Method for monitoring recurrence of bladder cancer after surgery based on ctDNA methylation profile

Abstract

The application relates to the field of biomedical detection technology, in particular to a bladder cancer postoperative recurrence monitoring method based on a ctDNA methylation spectrum, which comprises the following steps: collecting postoperative patient cell-free plasma and extracting circulating tumor DNA; performing bisulfite conversion treatment on the DNA; performing targeted amplification using a multiplex PCR primer group designed for a group of predetermined genomic methylation regions related to recurrence, constructing a sequencing library; performing high-throughput sequencing on the library, obtaining methylation level data of a plurality of CpG sites to form a sample methylation spectrum data matrix; the core of the application is that tumor heterogeneity is overcome through multi-marker combination detection, and a machine learning model is used to integrate multidimensional data to realize precise risk stratification, which has the advantages of non-invasiveness, high sensitivity, high specificity and quantifiable output, and provides an effective tool for individualized follow-up management of bladder cancer postoperation.

Description

Technical Field

[0001] This invention relates to the field of biomedical detection technology, and in particular to a method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles. Background Technology

[0002] Bladder cancer is a common malignant tumor of the urinary system worldwide, with non-muscle-invasive bladder cancer accounting for the vast majority of initial cases. Transurethral resection of bladder tumor (TURP) is currently the standard initial treatment. However, the disease has a high recurrence rate and a certain rate of progression, requiring long-term and frequent follow-up monitoring after surgery. Current standard monitoring protocols mainly rely on cystoscopy and urine cytology. Cystoscopy is invasive, causing physical discomfort and psychological burden for patients, and carries risks such as urinary tract infections and bleeding, while also increasing medical costs. Although urine cytology is non-invasive, its sensitivity is low, especially for low-grade tumors, making it prone to missed diagnoses.

[0003] In recent years, liquid biopsy technology, especially the detection of circulating tumor DNA (ctDNA), has provided new possibilities for non-invasive cancer monitoring. ctDNA is a DNA fragment released into the bloodstream after tumor cell apoptosis, necrosis, or secretion, carrying genetic variations and epigenetic information about the tumor. DNA methylation, as a key epigenetic modification, shows abnormalities in the early stages of cancer development and is tissue-specific. Therefore, detection based on ctDNA methylation markers can theoretically achieve highly sensitive and specific tumor monitoring.

[0004] However, applying ctDNA methylation analysis to monitor postoperative recurrence in bladder cancer still faces a series of technical challenges and unresolved key issues, constituting the shortcomings of existing technologies: First, most bladder cancer methylation biomarkers in existing studies are derived from tumor tissue, and their abundance, stability, and correlation with postoperative recurrence status in ctDNA lack systematic validation. Directly applying tissue-derived biomarkers may result in unstable detection in blood due to extremely low ctDNA levels, leading to monitoring failure. Second, conventional methylation detection methods, such as methylation-specific PCR, typically only detect single or a few methylation sites. Bladder cancer exhibits high heterogeneity, and single or a few biomarkers cannot comprehensively capture all methylation variation patterns associated with recurrence, resulting in insufficient sensitivity and specificity for monitoring, and an inability to accurately distinguish between high-risk and low-risk patients. Third, existing technical solutions often remain at the level of "detecting the presence of specific methylation biomarkers," lacking a standardized and computable discriminative model that integrates and quantifies information from multiple methylation sites and directly correlates it with clinical recurrence risk. This makes the interpretation of test results subjective and vague, making it difficult to provide a clear and guiding classification of recurrence risk in clinical practice.

[0005] Therefore, there is an urgent clinical need for a method to monitor postoperative recurrence of bladder cancer that can overcome the aforementioned shortcomings. This method needs to be able to stably detect low-abundance ctDNA methylation signals from a complex blood background; it needs to comprehensively utilize a set of rigorously screened multiple methylation biomarkers strongly correlated with recurrence to overcome tumor heterogeneity; and it needs to establish an objective recurrence risk prediction model based on multi-biomarker methylation data, thereby achieving accurate, non-invasive, quantifiable, and operable stratified monitoring of postoperative recurrence risk. Summary of the Invention

[0006] To achieve the above objectives, this invention provides a method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles, the monitoring method comprising the following steps: Step 1: Collect cell-free plasma samples from patients after bladder cancer surgery; Step 2: Extract circulating tumor DNA from the cell-free plasma sample; Step 3: The extracted circulating tumor DNA is subjected to bisulfite conversion treatment to convert unmethylated cytosine into uracil; Step 4: Using a set of multiplex PCR primers designed for a group of at least 5 to 50 genomic methylation regions pre-screened by differential methylation analysis and survival analysis, and combined with retrospective cohort studies and clinical validation, which are associated with postoperative recurrence of bladder cancer, the transformed circulating tumor DNA is targeted for amplification to construct a sequencing library. Step 5: Perform high-throughput sequencing on the sequencing library to obtain methylation level data of multiple CpG sites in each methylation region of the genome, forming a sample methylation spectrum data matrix; Step 6: Input the methylation level data corresponding to the core feature combination of the pre-trained recurrence risk prediction model from the sample methylation spectrum data matrix into the recurrence risk prediction model. The model is trained based on plasma samples and clinical recurrence status information of patients after bladder cancer surgery, and calculates the recurrence risk score of the patients after bladder cancer surgery. Step 7: Generate a recurrence risk monitoring report based on the recurrence risk score.

[0007] Preferably, in step one, the collection of cell-free plasma samples from patients after bladder cancer surgery specifically includes: Peripheral blood was collected from the patient using blood collection tubes containing cell stabilizers; Within 2 hours of collection, the collected peripheral blood was centrifuged to separate plasma and blood cell components. The separated plasma is then centrifuged in a second step to remove residual cells, cell debris, and platelets, resulting in a cell-free plasma sample. The cell-free plasma samples were aliquoted and stored in an ultra-low temperature environment of -70°C to -80°C.

[0008] Preferably, in step two, the extraction of circulating tumor DNA from the cell-free plasma sample specifically involves: The circulating tumor DNA was extracted using a magnetic bead-based extraction kit. The extraction process included lysis, binding, washing, and elution steps. The cell-free plasma sample was treated with proteinase K and lysis buffer during the lysis step to release circulating tumor DNA. The circulating tumor DNA is specifically adsorbed and released using surface-modified magnetic beads in the combined step; The washing step involves repeatedly washing the magnetic beads with the adsorbed circulating tumor DNA using washing buffers of different compositions to remove impurities such as proteins and salt ions. The purified circulating tumor DNA is eluted from the magnetic beads using elution buffer or water under heating conditions during the elution step. The concentration and total amount of the eluted circulating tumor DNA were determined using a highly sensitive fluorescence quantitative method.

[0009] Preferably, in step three, the bisulfite conversion treatment of the extracted circulating tumor DNA specifically includes: The extracted circulating tumor DNA was mixed with bisulfite conversion reagent and a conversion program including high-temperature denaturation and medium-temperature long-term incubation was performed in a thermal cycler to cause deamination of unmethylated cytosine. After the transformation reaction is completed, the transformed DNA is desalted and purified using a silica membrane-based purification column or magnetic beads to remove sulfites and other chemical reagents from the reaction system. The purified single-stranded DNA was dissolved in a weakly alkaline elution buffer, and its concentration was determined using a method suitable for the quantification of single-stranded DNA.

