Gastric cancer recurrence risk prediction method based on metabonomics characteristic analysis

Abstract

The invention discloses a gastric cancer recurrence risk prediction method based on metabonomics characteristic analysis. The method comprises the following steps: S1, constructing a metabonomics data set; s2, the collected metabonomics data set is preprocessed; s3, extracting metabolite characteristics related to gastric cancer recurrence; s4, based on the extracted metabolite features, using a metabolic pathway integration method to quantify disturbance of each metabolic pathway, and calculating the abnormal degree of metabolite concentration change in the metabolic pathways; s5, constructing a multi-dimensional feature matrix containing a metabonomics data set, metabolic pathway integrals and patient individual information; s6, inputting the multi-dimensional feature matrix into a trained machine learning model, and outputting a gastric cancer recurrence risk score of the patient; and S7, generating a prediction report according to the recurrence risk score. According to the method, by combining metabonomics characteristic analysis, the space-time coupling quantum optimization network and the nonlinear disturbance model, the problems of insufficient real-time performance and insufficient data integration in gastric cancer recurrence risk prediction in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of gastric cancer, and in particular to a method for predicting gastric cancer recurrence risk based on metabolomics feature analysis. Background Art

[0002] In the existing technology, the prediction methods of gastric cancer recurrence risk mostly rely on imaging and genomic data. Although imaging and genomic data can provide prognostic evaluation to a certain extent, they have obvious deficiencies in dynamics and real-time performance. Imaging methods usually rely on regular patient examinations and cannot reflect the dynamic changes in the patient's physiological state, resulting in a certain lag in recurrence prediction. In addition, although the analysis of genomic data can provide information on the patient's genetic level, it is often not sufficiently correlated with the actual physiological state, especially when reflecting gastric cancer recurrence, genomic data cannot track the metabolic process in the patient's body in real time.

[0003] Metabolomics, as an emerging bioinformatics technology, can dynamically monitor the changes in metabolites in individuals and reveal immediate changes in the physiological state of the human body. Therefore, it is of great research significance to apply metabolomics characteristics to the prediction of gastric cancer recurrence risk. However, the existing data analysis methods based on metabolomics are relatively simple. Most studies only monitor the concentration changes of individual metabolites, ignoring the dynamic changes of metabolites in different metabolic pathways and their comprehensive effects with clinical characteristics and treatment plans. Current prediction methods are also unable to flexibly respond to individual differences in patients, and often use fixed models for evaluation, lacking personalized processing capabilities.

[0004] Therefore, the main disadvantages of the prior art are:

[0005] Existing predictive methods based on imaging and genomic data analysis cannot dynamically reflect immediate changes in the patient's body, especially in real-time monitoring and analysis at the metabolic level. Existing technologies are obviously lagging behind and cannot provide accurate predictions of recurrence risk.

[0006] Existing methods rarely consider the multidimensional comprehensive analysis of metabolomics, clinical history and treatment plans and fail to effectively utilize the associations between multiple data sources, resulting in insufficient comprehensiveness and accuracy of the prediction models.

[0007] The models used in existing technologies cannot be adaptively optimized and are difficult to personalize based on the metabolic characteristics and treatment responses of different patients. The predictive applicability is limited and misjudgment or over-prediction problems are prone to occur.

[0008] In summary, the existing technologies in the field of gastric cancer recurrence risk prediction mainly have problems such as poor real-time performance, insufficient multidimensional data integration and lack of individualized prediction capabilities, which need to be improved urgently. Summary of the invention

[0009] One purpose of the present invention is to propose a method for predicting the risk of gastric cancer recurrence based on metabolomics feature analysis. The present invention solves the problems of insufficient real-time performance, insufficient data integration and lack of individualized processing in the existing technology for predicting the risk of gastric cancer recurrence by combining metabolomics feature analysis, spatiotemporal coupled quantum optimization network and nonlinear perturbation model.

[0010] A method for predicting gastric cancer recurrence risk based on metabolomics feature analysis according to an embodiment of the present invention comprises the following steps:

[0011] S1. Collect multiple metabolite concentration information of patients, including key metabolites in energy metabolism pathways, lipid metabolism pathways, and amino acid metabolism pathways related to gastric cancer, and construct a metabolomics dataset;

[0012] S2. Preprocess the collected metabolomics data set;

[0013] S3. Based on the preprocessed metabolomics dataset, extract metabolite characteristics associated with gastric cancer recurrence, including the concentration change patterns of key metabolites in the energy metabolism pathway, lipid metabolism pathway, and amino acid metabolism pathway, and analyze the dynamic changes of each metabolite in the metabolic pathway;

[0014] S4. Based on the extracted metabolite features, the metabolic pathway integration method is used to quantify the disturbance of each metabolic pathway and calculate the abnormal degree of metabolite concentration changes within the metabolic pathway. The metabolic pathway integration method is combined with each metabolite metabolomics data set for weight allocation and biochemical association analysis;

[0015] S5. Combine the patient's clinical history data and treatment plan data to construct a multidimensional feature matrix containing metabolomics data sets, metabolic pathway scores, and individual patient information;

[0016] S6. Inputting the multidimensional feature matrix into a trained machine learning model, the machine learning model performs a comprehensive assessment on the risk of gastric cancer recurrence, and outputting a gastric cancer recurrence risk score for the patient based on the comprehensive assessment of the machine learning model;

[0017] S7. Generate a prediction report based on the recurrence risk score, including the degree of abnormality of the metabolic pathway, the level of recurrence risk, and treatment adjustment recommendations based on the individual characteristics of the patient.

