Treatment response prediction method and system, storage medium and terminal

Abstract

The invention discloses a treatment response prediction method and system, a storage medium and a terminal, and the method comprises the steps: carrying out the heterogeneity analysis and cell division of first single cell sequencing data of a patient through employing a clustering algorithm, and obtaining a plurality of corresponding cell subgroups; performing feature extraction and biomarker recognition on all the cell subgroups to obtain corresponding cell subgroup features; wherein the cell subset characteristics comprise a cell development trajectory and a plurality of cell markers; and inputting the clinical information and the cell subset characteristics of the patient into the treatment response prediction model, so that the model performs multi-factor analysis and outputs a treatment response prediction result of the patient for each treatment scheme. By adopting the embodiment of the invention, the tiny key difference between the cells is captured, the cells are divided based on the difference between the cells, and then feature extraction and biomarker recognition are performed on each cell subset, so that the specific gene expression modes of different cell subpopulations corresponding to diseases can be recognized, and the treatment response prediction accuracy is further improved.

Description

Technical Field

[0001] The present invention relates to the field of therapeutic effect prediction, and in particular to a treatment response prediction method, system, storage medium and terminal. Background Art

[0002] In the field of biomedical research and clinical diagnosis, the realization of treatment response prediction is highly dependent on a deep understanding of cellular heterogeneity, which is crucial for revealing disease mechanisms, accurate diagnosis, and personalized treatment plans to achieve personalized medicine. However, traditional cell analysis techniques, due to their limitations in analyzing large cell populations, often fail to capture the tiny but critical differences between cells, limiting the in-depth understanding of individual cell characteristics.

[0003] The emergence of single-cell sequencing technology has brought breakthrough opportunities for the development of treatment response prediction and personalized medicine. Through single-cell sequencing technology, gene expression and genomic variation can be deeply analyzed at the level of individual cells, thereby providing a new perspective to understand the complex changes of cells under different biological states and disease conditions. This technology can reveal multi-dimensional information such as gene expression levels, DNA sequence variation, epigenetic modification, etc., providing a rich data foundation for treatment response prediction and personalized medicine. Although single-cell sequencing technology has great potential in theory, it faces multiple challenges such as the complexity of data analysis, high-dimensional data characteristics, heterogeneity between cells, technical noise, and data conversion. Therefore, the accuracy of treatment response prediction and personalized medicine based on single-cell sequencing technology is limited, and it is difficult to provide users with effective treatment recommendations. Summary of the invention

[0004] Embodiments of the present invention provide a treatment response prediction method, system, storage medium and terminal, which capture the tiny but critical differences between cells and identify the specific gene expression patterns of different cell subpopulations corresponding to the disease to improve the accuracy of treatment response prediction.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a method for predicting treatment response, comprising:

[0006] Obtaining the patient’s first single-cell sequencing data;

[0007] Using a clustering algorithm, performing heterogeneity analysis and cell division on the first single-cell sequencing data to obtain a corresponding plurality of cell subpopulations;

[0008] Performing feature extraction and biomarker identification on all the cell subpopulations to obtain corresponding cell subpopulation features; wherein the cell subpopulation features include cell development trajectories and a number of cell markers;

[0009] The patient's clinical information and the cell subpopulation characteristics are input into a pre-constructed treatment response prediction model, so that the treatment response prediction model performs multi-factor analysis and outputs the patient's treatment response prediction results for each treatment regimen.

[0010] In the implementation of the embodiment of the present invention, a clustering algorithm is used to perform heterogeneity analysis and cell division on the first single-cell sequencing data collected from the patient. Through heterogeneity analysis, not only can the small but critical differences between cells be captured, but also the occurrence and development mechanism of the disease can be understood more deeply, and the cells can be divided based on the differences between cells, thereby obtaining corresponding multiple cell subpopulations, so as to subsequently identify the specific gene expression patterns of different cell subpopulations corresponding to the disease, so that the treatment response prediction can fully consider factors such as the invasiveness and / or drug resistance of the disease, thereby improving the accuracy of the treatment response prediction, and then all cell subpopulations are subjected to feature extraction and biomarker identification to identify the specific gene expression patterns of different cell subpopulations, and obtain cell subpopulation characteristics including cell development trajectories and several cell markers, which provide a solid foundation for subsequent biological verification and treatment response prediction. In addition, the patient's clinical information and cell subpopulation characteristics are input into a pre-constructed treatment response prediction model so that the model performs multi-factor analysis and outputs the patient's treatment response prediction results for different treatment plans, assisting medical staff in selecting the most likely effective treatment plan, thereby improving the success rate and effect of treatment, and providing patients with better medical services.

