inhibition strategy and the change in the patient's biological state; Based on the fusion signature, a predictive index value is determined to indicate the effectiveness of the first inhibition strategy in inhibiting exosomes from liver cancer cells.

2. The method according to claim 1, characterized in that The method of obtaining a first inhibition strategy and its strategy data that matches the patient from a pre-built inhibition strategy library based on the patient's individualized data includes: Based on the clinical data, excluding inhibition strategies that meet preset mutually exclusive conditions from the inhibition strategy library to obtain an initial inhibition strategy; performing similarity calculation on the user vector corresponding to the individualized data and the strategy vector corresponding to the initial suppression strategy, and determining the suppression strategy having a similarity greater than a preset threshold as the first suppression strategy; The suppression strategy library is constructed based on the acquired instance samples.

3. The method according to claim 1, characterized in that Before extracting features from the individualized data and the strategy data respectively, the method further includes: Preprocessing the individualized data and the strategic data according to different data types; Align the preprocessed individualized data with the preprocessed strategy data.

4. The method according to claim 1, wherein The extracting features of the individualized data and the policy data to obtain user features and policy features respectively includes: Performing feature extraction on the individualized data through the first network branch to obtain user features including multi-dimensional heterogeneous data; Through the second network branch, feature extraction is performed on the strategy feature to obtain the strategy feature that integrates parameter information and strategy information.

5. The method according to claim 4, characterized in that The extraction of user features includes: Performing feature extraction on the molecular data in spatial and temporal dimensions to obtain a first feature vector; Performing spatial mapping processing of discrete features on the clinical data to obtain a second feature vector; For the microenvironment data, graph convolution aggregation processing is performed to obtain the third eigenvector; The first feature vector, the second feature vector, and the third feature vector are fused to obtain user features.

6. The method according to claim 4, characterized in that The extraction of the strategy features includes: Performing spatial attention processing on the strategy data to obtain a parameter feature vector; Performing semantic encoding on the policy data to obtain a semantic feature vector; The parameter feature vector and the semantic feature vector are fused to obtain a strategy feature.

7. The method according to claim 1, characterized in that The fusing of the user feature and the strategy feature to obtain a fusion feature related to the first inhibition strategy and the change in the patient's biological state includes at least one of the following: Performing feature interaction based on the user features and the strategy features to obtain an interaction matrix; performing attention processing based on the interaction matrix to obtain a weight matrix; Perform weighted fusion based on the weight matrix to obtain fusion features; Based on the graph corresponding to the user features, an interaction graph is constructed with the strategy features as node attributes of the graph; a graph convolution operation is performed on the interaction graph, and the updated interaction graph is aggregated to obtain a fusion feature.

8. The method according to claim 1, characterized in that Determining, based on the fusion signature, a predictive index value for indicating the effectiveness of the first inhibition strategy in inhibiting liver cancer cell exosomes comprises: The fusion features are transformed through a fully connected layer; The output vector of the fully connected layer is processed through the activation function to obtain the predicted index value.

9. The method according to claim 1, characterized in that The predictive index value includes at least one of exosome clearance rate, receptor-end signaling pathway inhibition rate, or tumor growth inhibition rate; The method further comprises: Based on the predictive indicator value, a second suppression strategy matched to the patient is generated.

10. The method according to claim 9, characterized in that The method is performed by an artificial intelligence (AI) network, and updating of the AI ​​network includes: Acquiring a current first liver cancer state and a target inhibition strategy employed, wherein the target inhibition strategy is determined based on the second inhibition strategy; When the change in the patient's liver cancer status reaches a preset condition, obtaining a corresponding second liver cancer status; Based on the second liver cancer state, the first liver cancer state and the target suppression strategy, a reward value is determined, and network parameters of the AI ​​network are updated based on the reward value.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the method according to any one of claims 1 to 10.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

13. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images