[0018] Optionally, the S1 specifically includes the following steps:

[0019] S11. Collect multiple metabolite concentration information of the patient, including key metabolites in the energy metabolism pathway, lipid metabolism pathway and amino acid metabolism pathway related to gastric cancer. Suppose the metabolite concentration data of the patient is in the matrix form X = {x ij}, where i represents the type of metabolite, j represents the samples collected at different time points, and x ij represents the concentration value of the i-th metabolite at the j-th time point;

[0020] S12. Classify the metabolite data and label the metabolites in the energy metabolism pathway as set M. E , metabolites in the lipid metabolism pathway are labeled as set M L , the metabolites in the amino acid metabolic pathway are labeled as set M A , the metabolomics dataset X is divided into:

[0021] X E ={x ij |i∈M E}, X L ={x ij |i∈M L}, X A ={x ij |i∈M A};

[0022] S13. Constructing energy metabolism pathway set X E , lipid metabolism pathway set X L and amino acid metabolic pathway set X A Metabolomics dataset D:

[0023] D={X E ,X L ,X A}.

[0024] Optionally, S3 specifically includes the following steps:

[0025] S31. Extract metabolite features related to gastric cancer recurrence based on the preprocessed metabolomics dataset D. Suppose the metabolite feature set is F = {f i}, i represents the type of metabolite, f i Represents the i-th metabolite feature:

[0026]

[0027] Among them, x ij represents the concentration value of the i-th metabolite at the j-th time point, μ i is the mean of the ith metabolite, σ i is the standard deviation, w ij is the weight corresponding to time point j, λ ij is the time decay factor;

[0028] S32, using dynamic time warping algorithm to analyze metabolite features f iAnalyze and evaluate the dynamic change trend of metabolite concentration over time, and generate the dynamic change pattern matrix T = {t ij}, t ij Indicates the change amplitude of the i-th metabolite at the j-th time point:

[0029]

[0030] Among them, f i (t k ) represents metabolite i at time point t k Concentration mode value, Δt k is the time interval, α ik is the weight factor for each time point, β ik Indicates the association weights between different time periods;

[0031] S33. Extract the characteristics of each metabolic pathway according to the dynamic changes of each metabolite in the metabolic pathway, including the energy metabolic pathway characteristics P E , lipid metabolism pathway characteristics are P L and amino acid metabolic pathway characteristics P A , each feature is calculated as follows:

[0032]

[0033] Among them, P X represents the total eigenvalue of metabolic pathway X, γ i represents the pathway weight of metabolite i, t ij represents the dynamic change amplitude of metabolite i at time point j, ω ij is the correction coefficient at the time point, T i is the overall temporal variation trend of metabolite i, η X is the pathway characteristic adjustment factor.

[0034] Optionally, the S4 specifically includes the following steps:

[0035] S41. Based on the extracted metabolic pathway features, the energy metabolic pathway features P E , lipid metabolism pathway characteristics are P L and amino acid metabolic pathway characteristics P A The metabolic pathway integration method is used for quantification, and the perturbation integral of each metabolic pathway is I X :

[0036]

[0037] Among them, w X (t) is the weight factor at time t, ξ X represents the characteristic correction coefficients of metabolic pathway X, t1 and tn is the starting point and end point of the time period;

[0038] S42. Perturbation integral according to metabolic pathway I X and metabolite characteristics f i Calculate the abnormality of each metabolic pathway A X .

[0039] Optionally, the S5 specifically includes the following steps:

[0040] S51, combined with the patient's clinical history data C = {c k}, where c k represents the patient's kth clinical characteristic data, including medical history, surgical records, and pathology report information;

[0041] S52, combined with the patient's treatment plan data T = {t l}, where t l Represents the relevant data of the first treatment plan, including the type of treatment, treatment dosage and the patient's response to the treatment;

[0042] S53, combining the metabolomics data set D, the patient's clinical history data C, and the treatment plan data T to construct a multidimensional feature matrix M:

[0043]

[0044] Among them, X i represents the concentration value of the ith metabolite in metabolic pathway X, is the weight factor of the pathway, c k is the kth clinical characteristic data, and are the mean and standard deviation of the clinical characteristics respectively; t l is the data of the first treatment plan, and are the mean and standard deviation of the treatment options respectively; α i , β k , γ kl and η i,k,l are the correction coefficients and weight factors of metabolomics, clinical characteristics and treatment regimens in the matrix, respectively.