[0010] Preferably, in step four, the primers of the multiplex PCR primer set target the genomic methylation region selected from CpG islands or differentially methylated regions of genes associated with the occurrence, development, or prognosis of bladder cancer; Each primer pair in the multiplex PCR primer set is designed to meet the following conditions: the amplicon length is between 80 and 250 base pairs to accommodate the fragmentation characteristics of circulating tumor DNA in plasma; the primer sequence is located in a segment of the sequence after bisulfite conversion that does not contain CpG dinucleotides; and the annealing temperature of all primers has been calculated and experimentally verified to ensure comparable amplification efficiency at the same set annealing temperature. The targeted amplification specifically includes two rounds of PCR amplification: the first round of PCR uses the multiplex PCR primer set to perform multiplex targeted amplification on the transformed DNA; the second round of PCR uses universal primers carrying sequencing platform compatible adapters and sample-specific index sequences to further amplify the products of the first round of PCR in order to construct the complete sequencing library.

[0011] Preferably, in step five, the high-throughput sequencing adopts a paired-end sequencing mode; The acquisition of methylation level data for multiple CpG sites in each of the genomic methylation regions is specifically achieved through the following bioinformatics methods: The raw sequence readings generated from sequencing are assigned to the corresponding samples based on the sample-specific index sequence. The assigned sequence readings are subjected to quality filtering and connector sequence trimming. The pruned sequence reads are compared with the human reference genome sequence that has undergone bisulfite conversion to determine the position of each sequence read on the genome and its corresponding predetermined genomic methylation region; The number of sequence reads supporting the methylation state at each CpG site within each predetermined genomic methylation region was compared with the total number of sequence reads. The methylation level of each CpG site is calculated as the percentage of sequence reads supporting a methylated state out of the total number of sequence reads.

[0012] Preferably, in step six, the method for constructing the pre-trained recurrence risk prediction model includes: Collect a batch of plasma samples from patients who have undergone bladder cancer surgery in the past, and perform steps one through five to obtain the methylation spectrum data matrix of the historical samples; Obtain the clinical recurrence status information of the patients who underwent bladder cancer surgery in the past within a preset follow-up period. The clinical recurrence status information includes recurrence and no recurrence. The feature selection algorithm is used to screen out the CpG site combination that is most strongly associated with the clinical relapse status information from the historical sample methylation spectrum data matrix, and this combination is used as the core feature combination. The methylation level data of the core feature combination is used as the input feature, and the corresponding clinical relapse status information is used as the output label. The model is trained using a machine learning classification algorithm. The parameters of the machine learning classification algorithm are optimized using cross-validation, and the model performance is evaluated using an independent validation set. Finally, the recurrence risk prediction model is obtained, which can output a quantitative recurrence risk score.

[0013] Preferably, the feature selection algorithm is selected from one or more combinations of analysis of variance, recursive feature elimination, minimum absolute value convergence and selection operator algorithms; The machine learning classification algorithm is selected from logistic regression, support vector machine, random forest or gradient boosting decision tree algorithm; The cross-validation method is either 50-fold or 100-fold cross-validation; The quantified relapse risk score is a value between 0 and 1, representing the probability of relapse.

[0014] Preferably, in step seven, generating a relapse risk monitoring report based on the relapse risk score specifically includes: The calculated recurrence risk score is compared with at least two preset risk thresholds to determine the recurrence risk level of the bladder cancer patient after surgery, and the risk level includes at least low risk and high risk. Based on the determined recurrence risk level, pre-defined differentiated clinical recommendations are matched, including recommended follow-up intervals and further examination methods for low-risk, intermediate-risk, and high-risk levels. The patient identification information, the detected methylation region information, the recurrence risk score, the recurrence risk level, and the matching clinical recommendations are integrated to generate a structured recurrence risk monitoring report document.

[0015] Preferably, the monitoring method is repeated during regular follow-up of patients who have undergone transurethral resection of bladder tumor after bladder cancer surgery, and the regular follow-up time points include the 3rd month, 6th month, 12th month after surgery, and once a year thereafter. The recurrence risk scores obtained from the same patient at different follow-up time points are dynamically compared, and the evolution of the recurrence risk of the bladder cancer patient after surgery is assessed based on the changing trend of the recurrence risk scores.

[0016] The beneficial effects of this invention are: 1. This invention optimizes the extraction and targeted enrichment techniques for low-abundance, fragmented circulating tumor DNA, enabling the stable capture of trace methylation signals associated with recurrence. Its sensitivity significantly surpasses that of traditional urine cytology. Furthermore, this method requires only peripheral blood collection, completely avoiding the invasive procedures of cystoscopy and its associated pain, infection risks, and medical costs, thus providing patients with a more compliant long-term monitoring method. 2. This invention abandons the single-marker detection mode, but instead selects a group of methylation regions strongly associated with recurrence to form a combination, and uses high-throughput sequencing for parallel detection, thereby comprehensively covering the heterogeneous methylation characteristics of tumors. Furthermore, a predictive model is constructed using machine learning algorithms, fusing multidimensional methylation data into an objective quantitative score, effectively distinguishing recurrence-specific signals from background noise, significantly reducing the false positive rate, and achieving highly specific discrimination of recurrence risk. 3. The output of this invention is not a simple qualitative result, but a quantitative recurrence risk score and a clear risk level calculated based on a multivariate model. This provides clinicians with an intuitive and objective basis for decision-making, enabling truly individualized postoperative management: early warning and enhanced intervention for high-risk patients, and reasonable extension of follow-up intervals and avoidance of overtreatment for low-risk patients, thereby optimizing the allocation of medical resources and improving patient prognosis and quality of life. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the steps of the method of the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0020] Please see Figure 1 This invention provides a method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles. The core of this method lies in the close coordination of multiple steps and the seamless connection of data flow, which together constitute a complete solution from biological samples to clinical reports.

[0021] First, obtaining qualified biological materials from the patient is fundamental. Therefore, it is necessary to collect and prepare high-quality cell-free plasma samples to preserve circulating tumor DNA to the greatest extent and remove genomic DNA contamination.

[0022] Secondly, trace amounts of circulating tumor DNA were efficiently extracted from the plasma and subjected to bisulfite conversion to transform epigenetic information into sequenced differences, which is a key step in information conversion.

[0023] Next, a set of pre-designed multiplex primers targeting relapse-related genomic regions were used to specifically target and amplify the transformed DNA, and a sequencing library was constructed. This step achieved effective enrichment of the target information.

[0024] Then, using high-throughput sequencing technology, the sequence information of all amplicon was read in parallel, and the methylation level of each target CpG site was quantified by bioinformatics analysis, thereby generating a multidimensional sample methylation spectrum data matrix, completing the transformation from biological signals to digital data.

[0025] Then, the key feature data in the data matrix is ​​input into a machine learning model pre-trained based on historical data. The model integrates multi-dimensional information through complex algorithms and finally outputs a quantitative recurrence risk score.

[0026] Finally, based on this score and referring to the preset clinical decision-making logic, a structured monitoring report containing a clear risk level and action recommendations is automatically generated. The entire process is interconnected. The combination of multiplex targeted amplification and high-throughput sequencing solves the problem of detecting low-abundance targets, while the integration and analysis of multi-marker methylation data by machine learning models is the core of achieving high-precision risk stratification. The two work together to achieve the technical effect of non-invasive and accurate monitoring.