[0011] As a preferred solution, feature extraction and biomarker identification are performed on all the cell subpopulations to obtain corresponding cell subpopulation features, specifically:

[0012] Identifying the cell state transition points of cells in each of the cell subpopulations, and drawing corresponding cell development trajectories based on all of the cell state transition points;

[0013] Using deep learning technology, feature extraction and pattern recognition are performed on all the cell subpopulations to obtain a corresponding plurality of differentially expressed genes, and a number of cell markers are screened out from all the differentially expressed genes according to preset rules;

[0014] The differentially expressed genes are genes whose expression change multiples are greater than a preset value.

[0015] A preferred scheme of implementing the embodiments of the present invention identifies the cell state transition points of cells in each cell subpopulation, and based on all cell state transition points, draws the corresponding cell development trajectory to facilitate understanding of how cells transition between different states and the relationship between state transition and disease development.

[0016] As a preferred solution, the deep learning technology is used to perform feature extraction and pattern recognition on all the cell subpopulations to obtain a corresponding plurality of differentially expressed genes, and a number of cell markers are screened out from all the differentially expressed genes according to preset rules, specifically:

[0017] Using deep learning technology, feature extraction and pattern recognition are performed on all the cell subpopulations to obtain corresponding multiple differentially expressed genes;

[0018] Combined with GO and KEGG, functional enrichment analysis and pathway analysis were performed on all the differentially expressed genes, and based on the analysis results, several cell markers were screened out from all the differentially expressed genes.

[0019] By implementing the preferred scheme of the embodiment of the present invention, combining GO and KEGG, pathway analysis of all differentially expressed genes can help understand how each differentially expressed gene plays a role in a specific signaling pathway, and functional enrichment analysis of all differentially expressed genes can determine whether a differentially expressed gene is enriched in a specific biological process, such as cell proliferation or apoptosis, and then screen out several cell markers from all differentially expressed genes.

[0020] As a preferred solution, the obtaining of the first single-cell sequencing data of the patient is specifically as follows:

[0021] Collecting the patient's second single-cell sequencing data;

[0022] The second single-cell sequencing data is preprocessed to obtain corresponding first single-cell sequencing data; wherein the preprocessing includes removing background noise, eliminating outliers, signal correction and normalization.

[0023] According to a preferred scheme of the embodiment of the present invention, after collecting the second single-cell sequencing data of the patient, the second single-cell sequencing data is preprocessed to remove background noise, which can effectively reduce the impact of noise and provide a more accurate data basis for subsequent analysis. The second single-cell sequencing data is preprocessed to remove outliers to avoid the impact of non-valid values ​​on data processing efficiency. The second single-cell sequencing data is preprocessed to correct the signal to ensure the accuracy and reliability of the data. The second single-cell sequencing data is preprocessed to normalize the data to eliminate the differences between different samples, making the data easier to compare and analyze.

[0024] As a preferred embodiment, the method for predicting treatment response further comprises:

[0025] A personalized treatment plan is formulated based on the patient's predicted treatment response to each treatment plan and the patient's clinical information.

[0026] A preferred solution of the embodiment of the present invention is implemented to formulate a personalized treatment plan based on the predicted results of the patient's treatment response to each treatment plan and the patient's clinical information, so as to assist the user in selecting a more appropriate treatment measure.

[0027] As a preferred solution, the clustering algorithm is used to perform heterogeneity analysis and cell division on the first single-cell sequencing data to obtain corresponding multiple cell subpopulations, specifically:

[0028] A clustering algorithm with an adaptive threshold is used to perform heterogeneity analysis on the first single-cell sequencing data, and each cell is divided into a corresponding cell subpopulation according to the heterogeneity analysis results.

[0029] A preferred scheme for implementing the embodiment of the present invention, compared with the traditional fixed threshold method, adopts an adaptive threshold clustering algorithm to perform heterogeneity analysis on the first single-cell sequencing data, and divides each cell into a corresponding cell subpopulation according to the heterogeneity analysis results, which can better handle noise and outliers in the data, thereby improving the accuracy and reliability of the clustering results.

[0030] In order to solve the same technical problem, an embodiment of the present invention further provides a treatment response prediction system, comprising:

[0031] A data acquisition module, used to acquire the first single-cell sequencing data of the patient;

[0032] A cell segmentation module, configured to use a clustering algorithm to perform heterogeneity analysis and cell segmentation on the first single-cell sequencing data to obtain a corresponding plurality of cell subpopulations;

[0033] An extraction and identification module is used to perform feature extraction and biomarker identification on all the cell subpopulations to obtain corresponding cell subpopulation features; wherein the cell subpopulation features include cell development trajectories and a number of cell markers;

[0034] The treatment response prediction module is used to input the patient's clinical information and the cell subpopulation characteristics into a pre-constructed treatment response prediction model, so that the treatment response prediction model performs multi-factor analysis and outputs the patient's treatment response prediction results for each treatment regimen.