[0045] Optionally, the S6 specifically includes the following steps:

[0046] S61. Input the constructed multidimensional feature matrix M into the spatiotemporal coupled quantum optimization network, encode the multidimensional feature matrix M in quantum state, and use quantum bits to encode the changes in each feature dimension. The quantum state of the metabolomics feature is q i (M);

[0047] S62. The multidimensional feature matrix M encoded in the quantum state is processed through the spatiotemporal coupling network to construct a spatiotemporal coupling relationship matrix. The spatiotemporal coupling network adaptively captures the changing trends of metabolomics, clinical history data and treatment response data at different time points to generate a spatiotemporal feature matrix S(M):

[0048]

[0049] Among them, α i (t) is the feature dimension M i The time weight, f(M i ) is the spatial distribution characteristics of each feature dimension in the multidimensional feature matrix;

[0050] S63. Use quantum optimization learning algorithm to globally optimize the spatiotemporal feature matrix S(M), use quantum approximate optimization algorithm to search for gastric cancer recurrence risk characteristics, and generate gastric cancer recurrence risk score R Q (S(M)):

[0051]

[0052] Among them, μ S and σ S are the mean and standard deviation of the spatiotemporal feature matrix S(M), γ 1i is the weight factor of different feature dimensions, κ i is the adjustment factor of the feature dimension;

[0053] S64. Perform nonlinear perturbation simulation on the input clinical history data and treatment plan data through the nonlinear pathological feature perturbation module to generate a nonlinear perturbation feature matrix N(M,t):

[0054]

[0055] Among them, C is the clinical history data, μ C and σ C are the mean and standard deviation of clinical historical data, μ T and σ T are the mean and standard deviation of the treatment data, δ k , η l are weight factors respectively;

[0056] S66, the quantum optimized gastric cancer recurrence risk score R Q (S(M)) is comprehensively evaluated with the nonlinear perturbation feature matrix N(M,t) to generate a gastric cancer recurrence risk prediction score R final (M):

[0057] R final (M)=RQ (S(M))+λ1·N(M,t);

[0058] Among them, λ1 is the weight coefficient combining nonlinear perturbation and quantum optimization results.

[0059] Optionally, the S7 specifically includes the following steps:

[0060] S71. Based on the generated recurrence risk score R final (M) Generate a prediction report based on the degree of abnormality of the metabolic pathway A X Perform abnormal analysis. The higher the abnormality, the greater the risk of recurrence.

[0061] S73, according to the recurrence risk score R final (M) Determine the recurrence risk level L R , set the threshold θ to divide the risks into different levels:

[0062]

[0063] Among them, θ1 and θ2 are the preset thresholds of risk levels;

[0064] S74. Generate treatment adjustment recommendations based on the individual characteristics of the patient, metabolic pathway characteristics, and clinical treatment response data. The treatment adjustment recommendations are optimized based on the abnormal degree of the patient's metabolite concentration and the dynamic trend of clinical data:

[0065] Adjust the dose and combination of the patient's current medication based on abnormal metabolic pathway scores and treatment response data;

[0066] Combined with the patient's quantum optimized risk score R final (M), it is recommended to regularly and dynamically monitor the patient's metabolic characteristics;

[0067] If the patient's metabolic pathway abnormalities are related to their immune system response, it is recommended to adjust the immunotherapy regimen based on the metabolic characteristics;

[0068] According to the patient's response characteristics in the multidimensional feature matrix, the treatment plan is adjusted in a timely manner to optimize the drug combination and dosage.

[0069] The beneficial effects of the present invention are:

[0070] (1) The present invention introduces a spatiotemporal coupled quantum optimization network to encode metabolomics data, clinical history data, and treatment response data in quantum states, and adaptively captures the changing trends of these data at different time points through the spatiotemporal coupled network. Compared with the traditional fixed model, the present invention can dynamically adjust the weights of each dimension, and use the quantum optimization algorithm to perform global optimization to quickly search for the best recurrence risk characteristics, thereby improving the accuracy of the prediction. The comprehensive prediction accuracy of gastric cancer recurrence risk is significantly improved through the spatiotemporal coupling mechanism, and it can effectively integrate multidimensional data, solving the problem of insufficient integration of different types of data in the existing technology.

[0071] (2) The present invention, by combining a nonlinear pathological feature perturbation analysis module, can capture the abnormalities of the patient's metabolic pathways in real time, and generate personalized treatment adjustment recommendations based on the individual's metabolite concentration changes and treatment response data. Unlike traditional static models, the present invention simulates the dynamic changes of various pathological data through nonlinear perturbations to accurately analyze the patient's adaptability to treatment and potential risks, provide clinicians with personalized treatment recommendations, and flexibly respond to individual differences among different patients, significantly improving the applicability of recurrence risk prediction and addressing the deficiencies of existing technologies in individualized processing capabilities.