[0027] In one possible implementation, the step begins with peripheral venous blood collection using a vacuum blood collection tube containing a specific cell stabilizer. The cell stabilizer works by rapidly penetrating the blood cell membrane after the blood leaves the body, halting cellular metabolism and stabilizing the cell membrane structure, thereby strongly inhibiting apoptosis and necrosis and preventing the release of intracellular genomic DNA into the plasma. This step is crucial because it minimizes the interference of background genomic DNA on subsequent trace circulating tumor DNA detection at the source.

[0028] Collected blood samples must be processed within a specified time window to further ensure sample quality. The first step, centrifugation, uses low-speed centrifugation conditions, such as centrifugation at 4°C with an acceleration of 1000 to 2000 times gravity for 10 to 20 minutes. The purpose of this step is to achieve preliminary stratification of the blood under gentle conditions, separating whole blood into a lower layer of blood cell sediment and an upper layer of plasma, while avoiding excessive centrifugation force that could cause cell rupture. Subsequently, the upper plasma layer is transferred to a new tube for the second step, high-speed centrifugation, under conditions such as centrifugation at 4°C with an acceleration of 16000 times gravity for 10 minutes. This high-speed step aims to thoroughly remove any small number of cells, platelets, cell debris, and large vesicles that may remain in the plasma after the first centrifugation, obtaining highly pure cell-free plasma.

[0029] Finally, the purified plasma is rapidly aliquoted and stored at an ultra-low temperature of -70°C to -80°C. Aliquoting avoids the damage to the integrity of circulating tumor DNA caused by repeated freeze-thaw cycles.

[0030] The benefits of this standardized preparation process are as follows: First, by combining chemical and physical methods, the representativeness and authenticity of circulating tumor DNA in plasma samples are ensured, reducing the risk of false negatives. Second, the extremely low background DNA interference lays a solid foundation for subsequent high-sensitivity detection. Third, the standardized operation ensures the comparability between samples at different time points and in different batches, which is crucial for long-term dynamic monitoring.

[0031] In one possible implementation, the extraction process is a multi-step, refined purification process. In the lysis step, a lysis buffer containing a high concentration of dissociative salt and proteinase K is added to the thawed plasma. The dissociative salt disrupts intermolecular hydrogen bonds and hydrophobic interactions, denaturing proteins and releasing them from DNA; proteinase K systematically digests various structural proteins and enzymes in the plasma, thereby fully releasing circulating tumor DNA that was originally bound to nucleoprotein complexes into the solution, forming a homogeneous lysis buffer.

[0032] Silanol-modified magnetic beads and a binding buffer of specific concentration and pH were added to the lysis buffer. Under optimized binding buffer conditions, the negatively charged phosphate backbone of circulating tumor DNA specifically binds to the magnetic beads via electrostatic adsorption and hydrogen bonding with the positively charged silanol groups on the bead surface. Meanwhile, impurities such as proteins, lipids, and carbohydrates exhibit weak affinity for the magnetic beads under the binding conditions and are mostly retained in the supernatant.

[0033] The washing step involves multiple washes using two different buffer solutions. The first wash buffer typically contains a high concentration of ionizing salt and ethanol to further remove co-adsorbed nonspecific impurities. The second wash buffer is usually a weak salt solution containing ethanol or a Tris buffer, used to remove residual salts and organic matter, ensuring the cleanliness of the magnetic beads. The elution step uses a low-ionic-strength, weakly alkaline Tris buffer or deionized water, incubated at 55°C to 60°C.

[0034] This condition weakens the electrostatic interaction between the DNA phosphate backbone and the positive charge on the surface of the magnetic beads, allowing high-purity circulating tumor DNA to desorb from the magnetic beads and dissolve in the elution solution.

[0035] The advantages of this method are as follows: First, the magnetic bead method has high binding capacity and high recovery rate, making it particularly suitable for capturing trace amounts of circulating tumor DNA in large volumes of plasma; second, based on the specific binding of surface chemistry, it can effectively remove PCR inhibitors that affect downstream enzymatic reactions; finally, the operation is easy to automate, improving throughput and consistency of results, and making it possible to process large-scale clinical samples.

[0036] In one possible implementation, the transformation process is the core chemical reaction linking DNA methylation status to sequence detection. First, extracted circulating tumor DNA is mixed with a freshly prepared high-concentration sodium bisulfite solution, along with necessary free radical scavengers and pH stabilizers. The reaction is performed in a thermal cycler under programmed temperature control. An initial high-temperature step, such as maintaining 98°C for 5 to 10 minutes, ensures complete thermal denaturation and dissociation of the double-stranded circulating tumor DNA into single strands, a prerequisite for the subsequent transformation reaction. This is followed by a prolonged medium-temperature incubation step, such as at 60°C to 65°C in the dark for 2.5 to 4 hours.

[0037] At this stage, bisulfite ions undergo a nucleophilic addition reaction with cytosine on single-stranded DNA, forming an unstable intermediate. For unmethylated cytosine, this intermediate undergoes hydrolysis and deamination in the presence of water, eventually transforming into uracil; while for methylated pentamethylcytosine, due to the steric hindrance and electronic effects of the methyl group, the reaction rate is extremely slow, thus remaining unchanged.

[0038] After the reaction, excess bisulfite must be thoroughly removed from the system, as it can interfere with subsequent PCR amplification. This is achieved using a silica membrane-based desalting purification column: under specific high-salt, high-ethanol binding buffer conditions, the transformed single-stranded DNA adsorbs onto the column membrane, while small molecule impurities such as bisulfite are washed away. A desulfonation step is then performed, typically using fresh sodium hydroxide solution to treat the DNA bound to the membrane to remove sulfonated cytosine intermediates that may have formed due to incomplete reaction. Finally, the purified single-stranded DNA is eluted with preheated, weakly alkaline, low-salt elution buffer.

[0039] First, it successfully transforms epigenetic methylation differences (C and 5mC) into stable DNA sequence differences (U and C), enabling subsequent sequence-based detection. Second, the optimized reaction and purification protocol minimizes DNA degradation loss during strong chemical treatment, ensuring a high detection success rate for low-volume samples. Third, effective desulfonation and purification ensure the quality of the transformed DNA as a PCR template.

[0040] In one possible implementation, the multiplex PCR primer set is designed for a set of genomic regions that have been pre-screened through bioinformatics analysis and clinical validation and are significantly associated with postoperative recurrence of bladder cancer. These regions are typically located in the promoter region, enhancer region, or first exon region of a gene, such as CpG islands in genes like RASSF1A, GSTP1, and APC.

[0041] Primer design follows strict principles: amplicon length is limited to between 80 and 250 base pairs, a range that matches the major fragment length distribution of circulating tumor DNA in plasma, ensuring the possibility of efficient amplification from highly fragmented templates. Primer sequences must be located in conserved regions that do not contain any CpG sites after bisulfite conversion. This means that regardless of whether the CpG sites in this region are methylated, the primers can bind and initiate amplification with the same efficiency, thus ensuring the accuracy of subsequent methylation level quantification and avoiding detection bias introduced by differences in primer binding efficiency.