[0035] As a preferred solution, the treatment response prediction system further includes:

[0036] The program formulation module is used to formulate a personalized treatment plan based on the patient's predicted treatment response results for each treatment plan and the patient's clinical information.

[0037] In order to solve the same technical problem, the present invention also provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the treatment response prediction method.

[0038] In order to solve the same technical problem, the present invention also provides a terminal, including a processor, a memory and a computer program stored in the memory; wherein the computer program can be executed by the processor to implement the treatment response prediction method. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 : A schematic diagram of a process flow of a treatment response prediction method provided in Example 1 of the present invention;

[0040] Figure 2 : A structural diagram of a treatment response prediction system provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] Embodiment one:

[0043] Please refer to Figure 1 , is a treatment response prediction method provided by an embodiment of the present invention, the method includes steps S1 to S4, each step is specifically as follows:

[0044] Step S1, obtaining the first single-cell sequencing data of the patient.

[0045] In this embodiment, the first single-cell sequencing data includes but is not limited to gene expression levels, DNA sequence variations, and epigenetic modifications.

[0046] As a preferred solution, step S1 includes step S11 to step S12, and each step is specifically as follows:

[0047] Step S11, collecting the second single-cell sequencing data of the patient.

[0048] Specifically, for 50 patients with a specific disease state, expression data of 100,000 single cells of each patient are collected by single-cell sequencing technology, and expression data of 5,000,000 single cells in total are collected to form the second single-cell sequencing data. The single-cell sequencing technology may be scRNA-seq.

[0049] As an example, FastQC software was used to perform quality control on the collected second single-cell sequencing data, and the quality threshold was set to Q30, which is expected to eliminate about 5% of low-quality reads.

[0050] Step S12: preprocess the second single-cell sequencing data to obtain corresponding first single-cell sequencing data.

[0051] Among them, preprocessing includes but is not limited to removing background noise, eliminating outliers, signal correction and normalization processing.

[0052] In this embodiment, the preprocessing includes steps S121 to S124, and each step is specifically as follows:

[0053] Step S121: removing background noise: using a Gaussian filter to smooth the second single-cell sequencing data to reduce random noise.

[0054] Among them, when choosing σ, larger values ​​will reduce noise but may blur important features, while smaller values ​​will retain more details but have more noise.

[0055] Step S122, outlier removal: by calculating the mean μ and standard deviation σ of the data set containing the second single-cell sequencing data, and then removing those points whose deviation from μ exceeds k times σ, where the selection of k depends on the sensitivity to outliers, usually k=2 or k=3.

[0056] Step S123, calibrate the signal.

[0057] Step S124, normalization processing: using Z-score or Quantile normalization method to ensure the consistency and comparability of the data, so that the mean of the data set is close to 0 and the standard deviation is close to 1, thereby improving the comparability of the data.

[0058] It should be noted that in the actual data processing process, data cleaning steps such as background noise removal, outlier removal, and signal correction are mainly to remove noise and outliers that may affect subsequent analysis to ensure data quality; while the data normalization step is to ensure that data from different samples or under different experimental conditions are comparable. Therefore, the data cleaning step is usually performed before the data normalization step. By sequentially performing the operations of background noise removal, outlier removal, signal correction, and normalization, unqualified data points can be eliminated, ensuring a high-quality foundation for sequencing data, and preparing a consistent data set for subsequent analysis.

[0059] Step S2: using a clustering algorithm to perform heterogeneity analysis and cell division on the first single-cell sequencing data to obtain a corresponding plurality of cell subpopulations.

[0060] As a preferred solution, step S2 includes step S21 to step S22, and each step is specifically as follows:

[0061] Step S21, using an adaptive threshold clustering algorithm to perform heterogeneity analysis on the first single-cell sequencing data.

[0062] Specifically, the adaptive threshold clustering algorithm can be density-based DBSCAN or graph-theory-based Louvain method. Among them, the parameter selection and optimization of the clustering algorithm are carried out by cross-validation method to ensure the robustness and reliability of the results. Cluster analysis, the DBSCAN algorithm identifies clusters in the data by setting and minPts. It defines the maximum distance between two points that are considered neighbors, and minPts is the minimum number of points required to form a dense area. Parameter optimization, cross-validation is used to evaluate the clustering effect of different and minPts, and select the parameters that make the clustering results most consistent with the actual biological significance.