[0072] (3) The present invention utilizes quantum optimization learning and spatiotemporal feature matrices combined with the dynamic monitoring capabilities of metabolomics to track the concentration changes of metabolites in real time and dynamically adjust them in combination with the patient's clinical data, thereby improving the real-time nature of recurrence risk prediction. In particular, risk trends can be discovered in a timely manner in the detection of metabolic pathway disturbances, thereby achieving early warning and reducing the risks caused by prediction delays. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0074] Figure 1 A flow chart of a method for predicting gastric cancer recurrence risk based on metabolomics feature analysis proposed by the present invention;

[0075] Figure 2 This is a schematic diagram of feature processing in metabolomics, clinical history data and treatment plan data of a spatiotemporal coupled quantum optimization network for a gastric cancer recurrence risk prediction method based on metabolomics feature analysis proposed in the present invention. DETAILED DESCRIPTION

[0076] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0077] refer to Figure 1-2 , a method for predicting gastric cancer recurrence risk based on metabolomics feature analysis, comprising the following steps:

[0078] S1. Collect multiple metabolite concentration information of patients, including key metabolites in energy metabolism pathways, lipid metabolism pathways, and amino acid metabolism pathways related to gastric cancer, and construct a metabolomics dataset;

[0079] S2. Preprocess the collected metabolomics data set;

[0080] S3. Based on the preprocessed metabolomics dataset, extract metabolite characteristics associated with gastric cancer recurrence, including the concentration change patterns of key metabolites in the energy metabolism pathway, lipid metabolism pathway, and amino acid metabolism pathway, and analyze the dynamic changes of each metabolite in the metabolic pathway;

[0081] S4. Based on the extracted metabolite features, the metabolic pathway integration method is used to quantify the disturbance of each metabolic pathway and calculate the abnormal degree of metabolite concentration changes within the metabolic pathway. The metabolic pathway integration method is combined with each metabolite metabolomics data set for weight allocation and biochemical association analysis;

[0082] S5. Combine the patient's clinical history data and treatment plan data to construct a multidimensional feature matrix containing metabolomics data sets, metabolic pathway scores, and individual patient information;

[0083] S6. Inputting the multidimensional feature matrix into the trained machine learning model, the machine learning model performs a comprehensive assessment on the risk of gastric cancer recurrence, and outputting a gastric cancer recurrence risk score for the patient based on the comprehensive assessment of the machine learning model;

[0084] S7. Generate a prediction report based on the recurrence risk score, including the degree of abnormality of the metabolic pathway, the level of recurrence risk, and treatment adjustment recommendations based on the individual characteristics of the patient.

[0085] In this implementation, S1 specifically includes the following steps:

[0086] S11. Collect multiple metabolite concentration information of the patient, including key metabolites in the energy metabolism pathway, lipid metabolism pathway and amino acid metabolism pathway related to gastric cancer. Suppose the metabolite concentration data of the patient is in the matrix form X = {x ij}, where i represents the type of metabolite, j represents the samples collected at different time points, and x ij represents the concentration value of the i-th metabolite at the j-th time point;

[0087] S12. Classify the metabolite data and label the metabolites in the energy metabolism pathway as set M. E , metabolites in the lipid metabolism pathway are labeled as set ML , the metabolites in the amino acid metabolic pathway are labeled as set M A , the metabolomics dataset X is divided into:

[0088] X E ={x ij |i∈M E}, X L ={x ij |i∈M L}, X A ={x ij |i∈M A};

[0089] S13. Constructing energy metabolism pathway set X E , lipid metabolism pathway set X L and amino acid metabolic pathway set X A Metabolomics dataset D:

[0090] D={X E ,X L ,X A}.

[0091] In this implementation, S3 specifically includes the following steps:

[0092] S31. Extract metabolite features related to gastric cancer recurrence based on the preprocessed metabolomics dataset D. Suppose the metabolite feature set is F = {f i}, i represents the type of metabolite, f i Represents the i-th metabolite feature:

[0093]

[0094] Among them, x ij represents the concentration value of the i-th metabolite at the j-th time point, μ i is the mean of the ith metabolite, σ i is the standard deviation, w ij is the weight corresponding to time point j, λ ij is the time decay factor;

[0095] S32, using dynamic time warping algorithm to analyze metabolite features f i Analyze and evaluate the dynamic change trend of metabolite concentration over time, and generate the dynamic change pattern matrix T = {t ij}, t ij Indicates the change amplitude of the i-th metabolite at the j-th time point:

[0096]

[0097] Among them, fi (t k ) represents metabolite i at time point t k Concentration mode value, Δt k is the time interval, α ik is the weight factor for each time point, β ik Indicates the association weights between different time periods;

[0098] S33. Extract the characteristics of each metabolic pathway according to the dynamic changes of each metabolite in the metabolic pathway, including the energy metabolic pathway characteristics P E , lipid metabolism pathway characteristics are P L and amino acid metabolic pathway characteristics P A , each feature is calculated as follows:

[0099]

[0100] Among them, P X represents the total eigenvalue of metabolic pathway X, γ i represents the pathway weight of metabolite i, t ij represents the dynamic change amplitude of metabolite i at time point j, ω ij is the correction coefficient at the time point, T i is the overall temporal variation trend of metabolite i, η X is the pathway characteristic adjustment factor.

[0101] In this implementation, S4 specifically includes the following steps:

[0102] S41. Based on the extracted metabolic pathway features, the energy metabolic pathway features P E , lipid metabolism pathway characteristics are P L and amino acid metabolic pathway characteristics P A The metabolic pathway integration method is used for quantification, and the perturbation integral of each metabolic pathway is I X :

[0103]

[0104] Among them, w X (t) is the weight factor at time t, ξ X represents the characteristic correction coefficients of metabolic pathway X, t1 and t n is the starting point and end point of the time period;

[0105] S42. Perturbation integral according to metabolic pathway I X and metabolite characteristics f i Calculate the abnormality of each metabolic pathway A X .