[0042] The annealing temperatures of all primers were homogenized by adjusting sequence length and GC content to ensure similar amplification kinetics at a single set annealing temperature. This is crucial for achieving balanced amplification of multiple targets in multiplex PCR. The targeted amplification employs a two-round PCR strategy. The first round of multiplex PCR amplifies the transformed DNA using the aforementioned specific primer set. The cycle number is optimized to achieve initial enrichment of the target region while avoiding plateau phase, thus reducing amplification bias. The second round of PCR uses the first-round product as a template and amplifies with universal primers carrying complete sequencing adapters and sample-specific barcode indexes. The purpose is to uniformly add the universal sequences and sample identification tags required for upstream and downstream sequencing to all amplicons.

[0043] The beneficial effects of this design are: First, multiplex PCR enables the simultaneous parallel enrichment of multiple target regions dispersed at different locations in the genome, greatly improving detection throughput and efficiency; Second, the two-round PCR strategy separates specific amplification from library construction, standardizing the library construction process and facilitating the introduction of indexes required for multi-sample mixed sequencing; Third, rigorous primer design ensures the fairness and quantitative accuracy of the detection, laying the foundation for subsequent reliable bioinformatics analysis.

[0044] In one possible implementation, the process begins with the decomposition of the raw sequencing data. Cluster signal images generated by the sequencer are converted into raw read files containing sequence information and quality values ​​by basic identification software. First, based on the barcode index sequence of each read, all reads are precisely split and grouped under their corresponding single sample names, achieving data separation after multi-sample mixed sequencing. Subsequently, each sample's read undergoes rigorous quality control and preprocessing. Algorithms are used to prune low-quality bases starting from the end of the read, typically removing consecutive base segments with quality values ​​below a certain threshold; simultaneously, sequencing adapter sequences that may remain at the end of the read are identified and removed to prevent interference with subsequent alignments. After filtering, high-quality reads with the required length and high average quality are retained and proceed to the core analysis steps.

[0045] Next comes sequence alignment. Since the reads originate from bisulfite-treated DNA, the cytosine may have been converted to thymine, therefore they cannot be directly aligned with a standard reference genome. This workflow employs specially designed alignment algorithms, such as performing a "virtual" C-to-T conversion on the reference genome to generate two derived reference sequences, or using a seed extension algorithm that allows C-to-T conversion, to efficiently and accurately locate the pre-processed reads onto the human genome reference sequence and record their alignment position and orientation. For successfully aligned reads, methylation sites are extracted.

[0046] The system identifies all CpG dinucleotide sites covered by the reads and determines methylation by comparing the bases in the reads with their corresponding positions in the original reference genome: if the read shows a C and the corresponding position in the reference genome is C, the CpG site is determined to be methylated in the original DNA sample; if the read shows a T and the corresponding position in the reference genome is C, the site is determined to be unmethylated. Finally, quantitative statistics are performed. For each CpG site within each target region, all valid reads covering that site are summarized, and the number of reads supporting methylation and the total number of covered reads are counted separately.

[0047] The methylation level is calculated as a percentage of the total number of covered readings.

[0048] The beneficial effects of this process are as follows: First, it ensures the purity and reliability of input data through precise data splitting and quality filtering; second, a dedicated alignment strategy solves the sequence complexity caused by bisulfite processing, ensuring the accuracy of localization; and third, a quantitative method based on multi-site coverage depth provides a statistically robust estimate of methylation levels with a precision far exceeding that of traditional methods.

[0049] In one possible implementation, this embodiment details the construction method of a recurrence risk prediction model, including the entire process of data preparation, feature engineering, algorithm training, and validation. The model construction is a data-driven machine learning process. First, historical data preparation is performed by collecting a cohort of post-bladder cancer patients with long-term, complete clinical follow-up. Their plasma samples undergo the aforementioned standardized process to obtain a historical sample methylation profile data matrix.

[0050] Simultaneously, information on the definite recurrence status of these patients within a specified postoperative time period is collected to form labels for model training. The core step in model construction is feature selection, which aims to sift through a massive number of CpG sites to identify the most predictive subset. This is typically achieved by calculating the statistical difference in methylation levels at each site between the recurrence and non-recurrence groups, for example, by using analysis of variance to calculate the F-value, or by using machine learning embedded methods such as Lasso regression for screening.

[0051] Recursive feature elimination is another effective strategy: First, an initial classifier is trained using all features. Features are then ranked by importance, and the least important features are recursively removed. The model performance is re-evaluated until an optimal subset of features, the core feature combination, is found. This combination typically contains dozens of sites that collectively characterize the methylation patterns most relevant to relapse. After feature selection, the model training phase begins. The core feature combination methylation data of historical samples is used as the input feature matrix, and the corresponding relapse states are used as the output label vector. This data is then fed into a selected machine learning classification algorithm for training. During training, the algorithm learns the complex mapping relationship from input features to output labels.

[0052] To prevent overfitting and find the optimal model parameters, cross-validation is employed. For example, 5-fold cross-validation involves randomly dividing historical data into five parts, using four parts as the training set and one part as the validation set, repeating the training and validation process five times. By adjusting the algorithm parameters, the model's average performance across the five validations is optimized. Finally, an independent sample set that was not used in training and cross-validation is used as the test set to evaluate the performance of the final model and confirm its generalization ability.

[0053] The beneficial effects of this model construction method are: First, by using feature selection to reduce dimensionality, redundant and noisy information is removed, enhancing the robustness and interpretability of the model; Second, by using cross-validation and independent test set evaluation, the stability and generalization ability of the model are strictly guaranteed, avoiding overfitting to the training data; Third, the final model is a powerful tool that can efficiently integrate multidimensional methylation data and transform it into a single risk score.

[0054] In one possible implementation, the feature selection algorithm uses ANOVA to quickly screen sites with significant differences in means between the two groups by calculating the ratio of the between-group variance to the within-group variance of the methylation level of each CpG site between the relapsed and non-relapsed groups. Recursive feature elimination is often used in conjunction with classifiers such as support vector machines or random forests, iteratively removing features that contribute the least to the model's decision boundary to obtain a concise and efficient feature subset. The minimum absolute value convergence and selection operator algorithm directly embeds feature selection during linear model training, adding the sum of the absolute values ​​of the model coefficients as a penalty term to its loss function, compressing many unimportant feature coefficients to zero, thus achieving automatic feature selection. These methods can be used individually or in combination to ensure the discriminative power of the selected core feature combination.

[0055] The machine learning classification algorithms described above each have their own characteristics: the logistic regression model outputs a probability value between 0 and 1, which has good interpretability; the support vector machine is dedicated to finding an optimal hyperplane that maximizes the margin between the two classes of samples, and performs well for high-dimensional small sample data; the random forest, by constructing a large number of decision trees and integrating their results, can effectively handle nonlinear relationships and has strong resistance to overfitting; the gradient boosting decision tree constructs decision trees in a serial manner, with each tree learning to correct the residuals of the previous tree, which usually achieves very high prediction accuracy.

[0056] The cross-validation process is the core of model tuning. Taking 5-fold cross-validation as an example, in each fold, the model is trained using the training data, and performance metrics are calculated on the validation data. For support vector machines, key parameters that need optimization include the penalty coefficient and kernel function parameters; for random forests, the number of trees and maximum depth need to be optimized. Optimization is performed in these parameter spaces using grid search or random search to select the set of parameters that yields the best average performance in cross-validation.