[0063] As an example, when applying the graph-theory based Louvain clustering algorithm, one can set the resolution parameter to 1.0 and expect to identify 5 to 10 different cell subpopulations.

[0064] Step S22: According to the heterogeneity analysis result of the first single-cell sequencing data, each cell is divided into a corresponding cell subpopulation.

[0065] Step S3: perform feature extraction and biomarker identification on all cell subpopulations to obtain corresponding cell subpopulation features.

[0066] Among them, cell subpopulation characteristics include cell development trajectories and several cell markers.

[0067] As a preferred solution, step S3 includes step S31 to step S32, and each step is specifically as follows:

[0068] Step S31, identifying the cell state transition points of cells in each cell subpopulation, and drawing the corresponding cell development trajectory based on all cell state transition points.

[0069] In this embodiment, using Monocle software, it is expected to construct a cell development trajectory and reveal at least 3 key cell state transition points. Based on the key cell state transition points, cell trajectory inference is performed to construct a clear cell development trajectory and identify at least 3 key differentiation or development stages.

[0070] Step S32, using deep learning technology to perform feature extraction and pattern recognition on all cell subpopulations, obtain corresponding multiple differentially expressed genes, and screen out several cell markers from all differentially expressed genes according to preset rules.

[0071] Among them, differentially expressed genes are genes whose expression change fold is greater than a preset value.

[0072] In this embodiment, the preset value is 2.

[0073] As a preferred solution, step S32 includes step S321 to step S322, and each step is specifically as follows:

[0074] Step S321, using deep learning technology to perform feature extraction and pattern recognition on all cell subpopulations to obtain corresponding multiple differentially expressed genes.

[0075] In this example, deep learning technology is used to extract features and perform pattern recognition on all cell subpopulations to obtain at least 100 significantly differentially expressed genes, of which at least 50 genes have a padj value less than 0.05, providing key clues for further biological and clinical research.

[0076] As an example, the deep learning technique may be a support vector machine (SVM) or a random forest (RF).

[0077] Step S322, combining GO and KEGG, performing functional enrichment analysis and pathway analysis on all differentially expressed genes, and screening out several cell markers from all differentially expressed genes based on the analysis results.

[0078] In this example, GO and KEGG can help understand the biological significance of differentially expressed genes, and then screen cell subpopulation-specific marker genes to help identify cell types. Among them, functional enrichment analysis will reveal at least 10 significantly enriched biological processes or molecular functions, further explaining the biological significance of cell subpopulations.

[0079] Step S4, inputting the patient's clinical information and cell subpopulation characteristics into a pre-constructed treatment response prediction model, so that the treatment response prediction model performs multi-factor analysis and outputs the patient's treatment response prediction results for each treatment regimen.

[0080] Specifically, in the process of constructing the treatment response prediction model, cell cycle, cell state and microenvironment factors were taken into account to comprehensively evaluate the potential impact of the treatment plan. The treatment response prediction model uses multivariate analysis and combines clinical data to predict the efficacy.

[0081] In this embodiment, the construction of the treatment response prediction model can use logistic regression or decision tree algorithm to predict the efficacy of different treatment plans. Among them, the model constructed by logistic regression is used to predict the probability of the patient's response to the treatment, and the coefficient β in the model is obtained by maximum likelihood estimation, which is used to quantify the effect of each feature on the treatment response; the model constructed by the decision tree algorithm is used to predict the treatment response, and the prediction error is minimized by selecting the optimal split point.

[0082] As a preferred solution, a treatment response prediction method provided in an embodiment of the present invention further includes step S5, which is specifically as follows:

[0083] Step S5, formulating a personalized treatment plan based on the predicted results of the patient's treatment response to each treatment plan and the patient's clinical information.

[0084] In this embodiment, step S5 includes step S51 to step S58, and each step is specifically as follows:

[0085] Step S51, integrating the patient's clinical information.

[0086] The patient's clinical information includes, but is not limited to, age, gender, medical history, and lifestyle-related data, such as eating habits, exercise frequency, sleep quality, and sleep patterns, which are critical to understanding the patient's overall health status.

[0087] Step S52, analyzing the genetic background of the patient.

[0088] Among them, genetic background includes but is not limited to family medical history, known gene mutations and polymorphisms. Through the following methods ① to ⑩, the genetic background of the patient can be systematically analyzed to accurately identify the patient's possible genetic susceptibility and provide data support for personalized treatment.

[0089] ① Detailed record of family medical history: Collect and record the medical history of the patient and his or her family members to determine the possible inheritance pattern.

[0090] ② High-throughput genome sequencing: Perform whole exome sequencing or whole genome sequencing to identify genetic variations in the patient's genome.