[0106] In this implementation, S5 specifically includes the following steps:

[0107] S51, combined with the patient's clinical history data C = {c k}, where c k represents the patient's kth clinical characteristic data, including medical history, surgical records, and pathology report information;

[0108] S52, combined with the patient's treatment plan data T = {t l}, where t l Represents the relevant data of the first treatment plan, including the type of treatment, treatment dosage and the patient's response to the treatment;

[0109] S53, combining the metabolomics data set D, the patient's clinical history data C, and the treatment plan data T to construct a multidimensional feature matrix M:

[0110]

[0111] Among them, X i represents the concentration value of the ith metabolite in metabolic pathway X, is the weight factor of the pathway, c k is the kth clinical characteristic data, and are the mean and standard deviation of the clinical characteristics respectively; t l is the data of the first treatment plan, and are the mean and standard deviation of the treatment options respectively; α i , β k , γ kl and η i,k,l are the correction coefficients and weight factors of metabolomics, clinical characteristics and treatment regimens in the matrix, respectively.

[0112] In this implementation, S6 specifically includes the following steps:

[0113] S61. Input the constructed multidimensional feature matrix M into the spatiotemporal coupled quantum optimization network, encode the multidimensional feature matrix M in quantum state, and use quantum bits to encode the changes in each feature dimension. The quantum state of the metabolomics feature is q i (M);

[0114] S62. The multidimensional feature matrix M encoded in the quantum state is processed through the spatiotemporal coupling network to construct a spatiotemporal coupling relationship matrix. The spatiotemporal coupling network adaptively captures the changing trends of metabolomics, clinical history data and treatment response data at different time points to generate a spatiotemporal feature matrix S(M):

[0115]

[0116] Among them, α i (t) is the feature dimension M i The time weight, f(M i ) is the spatial distribution characteristics of each feature dimension in the multidimensional feature matrix;

[0117] S63. Use quantum optimization learning algorithm to globally optimize the spatiotemporal feature matrix S(M), use quantum approximate optimization algorithm to search for gastric cancer recurrence risk characteristics, and generate gastric cancer recurrence risk score R Q (S(M)):

[0118]

[0119] Among them, μ S and σ S are the mean and standard deviation of the spatiotemporal feature matrix S(M), γ 1i is the weight factor of different feature dimensions, κ i is the adjustment factor of the feature dimension;

[0120] S64. Perform nonlinear perturbation simulation on the input clinical history data and treatment plan data through the nonlinear pathological feature perturbation module to generate a nonlinear perturbation feature matrix N(M,t):

[0121]

[0122] Among them, C is the clinical history data, μ C and σ C are the mean and standard deviation of clinical historical data, μ T and σ T are the mean and standard deviation of the treatment data, δ k , η l are weight factors respectively;

[0123] S66, the quantum optimized gastric cancer recurrence risk score R Q (S(M)) is comprehensively evaluated with the nonlinear perturbation feature matrix N(M,t) to generate a gastric cancer recurrence risk prediction score R final (M):

[0124] R final (M)=R Q (S(M))+λ1·N(M,t);

[0125] Among them, λ1 is the weight coefficient combining nonlinear perturbation and quantum optimization results.

[0126] In this implementation, S7 specifically includes the following steps:

[0127] S71. Based on the generated recurrence risk score R final (M) Generate a prediction report based on the degree of abnormality of the metabolic pathway A X Perform abnormal analysis. The higher the abnormality, the greater the risk of recurrence.

[0128] S73, according to the recurrence risk score R final (M) Determine the recurrence risk level L R , set the threshold θ to divide the risks into different levels:

[0129]

[0130] Among them, θ1 and θ2 are the preset thresholds of risk levels;

[0131] S74. Generate treatment adjustment recommendations based on the individual characteristics of the patient, metabolic pathway characteristics, and clinical treatment response data. The treatment adjustment recommendations are optimized based on the abnormal degree of the patient's metabolite concentration and the dynamic trend of clinical data:

[0132] Adjust the dose and combination of the patient's current medication based on abnormal metabolic pathway scores and treatment response data;

[0133] Combined with the patient's quantum optimized risk score R final (M), it is recommended to regularly and dynamically monitor the patient's metabolic characteristics;

[0134] If the patient's metabolic pathway abnormalities are related to their immune system response, it is recommended to adjust the immunotherapy regimen based on the metabolic characteristics;

[0135] According to the patient's response characteristics in the multidimensional feature matrix, the treatment plan is adjusted in a timely manner to optimize the drug combination and dosage.

[0136] Embodiment 1:

[0137] In order to verify the feasibility of the present invention in predicting the risk of gastric cancer recurrence, the experimenters used a real patient data set and conducted a comparative experiment. The data set came from the gastric cancer patient follow-up database of the oncology department of a large tertiary hospital and the public gastric cancer metabolomics data set. The public gastric cancer metabolomics data set contains metabolite concentration information collected from 2018 to 2021, the patient's clinical history and treatment plan data. In the experimental design, the experimenters combined the hospital's patient data set and the public data set to form a multidimensional feature data set. The data set contains follow-up information of 400 patients in total, and the metabolite concentrations include concentration data of more than 50 key metabolites related to energy metabolism pathways, lipid metabolism pathways and amino acid metabolism pathways. The experimenters divided the data set into training and test sets in a ratio of 7:3, of which 280 patient data were used for training and 120 patient data were used for testing.