[0057] The final model output is a quantified relapse risk score, a continuous numerical value typically calibrated or normalized to between 0 and 1. This value has a clear probabilistic interpretation, which can be understood as the predicted probability of future relapse for the patient given the current methylation profile data.

[0058] The beneficial effects of this specific implementation method are as follows: First, it provides a variety of mature algorithm selection and combination schemes, enabling model construction to be flexibly optimized for different data characteristics; second, the detailed cross-validation and parameter optimization process ensures the maximization of model performance and stability; third, it explicitly defines the model output as a continuous probability score, providing clinicians with richer and more refined risk information than simple binary classification.

[0059] In one possible implementation, the report generation process is a rule-based automated decision-making process. First, the system compares the calculated recurrence risk score with preset risk thresholds. For example, two thresholds are set: 0.3 and 0.7. If the score is less than 0.3, it is classified as "low risk"; if the score is greater than or equal to 0.3 and less than 0.7, it is classified as "medium risk"; and if the score is greater than or equal to 0.7, it is classified as "high risk". The thresholds can be set based on the distribution of risk scores in training data or determined through methods such as clinical decision curve analysis to balance sensitivity and specificity. Subsequently, the system matches corresponding differentiated clinical recommendations from a preset knowledge base based on the determined risk level. For the "low risk" level, the recommendation may include: indicating a low risk of recurrence and suggesting continued routine follow-up according to the standard protocol.

[0060] For the "medium risk" level, the recommendations might be: indicating a certain risk of recurrence, it is recommended to appropriately shorten the interval between the next cystoscopy or monitoring using this method, and to strengthen clinical observation. For the "high risk" level, the recommendations are more proactive, for example: indicating a high risk of recurrence, it is strongly recommended to have a follow-up cystoscopy within one month to confirm the diagnosis, and to immediately discuss with the clinician whether to start or adjust the adjuvant treatment plan.

[0061] Finally, the system integrates all the above information and automatically generates a structured electronic report. This report includes at least the following: basic patient identification information; a list of methylation biomarker combinations used in this test; the calculated recurrence risk score; a clearly defined risk level based on the threshold; and specific, actionable clinical follow-up and intervention recommendations for that risk level. The report can be clearly presented in text, tables, or a combination of text and graphics.

[0062] The beneficial effects of this step are: First, it transforms abstract mathematical model scores into risk levels and specific recommendations that are familiar and intuitive to clinicians, greatly improving the clinical usability and operability of the results; Second, it achieves standardized output of monitoring results, avoiding subjective differences in human interpretation; Third, the differentiated recommendations realize truly individualized medicine, enabling high-risk patients to receive timely intervention and low-risk patients to avoid overtreatment.

[0063] In one possible implementation, the method is not a one-time test, but rather embedded periodically within the patient's standard postoperative follow-up plan. Typically, after transurethral resection of bladder tumors, it is recommended to perform the first intensive monitoring at key time points, such as 3, 6, and 12 months postoperatively, followed by annual monitoring. Each monitoring session independently executes the complete workflow from step one to step seven, generating an independent recurrence risk score and monitoring report for each time point. The core value of dynamic monitoring lies in the longitudinal comparison and analysis of sequential scores for the same patient. By plotting the trend of the patient's recurrence risk score over time, its risk trajectory can be observed.

[0064] Ideally, the postoperative score is high, but it shows a stable downward trend over time and with possible adjuvant therapies, indicating effective treatment and a reduced risk of recurrence. Another scenario requiring caution is a score that remains stable at a low level for a period, then shows a significant increase at a follow-up point. This may indicate a molecular-level recurrence not yet detected by imaging or cystoscopy, providing an early warning signal for clinicians. Furthermore, combining dynamic score trends with patient clinical events and changes in treatment regimens allows for more in-depth analysis.

[0065] The beneficial effects of this implementation method go beyond a single test: First, it enables dynamic and continuous assessment of the patient's recurrence risk, capturing the process of risk change rather than just a static state at a certain point in time; second, dynamic trend analysis can provide earlier warning information than a single test, and has predictive value; third, through the accumulation of long-term monitoring data, it is possible to reverse-evaluate the impact of different treatment strategies on the disease at the molecular level, providing a reference for adjusting individualized treatment plans and forming a closed-loop management of "monitoring-assessment-intervention-remonitoring".

[0066] Example This embodiment uses a patient diagnosed with high-grade non-muscle-invasive bladder cancer who has undergone transurethral resection of bladder tumor as an example to demonstrate how to use the method of the present invention to monitor recurrence in the third month after surgery.

[0067] Step 1: Collect cell-free plasma samples from the patient after bladder cancer surgery; The patient returned to the hospital for routine follow-up three months post-surgery. The monitoring process began with standardized blood sample collection. Peripheral blood was collected from the patient's elbow vein using a 10mL vacuum blood collection tube containing a special cell stabilizer. This cell stabilizer effectively prevents the release of genomic DNA from blood cells into the plasma after blood is removed from the body, ensuring that the circulating tumor DNA extracted subsequently is primarily derived from true extracellular DNA. After collection, the blood collection tube was gently inverted eight times at room temperature and sent to the laboratory for processing within two hours of collection.

[0068] In the laboratory, blood collection tubes were placed in a centrifuge pre-cooled to 4 degrees Celsius and centrifuged at 1600 times the force of gravity for 10 minutes to complete the first centrifugation step. The purpose of this step is to initially separate the blood into plasma, white blood cell, and red blood cell layers under low-temperature conditions. Using a sterile pipette, the upper plasma layer was carefully aspirated and transferred to a 15 mL enzyme-free centrifuge tube, taking care to avoid aspirating the middle white blood cell layer. Then, the centrifuge tube containing plasma was centrifuged again at 4 degrees Celsius at 16000 times the force of gravity for 10 minutes to complete the second centrifugation step. This high-speed centrifugation aims to thoroughly remove residual cells, platelets, and cell debris from the plasma, obtaining highly purified cell-free plasma. The supernatant was aspirated, which is the final cell-free plasma sample. It was aliquoted into 1.5 mL cryopreservation tubes (1 mL per tube) and immediately immersed in liquid nitrogen for rapid freezing. Subsequently, it was transferred to an ultra-low temperature freezer at -80 degrees Celsius for long-term storage until the next step of DNA extraction.

[0069] Step 2: Extract circulating tumor DNA from cell-free plasma samples; A 1 mL vial of cell-free plasma was removed from a -80°C freezer and thawed on ice. Circulating tumor DNA (CTN) extraction was performed using a magnetic bead-based purification kit. The specific procedure was as follows: A lysis buffer containing proteinase K was added to the thawed plasma, and the mixture was gently incubated with shaking at 56°C for 30 minutes. This lysis step thoroughly digests proteins in the plasma and destroys any viral capsids present, allowing the protein-bound CTN to be fully released into the solution. Then, a binding buffer and silanol-modified magnetic beads were added to the lysis buffer, and the mixture was incubated at room temperature for 15 minutes. During this process, negatively charged CTN specifically adsorbs onto the positively charged magnetic bead surface under specific ion concentration conditions, while most proteins, lipids, and other impurities are not adsorbed.