[0091] ③Genetic variation detection and annotation: Use tools such as GATK and Samtools to identify SNPs, indels and CNVs, and perform functional annotation through ANNOVAR.

[0092] ④Disease association database query: Query databases such as ClinVar and OMIM to evaluate the association between variants and specific diseases.

[0093] ⑤Statistical association analysis: Use chi-square test or Fisher's exact test to determine the statistical association between the variant and the disease.

[0094] ⑥ Application of polygenic risk score: Based on known disease-related variants, the PRS method is used to calculate the patient's genetic risk score.

[0095] ⑦ Machine learning model development: Develop and train models such as support vector machines and random forests to improve the prediction accuracy of disease susceptibility.

[0096] ⑧ Comprehensive risk assessment model: Combine genetic scores with clinical data to establish a comprehensive assessment model to provide a basis for personalized treatment plans.

[0097] ⑨ Experimental validation of results: Ensure the reliability of the prediction results through laboratory functional validation or replication studies in independent cohorts.

[0098] ⑩Genetic counseling and result interpretation: Professional genetic counselors will interpret the analysis results and provide patients with personalized genetic counseling.

[0099] Step S53, obtaining the patient's response to different drugs during drug sensitivity testing.

[0100] Step S54, combining the data obtained from steps S51 to S53, uses deep learning technology to perform disease risk assessment and prognosis analysis to provide data support for formulating treatment strategies.

[0101] Specifically, step S54 includes step S541 to step S5410, and each step is specifically as follows:

[0102] Step S541, data integration and cleaning: Integrate the patient's clinical information, genetic data and response to different drugs, and perform cleaning, including missing value processing and outlier removal.

[0103] Step S542, feature engineering: extract key SNPs from genetic data, quantify and encode the patient's clinical information and response to different drugs.

[0104] Step S543, model architecture selection: select a suitable deep learning model according to the data type, such as LSTM for time series and CNN for image data.

[0105] Step S544, model training and tuning: use cross-validation to train the model and optimize performance by adjusting hyperparameters.

[0106] Step S545, feature importance assessment: determine the most critical factors affecting disease risk and prognosis.

[0107] It should be noted that in the feature importance evaluation stage, the gene expression data from single-cell sequencing is first integrated with the patient's detailed clinical records, including but not limited to age, gender, medical history, lifestyle habits, etc. Among them, correlation analysis is used to identify the gene expression variables that are most correlated with the disease state. Next, the model is trained to predict disease risk through the random forest algorithm, and the feature importance score provided by the algorithm is used to determine which gene expression variables contribute most to the predictive ability of the model. In addition, the importance of each feature is evaluated by randomly disrupting the order of the values ​​of a single feature and observing the changes in model accuracy. Through the above steps, it is ensured that key biomarkers that have a significant impact on disease risk and prognosis can be accurately identified.

[0108] Step S546, risk scoring system development: generating a personalized risk score for the patient based on the training model.

[0109] It should be noted that in the development of the risk scoring system, the logistic regression model was chosen as the basis because it not only has good prediction effects, but also can provide a clear weight for each feature, which can be directly converted into a risk score. The model is then trained using the training data set, paying special attention to the problem of overfitting, and limiting the complexity of the model by setting regularization parameters. Among them, 5-fold cross-validation is used to evaluate the performance of the model on different data partitions to ensure the stability and generalization ability of the model. Finally, the risk score of each patient is calculated by inputting their clinical information and gene expression data into the trained logistic regression model. This score quantifies the probability of the patient's future disease risk.

[0110] Step S547, prognosis model construction: using a deep learning model to predict disease progression and patient survival probability.

[0111] It should be noted that, in the construction of the prognostic model, considering that the disease progression may involve time series data, this embodiment uses a recurrent neural network (RNN) or its variants, such as a long short-term memory network (LSTM), because they can process sequence data and capture long-term dependencies. The single-cell sequencing data is then preprocessed, including removing outliers, normalization, and possible time series feature extraction. During model training, this embodiment uses mean square error as a loss function, and optimizes the training process by adjusting the learning rate and batch size. In order to prevent overfitting, this embodiment also applies dropout regularization technology. In the model validation stage, an independent test set is used to evaluate the prediction accuracy of the model, and the model performance is comprehensively evaluated by indicators such as confusion matrix, ROC curve, and precision-recall curve. In addition, this embodiment uses SHAP values ​​to explain the contributing factors of each prediction to ensure that the decision-making process of the model is transparent and explainable, which is crucial for clinical applications. Through this series of detailed steps, a prognostic model that can accurately predict disease progression and patient survival probability is constructed, providing a powerful tool for clinical decision-making.