[0138] In terms of data processing, the experimenters first performed standardized preprocessing for the multidimensional features of metabolomics data, including data normalization, denoising, and feature scaling to ensure the comparability of data dimensions between different patients. For the feature extraction of metabolite concentrations, the spatiotemporal coupled quantum optimization network of the present invention was used to encode the time series of each metabolite. At the same time, combined with clinical history and treatment plan data, the treatment response of individual patients was simulated through a nonlinear pathological perturbation module. During the model training process, the experimenters used the Adam optimizer, the initial learning rate was set to 0.001, the batch size was set to 32, and the model was trained for a total of 300 epochs. In order to reduce the risk of overfitting of the model, the experimenters introduced the L2 regularization term during the training process and used cross-validation technology.

[0139] In the experiment, the experimenters used several common evaluation indicators, including accuracy, root mean square error and weighted F1 score to evaluate the performance of the method of the present invention. In order to conduct comparative experiments, the experimenters also used traditional recurrence risk prediction models based on imaging and genomic data as a control group. Table 1 shows the comparison results of prediction accuracy and related evaluation indicators on the test set:

[0140] Table 1 Comparison of prediction accuracy and evaluation indicators

[0141]

[0142] As can be seen from Table 1, the method of the present invention is significantly superior to traditional imaging methods and genomic data methods in terms of prediction accuracy, root mean square error and weighted F1 score. In particular, the weighted F1 score of the method of the present invention reaches 0.85, which is much higher than 0.68 and 0.72 of the traditional method, which proves the high efficiency of the present invention in processing the gastric cancer recurrence risk prediction task.

[0143] In addition to the comparison of model performance, the experimenters also analyzed the early warning time of patient relapse. The experimenters counted the differences in early warning time before relapse. Table 2 shows the comparison of the three methods in early warning:

[0144] Table 2 Comparison of early warning time

[0145] method Average advance warning time (months) Traditional imaging methods 1.2 Methods based on genomic data 2.4 Method of the present invention 4.7

[0146] As can be seen from Table 2, the average advance warning time of the method of the present invention is 4.7 months, which is far longer than the 1.2 months of the traditional imaging method and the 2.4 months of the genomic method. This means that the present invention can detect the risk of recurrence earlier, help clinicians intervene in the early stage of recurrence, and improve the prognosis of patients.

[0147] In order to further verify the practical application effect of the present invention, the experimenters simulated specific patient cases during the training process. The following is a typical patient case:

[0148] Within 6 months after surgery, a 62-year-old male patient's metabolite concentration monitoring showed abnormal fluctuations in his lipid metabolism pathway. The spatiotemporal coupled quantum optimization network of the present invention detected that the patient's recurrence risk score was significantly elevated and indicated a high risk. The system recommended personalized treatment adjustments, and the doctor adjusted his chemotherapy drug combination according to the recommendations. In the subsequent 3-month follow-up, the patient's metabolite concentration tended to stabilize and the risk of recurrence was effectively controlled, while traditional imaging examinations failed to detect signs of recurrence in the patient at an early stage.

[0149] In addition, the experimenters compared the treatment response adjustment after the patient relapsed. The present invention generated personalized treatment plan recommendations by nonlinear perturbation simulation of clinical data and metabolomics characteristics. Among the patients who were adjusted using the method of the present invention, 70% of the patients had a significantly delayed relapse time after the treatment plan adjustment, and the survival period was extended by more than 6 months. Table 3 summarizes the effects of different treatment adjustment plans on the patient's relapse time:

[0150] Table 3 Comparison of treatment adjustment plans

[0151]

[0152] It can be seen from the above experimental results that the present invention not only has obvious advantages in prediction accuracy and early warning capabilities, but also has higher effectiveness in generating treatment plans. When dealing with complex gastric cancer recurrence risks, the space-time coupled quantum optimization network can capture recurrence signals earlier and generate effective treatment adjustment recommendations, greatly improving patients' survival expectations.

[0153] In summary, the present invention has demonstrated its feasibility and effectiveness in predicting the risk of gastric cancer recurrence by simulating real patient data. It is significantly superior to traditional imaging and genomic methods and has strong clinical application prospects.

[0154] The present invention introduces a spatiotemporal coupled quantum optimization network to encode metabolomics data, clinical history data and treatment response data in quantum states, and adaptively captures the changing trends of these data at different time points through the spatiotemporal coupled network. Compared with the traditional fixed model, the present invention can dynamically adjust the weights of each dimension, and perform global optimization through the quantum optimization algorithm to quickly search for the optimal recurrence risk characteristics, thereby improving the accuracy of the prediction. The comprehensive prediction accuracy of gastric cancer recurrence risk is significantly improved through the spatiotemporal coupling mechanism, and can effectively integrate multidimensional data, solving the problem of insufficient integration of different types of data in the existing technology.