[0070] Next, place the centrifuge tube on a magnetic rack. Once the magnetic beads are adsorbed onto the tube wall and the solution becomes clear, carefully discard the supernatant. Retain the magnetic beads and wash twice, sequentially, with two different ethanol washing buffers to remove residual salts, detergents, and other organic impurities. After washing, open the tube cap and let it stand at room temperature for several minutes to allow any residual ethanol to evaporate completely. Finally, add preheated low-salt elution buffer to the tube containing the dried magnetic beads and incubate at 55°C for 5 minutes to allow circulating tumor DNA to desorb from the magnetic beads and enter the solution. Place the centrifuge tube back on the magnetic rack and transfer the supernatant containing the purified circulating tumor DNA to a new enzyme-free centrifuge tube.

[0071] A highly sensitive quantitative detection method based on a fluorescent dye was used to quantify the eluted circulating tumor DNA. This fluorescent dye specifically binds to double-stranded DNA and emits fluorescence, with the fluorescence intensity proportional to the DNA concentration. By measuring the fluorescence value of the sample and comparing it with a standard curve of known concentrations, the total amount of circulating tumor DNA extracted in this example was calculated to be 15 nanograms. Simultaneously, fragment analysis of the extracted DNA was performed using microfluidic chip electrophoresis. The results showed that the main peak was concentrated around 170 base pairs, consistent with the typical characteristics of short fragmentation in circulating tumor DNA, confirming the quality of the extraction.

[0072] Step 3: Perform bisulfite conversion treatment on the extracted circulating tumor DNA; Take 10 nanograms of circulating tumor DNA extracted in step two and process it using a high-efficiency, low-loss bisulfite conversion kit to distinguish between methylated and unmethylated cytosine. Mix the circulating tumor DNA with the prepared bisulfite conversion reaction solution, with a total volume of 20 μL. Place the mixture in a thermal cycler and run a precisely temperature-controlled conversion program: first, maintain at 98°C for 10 minutes to completely denature the double-stranded DNA into single strands; then, incubate at 64°C in the dark for 2.5 hours. During this stage, bisulfite reacts with the single-stranded DNA, deamination of unmethylated cytosine to uracil, while methylated 5-methylcytosine remains unchanged; finally, lower the temperature to 4°C for storage.

[0073] After the transformation reaction, the reaction products were desalted and purified using a purification column based on the principle of silica membrane adsorption. All reaction solution was transferred to the purification column, and DNA was bound to the membrane by high-speed centrifugation. The cells were then washed sequentially with a specially formulated desulfonation buffer and washing buffer to remove residual sulfites, salt ions, and other reaction byproducts. Finally, the purified single-stranded DNA was eluted with pre-warmed, weakly alkaline, low-concentration Tris buffer to a final volume of 15 μL. At this point, unmethylated C atoms in the DNA sequence had been converted to U atoms, while methylated C atoms remained C atoms. The concentration of the elution buffer was determined using a fluorescent dye specifically for single-stranded DNA quantification, and the result was 1.2 ng / μL. This transformed DNA served as the template for subsequent amplification.

[0074] Step 4: Targeted amplification of transformed circulating tumor DNA using multiplex PCR primer sets; The goal of this step is to specifically enrich multiple target methylation regions associated with bladder cancer recurrence from a complexly transformed background. The multiplex PCR primer set targets differentially methylated regions of eight genes closely related to bladder cancer prognosis, identified through pre-screening. These regions are located, for example, in the RASSF1A gene promoter region, exon 1 region of the GSTP1 gene, and APC gene promoter 1A region. Primers for each target region are carefully designed: first, all amplicons are strictly controlled to a length between 120 and 200 base pairs to match the short fragment characteristics of circulating tumor DNA in plasma, ensuring amplification efficiency; second, each primer sequence is located in a conserved region that does not contain any CpG dinucleotides after bisulfite conversion, meaning that primer binding is unaffected by the methylation status of this region, enabling fair amplification of methylated and unmethylated alleles and avoiding amplification bias; finally, the primer sequences are adjusted using bioinformatics software to ensure that the theoretical annealing temperatures of all primers are concentrated within a range of 60°C ± 1°C, guaranteeing balanced amplification of all targets in the same reaction system.

[0075] The amplification process consisted of two rounds of PCR. The first round of PCR was a multiplex targeted amplification: 5 μL of the transformed DNA obtained in step three was used as a template, and a mixture containing primers for all the target regions, high-fidelity thermostable DNA polymerase, and an optimized buffer system were added. The reaction program was as follows: initial denaturation at 95°C for 3 minutes; followed by 18 cycles of amplification, each cycle consisting of denaturation at 95°C for 15 seconds, annealing at 60°C for 30 seconds, extension at 72°C for 30 seconds, and a final extension at 72°C for 5 minutes. This round of PCR achieved preliminary enrichment of the eight target regions.

[0076] The second round of PCR was used to construct a complete library for sequencing: the purified product from the first round of PCR was used as a template, and a pair of primers carrying universal adapter sequences for the sequencing platform were used for amplification. The 5' ends of these primers contained P5 and P7 sequences compatible with flow-channel bridging amplification in the sequencer, and an 8-base-long sample-specific index sequence was included near the template to distinguish different samples during mixed sequencing. This round of PCR was performed for only 8 cycles to add complete sequencing adapters and indexes to both ends of the first-round product. After the reaction, magnetic beads were used to size-select the product, removing primer dimers and excessively long fragments, ultimately yielding a purified DNA library of 250 to 350 base pairs in length, i.e., the sequencing library, with a concentration of 12 nanomolar by fluorescence quantification.

[0077] Step 5: Perform high-throughput sequencing on the sequencing library and obtain methylation data; The sequencing library constructed in this embodiment was mixed with libraries from other samples at an equimolar concentration and subjected to paired-end sequencing on a high-throughput sequencing platform. Paired-end sequencing refers to sequencing from both ends of a DNA fragment separately, generating a pair of matching sequence reads, which improves alignment accuracy and the reliability of methylation site quantification. In this embodiment, a 2x150 base pair sequencing strategy was used, with a target average sequencing depth of 5000-fold per region per sample. After sequencing, the raw sequence data was converted into methylation level data using the following bioinformatics methods: Data splitting: Based on the 8-base index sequence carried in each sequence read, the massive sequence reads generated by mixed sequencing are precisely split into data files corresponding to the samples in this embodiment.

[0078] Quality filtering and trimming: Software is used to perform quality control on the split sequence reads. First, low-quality bases at the 3' and 5' ends of the sequence are removed; then, residual sequencing adapter sequences are identified and removed; finally, sequence reads that are too short or have too low average quality values ​​after trimming are discarded.

[0079] Sequence alignment: High-quality sequence reads are aligned with a human reference genome sequence that has undergone insilico bisulfite conversion. The alignment algorithm used in this step is tolerant of sequence differences caused by C-to-U conversion. Successfully aligned sequence reads are located to their genomic origin and their corresponding target methylation regions are marked.

[0080] Methylation level extraction and calculation: For each CpG site aligned to each target region, the software analyzes it: if the site is still C in the sequencing reads, it is determined to be methylated in the original DNA; if it is T, it is determined to be unmethylated. The number of reads supporting methylation and the total number of covered reads are counted among all valid reads covering the site. The methylation level is calculated as: (Number of reads supporting methylation / Total number of covered reads) × 100%. Finally, the methylation level values ​​of all CpG sites in all target regions are compiled into a data matrix, i.e., the sample methylation spectrum data matrix. In this embodiment, this matrix contains methylation ratio data for 56 valid CpG sites in 8 regions.