[0112] Step S548, model interpretability analysis: Use SHAP or LIME to improve the transparency and credibility of model predictions.

[0113] Step S549, clinical validation: verify the accuracy and generalization ability of the model on an independent clinical cohort.

[0114] Step S5410, integrated learning application: combining deep learning with other models to form an integrated learning system to improve prediction performance.

[0115] It should be noted that ensemble learning is a machine learning paradigm that improves prediction performance by combining multiple different models, and can usually achieve better results than a single model. In addition, this embodiment needs to ensure that model development complies with data protection regulations and medical ethics standards, and that the model report contains necessary information for clinicians to understand and apply.

[0116] Specifically, step S5410 includes step S01 to step S09, and each step is specifically as follows:

[0117] Step S01, model selection: select a set of different deep learning models and traditional machine learning models based on the characteristics of the problem and the characteristics of the data.

[0118] In this embodiment, different types of neural networks (such as CNN, RNN, LSTM), support vector machines (SVM), random forests, gradient boosted tree (GBDT), etc. can be selected.

[0119] Step S02, data preprocessing: perform unified preprocessing on the data, including normalization, encoding conversion, processing of missing values, etc., to ensure that the data format of all model inputs is consistent.

[0120] Step S03, model training: using the same training data set, each model is trained separately.

[0121] In this embodiment, for deep learning models, more detailed parameter tuning may be required, such as selecting a suitable network architecture, activation function, optimizer, learning rate, etc. For traditional machine learning models, feature selection and parameter tuning may be required.

[0122] Step S04, model optimization: Use cross-validation to evaluate the performance of each model, such as accuracy, recall, F1 score, etc. Based on the results of cross-validation, adjust and optimize the model to improve its prediction performance.

[0123] Step S05, ensemble strategy design: determine the ensemble learning method.

[0124] In this embodiment, common ensemble learning methods include bagging, boosting and stacking. For bagging, an integrated version of a random forest or deep learning model can be used, such as dropout, which can be regarded as an integrated strategy. For boosting, algorithms such as AdaBoost and GradientBoosting can be used to focus on the errors of the previous model by iteratively training the model. For stacking, a meta-learner (such as linear regression, logistic regression) can be trained, which takes the prediction results of other base learners as input to make the final prediction.

[0125] Step S06, integrated model training: different models are combined according to the selected integration strategy.

[0126] In this embodiment, for Stacking, the base learner is first trained, and then the meta learner is trained using the prediction results of the base learner. For Bagging and Boosting, the corresponding algorithms can be directly used for training.

[0127] Step S07, performance evaluation: Use an independent test set to evaluate the performance of the integrated model. Compare the performance of the integrated model with other single models to verify the effectiveness of integrated learning.

[0128] Step S08, model tuning: further tune the parameters such as weights, learning rate and / or regularization parameters of the integrated model according to the evaluation results of the test set.

[0129] Step S09, model deployment: deploy the trained integrated model to the production environment for actual prediction tasks. Ensure the scalability and maintainability of the model so that it can be updated and iterated when new data arrives.

[0130] Through this series of detailed steps, a powerful ensemble learning system can be built, which combines the advantages of deep learning and other machine learning models to improve the model's predictive performance and generalization ability.

[0131] Step S55, formulate a personalized treatment plan to achieve the best treatment effect.

[0132] In this embodiment, the personalized treatment plan includes, but is not limited to, a medication plan, lifestyle adjustment recommendations, and preventive measures.

[0133] Step S56, based on the patient's feedback and monitoring data, the treatment plan is continuously optimized and adjusted to adapt to the patient's individual response and changes in health status.

[0134] Step S57, using visualization tools to graphically display the personalized treatment plan and treatment response prediction results, and provide patient education to help understand the personalized treatment plan.

[0135] Specifically, it transforms complex medical data and personalized treatment plans into intuitive charts and graphics, making it easier for patients and medical professionals to understand information and make decisions. It provides disease-related education to patients, helps them understand treatment plans, and provides necessary psychological and social support to enhance patients' compliance with treatment and quality of life.

[0136] Step S58, during the treatment process, continuously monitor the patient's health status, track the treatment effect, and adjust the personalized treatment plan as needed.

[0137] It should be noted that during the treatment process, a preliminary treatment plan is first formulated based on the patient's comprehensive situation. Subsequently, the patient's baseline health data, including biochemical indicators and quality of life assessment, is recorded in detail when the treatment is started. During the treatment, the patient's physiological data and drug response are collected regularly, and the patient is encouraged to record changes in self-perception and lifestyle. Among them, these data are tracked and analyzed in real time through the electronic health record system to evaluate the treatment effect. Regular multidisciplinary team meetings will discuss the patient's treatment progress and adjust the treatment plan in a timely manner based on clinical results and patient-reported results, including changes in drug dosage or treatment methods. In addition, by providing patients with necessary education and communication, ensure that they understand the purpose of treatment adjustment and participate in treatment decisions. At the same time, in accordance with the patient's wishes, wearable devices and remote monitoring technology can also be used to automatically collect and improve monitoring efficiency to ensure that patients receive the best treatment effects and continuous and real-time medical care.