[0155] By combining a nonlinear pathological feature perturbation analysis module, the present invention can capture the patient's metabolic pathway abnormalities in real time, and generate personalized treatment adjustment suggestions based on the individual's metabolite concentration changes and treatment response data. Unlike traditional static models, the present invention simulates the dynamic changes of various pathological data through nonlinear perturbations to accurately analyze the patient's adaptability to treatment and potential risks, provide clinicians with personalized treatment suggestions, and flexibly respond to individual differences among different patients, significantly improving the applicability of recurrence risk prediction and solving the shortcomings of existing technologies in individualized processing capabilities.

[0156] The present invention utilizes quantum optimization learning and spatiotemporal feature matrices combined with the dynamic monitoring capabilities of metabolomics to track changes in metabolite concentrations in real time and dynamically adjust in combination with the patient's clinical data, thereby improving the real-time nature of recurrence risk prediction, especially in the detection of metabolic pathway disturbances. Risk trends can be discovered in a timely manner, thereby achieving early warning and reducing risks caused by prediction lags.

[0157] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for predicting gastric cancer recurrence risk based on metabolomics feature analysis, characterized in that: The steps include: S1. Collect multiple metabolite concentration information of patients, including key metabolites in energy metabolism pathways, lipid metabolism pathways, and amino acid metabolism pathways related to gastric cancer, and construct a metabolomics dataset; S2. Preprocess the collected metabolomics data set; S3. Based on the preprocessed metabolomics dataset, extract metabolite characteristics associated with gastric cancer recurrence, including the concentration change patterns of key metabolites in the energy metabolism pathway, lipid metabolism pathway, and amino acid metabolism pathway, and analyze the dynamic changes of each metabolite in the metabolic pathway; S4. Based on the extracted metabolite features, the metabolic pathway integration method is used to quantify the disturbance of each metabolic pathway and calculate the abnormal degree of metabolite concentration changes within the metabolic pathway. The metabolic pathway integration method is combined with each metabolite metabolomics data set for weight allocation and biochemical association analysis; S5. Combine the patient's clinical history data and treatment plan data to construct a multidimensional feature matrix containing metabolomics data sets, metabolic pathway scores, and individual patient information; S6. Inputting the multidimensional feature matrix into a trained machine learning model, the machine learning model performs a comprehensive assessment on the risk of gastric cancer recurrence, and outputting a gastric cancer recurrence risk score for the patient based on the comprehensive assessment of the machine learning model; S7. Generate a prediction report based on the recurrence risk score, including the degree of abnormality of the metabolic pathway, the level of recurrence risk, and treatment adjustment recommendations based on the individual characteristics of the patient.

2. The method for predicting gastric cancer recurrence risk based on metabolomics feature analysis according to claim 1, characterized in that: The S1 specifically includes the following steps: S11. Collect multiple metabolite concentration information of the patient, including key metabolites in the energy metabolism pathway, lipid metabolism pathway and amino acid metabolism pathway related to gastric cancer. Suppose the metabolite concentration data of the patient is in the matrix form X = {x ij }, where i represents the type of metabolite, j represents the samples collected at different time points, and x ij represents the concentration value of the i-th metabolite at the j-th time point; S12. Classify the metabolite data and label the metabolites in the energy metabolism pathway as set M. E , metabolites in the lipid metabolism pathway are labeled as set M L , the metabolites in the amino acid metabolic pathway are labeled as set M A , the metabolomics dataset X is divided into: X E ={x ij |i∈M E },X L ={x ij |i∈M L },X A ={x ij |i∈M A }; S13. Constructing energy metabolism pathway set X E , lipid metabolism pathway set X L and amino acid metabolic pathway set X A Metabolomics dataset D: D={X E ,X L ,X A }。 3. The method for predicting gastric cancer recurrence risk based on metabolomics feature analysis according to claim 1, characterized in that: The S3 specifically includes the following steps: S31. Extract metabolite features related to gastric cancer recurrence based on the preprocessed metabolomics dataset D. Suppose the metabolite feature set is F = {f i }, i represents the type of metabolite, f i Represents the i-th metabolite feature: Among them, x ij represents the concentration value of the ith metabolite at the jth time point, μ i is the mean of the ith metabolite, σ i is the standard deviation, w ij is the weight corresponding to time point j, λ ij is the time decay factor; S32, using dynamic time warping algorithm to analyze metabolite features f i Analyze and evaluate the dynamic change trend of metabolite concentration over time, and generate the dynamic change pattern matrix T = {t ij }, t ij Indicates the change amplitude of the i-th metabolite at the j-th time point: Among them, f i (t k ) represents metabolite i at time point t k Concentration mode value, Δt k is the time interval, α ik is the weight factor for each time point, β ik Indicates the association weights between different time periods; S33. Extract the characteristics of each metabolic pathway according to the dynamic changes of each metabolite in the metabolic pathway, including the energy metabolic pathway characteristics P E , lipid metabolism pathway characteristics are P L and amino acid metabolic pathway characteristics P A , each feature is calculated as follows: Among them, P X represents the total eigenvalue of metabolic pathway X, γ i represents the pathway weight of metabolite i, t ij represents the dynamic change amplitude of metabolite i at time point j, ω ij is the correction coefficient at the time point, T i is the overall temporal variation trend of metabolite i, η X is the pathway characteristic adjustment factor.