[0081] Step Six: Calculate the relapse risk score using the relapse risk prediction model; This step is crucial in showcasing the inventiveness of this invention. The recurrence risk prediction model is pre-trained using historical data. Its construction process (though not included in this monitoring procedure, it is essential for understanding the model input) is briefly described below: Historical data preparation: Plasma samples and complete 5-year follow-up clinical data of 300 patients after bladder cancer surgery were collected (clearly indicating whether recurrence occurred and the time of recurrence). Steps one through five were performed on each historical sample, resulting in a methylation profile data matrix of 300 historical samples.

[0082] Feature selection: A recursive feature elimination method combined with a support vector machine (SVM) was used for feature selection. First, the methylation levels of all 56 CpG sites were used as initial features, and a preliminary SVM model was trained with the recurrence rate (within 2 years) as the label. This model assigned a weight coefficient to each feature (CpG site), with the absolute value of the weight representing the feature's contribution to distinguishing between relapsed and non-relapsed sites. Then, the features with the smallest weights were removed, and the model was retrained with the remaining features, and its performance was evaluated. This process was repeated recursively, removing the least important features each time, until a subset of features was found that optimally aligns with the model's performance (e.g., the AUC value obtained through five-fold cross-validation). Finally, 28 of the most discriminative CpG sites were selected from the 56 sites, forming the core feature set.

[0083] Model Training and Validation: Using the methylation level data of the aforementioned 28 core CpG sites as input features, the final model was trained using the Support Vector Machine (SVM) algorithm. The core idea of ​​SVM is to find an optimal hyperplane that best separates relapsed samples from non-relapsed samples in the 28-dimensional feature space. Model parameters were optimized using five-fold cross-validation: 300 historical samples were randomly divided into five similarly sized subsets. Four subsets were used as the training set, and one subset as the validation set, repeating the training and validation process five times. By adjusting the SVM's penalty parameter C and the radial basis function kernel parameter gamma, the model's average performance across the five validations was optimized. This resulted in a stable and highly generalizable predictive model. This model can accept data from the 28 core sites of a new sample and output a continuous value between 0 and 1, i.e., a relapse risk score. This score can be intuitively understood as the probability of relapse for the patient corresponding to that sample in the future.

[0084] In this embodiment, the methylation level data corresponding to the above 28 core feature combinations are extracted from the methylation spectrum data matrix of the patient sample obtained in step five and input into the trained and saved prediction model. After internal calculation, the model outputs a recurrence risk score of 0.78 for the patient in this monitoring.

[0085] Step 7: Generate a recurrence risk monitoring report based on the recurrence risk score; The report generation process is a crucial step in translating quantitative scores into clinically actionable recommendations. The system has two preset risk thresholds: 0.3 and 0.7. Risk levels are determined according to the following rules: a recurrence risk score less than 0.3 is classified as "low risk"; a score greater than or equal to 0.3 and less than 0.7 is classified as "medium risk"; and a score greater than or equal to 0.7 is classified as "high risk".

[0086] In this embodiment, the patient's score of 0.78 is greater than the high-risk threshold of 0.7, and therefore they are classified as "high-risk." The system automatically matches differentiated clinical recommendations corresponding to this risk level based on a preset template. For the "high-risk" level, the recommendation is: "High risk of recurrence; it is recommended to undergo cystoscopy within one month to determine if a recurrence is present, and to consider adjuvant therapies such as intravesical chemotherapy in consultation with the clinician. It is recommended to shorten the next non-invasive monitoring based on this method to the 6th month post-surgery." Finally, a structured relapse risk monitoring report is generated. The report includes: patient name and unique identifier, sampling date, name of the testing method, a list of the eight target methylation regions detected, the calculated relapse risk score (0.78), the determined risk level (high risk), and the specific clinical follow-up and intervention recommendations mentioned above. The report is presented with clear items and format, facilitating quick understanding and decision-making by clinicians.

[0087] To demonstrate the superiority of the method of this invention, we designed a comparative study. A cohort of 100 postoperative bladder cancer patients with conditions similar to those in the examples was selected, and all patients underwent monitoring using three methods simultaneously at the third month post-surgery. The method of this invention is to perform steps one through seven as described above.

[0088] Comparative Example 1 (Traditional Standard Method): The patient underwent standard white light cystoscopy, with a diagnostic report issued by an experienced urologist; simultaneously, a urine cytology examination was performed, and the results were interpreted by a cytopathologist. A positive result for either method was considered a positive monitoring result.

[0089] Comparative Example 2 (Single Marker Detection Method): Only one commonly used bladder cancer methylation marker (such as a single CpG island in the RASSF1A gene) was detected. Methylation-specific quantitative PCR based on TaqMan probes was used for detection. A fixed cycle threshold was set; if a fluorescence signal appeared in the sample's Ct value before this threshold, the methylation at that site was considered positive, i.e., a positive monitoring result. All patients were subsequently followed up for at least 12 months, with cystoscopic biopsy pathology results used as the gold standard to confirm whether a true recurrence had occurred. The monitoring performance comparison of the three methods is shown in the table below: Table Analysis: Sensitivity: The method of this invention is significantly higher than that of the two comparative studies. This demonstrates that by combining multiple biomarkers and high-depth sequencing, this invention can more effectively detect minute residual lesions or early recurrences, overcoming the limitations of traditional cystoscopy, which is limited by visual observation, has low sensitivity in urine cytology, and is prone to missed detection by single biomarkers due to tumor heterogeneity.

[0090] Specificity: The method of this invention is also superior to Comparative Example 2, and comparable to or even slightly superior to Comparative Example 1. This indicates that by integrating information from multiple sites, the machine learning model can more accurately distinguish between tumor-derived specific methylation signals and non-specific methylation changes caused by background noise or benign diseases, thus reducing false positives.

[0091] Positive / Negative Predictive Values: The method of this invention has the highest positive and negative predictive values. This means that when the method indicates "high risk," the patient's actual relapse probability is very high (85.7%), and clinicians can take more proactive intervention with greater confidence; when it indicates "low risk," the patient is basically safe in the near term (97.4%), which can effectively avoid unnecessary invasive examinations, reduce the burden on patients and medical costs.

[0092] In summary, this embodiment fully and comprehensively demonstrates the entire process of the method of the present invention, from sample collection to report generation. Through comparison with traditional and simplified methods, it fully verifies the significant advancements and inventiveness of the present invention in terms of sensitivity, specificity, and clinical guidance value in non-invasive monitoring of bladder cancer postoperatively.