[0138] Furthermore, this embodiment also provides a user interaction interface, and the functions of the user interaction interface are as follows:

[0139] 1) Data upload, after the user uploads the data through the interface, the system automatically performs the data cleaning and standardization process, and processes the data using the formulas and algorithms described above.

[0140] 2) Analysis execution: After the user starts the analysis, the system background automatically performs steps such as clustering, marker identification, and prediction model construction, and uses corresponding algorithms and formulas for calculations.

[0141] 3) Results presentation: The analysis results are presented through a graphical interface, such as using a heat map to display gene expression differences and a decision tree diagram to display the logical process of treatment response prediction.

[0142] Please refer to Figure 2 , is a schematic diagram of the structure of a treatment response prediction system provided by an embodiment of the present invention, the system includes a data acquisition module M1, a cell segmentation module M2, an extraction and recognition module M3 and a treatment response prediction module M4, and the specifics of each module are as follows:

[0143] A data acquisition module M1 is used to acquire the first single-cell sequencing data of the patient;

[0144] The cell segmentation module M2 is used to perform heterogeneity analysis and cell segmentation on the first single-cell sequencing data using a clustering algorithm to obtain a corresponding plurality of cell subpopulations;

[0145] The extraction and identification module M3 is used to extract features and identify biomarkers for all cell subpopulations to obtain corresponding cell subpopulation features; wherein the cell subpopulation features include cell development trajectories and several cell markers;

[0146] The treatment response prediction module M4 is used to input the patient's clinical information and cell subpopulation characteristics into a pre-built treatment response prediction model, so that the treatment response prediction model performs multi-factor analysis and outputs the patient's treatment response prediction results for each treatment regimen.

[0147] As a preferred option, please refer to Figure 2 The treatment response prediction system provided by the embodiment of the present invention further includes a program formulation module M5, which is specifically as follows:

[0148] The treatment plan formulation module M5 formulates a personalized treatment plan based on the patient's predicted treatment response results for each treatment plan and the patient's clinical information.

[0149] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0150] An embodiment of the present invention further provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a treatment response prediction method described in Embodiment 1.

[0151] An embodiment of the present invention further provides a terminal, comprising a processor, a memory, and a computer program stored in the memory; wherein the computer program can be executed by the processor to implement a treatment response prediction method described in Embodiment 1.

[0152] Preferably, the computer program can be divided into one or more modules / units (such as computer program, computer program), one or more modules / units are stored in the memory and executed by the processor to complete the present invention. One or more modules / units can be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal.

[0153] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor can also be any conventional processor. The processor is the control center of the terminal, and various parts of the terminal are connected using various interfaces and lines.

[0154] The memory mainly includes a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function, etc., and the data storage area can store related data, etc. In addition, the memory can be a high-speed random access memory, or a non-volatile memory, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, and a flash card (Flash Card), etc., or the memory can also be other volatile solid-state storage devices.

[0155] It should be noted that the above-mentioned terminal may include, but is not limited to, a processor and a memory. Those skilled in the art will understand that the above-mentioned terminal is merely an example and does not constitute a limitation on the terminal. It may include more or fewer components, or a combination of certain components, or different components.

[0156] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0157] The present invention provides a treatment response prediction method, system, storage medium and terminal, which use a clustering algorithm to perform heterogeneity analysis and cell division on the first single-cell sequencing data collected from patients. Through heterogeneity analysis, not only can the tiny but critical differences between cells be captured, but also the occurrence and development mechanism of the disease can be more deeply understood, and cell division is performed based on the differences between cells, thereby obtaining corresponding multiple cell subpopulations, so as to subsequently identify the specific gene expression patterns of different cell subpopulations corresponding to the disease, so that the treatment response prediction can fully consider factors such as the invasiveness and / or drug resistance of the disease, thereby improving the accuracy of treatment response prediction, and then feature extraction and biomarker identification are performed on all cell subpopulations to identify the specific gene expression patterns of different cell subpopulations, and obtain cell subpopulation characteristics including cell development trajectories and several cell markers, providing a solid foundation for subsequent biological verification and treatment response prediction. In addition, the patient's clinical information and cell subpopulation characteristics are input into a pre-built treatment response prediction model so that the model can perform multi-factor analysis and output the patient's treatment response prediction results for different treatment options, assisting medical staff in selecting the most likely effective treatment option, thereby improving the success rate and effectiveness of the treatment and providing patients with better medical services.