4. The method for predicting gastric cancer recurrence risk based on metabolomics feature analysis according to claim 1, characterized in that: The S4 specifically comprises the following steps: S41. Based on the extracted metabolic pathway features, the energy metabolic pathway features P E , lipid metabolism pathway characteristics are P L and amino acid metabolic pathway characteristics P A The metabolic pathway integration method is used for quantification, and the perturbation integral of each metabolic pathway is I X : Among them, w X (t) is the weight factor at time t, ξ X represents the characteristic correction coefficients of metabolic pathway X, t1 and t n is the starting point and end point of the time period; S42. Perturbation integral according to metabolic pathway I X and metabolite characteristics f i Calculate the abnormality of each metabolic pathway A X .

5. The method for predicting gastric cancer recurrence risk based on metabolomics feature analysis according to claim 1, characterized in that: The S5 specifically includes the following steps: S51, combined with the patient's clinical history data C = {c k }, where c k represents the patient's kth clinical characteristic data, including medical history, surgical records, and pathology report information; S52, combined with the patient's treatment plan data T = {t l }, where t l Represents the relevant data of the first treatment plan, including the type of treatment, treatment dosage and the patient's response to the treatment; S53, combining the metabolomics data set D, the patient's clinical history data C, and the treatment plan data T to construct a multidimensional feature matrix M: Among them, X i represents the concentration value of the ith metabolite in metabolic pathway X, is the weight factor of the pathway, c k is the kth clinical characteristic data, and are the mean and standard deviation of the clinical characteristics respectively; t l is the data of the first treatment plan, and are the mean and standard deviation of the treatment options respectively; α i , β k , γ kl and η i,k,l are the correction coefficients and weight factors of metabolomics, clinical characteristics and treatment regimens in the matrix, respectively.

6. The method for predicting gastric cancer recurrence risk based on metabolomics feature analysis according to claim 1, characterized in that: The S6 specifically comprises the following steps: S61. Input the constructed multidimensional feature matrix M into the spatiotemporal coupled quantum optimization network, encode the multidimensional feature matrix M in quantum state, and use quantum bits to encode the changes in each feature dimension. The quantum state of the metabolomics feature is q i (M); S62. The multidimensional feature matrix M encoded in the quantum state is processed through the spatiotemporal coupling network to construct a spatiotemporal coupling relationship matrix. The spatiotemporal coupling network adaptively captures the changing trends of metabolomics, clinical history data and treatment response data at different time points to generate a spatiotemporal feature matrix S(M): Among them, α i (t) is the feature dimension M i The time weight, f(M i ) is the spatial distribution characteristics of each feature dimension in the multidimensional feature matrix; S63. Use quantum optimization learning algorithm to globally optimize the spatiotemporal feature matrix S(M), use quantum approximate optimization algorithm to search for gastric cancer recurrence risk characteristics, and generate gastric cancer recurrence risk score R Q (S(M)): Among them, μ S and σ S are the mean and standard deviation of the spatiotemporal feature matrix S(M), γ 1i is the weight factor of different feature dimensions, κ i is the adjustment factor of the feature dimension; S64. Perform nonlinear perturbation simulation on the input clinical history data and treatment plan data through the nonlinear pathological feature perturbation module to generate a nonlinear perturbation feature matrix N(M,t): Among them, C is the clinical history data, μ C and σ C are the mean and standard deviation of clinical historical data, μ T and σ T are the mean and standard deviation of the treatment data, δ k , η l are weight factors respectively; S66, the quantum optimized gastric cancer recurrence risk score R Q (S(M)) is comprehensively evaluated with the nonlinear perturbation feature matrix N(M,t) to generate a gastric cancer recurrence risk prediction score R final (M): R final (M)=R Q (S(M))+λ1·N(M,t); Among them, λ1 is the weight coefficient combining nonlinear perturbation and quantum optimization results.

7. The method for predicting gastric cancer recurrence risk based on metabolomics feature analysis according to claim 1, characterized in that: The S7 specifically comprises the following steps: S71. Based on the generated recurrence risk score R final (M) Generate a prediction report based on the degree of abnormality of the metabolic pathway A X Perform abnormal analysis. The higher the abnormality, the greater the risk of recurrence. S73, according to the recurrence risk score R final (M) Determine the recurrence risk level L R , set the threshold θ to divide the risks into different levels: Among them, θ1 and θ2 are the preset thresholds of risk levels; S74. Generate treatment adjustment recommendations based on the individual characteristics of the patient, metabolic pathway characteristics, and clinical treatment response data. The treatment adjustment recommendations are optimized based on the abnormal degree of the patient's metabolite concentration and the dynamic trend of clinical data: Adjust the dose and combination of the patient's current medication based on abnormal metabolic pathway scores and treatment response data; Combined with the patient's quantum optimized risk score R final (M), it is recommended to regularly and dynamically monitor the patient's metabolic characteristics; If the patient's metabolic pathway abnormalities are related to their immune system response, it is recommended to adjust the immunotherapy regimen based on the metabolic characteristics; According to the patient's response characteristics in the multidimensional feature matrix, the treatment plan is adjusted in a timely manner to optimize the drug combination and dosage.

Images