[0093] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0094] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles, characterized in that, The monitoring method includes the following steps: Step 1: Collect cell-free plasma samples from patients after bladder cancer surgery; Step 2: Extract circulating tumor DNA from the cell-free plasma sample; Step 3: The extracted circulating tumor DNA is subjected to bisulfite conversion treatment to convert unmethylated cytosine into uracil; Step 4: Using a set of multiplex PCR primers designed for a group of at least 5 to 50 genomic methylation regions pre-screened by differential methylation analysis and survival analysis, and combined with retrospective cohort studies and clinical validation, which are associated with postoperative recurrence of bladder cancer, the transformed circulating tumor DNA is targeted for amplification to construct a sequencing library. Step 5: Perform high-throughput sequencing on the sequencing library to obtain methylation level data of multiple CpG sites in each methylation region of the genome, forming a sample methylation spectrum data matrix; Step 6: Input the methylation level data corresponding to the core feature combination of the pre-trained recurrence risk prediction model from the sample methylation spectrum data matrix into the recurrence risk prediction model. The model is trained based on plasma samples and clinical recurrence status information of patients after bladder cancer surgery, and calculates the recurrence risk score of the patients after bladder cancer surgery. Step 7: Generate a recurrence risk monitoring report based on the recurrence risk score.

2. The method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles according to claim 1, characterized in that, Step one, specifically the collection of cell-free plasma samples from patients after bladder cancer surgery, includes: Peripheral blood was collected from the patient using blood collection tubes containing cell stabilizers; Within 2 hours of collection, the collected peripheral blood was centrifuged to separate plasma and blood cell components. The separated plasma is then centrifuged in a second step to remove residual cells, cell debris, and platelets, resulting in a cell-free plasma sample. The cell-free plasma samples were aliquoted and stored in an ultra-low temperature environment of -70°C to -80°C.

3. The method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles according to claim 1, characterized in that, In step two, the extraction of circulating tumor DNA from the cell-free plasma sample specifically involves: The circulating tumor DNA was extracted using a magnetic bead-based extraction kit. The extraction process included lysis, binding, washing, and elution steps. The cell-free plasma sample was treated with proteinase K and lysis buffer during the lysis step to release circulating tumor DNA. The circulating tumor DNA is specifically adsorbed and released using surface-modified magnetic beads in the combined step; The washing step involves repeatedly washing the magnetic beads adsorbed with the circulating tumor DNA using washing buffers of different compositions to remove proteins and salt ions. The purified circulating tumor DNA is eluted from the magnetic beads using elution buffer or water under heating conditions during the elution step. The concentration and total amount of the eluted circulating tumor DNA were determined using a highly sensitive fluorescence quantitative method.

4. The method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles according to claim 1, characterized in that, Step three, specifically the bisulfite conversion treatment of the extracted circulating tumor DNA, includes: The extracted circulating tumor DNA was mixed with bisulfite conversion reagent and a conversion program including high-temperature denaturation and medium-temperature long-term incubation was performed in a thermal cycler to cause deamination of unmethylated cytosine. After the transformation reaction is completed, the transformed DNA is desalted and purified using a silica membrane-based purification column or magnetic beads to remove sulfites and other chemical reagents from the reaction system. The purified single-stranded DNA was dissolved in a weakly alkaline elution buffer, and its concentration was determined using a method suitable for the quantification of single-stranded DNA.

5. The method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles according to claim 1, characterized in that, In step four, the primers of the multiplex PCR primer set target the genomic methylation region selected from CpG islands or differentially methylated regions of genes associated with the occurrence, development, or prognosis of bladder cancer. Each primer pair in the multiplex PCR primer set is designed to meet the following conditions: the amplicon length is between 80 and 250 base pairs to accommodate the fragmentation characteristics of circulating tumor DNA in plasma. The primer sequences are located in the CpG dinucleotide-free region of the sequence after bisulfite conversion; the annealing temperatures of all primers have been calculated and experimentally verified to ensure comparable amplification efficiencies at the same set annealing temperature. The targeted amplification specifically includes two rounds of PCR amplification: the first round of PCR uses the multiplex PCR primer set to perform multiplex targeted amplification on the transformed DNA; the second round of PCR uses universal primers carrying sequencing platform compatible adapters and sample-specific index sequences to further amplify the products of the first round of PCR in order to construct the complete sequencing library.

6. The method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles according to claim 1, characterized in that, In step five, the high-throughput sequencing adopts a paired-end sequencing mode; The acquisition of methylation level data for multiple CpG sites in each of the genomic methylation regions is specifically achieved through the following bioinformatics methods: The raw sequence readings generated from sequencing are assigned to the corresponding samples based on the sample-specific index sequence. The assigned sequence readings are subjected to quality filtering and connector sequence trimming. The pruned sequence reads were compared with the human reference genome sequence that had undergone bisulfite conversion to determine the location of each sequence read on the genome and its corresponding methylated region. The number of sequence reads supporting the methylation state at each CpG site within each methylation region of the genome was statistically compared with the total number of sequence reads; The methylation level of each CpG site is calculated as the percentage of sequence reads supporting a methylated state out of the total number of sequence reads.

7. The method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles according to claim 1, characterized in that, Step six, the method for constructing the pre-trained recurrence risk prediction model includes: Collect a batch of plasma samples from patients who have undergone bladder cancer surgery in the past, and perform steps one through five to obtain the methylation spectrum data matrix of the historical samples; Obtain the clinical recurrence status information of the patients who underwent bladder cancer surgery in the past within a preset follow-up period. The clinical recurrence status information includes recurrence and no recurrence. The feature selection algorithm is used to screen out the CpG site combination that is most strongly associated with the clinical relapse status information from the historical sample methylation spectrum data matrix, and this combination is used as the core feature combination. The methylation level data of the core feature combination is used as the input feature, and the corresponding clinical relapse status information is used as the output label. The model is trained using a machine learning classification algorithm. The parameters of the machine learning classification algorithm are optimized using cross-validation, and the model performance is evaluated using an independent validation set. Finally, the recurrence risk prediction model is obtained, which can output a quantitative recurrence risk score.

8. The method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles according to claim 7, characterized in that, The feature selection algorithm is selected from one or more combinations of analysis of variance, recursive feature elimination, minimum absolute value convergence and selection operator algorithm; The machine learning classification algorithm is selected from logistic regression, support vector machine, random forest or gradient boosting decision tree algorithm; The cross-validation method is either 50-fold or 100-fold cross-validation; The quantified relapse risk score is a value between 0 and 1, representing the probability of relapse.

9. The method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles according to claim 1, characterized in that, Step seven, specifically generating a recurrence risk monitoring report based on the recurrence risk score, includes: The calculated recurrence risk score is compared with at least two preset risk thresholds to determine the recurrence risk level of the bladder cancer patient after surgery, and the risk level includes at least low risk and high risk. Based on the determined recurrence risk level, pre-defined differentiated clinical recommendations are matched, including recommended follow-up intervals and further examination methods for low-risk, intermediate-risk, and high-risk levels. The patient identification information, the detected methylation region information, the recurrence risk score, the recurrence risk level, and the matching clinical recommendations are integrated to generate a structured recurrence risk monitoring report document.

10. The method for monitoring postoperative recurrence of bladder cancer based on ctDNA methylation profiles according to any one of claims 1 to 9, characterized in that, The monitoring method was repeated during regular follow-up of patients who underwent transurethral resection of bladder tumor after bladder cancer surgery. The regular follow-up time points included the 3rd month, 6th month, 12th month after surgery, and once a year thereafter. The recurrence risk scores obtained from the same patient at different follow-up time points are dynamically compared, and the evolution of the recurrence risk of the bladder cancer patient after surgery is assessed based on the changing trend of the recurrence risk scores.

Images