[0158] Furthermore, combining GO and KEGG, pathway analysis of all differentially expressed genes can help understand how each differentially expressed gene plays a role in a specific signaling pathway, while functional enrichment analysis of all differentially expressed genes can determine whether a differentially expressed gene is enriched in a specific biological process, such as cell proliferation or apoptosis, and then screen out several cell markers from all differentially expressed genes.

[0159] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting treatment response, characterized in that: include: Obtaining the patient’s first single-cell sequencing data; Using a clustering algorithm, performing heterogeneity analysis and cell division on the first single-cell sequencing data to obtain a corresponding plurality of cell subpopulations; Performing feature extraction and biomarker identification on all the cell subpopulations to obtain corresponding cell subpopulation features; wherein the cell subpopulation features include cell development trajectories and a number of cell markers; The patient's clinical information and the cell subpopulation characteristics are input into a pre-constructed treatment response prediction model, so that the treatment response prediction model performs multi-factor analysis and outputs the patient's treatment response prediction results for each treatment regimen.

2. A method for predicting treatment response according to claim 1, characterized in that: The feature extraction and biomarker identification are performed on all the cell subpopulations to obtain the corresponding cell subpopulation features, specifically: Identifying the cell state transition points of cells in each of the cell subpopulations, and drawing corresponding cell development trajectories based on all of the cell state transition points; Using deep learning technology, feature extraction and pattern recognition are performed on all the cell subpopulations to obtain a corresponding plurality of differentially expressed genes, and a number of cell markers are screened out from all the differentially expressed genes according to preset rules; The differentially expressed genes are genes whose expression change multiples are greater than a preset value.

3. A method for predicting treatment response according to claim 2, characterized in that: The deep learning technology is used to perform feature extraction and pattern recognition on all the cell subpopulations to obtain a corresponding plurality of differentially expressed genes, and a number of cell markers are screened out from all the differentially expressed genes according to preset rules, specifically: Using deep learning technology, feature extraction and pattern recognition are performed on all the cell subpopulations to obtain corresponding multiple differentially expressed genes; Combined with GO and KEGG, functional enrichment analysis and pathway analysis were performed on all the differentially expressed genes, and based on the analysis results, several cell markers were screened out from all the differentially expressed genes.

4. A method for predicting treatment response according to claim 1, characterized in that: The obtaining of the first single-cell sequencing data of the patient is specifically: Collecting the patient's second single-cell sequencing data; The second single-cell sequencing data is preprocessed to obtain corresponding first single-cell sequencing data; wherein the preprocessing includes removing background noise, eliminating outliers, signal correction and normalization.

5. A method for predicting treatment response according to claim 1, characterized in that: Also includes: A personalized treatment plan is formulated based on the patient's predicted treatment response to each treatment plan and the patient's clinical information.

6. A method for predicting treatment response according to claim 1, characterized in that: The clustering algorithm is used to perform heterogeneity analysis and cell division on the first single-cell sequencing data to obtain corresponding multiple cell subpopulations, specifically: A clustering algorithm with an adaptive threshold is used to perform heterogeneity analysis on the first single-cell sequencing data, and each cell is divided into a corresponding cell subpopulation according to the heterogeneity analysis results.

7. A treatment response prediction system, characterized in that: include: A data acquisition module, used to acquire the first single-cell sequencing data of the patient; A cell segmentation module, configured to use a clustering algorithm to perform heterogeneity analysis and cell segmentation on the first single-cell sequencing data to obtain a corresponding plurality of cell subpopulations; An extraction and identification module is used to perform feature extraction and biomarker identification on all the cell subpopulations to obtain corresponding cell subpopulation features; wherein the cell subpopulation features include cell development trajectories and a number of cell markers; The treatment response prediction module is used to input the patient's clinical information and the cell subpopulation characteristics into a pre-constructed treatment response prediction model, so that the treatment response prediction model performs multi-factor analysis and outputs the patient's treatment response prediction results for each treatment regimen.

8. A treatment response prediction system according to claim 7, characterized in that: Also includes: The program formulation module is used to formulate a personalized treatment plan based on the patient's predicted treatment response results for each treatment plan and the patient's clinical information.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program; wherein, when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute a treatment response prediction method as described in any one of claims 1 to 6.

10. A terminal, characterized in that: It comprises a processor, a memory and a computer program stored in the memory; wherein the computer program can be executed by the processor to implement a treatment response prediction method as described in any one of claims 1 to 6.

Images